Behavioral gerontology and gambling: The jackalope ...

ISSN: 1942-6453

Volume 4 Number 1 Summer 2010

The Analysis of Gambling Behavior (AGB) is a peer-reviewed publication that contains original general interest and discipline specific articles related to the scientific study of gambling

E D IT OR

A SS O C IAT E E D IT OR

Jeffrey N. Weatherly University of North Dakota

Mark R. Dixon Southern Illinois University

E DIT ORIAL B O ARD MEMBERS Lewis Bizo Southern Cross University

Jennifer Austin Swansea University Andrew Brandt Ohio Wesleyan University

John C. Borrero University of Maryland, Balti more County

Andrew Cooper Swansea University

Simon Dymond Swansea University Edmund Fantino University of California, San Diego

Patrick M. Ghezzi University of Nevada

Donald A. Hantula Te mple University

Becky Nastally Southern Illinois University

Eric A. Jacobs Southern Illinois University

John Haw Souther Cross University

Charles A. Lyons E astern O regon University

Otto H. MacLin University of Northern Iowa

Gregory J. Madden Utah State University

Western Michigan University

Nancy Petry University of Connecticut

Brady Phelps South D akota S tate University

Cynthia J. Pietras Western Michigan University

Bryan Roche National University of Ireland, Maynooth

Robert Whelan University College, Dublin

Richard Malott

Content of the Analysis of Ga mbling Behavior The Analysis of Gambling Behavior (AGB) contains general interest and discipline specific articles related to the scientific study of gambling. Articles appropriate for the journal include a) full-length research articles, b) research reports, c) clinical demonstrations, d) technical articles, and e) book reviews. Each category is detailed below along with submission guidelines:

Research Articles – a manuscript of full length (20-30 doublespaced pages approximately), which may contain multiple experiments, and are original contributions to the published literature on gambling.

Research Reports – a manuscript of reduced length (no more than 10 double-spaced pages and a single figure or table page), which may be less experimentally rigorous than a Research Article, a replication of or failure to replicate a prior published article, or pilot data that demonstrates a clear relationship between independent and dependent variable(s). The Results and Discussion sections of Reports should be combined.

Clinical Demonstrations – a manuscript of reduced length (no more than 8 double-spaced pages and a single figure or table page) which lack the rigor of a true experimental design, yet do demonstrate behavior change of persons with gambling disorders under clinical care. This manuscript should contain an Introduction, Methods/Treatments, Results, and Discussion sections. The Results and Discussion sections of Clinical Demonstrations should be combined.

Technical Article – a manuscript of either full or reduced length, depending on necessity, that describes either a new technology available that would be of interest to researchers or a taskanalysis style description of how to utilize existing technology for the conducting of research. Examples of appropriate topics may include, but are not limited to, the rewiring of a slot machine for the collection of data or controlling of win/losses, how to use computer software to simulate a casino game, or the way in which neuroimaging devices may interfaced with an experimental apparatus.

Book Review – a review of a contemporary book related to gambling not more than three years after the publication data of the book to be reviewed. The review should be no more than 15 doubled-spaced pages in length.

ANALYSIS OF GAMBLING BEHAVIOR Volume 4, Number 1, Summer 2010 ISSN: 1942-6453

Contents Comments from Incoming Editor Weatherly, J.N. Upward and onward.

3

Invited Papers

Baker, J.C. Behavioral gerontology and gambling: The jackalope of behavior analysis.

5

Nastally, B.L., & Dixon, M.R. The effect of relational training on the nearmiss effect in slot machine players.

16

Weatherly, J.N. Temporal discounting and gambling: A meaningful relationship?

27

Dymond, S., & Roche, B. The impact of derived relational responding on gambling behavior.

38

Dixon, M.R. The roulette near-miss effect.

54

Research Article

Miller, J.C., Dixon, M.R., Parker, A., Kulland, A.M., & Weatherly, J.N. Concurrent validity of the Gambling Functional Assessment (GFA): Correlations with the South Oaks Gambling Screen (SOGS) and indicators of diagnostic efficiency.

61

Guest Reviewers

76

1

2

Analysis of Gambling Behavior

2010, 4, 3–4

Number 1 (Summer2010)

UPWARD AND ONWARD Jeffrey N. Weatherly University of North Dakota -------------------------As of the present issue, I have assumed the role of Executive Editor of Analysis of Gambling Behavior. I am pleased that Dr. Dixon has agreed to remain on the board and serve in the role of Associate Editor. His guidance and expertise will no doubt be beneficial to my attempts to fulfill my new duties. The core goal of Analysis of Gambling Behavior will not change. We will still be interested in the study of gambling behavior from a behavioral perspective. In past issues, we have published theoretical papers that were accompanied by commentaries by noted scholars in the field. I anticipate that we will continue that practice as it helps to promote discussion, debate, and overall interest in the field. I am also hopeful that we will also continue to regularly publish experimental research on gambling behavior. In my opinion, given the paucity of such research on gambling in the field of psychology overall, I think these publications make Analysis of Gambling Behavior a unique and important journal. With that said, we will continue to consider publication of other types of works (e.g., review papers, book reviews) that promote the understanding of gambling from a behavior-analytic viewpoint. It is a somewhat daunting task and I have some big shoes to fill. However, what I wished for years ago has become a reality. There have been a number of individuals who have been instrumental in making that outcome possible. I thank them and look forward to their continued support of the journal. I believe they are, as I am, committed to its original mission. So, without further ado, upward and onward.

As the saying goes, be careful what you wish for because you might just get it. Well, apparently I got it. Back in 2006, my colleague Mark Dixon approached me with an idea for a scholarly journal that would be focused on behavioranalytic research in the area of gambling. Despite the expanding field of behavior analysis, few behavior analysts were pursuing research and treatment options for gambling and/or gambling problems and were instead focusing their attention on the more rare disorder of Autism. He argued that a refereed, peerreviewed journal in the area of gambling would raise awareness of the topic, as well as give behavior analysts a venue to showcase their research on gambling rather than needing to conform to the expectations of other gambling journals in psychology that promote the medical-model myth, routinely forward mentalistic explanations for behavior, and rely nearly exclusively on self-report data. I agreed that it was a good idea, promised to support his efforts as much as I could, and wished for the best. Now 2010, we are publishing the fourth volume of Analysis of Gambling Behavior. Our editorial board spans the globe, as have the authors who have published in the journal across the first three volumes. Our readership is also worldwide. The journal can be found on the shelves of scholars as well as in the periodical sections of university libraries. We are working towards having the journal included in major research literature search engines such as PsycInfo. In short, Dr. Dixon’s vision was a good one and, if I may be allowed to pass judgment, his efforts have been a success. 3

4

EDITORIAL

The Present Issue The present issue represents the work and research of individuals who participated in the 2010 conference of the Behaviorists Interested in Gambling Special Interest Group (BIG SIG) of the International Association of Behavior Analysis, which was held in Reno, NV. Presenters were invited to submit their papers for consideration for publication in this issue of Analysis of Gambling Behavior, each of which went through the peer-review process. The result is what I believe to be a representative sample of the different contributions that behavior analysts can make to the study of gambling behavior. The contributions range from theoretical perspectives of how gambling can and cannot be studied to original empirical contributions. I am confident that readers will appreciate the breadth of perspectives. I am especially pleased to present these papers in the first issue for which I am serving as executive editor. Each of these papers could potentially serve as a springboard for additional research, just as each makes a significant contribution to the existing literature. I thank the authors for both their contribution to the conference and this issue. I look forward to working with them, and other researchers, in the days and years ahead. Jeffrey N. Weatherly Executive Editor Analysis of Gambling Behavior


2010, 4, 5–15


BEHAVIORAL GERONTOLOGY AND GAMBLING: THE JACKALOPE OF BEHAVIOR ANALYSIS Jonathan C. Baker

Southern Illinois University Older adults constitute over one third of all gamblers in the United States. As the baby-boom generation continues to reach older adulthood, this proportion is likely to grow. To date, behavior-analytic research on gambling has focused on younger populations. Although such research is necessary and important, the present account will suggest that additional research should focus on studying older gamblers. The purpose of the present account is to review the literature that exists on typical behavior changes observed in older-adult populations and the implications for those changes related to current behavior-analytic research in gambling. Keywords: Behavioral Gerontology, Gambling, Behavior Analysis

---------------------------------Behavior analysts have long noted the importance of conducting research with adults over the age of 65 (Lindsley, 1964). Generally referred to as older adults, this group is typically split into three categories: (a) the youngold (those age 65 to 74); (b) old or middle-old (those age 75 to 84); and (c) old-old or oldestold (those 85 or older). Behavioral gerontology focuses on the application of behavioranalytic principles to address changes related to aging and older adults (Adkins & Mathews, 1999). Over the past 46 years, behavioral gerontologists have addressed issues in the basic understanding of behavior principles with older adults, the ways in which clinical applications can ameliorate behavioral excesses and reinstitute behavioral deficits, and how organizational behavior management can improve systems that serve older adults (LeBlanc, Raetz, & Feliciano, in press). Despite a steady (albeit fairly low) flow of research in Address all correspondence to: Dr. Jonathan C. Baker Rehabilitation Services Program Rehabilitation Institute Southern Illinois University Carbondale, IL 62901 email:jonathan.c.baker@siu.edu

behavioral gerontology (Buchanan, Husfeldt, Berg, & Houlihan, 2008), one area that has not been addressed is gambling. The study of gambling behavior in older adults can be approached from two different angles: a) the benefits of recreational gambling and b) pathological gambling. Although behavior analysts have not addressed the gambling behavior of older adults, a rich and growing body of literature focusing on behavior analysis and gambling provides a solid foundation upon which to build the field’s understanding of such behavior. This proposed combination of research focusing on older adults and gambling is truly the Jackalope of behavior analysis. A Jackalope is a mythical creature believed to be the result of a crossbreed of deer or antelope and a jackrabbit (that is sometimes described as being killer). Despite the wealth of fiction related to Jackalopes, there is some fact to the existence of the creature itself, as a form of the papillomavirus that affects rabbits, called cottontail rabbit papillomavirus (CRPV; Christensen, 2005) can cause warts that become bonelike in nature (Giri, Danos, & Yaniv, 1985), and could be mistaken for antlers in a jackrabbit. Although interesting, it is 5

6

BEHAVIORAL GERONTOLOGY AND GAMBLING

quite saddening that more empirical research exists related to a rare breed of an extinct pygmy-deer and a species of killer-rabbit than on the gambling behavior of older adults. The purpose of this paper is to propose a combination of two relatively small, yet important, areas of behavior-analytic research: research on gambling and research on older adults. This is not to say that the behavior of older adults is in some way different from the operant and respondent behavior of any other organism, but that there are biological changes (e.g., pain related to chronic illness can create abolishing operations for engaging in once preferred tasks that involve physical activity) and environmental changes (e.g., environmental contingencies that support dependence rather than independence and the decreased salience of discriminative stimuli) that occur specific to older-adult populations and affect the ways in which behaviors occur (LeBlanc, Raetz, Feliciano, 2008; Skinner, 1983). Indeed, Skinner argued that contingencies of reinforcement tend to support different behaviors as adults age and that stimulus control weakens as adults age. As such, the study of older-adult behavior would yield important information. Despite the many potential benefits of such research, to date there have been few, if any, such studies. The focus of the paper will be to first cover what is currently known about the behavior of older adults and how that can impact current research on gambling. The subsequent review will focus on three areas: a) activities and engagement in aging; b) principles of reinforcement and stimulus control related to aging; and finally c) pathological gambling in older adults. Research on Gambling with Older Adults Reports (National Research Council, 1999) estimate the proportion of gamblers over the age of 65 to be about 27% in the United States. The highest proportion of gamblers is those age 50 – 65, which accounts for over 30%. Thus, gamblers age 50 and over

account for more than half of all gamblers. Within the gerontology literature, researchers (e.g., Preston, Shapiro, & Keene, 2007) have noted that successful aging for those over the age of 65 involves minimizing illness and loss of function (both physical and cognitive) as well as maximizing engagement in activities within the community. Research supports the idea that engaging in activities within the community can actually help to decrease the chances of illness and loss of function (Preston et al., 2007). However, as adults age the chances of becoming socially isolated increase (Vander Bilt, Dodge, Pandav, Shaffer, & Ganguli, 2004). Recreational gambling activities (e.g., going to Bingo or a casino) provide older adults with opportunities for social interaction within the community and cognitive stimulation in the form of engagement in mathematical tasks (National Research Council, 1999; Vander Bilt et al., 2004). Indeed, researchers have found that gambling can result in improved physical and mental health for older adults (Desai, Maciejewski, Dausey, Caldarone, & Potenza, 2004; Vander Bilt et al., 2004). For example, older adults who engage in regular recreational gambling activities appear to have lower incidence of depression, greater social support, and higher cognitive functioning (Vander Bilt et al., 2004). Thus, by maintaining activities within the community that provide stimulation and deter physical and cognitive decline, it is possible for older adults who engage in recreational gambling to be seen as aging successfully (Preston et al., 2007; Quadagno, 2005). Although there are many benefits to gambling, there is also a potential for abuse (Zaranek & Litchenberg, 2008). Research indicates that pathological gambling does exist among older adults. Studies (National Research Council, 1999) indicate that those over the age of 65 as a whole have the lowest levels of pathological gambling. However, older adults who do engage in pathological gambling are likely to have decreased physical and mental health

Jonathan C. Baker (Erickson, Molina, Ladd, Pietrzak, & Petry, 2005). In addition, they are likely to be of lower socio-economic status, which is often exacerbated by losing money during gambling (National Research Council, 1999). Despite the fact that gerontologists have begun to focus their research efforts on the study of older gamblers, examples of such research in behavior analysis are scarce. Indeed, at the time of this publication it is difficult to find even one study in behavior analysis that has focused on older adults specifically as the target populations. One study soon to become public by Dixon, Nastally, and Waterman (in press) demonstrates a very simple application of behavior analysis to the gambling behavior of older adults. The study, conducted in a nursing home, focused on indices of happiness during gambling activities. Participants were first exposed to different stimuli (animals, food, letters, people, and casino games) in a visual paired-choice format preference assessment. Following the preference assessment, participants were exposed to games on a laptop computer that simulated analog gambling. Data on indices of happiness indicated that all participants displayed higher percentages of intervals with indices of happiness during engagement in gambling activities than during baseline, though the effects were not observed once the activities were concluded (Dixon et al., in press). In sum, a search of published behavioranalytic research focusing on the gambling behavior of older adults yields few results. Research on the gambling behavior of older adults could first and foremost benefit older adults by expanding current technology for providing preferred activities. In addition, methodologies used for gambling research could be utilized to provide valuable insight into reinforcement and stimulus control changes that occur with aging, leading to improvements in interventions that could be used to treat pathological gambling. Finally,

7

such research could help to expand both the fields of behavioral gerontology and behavioral analysis of gambling. The following section provides some background information related to three areas that might benefit from behavior-analytic research on gambling with older adults: a) activities and engagement; b) understanding the effects of reinforcement and stimulus control in older adults; and c) the behavior of pathological older adult gamblers. Current Research on Older Adults and the Impact for Behavior-Analytic Research on Gambling Activities and Engagement A number of behavior-analytic studies have focused on increasing engagement in activities by older adults (e.g., Carstensen & Erickson, 1986; Gallagher & Keenan, 2000ab; McClannahan & Risley, 1975). Much of the research began as antecedent interventions that could supplement the living environment to foster engagement in activities (e.g., rearranging the room in which activities occurred, serving cookies during activities, etc.). Nursing homes, in particular, often have low levels of engagement. For example, McClannahan and Risley (1975) conducted a study to increase activity engagement in nursing home settings and found that during baseline, social interaction averaged 13% and activity engagement averaged about 36% (observations were conducted once per hour for 13 hours, 5 days a week for 2 weeks). Older adults with dementia in particular often engage in few activities. More recently, researchers have moved from the physical environment arrangement toward utilizing preferenceassessment methodology (Hagopian, Long, & Rush, 2004) to increase engagement in nursing home residents. LeBlanc, Cherup, Feliciano, and Sidener (2006) demonstrated items identified using a pair-stimulus preference-assessment methodology could effectively lead to engagement in older adults. LeBlanc, Raetz, Baker, Stroebel, and Feeney

8


(2008) demonstrated that an informant based preference assessment could also identify activities that lead to engagement. One limitation of many of the items that older adults (with or without dementia) might engage with at a nursing home is that access to items is typically staff controlled. Although research has shown that written feedback and training can increase the number of activities offered to staff (Engelman, Altus, & Mathews, 1999), there are still times when staff cannot be available to interact with residents. In addition, nursing home staff are typically expected to focus more on tasks related to care (e.g., toileting, feeding, bathing, transportation) than on providing activities. Gambling activities, such as the video-based slot machines, standard video poker, roulette, blackjack, and craps offered in Dixon et al. (in press), could serve as activities that residents might engage in with minimal staff involvement (e.g., in times when staff must provide care for other residents). A similar version of this currently exists in nursing homes – Bingo. However, even during Bingo, one staff member must call the numbers while others assist those who need it (e.g., helping to put chips down when needed, calling out “Bingo”, etc.). Automated simulated1 gambling games, which require little to no staff involvement and therefore offer prolonged engagement opportunities might prove beneficial in nursing home settings. Such activities can be engaged across a wide range of functioning levels, such that more residents may be able to engage in the activities (e.g., those with dementia). The preliminary reports from Dixon et al. (in press) suggest that older adults not only like engaging in simulated gambling, but that they will do so for as much

1 Although one of the potential reinforcers associated with gambling is the chance to win money, many nursing homes have restrictions on money related to Medicaid payments, potential hoarding of money, and disputes that might arise when two residents claim that money belongs to them and not the other person.

as 20 minutes at a time. Future studies, similar to those conducted by LeBlanc and colleagues, that focus on level of engagement without staff mediation with longer durations (i.e., more than 5 minutes) might help to determine whether activities like gambling might serve as alternatives to the more standard “group” activities typically offered at nursing homes. Although one benefit of such activities is that they involve less social interaction from staff, it would be important for researchers and clinicians to stress that such activities should not be used as a substitute for staff involvement. Such substitution might result in even lower levels of staff engagement than currently exist. Reinforcement and Stimulus Control The overall body of literature on basic research with older adults, specifically related to reinforcement and stimulus control, is limited (LeBlanc et al., in press). However, some trends have emerged as a result of the research that has been conducted. Two areas where some trends have emerged are related to the effects of reinforcement on the behaviors of older adults and the impact of stimuli on those behaviors, specifically that the behavior of older adults is sensitive to reinforcement (though perhaps differently than younger adults) and that stimulus control, although perhaps not as strong, is still possible. The following section reviews the literature supporting these findings and discusses how these findings could be important to gambling research. Plaud, Plaud, and Von Duvillard (1999) examined the effects of reinforcement on the behavior of older adults (ranging in age from 60 to 79) in the context of behavioral momentum. That is, following a period of reinforcement for a specific response, they altered the amount of reinforcement provided to determine the effect on behavior. Fifteen older adults served as participants for the study. Each participant was seated in front of a com-

Jonathan C. Baker puter and instructed to press the F1 key or the F12 key. A large green disc, presented on the screen, was associated with 10 tokens and a large red disc, also presented on the screen, was associated with 1 token (both keys were on a fixed-interval (FI) 45-s schedule). The two discs were associated with either the F1 or F12 key, depending on group assignment (i.e., for one group the F1 key was associated with the green disc whereas for the other group it was the F12 key). Following a threeweek training, participants were placed into one of five experimental conditions (i.e., the schedule on each button went from a FI 45-s schedule to the following): a) multiple schedule variable-interval (VI) 30 s; b) multiple schedule VI 60 s; c) multiple schedule variable-time (VT) 30 s; d) multiple schedule VT 60 s; and d) extinction (EXT). Overall, participants made significantly more responses on the green disc than on the red disc in the experimental condition, indicating that older adult behavior was sensitive to reinforcement density. In turn, even when reinforcement was no longer available for any response (as in the case of the VT & EXT schedules), participants still responded more on the green key than the red key (Plaud et al., 1999). Plaud et al. (1999) also compared the results of their study with the results of a previous study (Plaud, Gaither, & Lawrence, 1997) that involved first-year college students. They found that the older adults allocated less overall responding to the keys than college students and that more older adults responses were biased toward the green key (i.e., allocated more responding to the green key than the red key). These results indicate that the behavior of the older adults was more sensitive to the changes in schedules (e.g., when extinction was implement, older adults tended to respond less than college students), but persisted longer on the key that had been associated with higher levels of reinforcement (i.e., although they responded less, more of their responses were allocated to the key as-

9

sociated with the green disc rather than the red disc). A few studies have examined sensitivity to reinforcement and stimulus control within more complex preparations. These have typically been conducted using conditional discriminations in the form of stimulus equivalence or a signal preparation related to Signal Detection Theory (SDT; see below for description). Three studies have looked at performance of older adults in the context of stimulus equivalence. Stimulus equivalence refers to a summary of observed regularities with three formal properties: reflexivity, symmetry, and transitivity (Sidman, 1997). Teaching conditional discriminations results in the emergence of untaught conditional discriminations that conform to these properties (Sidman, Wayne, Macguire, & Barnes, 1989). When reflexivity (A=A), symmetry (if A=B, then B=A), and transitivity (if A=C and B=C, then A=C) are reliably shown between stimuli, then they are said to be part of the same equivalence class (Sidman & Tailby, 1982). Wilson and Milan (1995) studied stimulus class formation in 20 adults over the age of 62 (ranging in age from 62 to 81) and compared their results to 20 participants between the age of 19 and 22. Only 9 of the older adults demonstrated equivalence. Overall trials to criterion were higher for the older adult group, though the 9 older adults who demonstrated equivalence actually had lower trials to criterion than the younger adults who demonstrated equivalence, even though their response latencies were higher. Wilson and Milan noted that there may have been other stimuli that affected responding, including fatigue, attending to inappropriate stimuli, and decreases in memory. In another study, PerezGonzalez and Moreno Sierra (1999) included 6 participants over the age of 64 (ranging in age from 65 to 74) in their study on the formation of equivalence relations. All 6 demonstrated symmetry, reflexivity and transitivity, though they typically had more errors during

10


both training and testing, as well as took longer to master the baseline conditional discriminations, than the four participants under 64. Finally, Saunders, Chaney, and Marquis (2005) attempted to demonstrate equivalence in 12 older adults (ranging in age from 56 to 89). Following training, 9 of the 12 participants demonstrated equivalence. In a second experiment, 6 additional older adults were trained using a 0-s delay following the presentation of the sample stimulus and the response options. This modification resulted in fewer trials needed to demonstrate equivalence. Another preparation that researchers have used to assess the effects of reinforcement and stimulus control with older adults is SDT. There are three main variables that can be manipulated in a SDT preparation: a) the probability of the signal; b) the reinforcer or punisher ratio; c) and the signal strength (Nevin, 1969). The typical SDT preparation involves a simple discrimination task presented in discrete trials. In each trial, the participant is presented with one of two or more forms of stimuli: a noise stimulus (S0) and one or more noise-plus-signal stimuli (S1, S2,…Sm). In an auditory preparation, for example, the S0 might be an 8000 Hz tone, whereas the S1 might be the same 8000Hz tone, but also a 3000 Hz tone (an S2 might be a 12000 Hz tone and so on). The participant has two or more forms of responding (typical operandum is a button or key), corresponding to each form of stimulus; for S0, the correct response would be R0 (the experimenter would determine a priori which response is associated with which button) and for S1 the correct response would be R1. Correct responses result in a putative reinforcer, sometimes on a fixed-ratio 1 or on a VI schedule. Plaud, Gillund, and Ferraro (2000) provide one demonstration of the effects of reinforcement and stimulus control on older adult participants using SDT. In their study, six participants (ranging in age from 62 to 74) were presented with a computer and keyboard.

When the computer screen displayed a white circle, participants were to press the F1 key (which was reinforced with $0.10 and verbal praise on a VI 30-s schedule). When the computer screen displayed a red letter “A”, they were to press the F12 key (which was reinforced with $0.10 and verbal praise on a VI 60-s schedule). The response rates of the participants indicated that all of the participants demonstrated increased correct responding (i.e., reinforcement effect). Three of the six allocated responding to denser schedule (i.e., the VI 30 s) and two allocated responding to the leaner schedule (i.e., the VI 60 s). The final participant did not demonstrate statistically significant differential responding. These results seem to support the findings of other studies in that older adults’ behavior is sensitive to reinforcement but perhaps not as sensitive to supplemental stimuli used to establish stimulus control. In sum, the above findings related to the effects of reinforcement and stimulus control demonstrate that, overall, older-adult behavior is sensitive to reinforcement. Plaud et al. (1999) demonstrated that older adults respond appropriately to differing contingencies. They also found that, although older adults responded less, they were more likely to bias responding to previous schedules of reinforcement. The results of the above studies also indicate that stimuli correlated with the differential availability of reinforcement do control responding, though the impact of stimulus control appears to lessen. For instance, Wilson and Milan (1995) found that stimuli associated with correct responding had less of an impact with older-adult responding than other stimuli. Saunders et al (2005) used a 0-s delay and found that it resulted in fewer trials necessary to meet criteria. One focus of future research would be whether these findings relate to all groups of older adults. That is, the majority of participants in these studies could be classified as young-old (i.e., 65 to 74 years old) and there were not enough middle-

Jonathan C. Baker old or old-old participants to begin to determine if additional changes occur past the age of 75. If additional changes exist past the age of 75, researchers might seek to determine whether these are the result of age related changes or cohort effects. Whether these findings related to only the young-old or other groups, the findings are particularly relevant to research on gambling, where schedules of reinforcement and stimulus control have been hypothesized to play a crucial role in gambling behavior. Rachlin (1990) suggested that the unit of analysis for gambling might be a string of responses related to ratio. Specifically he said, “A history of [responses without reinforcement under large variable-ratio schedules] might conceivably characterize compulsive gamblers” (p. 297). He went on to suggest that the addition of counters or other supplemental stimuli might serve to lessen pathological gambling, as the effects of the gamblers behavior might become more apparent. Such a hypothesis would be interesting to test with older adult gamblers, who appear to respond to varying contingencies more effectively than younger adults (Plaud et al., 1999). Indeed, a gambling preparation might be an excellent platform to provide further evidence related to older-adult sensitivity to reinforcement. Given that gambling is a preferred activity in many older adults, participants might be more willing to sit for the long sessions needed to establish asymptotic responding that are characteristic of more basic preparations. Additionally, the amount and intensity of supplemental stimuli in gambling activities can be controlled through the context of the program used. It might be possible for researchers to add additional stimuli. In the case of slot machines, it may be possible to add additional chances to win to make detection of a “win” more difficult, thus assessing the discriminability of the signal. In addition to basic preparations, a number of recent studies have looked at derived

11

relations as a potential intervention for pathological gamblers. Given the current research on stimulus equivalence with older adults and the difficulties associated with demonstrating equivalence, it is unclear how interventions like those used by Zlomke and Dixon (2006) or Hoon, Dymond, Jackson, and Dixon (2008) would work with older populations. In both studies, participants were trained relational responding based on the cues of more than and less than. Following training, participants allocated responding to slot machines associated with the more than stimuli, even though the schedule of reinforcement was the same for both slot machines. Whether such a preparation would work with older adults is a yet unanswered question. In addition to the potential difficulty with establishing derived relations, current research indicates that older adults are more likely to demonstrate biased responding, which could provide further confounds for such research. Pathological Gambling As noted earlier, adults over the age of 65 appear to have the lowest levels of pathological gambling (National Research Council, 1999). There are, however, still pathological older gamblers. Much of the research on pathological older gamblers focuses on the deleterious effects pathological gambling but presently little has been done to address intervention strategies (Zaranek & Litchenberg, 2008). Behavior-analytic interventions for gambling have begun to move toward a function-based approach for treatment. For example, Dixon and Johnson (2007) developed the gambling functional assessment (GFA) to identify possible variables maintaining gambling behaviors in pathological gamblers. Behavioral gerontology has moved toward a more function-based account of many problem behaviors seen in older adults with dementia (Baker & LeBlanc, in press) and the use of functional assessment methodology for older adult gamblers would be both a natural

12


and valuable progression. For example, it is unknown whether the functions that maintain gambling in younger gamblers do so for older adults. Miller, Meier, Muehlenkamp, and Weatherly (2009) noted that escape scores on the GFA were strongly related to total GFA scores. Zaranek and Litchenberg (2008) argued that, in urban populations, as much as 30% of older adults are widowed or on government assistance gamble. Older adults, who are more likely to be socially isolated or on a fixed budget (Vander Bilt et al., 2004), might presumably be more likely to engage in gambling for social or tangible functions. In the event that gambling is maintained by social functions, interventions that help adults identify other preferred activities and potential social companions might be prudent. However, if gambling is maintained by tangible functions (i.e., money), interventions designed to enhance stimulus control (i.e., make the amount of money the older adult is losing more salient) and focusing on mediating verbal behavior (see Dixon, 2010, in this issue for a cogent account of remediating verbal behavior associated with near misses) might prove useful. In addition to adults over the age of 65, those ages 50 – 64 might also benefit from such interventions. Indeed, the group of adults age 50-64 might have additional influences to gamble – the need to gamble to supplement or replace retirement funds. Unfortunately, however, at this point there is simply not enough research on older-adult gamblers to make predictions about which interventions might be prudent or effective. Conclusion Behavior-analytic research on older adult gambling is the Jackalope of behavior analysis but has great potential. Behavioral gerontologists have demonstrated that many of the current practices in behavior analysis are easily applied to older-adult populations, including preference assessment methodology (LeBlanc et al., 2006; LeBlanc et al., 2008),

basic human operant research (Plaud et al., 1999), and functional analysis (Baker & LeBlanc, in press). Gambling behavior in older adults, however, remains relatively unstudied. Current behavior-analytic research on gambling has begun to provide valuable information about the preferences of gamblers and the factors that maintain gambling. Further behavioral research on gambling that focuses on older adults could benefit older adult populations by extending preference and engagement technology to activities that provide cognitive and health benefits. In addition, researchers could begin to identify changes in reinforcement and stimulus control that could directly impact behavioral interventions used to ameliorate aberrant behavior and promote pro-social behaviors. Also, research on pathological older gamblers might not only improve the quality of life for older gamblers, but may provide valuable information as to why pathological gambling is less common among older adults (i.e., information that might begin to parse out cohort effects from aging effects). In addition to helping older adults, behavior analysts who study gambling stand to benefit in a number of ways when working with older adults. First, older adults constitute a potentially large subject pool that is likely to enjoy gambling studies (i.e., participating in a study could be seen as access to a preferred activity). Second, by extending studies beyond college students, researchers can extend the external validity of their studies. Finally, as the baby-boom generation continues to age, the number of gamblers over the age of 65 will continue to grow and skew the average of the typical gambler. Behavior analysts who begin to answer questions about the behavior of older adults related to gambling will be able to provide answers that no other discipline has been able to provide and put behavior analysis on the forefront of treatment for something that could soon become much more pertinent in the public’s eye. Such a move would allow behavior analysts to pro-

Jonathan C. Baker vide socially relevant treatment and help to move behavior-analytic research on older adults and gambling beyond the mythical realm of Jackalopes and into a respected and sought after science of human behavior. REFERENCES Adkins, V., & Mathews, M. (1999). Behavioral gerontology: State of the science. Journal of Clinical Geropsychology, 5, 39-49. Baker, J. C., & LeBlanc, L. A. (in press). Assessment and treatment of hoarding in an individual with dementia. Behavior Therapy. Buchanan, J., Husfeldt, J., Berg, T., & Houlihan, D. (2008). Publication trends in behavioral gerontology in the past 25 years: Are the elderly still an understudied population in behavioral research? Behavioral Interventions, 23, 65-74. Carstensen, L. L., & Erickson, R. J. (1986). Enhancing the social environments of elderly nursing home residents: Are high rates of interaction enough? Journal of Applied Behavior Analysis, 19, 349-355. Christensen, N. D. (2005). Cottontail rabbit papillomavirus (CRPV) model system to test antiviral and immunotherapeutic strategies. Antiviral Chemistry & Chemotherapy, 16, 355–362. Desai R. A., Maciejewski, P. K., Dausey, D. J., Caldarone, B. J., & Potenza, M. N. (2004). Health correlates of recreational gambling in older adults. American Journal of Psychiatry, 161, 1672-1679. Dixon, M. R., Nastally, B., L., & Waterman, A. (in press). The effect of gambling activities on happiness indices of nursing home residents. Journal of Applied Behavior Analysis.

13

Dixon, M. R., & Johnson, T. E. (2007). The gambling functional assessment (GFA): An assessment device for identification of the maintaining variables of pathological gambling. Analysis of Gambling Behavior, 1, 44-49. Engelman, K. K., Altus, D. E., & Mathews, R. M. (1999). Increasing engagement in daily activities by older adults with dementia. Journal of Applied Behavior Analysis, 32, 107-110. Erickson, L., Molina, C. A., Ladd, G. T., Pietrzak, R. H., & Petry, N. M. (2005). Problem and pathological gambling are associated with poorer mental health in older adults. International Journal of Geriatric Psychiatry, 20, 754-759. Gallagher, S. M., & Keenan, M. (2000a). Independent use of activity materials by the elderly in a residential setting. Journal of Applied Behavior Analysis, 33, 325-328. Gallagher, S. M., & Keenan, M. (2000b). Extending high rates of meaningful interaction among the elderly in residental care through participation in a specifically designed activity. Behavioral Interventions, 15, 113-120. Giri, I., Danos, O., & Yaniv, M. (1985). Genomic structure of the cottontail rabbit (Shope) papillomavirus. Proceedings of the National Academy of Science, 82, 1580-1584. Hagopian, L. P., Long, E. S., Rush, K. S. (2004). Preference assessment procedures for individuals with developmental disabilities. Behavior Modification, 28(5), 668-677. Hoon, A., Dymond, S., Jackson, J. W., & Dixon, M. R. (2008). Contextual control of slot-machine gambling: Replication and extension. Journal of Applied Behavior Analysis, 41, 467-470.

14


LeBlanc, L. A., Cherup, S. M., Feliciano, L. & Sidener, T. M. (2006). Using choice making opportunities to increase activity engagement in individuals with dementia. American Journal of Alzheimer’s Disease and Other Dementias, 21, 318-325. LeBlanc, L. A., Raetz, P. B., Baker, J. C., Stroebel, M. J., & Feeney, B. A. (2008). Assessing preference in elders with dementia using multi-media and verbal Pleasant Events Schedules. Behavioral Interventions, 23, 213-225. LeBlanc, L. A., Raetz, P. B., & Feliciano, L. (in press). Behavioral gerontology. In W. W. Fisher, C. C. Piazza, and H. S. Roane (Eds.), Handbook of applied behavior analysis. New York: Guilford. Lindsley, O. R. (1964). Geriatric behavioral prosthetics. In R. Kastenbaum (Ed.), New Thoughts in Old Age (pp. 41-61). New York, NY: Springer. McClannahan, L. E., & Risley, T. R. (1975). Design of living environments for nursing home residents: Increasing participation in recreation activities. Journal of Applied Behavior Analysis, 8, 261-268. Miller, J. C., Meier, E., Muehlenkamp, J., & Weatherly, J. N. (2009). Testing the construct validity of Dixon and Johnson’s (2007) gambling functional assessment. Behavior Modification, 33, 156-174. National Research Council. (1999). Pathological gambling: A critical review. Washington, DC: National Academy Press. Nevin, J. A. (1969). Signal detection theory and operant behavior: A review of David M. Green and John A. Swets’ Signal Detection Theory and Psychophysics. Journal of the Experimental Analysis of Behavior, 12, 475–480. Perez-Gonzalez, L. A., & Moreno-Sierra, V. (1999). Equivalence class formation in elderly persons. Psicothema, 11, 325336.

Plaud, J. J., Gaither, G. A., & Lawrence, J. B. (1997). Operant schedule transformations and human behavioral momentum. Journal of Behavior Therapy and Experimental Psychiatry, 28, 169-179. Plaud, J. J., Gillund, B., & Ferraro, F. R. (2000). Signal detection analysis of choice behavior and aging. Journal of Clinical Geropsychology, 6, 73 – 81. Plaud, J. J., Plaud, D. M., & Von Duvillard, S. (1999). Human behavioral momentum in a sample of older adults. Journal of General Psychology, 126, 165-175. Preston, F. W., Shapiro, P. D., & Keene, J. R., (2007). Successful aging and gambling: Predictors of gambling risk among older adults in Las Vegas. American Behavioral Scientist, 51, 102-121. Quadagno, J. (2005). Aging and the life course (3rd edition). New York, NY: McGraw Hill. Rachlin, H. (1990). Why do people gamble and keep gambling despite heavy losses? Pschological Science, 1, 294 – 297. Saunders, R.R., Chaney, L., & Marquis, J.G., (2005). Equivalence class establishment with two-, three-, and four-choice matching to sample by senior citizens. The Psychological Record, 13, 539-559. Sidman, M. (1997). Equivalence relations. Journal of the Experimental Analysis of Behavior, 68, 258-266. Sidman, M., & Tailby, W. (1982). Conditional discrimination vs. matching to sample: An expansion of the testing paradigm. Journal of the Experimental Analysis of Behavior, 36, 5-22. Sidman, M., Wayne, C. K., Macguire, R. W., & Barnes, T. (1989). Functional classes and equivalence relations. Journal of the Experimental Analysis of Behavior, 52, 261-274. Skinner, B. F. (1983). Intellectual selfmanagement in old age. American Psychologist, 38, 239-244.

Jonathan C. Baker Vander Bilt, J. V., Dodge, H. H., Pandav, R., Shaffer, H. J., & Ganguli, M. (2004). Gambling participation and social support among older adults: A longitudinal community study. Journal of Gambling Studies, 20, 373- 390. Wilson, K. M., & Milan, M. (1995). Age differences in the formation of equivalence classes. Journals of Gerontology--Series B: Psychological Sciences & Social Sciences, 50B, 212-218. Zaranek, R. R., & Litchenberg, P. A. (2008). Urban elders and casino gambling: Are they at risk of a gambling problem? Journal of Aging Studies, 22, 13-23. Zlomke, K. R., & Dixon, M. R. (2006). Modification of slot-machine preferences through the use of a conditional discrimination paradigm. Journal of Applied Behavior Analysis, 39, 351-361. Action Editor:

Jeffrey N. Weatherly

15


2010, 4, 16–26


The Effect of Relational Training on the Near-Miss Effect in Slot Machine Players Becky L. Nastally & Mark R. Dixon Southern Illinois University In the current study, six slot machine players were exposed to two concurrently available computer simulated slot machines (one yellow and one blue). The blue slot machine produced a high frequency of near-miss outcomes and the yellow slot produced no such outcomes. Both machines produced reinforcement on a random-ratio 10 schedule and response options were presented in a free operant paradigm. After a 50-trial exposure, participants completed multiple exemplar training and testing as well as a stimulus-sort task to form a relation between the color blue and ‘worse-than’ and then were re-exposed to the slot machine task for another 50 trials. Results indicated that four of six participants initially showed a preference for the near-miss slot machine. However following training and testing phases, four of six participants’ response allocation toward this slot decreased. The results are discussed in terms of the formal and functional properties of what is termed as the ‘near-miss’ effect. Keywords: Near-miss effect, Gambling, Preference, Verbal behavior

--------------------------------------------------

The near-miss effect is a widely investigated concept in the gambling literature. It serves as a prime example of a variable other than winning that may work to maintain gambling behavior. Although it is primarily referred to as a ‘near miss’, it may be more clearly conceptualized as ‘almost winning’ or ‘very close to winning’ as previous research has shown (Dixon & Schreiber, 2004). On a slot machine, for example, a near miss is often defined as two of three slot machine reels stopping on identical symbols while the third or last reel stops on a different symbol, suggesting a win is just out of reach, even though this is not the case. This effect is not exclusive to slot machines, as recent research has shown parallels of almost winning in the game of blackjack (Dixon, Nastally, Hahs, HornerKing & Jackson, 2009) and roulette (Hahs &

Dixon, manuscript in preparation). Explanations of this observed effect have been offered both outside and within the field of behavior analysis. Those from the cognitive perspective have described the near miss as a cognitive fallacy (Griffiths, 1991) and speculated that this outcome can strengthen particular strategies and increase beliefs about a future success (Reid, 1986). Behavior-analytic interpretations have pointed to the effects of conditioned reinforcement through stimulus generalization (Skinner, 1957) and research has provided evidence of the role of verbal behavior (Dixon, Nastally, Jackson, & Habib, 2009). Additionally, recent research has attempted to analyze this effect at the physiological level and it seems there are neurological differences in how pathological and non-pathological gamblers respond to near misses (Habib & Dixon, in press). A study conducted by Kassinove and Schare (2001) investigated the effect of different rates of exposure to near-miss slotmachine outcomes (15%, 30%, and 45%) on gambling persistence in 180 undergraduate

Address all correspondence to: Mark R. Dixon Behavior Analysis and Therapy Program Rehabilitation Institute Southern Illinois University Carbondale, IL 62901 Email:mdixon@siu.edu

16

Becky L. Nastally and Mark R. Dixon

17

participants. Statistical analyses showed that there was a significant relationship between persistence and rate of near misses, the strongest occurring during the 30% nearmiss exposure. Another study examined these same rates of near-miss exposure, but all three response options were presented concurrently (MacLin, Dixon, Daugherty, & Small, 2007). These results showed that both prior to and during extinction, participant response allocation to the different slot machines was linearly related to the amount of near-miss frequency (i.e., participants allocated the majority of their responses toward the slot that produced near-misses 45% of the time). Dixon et al. (2009) recently investigated the effect of the formation of verbal relations on changes in the near-miss effect as measured by subjective ratings. Using a simple procedure, participants were first exposed to various pictures of slot machine outcomes (wins, near-misses, and total losses) and asked to rate them in terms of their closeness to winning. As was expected, most participants rated near-miss outcomes substantially higher than total losses but not as high as wins. Upon conditional discrimination training, wherein a relation between the word “Loss” and a picture of a near-miss outcome was formed, 10 of 16 participants (specifically those who met criterion in the relational-training portion of the experiment) decreased their closeness to win ratings when they were again presented with the picture. There is evidence to suggest that conditional discrimination procedures similar to the one utilized above can also have an effect on response allocation to concurrently available gambling options (Hoon, Dymond, Dixon, & Jackson, 2008; Nastally, Dixon, & Jackson, 2009; Zlomke & Dixon, 2006). For example, Zlomke and Dixon (2006) demonstrated that a yellow slot machine was preferred over a

concurrently available blue slot machine producing the same win rate following a training procedure that resulted in the formation of a rule between the color yellow and ‘greater-than’. As the two response options produced equal exposure to wins and losses in this study, it is unknown whether the verbal rule formation would override differing programmed contingencies produced by the concurrently available response options. The purpose of the current experiment was to extend previous findings on the nearmiss effect and verbal rule adherence as it relates to gambling. Specifically, we sought to evaluate the influence of non-arbitrary multiple exemplar training on response allocation toward two concurrently available simulated slot machines, one of which produced a high frequency of near-miss outcomes while the other produced no such outcomes. METHOD Participants and Setting Six graduate students (aged 22-42; 5 F, 1 M) were recruited to participate in the study for course extra credit. The participants were enrolled in a behavior analysis program, but they all reported being unfamiliar with the behavior-analytic literature on gambling. Participants were screened for potential gambling pathology using the South Oaks Gambling Screen (SOGS) (Lesieur & Blume, 1987). The entire experiment took approximately one hour to complete and took place in a university human operant laboratory. The specific room used was approximately 4 ft. by 6 ft that contained a computer, desk, and chair. Apparatus and Experimental Stimuli All phases of the current study were conducted on a Dell Precision 690 PC equipped with a 22” monitor and a mouse.

18

THE EFFECT OF RELATIONAL TRAINING

All experimental procedures were programmed using Microsoft Visual Basic .NET. Phase 2, or multiple exemplar training with non-arbitrary stimuli, involved the use of 10 different word pairs and two color stimuli (blue and yellow) that were incorporated to train multiple exemplars of better- or worse-than relations. All stimuli were presented as 2 in. by 2 in. graphical images containing one word or color. The textual stimuli included both word pairs that were associated with gambling and were not necessarily associated with gambling. These word pairs were “alive-dead”, “rich-poor”, “healthy-sick”, “winner-loser”, “successfulfailure”, “attractive-ugly”, “intelligentstupid”, “interesting-boring”, “happydepressed”, and “strong-weak”. The rest of the stimuli used in Phase 2 were two colors: blue and yellow. Stimuli used during Phase 3 (the stimulus sort task) consisted of 14 2in. by 3-in. pictures (one blue slot, one yellow slot, and 12 words). This phase was a table top procedure so that pictures were printed out and cut to be uniform in size. Experimental Design and Procedure The study utilized a within-subjects pre/post-test design. Upon beginning the experiment, participants signed the consent form. Next, the experimenter led the participant into the room in which the experiment took place and the participant completed a computerized version of the SOGS (Lesieur & Blume, 1987). Upon completing this questionnaire, Phase 1 began. Phase 1: Slot Machine Exposure. In Phase 1, the following instructions were first read to the participant: “Today you have the opportunity to play these two computerized slot machines and switch back and forth between them as you so choose. You may choose by clicking the mouse on the picture of the slot machine on which you would like to play (prompt the participant to make their

first choice.) Each time you will bet by pressing the ‘bet one credit’ button. Upon clicking on the betting button, the spin button will be activated. After each spin, your credits will be cashed out and you will again return to the choice screen and can choose freely each time which machine you would like to play on. In addition to earning your extra credit today, there are some additional contingencies in place for playing. For example for every time that you win on the cherries on up through the double bars, we will enter your name into a drawing for a $25 gift certificate. The 3 ‘Exit Signs’ are our jackpot today. Upon getting this win you will not only receive extra credit, but you’ll get to leave the study immediately and we will give you the $25 gift certificate. When you are ready to begin click on the BEGIN button.”

During this phase, participants were given the opportunity to play “Slot Machine 1” or “Slot Machine 2” which were identical except for their base color. The background of Slot Machine 1 was blue and the background of Slot Machine 2 was yellow. Both machines were represented on a choice screen with the question “You may play on either Slot Machine 1 or Slot Machine 2. Which slot machine would you like to play on?” The position of the slot machines on this screen was randomized in order to control for position bias. Upon clicking on one of the slot machines, participants were exposed to a screen containing that slot machine in the center of the screen. When this screen appeared, 200 credits were transferred to the ‘total credit’ window and the ‘bet one’ and ‘bet max’ buttons were highlighted. The participant was only able to bet one credit on each trial. Upon clicking on this button, the ‘spin’ button lit up and after it was clicked the three reels of the slot machine spun for approximately three seconds. After the reels stopped, depending on the outcome, one credit was either added or deducted from the ‘total credit’ window (reinforcement


19

magnitude was held constant at one credit gain or loss on each trial). Then a prompt appeared on the screen in the form of an arrow pointing at the ‘cash out’ button that read “Click on the Cash Out button to continue”. Upon clicking on the cash-out button, participants were exposed to a screen that informed them of the amount of total credits they had. An observing response was required (clicking on a button that read “Click Here”) in order to return to the choice screen. This observing response was instated to reinforce attending to the stimuli on the screen. Phase 1 consisted of 50 trials. During this phase, wins were programmed on a random-ratio (RR) 10 schedule on each of the two slot machines (i.e. each slot machine produced wins according to this schedule independently of the other machine) and near-miss outcomes were programmed on a RR 40 on the blue slot machine. As with real slot machines, the outcome of each trial (win or various loss types) occurred independently of past or future trials. In an attempt to increase the value of ‘winning’ and thus the value of a near-miss outcome, participants were told that each time they won on the first nine possible winning combinations (cherries through double bars), their name would be entered into a drawing for a $25 gift certificate. A ‘jackpot’ was also created in that participants were told that obtaining three identical ‘EXIT’ signs would result in getting to leave the experiment immediately and receiving the $25 gift certificate directly. The jackpot winning combination was programmed never to occur, so participants never actually contacted this contingency. In addition to these contingencies, Slot Machine 1 (i.e., the blue slot machine) was programmed to produce near-miss outcomes 40% of the time while Slot Machine 2 (i.e., the yellow slot machine) produced zero near-miss outcomes. A near miss was defined as two

identical symbols appearing on the payout line in either the first and second, first and third, or second and third position with a different symbol appearing in the remaining position. Following 50 trials of simulated-slotmachine play, the experimenter asked participants to answer two questions. First, to estimate how many times they won on both the yellow and blue slots and second, if they were given the opportunity to play on only one of the slot machines for the next 100 trials, which one would they prefer. Phase 2: Multiple Exemplar Training with Non-Arbitrary Stimuli. In Phase 2, a non-arbitrary relational training procedure was presented to participants (see Dixon, Bihler, & Nastally, in press; Reilly, Whelan, & Barnes-Holmes, 2005) to establish relations of ‘better-than’ and ‘worse-than’ in the presence of the colors yellow and blue. In the training, there were five pairs of textual stimuli, or written words, the first of which represented the ‘better-than’ relation and the second represented the ‘worse-than’ relation. These stimuli consisted of the following pairs: “alive-dead”, “rich-poor”, “healthy-sick”, “winner-loser”, and “successful-failure”. Because each pair appeared in the presence of both colors, there were 10 different trial type combinations. During a trial, either of the two colors first appeared toward the top of the screen (as the sample stimulus) followed by two words underneath the figure side by side (the comparison stimuli). Differential reinforcement for clicking on the appropriate comparison stimulus given the sample was provided in the form of auditory feedback consisting of a pleasant auditory sound (short tada .wav file) following a correct response or a neutral auditory sound (tone .wav file) following an incorrect response. The presence of the colors

20


determined which selection of the two comparison stimuli was reinforced. For example, in the presence of yellow, selecting the stimulus representing the relation ‘better-than’ (i.e. “alive”, “rich”, “healthy”, “winner”, “successful”) was reinforced. In the presence of blue, selecting the stimulus representing the relation ‘worse-than’ (i.e. “dead”, “poor”, “sick”, “loser”, “failure”) was reinforced. There were a total of 40 trials and participants needed to reach a criterion of 90% correct responding (36 out of 40 trials). If a participant did not reach criterion responding, exposure to the training blocks continued. If a participant did not meet criterion after exposure to five successive trial blocks following initial exposure, he or she was thanked for participating, provided course extra credit, and excused from the experiment. Once participants met criterion in the training portion, they immediately moved into a testing phase which was the same as the training phase including stimuli presented, number of trials and correct responding criterion in order to advance to the next phase. The only difference was the trials were presented in the absence of any feedback and the following novel word sets were used: “attractive-ugly”, “intelligentstupid”, “interesting-boring”, “happydepressed”, and “strong-weak”. Following Phase 2, participants immediately proceeded to Phase 3. Phase 3: Stimulus Sort Task. Upon successful completion of the multiple exemplar training/testing phase, to verify that the proper relations were formed participants were exposed to a stimulus sort task that incorporated three word-sets randomly selected from the training phase (alive-dead, attractive-ugly, happy-sad) and three novel word sets related to gambling (good-bad, gamble-save, jackpot-bankrupt).

The selection of stimuli for the sort task was based on prior research on transfer of function in a gambling context (Hoon et al., 2008; Zlomke & Dixon, 2006). The sort task was a table-top procedure in which two 2-in. by 3-in. pictures of the slot machines (exactly as they appeared in the visual basic program) were presented in front of the participant. The participant was then given 2-in. by 3-in. cut-out cards with the individual words from each set typed in bold font and asked to place the card underneath the picture of the slot machine it went with. The experimenter informed the participant that no feedback would be delivered during this phase. In all, there were 12 individual words presented a total of three times each making the phase consist of 36 trials. The order in which the words were presented was determined randomly. Phase 4: Slot Machine Re-exposure. After completing Phase 3, participants were re-exposed to the slot-machine task as described above. In addition to the two questions asked of participants during Phase 1, the experimenter also asked them “Overall, in deciding whether to play on the yellow or blue slot machine, what were you attending to more, the color of the slot machine or the number of times it was winning?” Dependent Measures and Reliability Three response measures were of interest in the current study. First, response allocation to the slot machine producing a higher proportion of near-miss outcomes (the blue slot machine – Slot Machine 1) before and after the multiple exemplar training indicated whether the verbal relations that emerged affected overt responding. Second, the percentage of correct responses within a set of test-trial blocks during multiple exemplar training


21

(Phase 2) indicated whether the functions of ‘better-than’ and ‘worse-than’ transferred to the colors. And third, the number of correct responses during the stimulus sort task was measured as another indicator of the strength of the relation. To ensure reliability of response measurement, a second observer reviewed

50% of the total output files that contain permanent records of responding produced by the programming software and IOA was calculated as 100%. An independent observer also scored 50% of the table-top stimulus sort task sessions and IOA for this measure was 91%.

Figure 1. Response allocation to the near-miss slot during Phase 1 (black) and Phase 4 (white).

RESULTS The SOGS scores of Participants 001, 002, 003, 005, 006 and 007 were 0, 3, 0, 0, 0, and 0 respectively. Percent response allocation toward the blue (or near miss; NM) slot machine for each participant during the initial and final slot machine exposure are presented in Figure 1. During Phase 1, the initial slot machine exposure, Participants 001, 003, 006, and 007 (four out of six) showed a preference for the slot machine that was producing near-miss outcomes at a rate of 40% as opposed to the other slot which was producing no near-miss outcomes. Preference was defined as percent response allocation (of a total of 50 trials)

exceeding 55%. The other two participants (002 and 005) emitted response allocation to the blue slot machine 14 and 42% of the time respectively. Overall, the median percentage of responses allocated to the NM slot machine across all participants during Phase 1 was 61% . The next phases of the experiment were multiple exemplar training and the stimulus sort task. Trial blocks to criteria during multiple exemplar training for Participants 001, 002, 003, 005, 006 and 007 were 1, 4, 1, 1, 1, and 1 respectively. Performance during the stimulus sort task was 92, 92, 100, 92, 11, and 83% correct for Participants 001 through 007 respectively.

22


Following the stimulus sort task, participants were re-exposed to the simulated slot machines for another 50 trials. Of the participants who showed an initial preference for the blue slot machine (001, 003, 006, and 007), all except Participant 006 showed a decrease in response allocation toward the same machine following training and stimulus sort task phases (see Figure 1). Participant 006 actually showed a 6% increase toward the blue slot machine. Interestingly, even though Participants 002 and 005 did not show an initial preference for the blue slot, Participant 002’s response allocation toward this slot increased substantially following the training and sorting phases (specifically by 50%). Participant 005’s response allocation toward the blue slot, however, started out low and decreased slightly during re-exposure. Total median response allocation toward the blue slot machine during Phase 1 (pre) and Phase 4 (post) was 61% and 32% respectively. In regard to the measures of verbal behavior across participants following Phases 1 and 4, participants were fairly accurate in estimating how many times they won on each of the two slot machines. The greatest discrepancy between actual and reported win frequencies never exceeded three instances. The question of which slot machine they would play on for the next 100 trials, given the opportunity, sought to gain a measure of subjective preference. Interestingly, during the initial slot-machine exposure, there was no discrepancy between which slot participants reported preferring and to which slot they allocated a greater proportion of responses. However, during the final exposure, differences were observed. For example, Participants 001 and 007 both reported preferring the blue slot, but allocated a greater percentage of responding toward the yellow slot machine. Lastly, five of six participants reported

attending to the win rate of slot machines to a greater degree than their color when choosing to play on either the blue or yellow one. Because each slot machine in the current study produced reinforcement on a true RR schedule, a contingency table summarizing the exact magnitude of reinforcement produced by both the yellow (Y) and blue (B) slot machines for each participant is presented in Table 1. Also found in this table are the total losses during Phase 1, or pre-test (Pr), and Phase 4, or post-test (Po), produced by both machines for each participant. The total number of near-miss outcomes produced by the blue slot only and how many times the two identical symbols appeared in the first two (NMright), first and third (NMmid), or second and third reel positions (NMleft) are also depicted here. DISCUSSION The results of the current study showed that the majority of participants initially preferred a simulated slot machine that produced a fairly high frequency of nearmiss outcomes (40%) over one that produced no near-miss outcomes at all. The results of Phase 1 by itself extend the findings of previous studies on the near-miss effect in that preference was measured as response allocation to two concurrently available slot machines, one of which was producing a higher rate of near-miss outcomes. Most studies on the near-miss using objective measures have manipulated this variable in the context of prolonged play on one available slot machine using group design methodology (Cote et al., 2003; Daugherty & MacLin, 2007; Ghezzi et al., 2006; Kassinove & Schare, 2001) and we are aware of only one other study that has examined it in a concurrent operant paradigm (MacLin et al., 2007). If the nearmiss effect is defined in terms of


23

conditioned reinforcement, it would be expected that it would not only produce gambling persistence but also greater responding in the context of choice. In addition, the results observed in Phase 1 extend the findings of studies that have utilized verbal behavior as the primary measure of preference for near-miss outcomes (Dixon & Schreiber, 2004; Dixon, et al., 2009). Specifically, they provide evidence of a correspondence between what participants said they prefer and the overt choices they made related to gambling

response options. Such a correspondence is essential if valid assumptions are to be made solely based on verbal behavior as a dependent variable and further replication of this correspondence is needed to strengthen the validity of self-report measures in general. In essence, there is a need to increase investigation of whether people ‘do as they say they do’ as it were. The results of the multiple exemplar training and stimulus-sort task, as well as the participants’ SOGS scores, as they relate to the overt responding in Phases 1 and 4 are

Table 1

Contingency Summary Across Trial Types for Each Participant Total Loss Total Wins Total NM NMright

NMmid

NMleft

Part # Pr Po Pr Po Y B Y B Y B Y B

Pr B

Po B

Pr B

Po B

Pr B

Po B

Pr B

Po B

001

21 18 23 16 1 1

2 4

7

5

5

4

2

1

0

0

002

37 5

0 3

1

12

1

9

0

1

0

2

003

18 18 33 9

2 3

3 0

9

5

7

5

0

0

2

0

005

28 11 30 9

0 3

2 1

7

8

7

5

0

2

0

0

006

16 16 19 21 5 3

0 1

10

9

9

9

0

0

1

0

18 16 6 1

007 9 18 19 14 2 6 6 2 15 9 12 6 1 1 2 2 ______________________________________________________________________________ note-worthy for two reasons. First, all participants, except for 002 who required four training blocks, met correction criterion during the multiple exemplar training within one trial block. Participant 002 also had the highest SOGS score. Even though she did not score in the range of a potential pathological gambler, a score of 3 does indicate some evidence of a potential gambling problem (Lesieur & Blume, 1987). It could be that, as other studies on gambling

have speculated (Nastally et al., 2009), individuals with a history of problem gambling adhere to self-rules to a greater degree than do individuals with no evidence of pathology and because of this adherence it was more difficult for her to learn the rule provided by the training. This hypothesis seems to be supported by the fact that even though the relations were eventually derived, she did not respond in accordance with them as demonstrated by the

24


subsequent increase in her responding toward the near-miss slot. Second, an interesting finding comes from the results of the stimulus-sort task. For example, Participant 006 responded correctly on only 11% of the sort task trials, but reached criterion responding in only one trial during multiple exemplar training. It appears that she was responding in accordance with a rule involving the exact opposite contingencies between the multiple exemplar training and the sort task. This participant showed an initial preference for the near-miss slot, which increased slightly during re-exposure suggesting the rule derived during the sort-task had greater control over her responding. However, it is difficult to attribute such an influence of the sort-task by itself when Participant 007 produced the next lowest score (83%) and subsequently reduced responding toward the blue slot substantially. As stated in the results, of the four participants who demonstrated an initial preference for the blue slot, three of them showed a decrease in preference following the training and sort tasks. Although Participant 005 did not show an initial preference for the blue slot, her response allocation also decreased following the training phases. These findings extend those of prior studies on transfer of function in a gambling context (Hoon et al., 2008, Nastally et al., 2009; Zlomke & Dixon, 2006) in that the slot machines in the current experiment differed in both color and loss type frequency. The finding that initial differences in responding can be predicted based on loss type (a greater frequency of near-miss as opposed to total loss outcomes), and those differences can be reversed as a result of reinforced rule following is a meaningful contribution to this particular body of research. At the same time, although it can be said that four of six participants constituted

a majority in the present experiment, it is worth noting that these outcomes are not vastly greater than those that would be expected based on chance alone. Other investigations of the near-miss effect using a concurrent operant set-up have produced similarly less-than-extreme demonstrations of preference (MacLin et al., 2007) and there have also been instances of no nearmiss effect being observed whatsoever (Ghezzi et al., 2006; Whitton & Weatherly, 2009). Given this fluctuation in pronouncement of the near miss-effect and the range of methodology designed to study it, it is possible that procedural variations (e.g. availability of one or more response options, schedule of reinforcement, forced trials vs. free operants, etc.) contribute at least in part to its occurrence. In terms of ways this experiment differs from and extends the conclusions that can be drawn from previous gambling studies on transfer of function (Hoon et al., 2008, Nastally et al., 2009; Zlomke & Dixon, 2006), the role of reinforcement in the current study is also worth comment. While these former studies have incorporated a matched schedule of reinforcement for each of the concurrently available slot machines, the current set of slot machines produced reinforcement on a true RR schedule of reinforcement. As a consequence, it is impossible to completely rule out reinforcement as a potential determiner of participant response allocation. At the same time, however, it lends greater external validity to the current methodology. Additionally, as shown by the data in Table 1, the magnitude of reinforcement produced by the two machines did not differ substantially and this difference must be considered in terms of the actual percentage of response allocation that allowed the participant to experience such reinforcement.


25

The current experiment is not without limitations. For example, pre-existing histories associated with the specific colors of the slot machines cannot be accounted for in the current study because the color slot producing the greater frequency of nearmiss outcomes was not counterbalanced across participants. However, if history did in fact play a role, one would expect more exclusive preference for either the yellow or blue slot across participants during pre-test. Still, future studies incorporating similar methodology should control for this factor. Likewise, the pre-test/post-test design methodology does not rule out all confounding variables (e.g. reinforcement). However, the present experiment should be viewed as preliminary as it is one of the first to attempt this type of investigation of the near-miss effect, and subsequent studies should employ more conservative design methodology to support the present findings. The present study addressed the roles of both verbal rule adherence and what has been termed the near-miss effect in influencing the choice making of gamblers. Within both of these areas there are a number of potential research questions to pursue. For example, to further illustrate the role of verbal behavior in gambling, future studies should continue to pit derived or directly introduced rules against a variety programmed contingencies within the context of choice making. Varying the loss type in the present analysis was one example, but several variations of mixed schedules of reinforcement produced by each response option could also be utilized. In terms of the near-miss effect, future studies could attempt to treat such maladaptive rule following in individuals with a history of problem gambling using a brief clinical intervention such as providing accurate information about near-miss outcomes. There is increasing support for reducing gambling behavior in non-

pathological gamblers through such interventions (Mui & Dixon, under review; Weatherly & Meier, 2008) and it would seem to follow that such treatment strategies could also be effective in reducing the rulegoverned behavior of real gamblers. In summary, the near-miss effect represents an important area of gambling research. Although it has been conceptualized in a number of ways within the gambling literature, most researchers are in agreement about its harmful effect on the problem gambler. Evidence of this effect on both cognition and behavior is well documented. This effect represents just one of the strategies that are used by the gambling industry to perpetuate gambling behavior and more research is necessary to identify the most effective way to counteract these efforts. REFERENCES Daugherty, D., & MacLin, O. H. (2007). Perceptions of luck: Near wins and near loss experiences. Analysis of Gambling Behavior, 1, 123-133. Dixon, M. R., Bihler, H., & Nastally, B. L. (in press). Slot machine preferences of pathological and recreational gamblers are verbally constructed. The Psychological Record. Dixon, M. R. Nastally, B. L., Hahs, A. D., Horner-King, M., & Jackson, J. W. (2009). Black-jack players demonstrate the near-miss effect. Analysis of Gambling Behavior, 3, 56-61. Dixon, M. R., Nastally, B. L., Jackson, J. W., & Habib, R. (2009). Altering the “near-miss effect” in slot machine gamblers. Journal of Applied Behavior Analysis, 42, 913-918. doi:10.1901/jaba.2009.42-913

26


Dixon, M.R., & Schreiber, J.E. (2004). Near-miss effects on response latencies and win estimations of slot machine players. The Psychological Record, 53, 335-348. Ghezzi, P.M., Wilson, G.R., & Porter, J.C.K. (2006). The near-miss effect in simulated slot machine play. In P.M. Ghezzi, C.A. Lyons, M.R. Dixon, & G.R. Wilson (Eds.) Gambling: Behavior Theory, Research, and Application (pp. 155-189). Reno: Context Press. Griffiths, M. (1991). Psychobiology of the near-miss in fruit machine gambling. The Journal of Psychology, 125, 347357. Hahs, A. D., & Dixon, M. R. (manuscript in preparation). Investigating the nearmiss effect in the game of roulette. Hoon, A., Dymond, S., Dixon, M.R., & Jackson, J.W. (2008). Contextual control of slot machine gambling: Replication and extension. Journal of Applied Behavior Analysis, 41(3), 467470. doi:10.1901/jaba.2008.41-467 Kassinove, J.I., & Schare, M.L. (2001). Effects of the “near miss” and the “big win” on persistence at slot machine gambling. Psychology of Addictive Behaviors, 15, 155-158. Lesieur, H. R., & Blume, S. B. (1987). The south oaks gambling screen (SOGS): A new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144(9), 11841188. MacLin, O. H., Dixon, M. R., Daugherty, D., & Small, S. L. (2007). Using a computer simulation of three slot machines to investigate a gambler’s preference among varying densities of near-miss alternatives. Behavior Research Methods, 39, 237-241.

Mui, N., & Dixon, M. R. (under review). Examining effects of different types of information on slot machine play. Analysis of Gambling Behavior. Nastally, B. L., Dixon, M. R., & Jackson, J. W. (2010). Manipulating slot machine preference in problem gamblers through contextual control. Journal of Applied Behavior Analysis, 43, 125-129. doi:10.1901/jaba.2010.43-125 Reid, R.L. (1986). The psychology of the near miss. Journal of Gambling Behavior, 2, 32–39. Reilly, T., Whelan, R., & Barnes-Holmes, D. (2005). The effect of training structure on the latency of responses to a five-term linear chain. The Psychological Record, 55, 233-249. Skinner, B. F. (1957) Verbal Behavior. Cambridge: Prentice-Hall. Weatherly, J. N., & Meier, E. (2008). Does providing accurate information about slot machines alter how participants play them? Analysis of Gambling Behavior, 2, 2-12. Whitton, M., & Weatherly, J. N. (2009). The effect of near-miss rate and card control when American Indians and nonIndians gamble in a laboratory situation: The influence of alcohol. American Indian and Alaska Native Mental Health Research, 16, 28-42. Zlomke, K.R., & Dixon, M.R. (2006). Modification of slot-machine preferences through the use of a conditional discrimination paradigm. Journal of Applied Behavior Analysis, 39,351-361. doi:10.1901/jaba.2006.10904 Action Editor: Jeffrey N. Weatherly


2010, 4, 27–37


Temporal Discounting and Gambling: A Meaningful Relationship? Jeffrey N. Weatherly

University of North Dakota Pathological gambling is an important and large societal problem. Theorists and researchers have linked pathological gambling to rates of temporal discounting, although not all attempts to do so have been successful. Unfortunately, popular measures of temporal discounting each have weaknesses, and studies of discounting have tended to focus on one particular commodity – hypothetical monetary rewards. Evidence exists to suggest that problem and pathological gambling is also linked to escape contingencies. If so, these findings could potentially explain the link that has been found between temporal discounting and gambling. Implications and predictions of this possibility are discussed. Keywords: Gambling, Temporal discounting, Escape

________________________________

Pathological gambling is a large societal problem, with around 2% of the adult population displaying the disorder and an additional 5 – 8% displaying sub-clinical symptoms (i.e., problem gambling; Petry, 2005). According to Petry (2005), there are six known risk factors associated with pathological gambling. One is gender, with males displaying the disorder significantly more frequently than females. Another is ethnicity, with higher rates of pathological gambling being found in minority populations than in the majority population. The third is age, with young adults being most likely to display pathological gambling and the likelihood of the disorder decreasing with age. The fourth factor is socio-economic status. Those low in socio-economic status are more likely to be pathological gamblers than are those high in socio-economic status. The penultimate factor is marital status, with pathological gamblers more likely to be single or divorced than be married. The final, and by far the biggest, risk factor is drug use

and abuse. The comorbidity rate of substance dependence and pathological gambling is so high that it is recommended that mental-health-care professionals working with one population screen for the other disorder (Petry, 2005). These factors are not the only ones that have garnered research attention in the study of gambling, however. Other factors have included psychological disorders (e.g., depression; Dannewitz & Weatherly, 2007) or personality characteristics (e.g., sensation seeking; Gillis, McDonald, & Weatherly, 2008). A factor that has received a great deal of research attention is the rate of temporal discounting (see Petry & Madden, 2010). Temporal discounting occurs when the subjective value of an outcome or consequence (e.g., a sum of money) is lessened because it is delayed in time. Phrased differently, individuals will typically take less than the full amount of the outcome or consequence in order to get it immediately rather than having to wait for the full amount (e.g., Baker, Johnson, & Bickel, 2003), with the amount that is acceptable immediately decreasing as the delay to the full amount increases (e.g., Smith & Hantula, 2008). Research on temporal discounting has found that pathological gamblers discount hypo-

Jeffrey N. Weatherly, Ph.D. Department of Psychology University of North Dakota Grand Forks, ND 58202-8380 Phone: (701) 777-3470 Fax: (701) 777-3454 Email: jeffrey.weatherly@und.edu

27

28

DISCOUNTING AND GAMBLING

thetical monetary rewards to a greater degree than do their non-pathological counterparts (e.g., Dixon, Jacobs, & Sanders, 2006; Dixon, Marley, & Jacobs, 2003). The idea that temporal discounting may play a role in the development and maintenance of pathological gambling is not new (e.g., Fantino & Stolarz-Fantino, 2008; and see Petry, 2005, for a review) nor is the potential application of discounting isolated to gambling (e.g., see Rachlin, 1997). In fact, Weatherly and Dixon (2007) made temporal discounting an integral component of their behavioral model of gambling. Specifically, they argued that certain of the risk factors for pathological gambling (described above) could serve as setting events or establishing operations, thus altering the subjective value of money. When the value of money is altered, you would expect to get changes in how money is discounted when it is delayed in time. According to Weatherly and Dixon (2007), those changes would ultimately lead the gambler down the road of pathology. After proposing this model, our laboratory set about testing its premises and predictions. To some extent, the results from those attempts supported the model. For instance, Weatherly, Marino, Ferraro, and Slagle (2008) recruited non-pathological individuals across a wide age range to participate in a laboratory gambling study. Participants completed a number of paperpencil measures, including a temporaldiscounting task. They were then staked with $10 in tokens that could be gambled, across a 15-min session, on a slot machine. The results showed that the only significant predictor of how many tokens the participants gambled across the session was the rate of discounting they had displayed on the temporal-discounting task. The other factors (gender, age, annual income) were not related to rates of gambling. Thus, this study became the first to demonstrate that rates of discounting were predictive of actual gam-

bling behavior (vs. self reports or hypothetical situations). Furthermore, the results were observed in non-pathological participants, suggesting that the relationship between gambling and temporal discounting did not require the presence of pathology. Other research, however, was not so supportive. For instance, Weatherly, Derenne, and Chase (2008) investigated the idea that the risk factors for gambling would be related to temporal discounting. Specifically, they collected demographic information from 236 college students who then completed a temporal discounting task, the South Oaks Gambling Screen (SOGS), which measures lifetime gambling behavior (Lesieur & Blume, 1987), and the Gambling Functional Assessment (GFA), which measures the contingencies that maintain gambling behavior (Dixon & Johnson, 2007). The study was designed to test the following predictions: that the risk factors for gambling would be related to rates of temporal discounting, that rates of temporal discounting would be related to the extent to which people displayed symptoms of pathological gambling (as measured by the SOGS), and that whether or not peoples’ gambling behavior was maintained by monetary consequences would be related to both symptoms of gambling problems and rates of temporal discounting. However, none of these predictions were supported. The value of the data reported by Weatherly, Derenne, and Chase (2008) was not necessarily related to the association between temporal discounting and gambling behavior. Rather, the interesting outcome in their study was the temporal-discounting data themselves. Their study employed a paper-pencil, binary-choice temporal-discounting task in which participants were asked a series of questions that required them to choose between two monetary options (i.e., $1,000 available after a delay or a lesser amount available immediately). With

Jeffrey N. Weatherly this procedure, rate of temporal discounting is determined by identifying the point at which, at each different delay, the participant switches from preferring the full, but delayed amount to preferring the lesser, more immediate amount. To minimize the number of questions that needed to be asked, and to combat order effects, the temporaldiscounting questions were randomized. Thus, from question to question, participants were presented with changes in both the delay to the full amount and the size of the immediately available amount. Despite the scientifically sound practice of randomization, this manipulation wreaked havoc with the data. Specifically, nearly 65% of the sample displayed multiple switch points at at least one delay. For instance, at a delay of one month, a person might in one question choose $900 rather than waiting for $1,000, but when faced with another question (later in the survey) choose to wait the month to get the $1,000 rather than accepting $950 immediately. When this inconsistency occurs, the researcher is faced with a number of decisions. For one, did these participants understand the task or take it seriously? If not, then perhaps their data should be discarded. If there is no reason to believe that the data were corrupt in some way, then how does one go about estimating or determining what the indifference point should be when there are multiple switch points? Ultimately, three different data sets were constructed; one that included only participants who did not display multiple switchovers, a second that included participants who made one or less multiple switchovers, and a third that included participants who made two or less multiple switchovers. When a multiple switchover did occur at a particular delay, the indifference point for that delay was determined as the midpoint between the two switch points. The ultimate conclusion that could be drawn from the data from Weatherly, Der-

29

enne, and Chase (2008) was that temporal discounting varied systematically across the three data sets. Specifically, the rate of temporal discounting became increasingly steeper as participants who had made multiple switchovers were added to the data set. Given that these were potentially the individuals who did not understand the task or take it seriously, that their discounting rates were determined partially by estimations based on their multiple switchover responses, and the fact that pathological gambling is related to steeper rates of temporal discounting, these results were rather disconcerting. Measures of Temporal Discounting Several different methods exist to measure temporal discounting. One popular technique, and the one employed by Weatherly, Derenne, and Chase (2008), is to fit the indifference (i.e., switchover) points to a hyperbolic equation (Mazur, 1987): V = A / (1 + kD) (Equation 1) When using Equation 1, V represents subjective value of the delayed consequence, A represents the amount of the consequence, D represents the delay, and k is a free parameter that describes the rate that temporal discounting occurs. When Equation 1 is used, k is employed as the dependent measure for discounting with higher values of k being indicative of steeper rates of discounting. Phrased differently, previous research has shown that pathological gamblers display higher k values than non-pathological gamblers (e.g., Dixon et al., 2003). A second technique for measuring temporal discounting is to determine the area under the curve (AUC) created by the indifference points across delays and assuming that the commodity is at its full value when there is no delay (Myerson, Green, & Waru-

30


sawitharana, 2001). The AUC can be calculated using the following equation: (x2 – x1)[(y1 + y2)/2] (Equation 2) When using Equation 2, AUC can vary between 0.0 and 1.0, with the rate of temporal discounting being inversely related to AUC value. Phrased differently, if pathological gamblers discount more than nonpathological gamblers, then one would expect them to display lower AUC values than non-pathological gamblers. Although other formulas have been proposed to measure temporal discounting (e.g., Green, Fry, & Myerson, 1994; and see Killen, 2009, for a discussion), Equations 1 and 2 are commonly found in discounting studies. Each has their weaknesses. Equation 1 is, at best, an estimation of discounting and the processes that are involved in it. That is, the resulting dependent measure, k, is estimated given the responses the participant/subject provides, at which point the actual data are no longer considered. Studies that employ Equation 1 therefore also report how well it fit the data in terms of the variance for which it account (i.e., R2). Often these values are quite high (e.g., Smith & Hantula, 2008). However, sometimes they are not (e.g., Weatherly, Derenne, & Terrell, 2010; Weatherly, Terrell, & Derenne, 2010). Next, implicit in the use of Equation 1 is that temporal discounting is hyperbolic in nature. Although Equation 1 has adequately fit many data sets in the literature, the theoretical reasons for why discounting should be hyperbolic in nature have been elusive (see Killeen, 2009, for a thorough discussion). Furthermore, temporal discounting data are not always that “clean.” Some participants do not decrease the value of the commodity as it is delayed (non-discounters; e.g., see Beck & Triplett, 2009). Others might in fact display the inverse of discounting (i.e., expecting more of the immediately

available amount with increasing delays to the full amount). A typical reaction to such patterns of responding is to exclude them from data analyses and it is common to see 10 – 15% of a data set excluded for this reason (e.g., see Beck & Triplett, 2009). This practice is often done without much comment. An assumption is made that these individuals did not understand the task or questions. However, one could argue that these data are excluded because they do not fit with the researchers’ assumptions, which is troubling. Even if one could make a reasonable defense of this practice from a scientific basis, it is still troubling. If the relationship between pathological gambling and temporal discounting is a meaningful one, it seems odd that we should need to routinely exclude 1 out of every 7 - 10 participants in temporal discounting studies in our attempt to explain the 1 in every 50 individuals who suffer from the disorder. AUC values, on the other hand, directly represent the responses provided by the participants/subjects. It is also atheoretical in terms of the form temporal discounting should take. However, that is not necessarily a good thing. That is, it is potentially possible for the responses of two individuals to generate the same AUC value by displaying two distinctly different patterns of responding (e.g., one accepting increasingly less of the commodity as it is delayed and the other expecting increasingly more of the commodity as it is delayed; and see Smith & Hantula, 2008, for another example). Thus, one cannot determine by looking at an AUC value, as one can with a k value, the form of the participant’s/subject’s responses. It is also the case that, in typical studies of temporal discounting, AUC values will be highly correlated with participants’/subjects’ responses at long delays. Research on temporal discounting has historically found that individuals display steep rates of discounting across short delays and discounting rates

Jeffrey N. Weatherly flatten at longer delays (the delay effect; see Chapman, 1996, for a discussion). For this reason, studies of temporal discounting often have an overabundance of short delays and a few long delays. Because discounting is measured as a function of time, the long delays will constitute much of the overall AUC value whereas each of the short delays will potentially make up a lesser amount of the AUC value. In other words, if one uses Equation 2 and AUC as the dependent measure for temporal discounting, it is possible that the delay effect may get masked. Variations in Temporal Discounting Methodology and Interpretation As noted above, the meaningfulness of the relationship between temporal discounting and gambling has been driven by the finding that rates of discounting have been shown to differ as a function of gambling status (e.g., Dixon et al., 2003, 2006). These studies have found greater rates of temporal discounting in gamblers than in non gamblers. However, it is worth noting that the opposite finding has also been reported (Holt, Green, & Myerson, 2003). One issue that has not received much, if any, research attention is the commodity that the participants in these studies are asked to discount. The commodity in these studies, and most studies of temporal discounting in general, is a hypothetical amount of money. Two questions can be asked about this particular commodity. First, is discounting of this particular commodity indicative of an individual’s temporal discounting of all commodities? Second, if the answer to the first question is “no,” then is it the best commodity to use in such studies? The answer to the first question does indeed appear to be “no.” Recent research from our laboratory has asked just such a question. Weatherly, Terrell, and Derenne (2010) had 648 college students complete a temporal-discounting task that included five

31

commodities. There were two sets of commodities (two monetary values, cigarettes, a dating partner, & one’s own body image or two monetary values, retirement income, medical treatment, and federal education legislation). For both data sets, two outcomes were observed. First, significant differences in rates of discounting were observed across the five commodities (AUC was the dependent measure because Equation 1 provided a poor fit to the data). Second, a factor analysis of each data set resulted in a two-factor solution. Germane to the present topic, the monetary commodities loaded on to one of the factors while other commodities loaded on to a second, independent factor. Phrased differently, results from both data sets indicated that knowing how participants discounted money did not provide the information necessary to predict how they discounted all other commodities. Given that rates of discounting hypothetical monetary rewards are not universally predictive of an individual’s rate of discounting of all commodities, is the commodity of hypothetical monetary rewards the one we should be studying? Here the research literature is relatively silent (but see Yi, Mitchell, & Bickel, 2010, for a discussion). As noted above, the majority of studies on discounting have used this particular commodity. One could ask whether temporal discounting of hypothetical monetary rewards is similar to temporal discounting of real monetary rewards. Although it is impractical to use real monetary rewards of the size typically used in studies that employ hypothetical ones (e.g., $1,000) or use the same time delays (e.g., 10 years), research that has attempted to compare discounting of real and hypothetical monetary rewards have found similar rates of discounting between the two (e.g., Dixon, Mui, Green, & Myerson, in press; Madden, Begotka, Raiff, & Kastern, 2003). One might assume that because gamblers gamble money, that money

32


is the correct commodity to study. That is, however, an assumption. Future research is needed to determine if temporal discounting of other commodities might be just as strongly, if not more strongly, associated with gambling behavior as is that of hypothetical monetary amounts. A related issue (and one that is beyond the scope of the present discussion) is not whether money is the correct commodity to be studying, but rather whether temporal discounting is the correct type of discounting to be studying. That is, probability and temporal discounting are two potentially distinct phenomena (see Green & Myerson, 2004). Given that gambling involves risking something of value on a probabilistic outcome, the field might be better served to pursue the potential relationship between probability discounting and gambling rather than temporal discounting and gambling (see Petry & Madden, 2010, or Weatherly & Flannery, 2008, for a discussion). Even if the study of temporal discounting of hypothetical monetary rewards turns out to be the correct one in relationship to gambling behavior, the relationship between temporal discounting and gambling, as it stands today, is a correlational one. That is, studies that have shown a relationship between rates of temporal discounting and pathological gambling have done so in preexisting populations (e.g., pathological gamblers). Thus, it is not possible to tell whether changes in one’s gambling behavior led to changes in temporal discounting, whether changes in temporal discounting led to changes in one’s gambling behavior, or whether both phenomena are related to some third, yet unidentified, factor or process. Now for Something Slightly Different Dixon and Johnson (2007) proposed the GFA. The GFA is a paper-pencil measure intended to identify the contingency that is maintaining a person’s gambling behavior.

It was adapted from a similar measure that was designed to measure self-injurious behavior (Durand & Crimmins, 1988) and represents the first functional-assessment tool created for gambling behavior. It attempts to identify four maintaining consequences for gambling: tangible (i.e., money), social attention, sensory experience, and escape. There are 20 questions total, with five questions associated with each of the four consequences. The respondent can endorse each question from 0 (never) to 6 (always). Summing the scores of all responses gives one a total score on the GFA (maximum = 120). Summing the scores in each category is intended to identify the primary maintaining contingency (i.e, the consequence receiving the highest score; maximum = 30 for each consequence). Miller, Meier, Muehlenkamp, and Weatherly (2009) attempted to test the validity of the GFA by giving it to 949 undergraduate students. This sample was randomly divided into two groups, one on which an exploratory factor analysis was conducted and the second on which a confirmatory factor analysis was conducted. The results of both analyses were similar. Although the GFA was designed to identify four possible maintaining contingencies for gambling, both analyses identified only two factors. Factor loadings for the individual items on the GFA grouped in a logical fashion. Those items intended to measure tangible, social attention, and sensory experience consequences tended to load on one factor, which was labeled positive reinforcement. Those items intended to measure escape loaded on the second factor, which was labeled negative reinforcement. Thus, although the GFA was designed to identify four separate contingencies, these data suggested that it, in fact, measured only two. Further analysis of the data, however, revealed a potentially intriguing finding. If one looked at the respondents’ factor scores

Jeffrey N. Weatherly on the two factors, a distinct linear pattern was observed for both. As one’s total score on the GFA increased, one’s factor one score (i.e., positive reinforcement) increased accordingly. This result was not necessarily surprising given that the majority of the questions on the GFA were related to factor one. Thus, if one scored high on the GFA overall, one would expect to see high scores on factor one. However, the same result was not observed for factor two (negative reinforcement). For both males and females in both the exploratory and confirmatory data sets, as overall scores on the GFA increased, scores on factor two tended to be zero. However, there were a number of outliers. The intriguing finding was the placement of those outliers. Participants who scored high on factor two also tended to be the individuals who scored quite high on the GFA overall. In other words, those who gambled as an escape tended to score high on the measure as a whole. Our question was whether these individuals were potentially the problem or pathological gamblers in the data set? To test this possibility, Miller, Dixon, Parker, Kulland, and Weatherly (this issue) administered the GFA and the SOGS (Lesieur & Blume, 1987) to 204 people on the streets of Las Vegas and Wendover, Nevada and to 101 people in two sports bars in Rockford, Illinois. The SOGS is the most widely used screening measure for the potential presence of pathological gambling, with a score of 5 or more on the SOGS indicative of the potential presence of pathology. The question was whether scores in the escape category on the GFA would identify those individuals who scored 5 or more on the SOGS. Using an overall score of 8 or more in the escape category as the cutoff, the GFA correctly identified individuals scoring 5 or more on the SOGS in 20% of the cases in the Nevada sample and in over 50% of the cases in Illinois sample. Thus, although the GFA was designed to measure

33

the consequences that maintain gambling behavior, it also appears to do a decent job as a diagnostic tool. If one scores high in the escape category of the GFA, then it would be wise to screen the person for pathological gambling. What Does This Information Have to do With Temporal Discounting? The data from Miller et al. (2009, this issue) suggest that, at least for a fair number of potential pathological gamblers, the contingency maintaining their gambling behavior is negative reinforcement (i.e., escape). That connection should not be overly surprising given that gambling as an escape is an official symptom of pathological gambling (American Psychiatric Association, 2003). The connection between pathological gambling, gambling as an escape, and temporal discounting, however, may not be as clear. Research on temporal discounting has shown a finding that has come to be known as the magnitude effect (e.g., Chapman, 1996; Thaler, 1981). Specifically, the greater the size or value of the full commodity, the less participants/subjects tend to discount it when it is delayed. For example, you might be willing to accept $900 today rather than waiting one year for $1,000. However, you might be unwilling to accept $90,000 today and instead wait a year to get $100,000. Thus, although the rate of discounting in the former example is at least 10% over a year, when the magnitude of the commodity is increased (i.e., the latter example) the discounting rate is less than 10%. For this reason, rates of temporal discounting have also been used as a means for measuring the subjective value of a particular commodity (e.g., Weatherly, Derenne, & Terrell, 2010). If problem and pathological gamblers differ from their non-problem and nonpathological counterparts in the reason why

34


they are gambling, then finding differences in how they temporally discount money would be expected. That is, if someone is gambling for a reason other than winning money, then it would seem to be a reasonable assumption that winning money holds less subjective value for this person than it does for someone who is gambling for monetary gain. Likewise, if monetary gain holds less value for this individual than it does for another, one would expect a greater rate of temporal discounting for hypothetical monetary rewards for this individual than for another. Thus, is there a relationship between gambling and temporal discounting? The answer is likely “yes.” Is it a meaningful relationship? The answer to that question is less clear. For some individuals, it might indeed be a meaningful relationship. However, for others, the relationship may be the outcome of a third, independent factor or process. That is, pathological gamblers do not hold in high value (at least relative to non-pathological gamblers) the commodity that researchers are using in their studies of temporal discounting. PREDICTIONS The present idea would seem to be consistent with existing data. Clearly, however, its predictions need to be tested before it is accepted. Below three predictions are outlined that could potentially support or disconfirm the argument made in the present paper. First, gambling behavior should be related to one’s escape score on the GFA. As noted above, Weatherly, Marino, Ferraro, and Slagle (2008) demonstrated that participants’ gambling on a slot machine was predicted by their rate of temporal discounting. They did not specifically test, however, whether escape scores on the GFA were equally or more predictive. If rates of temporal discounting by pathological gamblers

are being lowered indirectly because they are gambling as an escape, then it would be reasonable to predict that rates and levels of gambling would be at least as, if not more highly, correlated with escape scores on the GFA than with rates of temporal discounting. Second, pathological gamblers will not always display greater rates of temporal discounting than non gamblers. The current argument is that most studies of temporal discounting have employed hypothetical monetary amounts as the commodity and this commodity might have a lowered value for pathological gamblers if they are indeed gambling as an escape. If this argument is correct, a temporal-discounting study that employs a commodity that potentially serves as an escape (e.g., winning a video game or a trip to a theme park, both of which could provide competing forms of escape) may find that pathological gamblers discount that commodity less than non gamblers because that commodity would hold a greater subjective value to them than for non gamblers. Third, for individuals whose pathological gambling is maintained by escape, therapies that involve finding alternative mechanisms to achieve that escape may prove successful in treating their gambling. However, if that is the case, you would not necessarily expect to eliminate the difference observed between that person’s temporal discounting of hypothetical monetary rewards relative to his/her non-pathological counterpart because finding an alternative escape contingency would not address/alter the subjective value of money for that person. CONCLUSION Is the relationship between temporal discounting and gambling a meaningful one? It may be. Certainly, there are many researchers out there, including myself (e.g., Weatherly & Dixon, 2007), who have argued that it is. However, the results are not

Jeffrey N. Weatherly universally supportive of the idea, our techniques for studying temporal discounting have not been perfected or extensively explored, and it remains to be determined whether we are even pursuing the correct type of discounting when it comes to studying gambling. Furthermore, I have attempted to outline a scenario in which the relationship between temporal discounting and gambling is related to a third factor or process. With all of the emphasis one can find on temporal discounting in the literature today, the field would be sage to give at least as much attention to the possibility that the relationship is perhaps less meaningful than once thought as it does to the possibility that the relationship is in fact a meaningful one. REFERENCES American Psychiatric Association. (2003). Diagnostic and Statistical Manual of Mental Disorders (4th ed., Text Revision). American Psychiatric Association: Washington, D. C. Baker, F., Johnson, M.W., & Bickel, W.K. (2003). Decision-making in state lotteries: Half now or all of it later? Psychonomic Bulletin & Review, 10, 965970. Beck, R.C., & Triplett, M.F. (2009). Testretest reliability of a group-administered paper-pencil measure of delay discounting. Experimental and Clinical Psychopharmacology, 17, 345-355. Chapman, G.B. (1996). Temporal discounting and utility for health and money. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 771-791. Dannewitz, H., & Weatherly, J.N. (2007). Investigating the illusion of control in mildly depressed and nondepressed individuals during video-poker play. The Journal of Psychology, 141, 307-319.

35

Dixon, M.R., Jacobs, E.A., Sanders, S. (2006). Contextual control of delay discounting by pathological gamblers. Journal of Applied Behavior Analysis, 39, 413-422. Dixon, M.R., & Johnson, T.E. (2007). The gambling functional assessment (GFA): An assessment device for identification of the maintaining variables of pathological gambling. Analysis of Gambling Behavior, 1, 44-49. Dixon, M.R., Marley, J., & Jacobs, E.A. (2003). Delay discounting by pathological gamblers. Journal of Applied Behavior Analysis, 36, 449-458. Dixon, M.R., Mui, N., Green, L., & Myerson, J. (in press). Delay discounting of hypothetical and real money: The effect of holding reinforcement rate constant. Journal of Applied Behavior Analysis. Durand, V.M., & Crimmins, D.B. (1988). Identifying the variables maintaining self-injurious behavior. Journal of Autism and Developmental Disorders, 18, 99-117. Fantino, E., & Stolarz-Fantino, S. (2008). Gambling: Sometimes unseemly; Not what it seems. Analysis of Gambling Behavior, 2, 61-68. Gillis, A., McDonald, J.D., & Weatherly, J.N. (2008). American Indians and nonIndians playing a slot-machine simulation: Effects of sensation seeking and payback percentage. American Indian and Alaska Native Mental Health Research: The Journal of the Nation Center, 14, 59-74. Green, L., Fry, A.F., & Myerson, J. (1994). Discounting of delayed rewards: A lifespan comparison. Psychological Science, 5, 33-36. Green, L., & Myerson, J. (2004). A discounting framework for choice with delayed and probabilistic rewards. Psychological Bulletin, 130, 769-792.

36


Holt, D.D., Green, L., & Myerson, J. (2003). Is discounting impulsive? Evidence from temporal and probability discounting in gambling and non-gambling college students. Behavioural Processes, 64, 355-367. Killeen, P.R. (2009). An additive-utility model of delay discounting. Psychological Review, 116, 602-619. Lesieur, H.R., & Blume, S.B. (1987). The South Oaks Gambling Screen (SOGS): a new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144, 1184-1188. Madden, G.J., Begotka, A.M., Raiff, B.R., & Kastern, L.L. (2003). Delay discounting real and hypothetical rewards. Experimental and Clinical Psychopharmacology, 11, 139-145. Mazur, J.E. (1987). An adjusting procedure for studying delayed reinforcement. In M.L. Commons, J.E. Mazur, J.A. Nevin, & H. Rachlin (Eds.), Quantitative Analyses of Behavior: Vol. 5. The Effect of Delay and Intervening Events on Reinforcement Value (p. 55-73. Hillsdale, NJ: Erlbaum. Miller, J.C., Dixon, M.R., Parker, A., Kulland, A.M., & Weatherly, J.N. (in press) Concurrent validity of the gambling function assessment (GFA): Correlations with the South Oaks Gambling Screen (SOGS) and indicators of diagnostic efficiency. Analysis of Gambling Behavior, 4, 61-75. Miller, J.C., Meier, E., Muehlenkamp, J., & Weatherly, J.N. (2009). Testing the validity of Dixon & Johnson’s (2007) gambling functional assessment. Behavior Modification, 33, 156-174. Myerson, J., Green, L., & Warusawitharana, M. (2001). Area under the curve as a measure of discounting. Journal of the Experimental Analysis of Behavior, 76, 235-243.

Petry, N.M. (2005). Pathological Gambling: Etiology, Comorbidity, and Treatment. Washington, D.C.: American Psychological Association. Petry, N.M., & Madden, G.J. (2010). Discounting and pathological gambling. In G.J. Madden and W.K. Bickel (Eds.) Impulsivity: The Behavioral and Neurological Science of Discounting (pp. 273294). Washington, D.C.: American Psychological Association. Rachlin, H. (1997). Four teleological theories of addiction. Psychonomic Bulletin & Review, 4, 462-473. Smith, C.L., & Hantula, D.A. (2008). Methodological considerations in the study of delay discounting in intertemporal choice: A comparison of tasks and modes. Behavior Research Methods, 40, 940-953. Thaler, R.H. (1981). Some empirical evidence on dynamic inconsistency. Economic Letters, 8, 201-207. Weatherly, J.N., Derenne, A., & Chase, S. (2008). Do the risk factors for pathological gambling predict temporal discounting? Analysis of Gambling Behavior, 2, 25-34. Weatherly, J.N., Derenne, A., & Terrell, H.K. (2010). College students discount money “won” more than money “owed.” The Psychological Record, 60, 463-472. Weatherly, J.N., & Dixon, M.R. (2007). Toward an integrative behavioral model of gambling. Analysis of Gambling Behavior, 1, 4-18. Weatherly, J.N., & Flannery, K.A. (2008). Facing the challenge: The behavior analysis of gambling. Behavior Analyst Today, 9, 130-142. Weatherly, J.N., Marino, J.M., Ferraro, F.R., & Slagle, B. (2008). Temporal discounting predicts how people gamble on a slot machine. Analysis of Gambling Behavior, 2, 135-141.

Jeffrey N. Weatherly Weatherly, J.N., Terrell, H.K., & Derenne, A. (2010). Delay discounting of different commodities. The Journal of General Psychology, 137, 273-286. Yi, R., Mitchell, S.H., & Bickel, W.K. (2010). Delay discounting and substance abuse-dependence. In G.J. Madden and W.K. Bickel (eds.) Impulsivity: The Behavioral and Neurological Science of Discounting (p. 191-211). Washington, D.C.: American Psychological Association. Action Editor: Mark R. Dixon

37


2010, 4, 38-53


The Impact of Derived Relational Responding on Gambling Behavior Simon Dymond1 & Bryan Roche2 1

2

Swansea University National University of Ireland, Maynooth

The present article describes existing research on the impact of derived relational responding on gambling behavior. First, it is argued that a greater understanding of the role of verbal behavior in gambling behavior is made possible by research findings and theoretical advances in research on derived relational responding generally, and the transformation of stimulus functions in particular. Second, the findings of several recent studies are described in order to describe the key features of this contemporary approach for verbal events. Finally, implications for the verbally based treatment of disordered gambling are outlined. Keywords: gambling behavior, verbal behavior, derived relational responding, transformation of functions, verbally based interventions.

-----------------------------A key challenge for contemporary efforts to design effective treatment services for the cluster of repertoires often referred to as “disordered gambling” (Petry, 2009) is the need for them to be empirically validated through initial basic research before eventual, applied intervention. Empirical research on gambling behavior is growing (Weatherly & Dixon, 2007) and several treatment approaches have been devised that are based on empirical findings (Petry, 2009). However, considerably more research effort is now needed if behavior scientists are to be at the forefront of the development of effective technologies for altering disordered gambling (Fantino, 2008; Mace & Critchfield, 2010). “Pure basic” and “pure applied” research (Mace & Critchfield, 2010) have important roles to play in furthering our understanding of the basic behavioral processes involved in gambling and in the development of effective interventions for disordered gambling. A similar effort is needed to facilitate “translational research” on gambling, which is usefully defined as “an essential complement to “pure basic” behavioral research because it explicitly considers the generality and everyday relevance of fundamental behavior principles” (Mace & Critchfield, 2010, p. 296).

Gambling is an activity enjoyed by many, yet is one that is increasingly becoming problematic for a growing proportion of the world’s population. Evidence from several countries now shows that the increased availability of opportunities to gamble, in a diverse and growing range of formats, is often followed by increases in the prevalence rates of problem and pathological gambling, and in the number of people seeking treatment (Petry, 2005; Wardle et al., 2007). For instance, in the United States, a 1974 telephone survey found that 0.7% of a national sample was classified as pathological gamblers and 2.3% as problem gamblers (Kallick, Suits, Dielman, & Hybels, 1976). Some decades later, the lifetime prevalence rates of pathological and problem gambling were 1.4% and 5.1%, respectively (Volberg, 1994, 1996). More recently, the lifetime prevalence rate of pathological gambling has been estimated to range between 1 and 3% (Petry, 2005). Address correspondence to: Simon Dymond, Department of Psychology. Swansea University, Singleton Park, Swansea SA2 8PP, United Kingdom s.o.dymond@swansea.ac.uk

38

Simon Dymond and Bryan Roche To continue to inform effective treatment, basic, applied and translational research should seek to address how, for instance, “verbal rules can augment the actual contingencies of games of chance to further promote future gambling” or “completely overcome those contingencies altogether” (Weatherly & Dixon, 2007, p. 13). A key challenge, therefore, for any contemporary account of gambling behavior is to identify the role of verbal behavior. After all, “gambling is a behavior that is engaged in by verbal humans…The verbal human is exposed to a variety of contingencies and verbal stimuli when engaging in a gamble ... assuming such verbal stimuli do not exist, or arranging artificial laboratory conditions to eliminate verbal stimuli from the environment seems counterproductive to understanding why a gambler engages in the behavior he/she does” (Dixon & Delaney, 2006, pp.173174). This article will outline an approach to understanding the role of verbal behavior in gambling behavior based on research conducted on derived relational responding and the transformation of stimulus functions. It will describe the features of this approach for explaining verbal events and suggest possible verbally based interventions based on the approach. Before this, it is necessary to consider existing research aimed at developing an experimental analysis of the role of verbal behavior in gambling behavior. Towards a Contemporary Analysis of the Impact of Derived Relational Responding on Gambling Behavior Nonarbitrary Relational Responding In research on slot-machine gambling, procedures based on nonarbitrary relational responding have been shown to systematically alter gamblers’ preferences over and above that predicted by the underlying reinforcement schedule. For instance, Hoon, Dymond, Jackson, and Dixon (2008) showed that recreational gamblers’ choices of one of two simultaneously presented slot machines of equal

39

payout probability (0.5) and reinforcement magnitude could be altered when a structural characteristic of one of the machines, such as background color, was established as a contextual cue for “greater than” relations. The identical reinforcement schedules operating with either slot machine should have resulted in relatively equal levels of response allocation (i.e., a “matching” of responding with relative reinforcement rates; Baum, 1979), and this is indeed what participants tended to do during a pretest phase. Following the relational intervention, however, the same slot machines were re-presented and participants’ allocated a significantly greater proportion of their responses to the slot machine associated with the “greater than” cue, despite the identical reinforcement schedules. In the Hoon et al. (2008) and related studies (Hoon, Dymond, Jackson, & Dixon, 2007; Johnson & Dixon, 2008; Nastally, Dixon, & Jackson, 2010; Zlomke & Dixon, 2006), a conditional discrimination procedure was used to train the two colors (blue and red) as contextual cues for more than and less than nonarbitrary relational responding, respectively. With such a procedure, correct selections are conditional on the presence of a particular stimulus. For instance, participants were presented with two comparison stimuli of differing physical quantities, such as three apples and six apples, and reinforcement delivered for selecting the three apples in the presence of the contextual cue for less than (i.e., background color of blue), and for selecting the six apples in the presence of the contextual cue for more than (i.e., background color of red). On reaching criterion, participants were presented with novel stimulus sets in the absence of feedback to test whether the cues were functioning as contextual cues for “more than” and “less than”, respectively. The findings of several studies on this topic have now consistently shown that is possible to alter preferences by juxtaposing background, situational characteristics such as colors

40

DERIVED RELATIONAL RESPONDING

with concurrently available slot machines of identical reinforcement probability, density and magnitude (Hoon et al., 2007; Nastally et al., 2010; Zlomke & Dixon, 2006). Recently, Johnson and Dixon (2009) extended this approach to show how an experimental history can lead to gambling behavior that appears to indicate "the presence of erroneous beliefs” (Delfabbro, Lambos, King, & Puglies, 2009) and override programmed reinforcement contingencies. Children, aged 7 to 10 years, played a simulated board game in which they could choose, on each turn, either of two concurrently presented dice that differed only by color (one red, one blue). Each die was programmed to roll a random number between 1 and 6, and each child’s preselected game piece then moved the corresponding number of spaces along the on-screen racetrack. Next, in a relational training and testing phase, children were taught to select stimuli of differing physical quantities in the presence of a contextual cue for more than (red background color) and a contextual cue for less than (blue background color), before being tested with novel stimulus sets. Then the children played the simulated board game again. Although the contingencies governing dice rolling were unchanged, all but one child showed increased use of the die whose color served as the more-than contextual cue (red). In the language of stimulus relations, these results show how, through relational experience, contingency-irrelevant features of a game of chance can come under nonarbitrary contextual control by formal features (such as dice colors). Procedures such as these (e.g., Dymond & Barnes, 1995; Whelan, BarnesHolmes, & Dymond, 2006) are considered demonstrations of nonarbitrary relational responding because the relational response of picking the smaller or larger comparison is controlled by formal, physical features of the particular stimuli involved (Stewart & McAlwee, 2009). Nonarbitrary

relational responding is entirely bound by the formal properties of the related events and is said to occur when for instance, in the absence of reinforcement, an organism selects the larger of two stimuli based on a history with multiple stimulus sets and contexts. Nonhumans are readily capable of acquiring nonarbitrary relational responding, and a nonhuman model of this generalized nonarbitrary performance is possible, and may even be desirable (Madden, Ewan, & Lagorio, 2007). However, burgeoning empirical evidence now shows that verbally able humans can also learn to respond relationally to objects and events when the relation is defined by the physical properties of the objects but rather by additional contextual cues (Hayes, Barnes-Holmes, & Roche, 2001). For example, consider a young child who learns that ‘‘X is taller than Y.’’ Subsequently, he or she may when asked, ‘‘which is shorter?’’ respond ‘‘Y,’’ without any further training. According to relational frame theory (RFT), this response, which is controlled solely by the contextual cues ‘‘taller’’ and ‘‘shorter’’ and not by any physical relations, is arbitrarily applicable because it can be applied to any stimuli regardless of their physical properties. An unequivocal demonstration of a nonhuman model of arbitrarily applicable (derived) relational responding has yet to emerge. Therefore, a degree of caution is needed when interpreting the findings showing contextual control of altered preferences in slot-machine simulations and pregambling tasks because the findings do not demonstrate derived, verbal control. The combined performances of the gamblers, recreational gamblers, nongamblers, and children in these studies were not derived, in the technical sense that the color contextual cues did not participate in derived stimulus relations. What, then, are derived relations? Derived Relational Responding Research on derived relational responding may provide a behavioral model

Simon Dymond and Bryan Roche of how verbal processes might interact with, and overcome, the directly experienced contingencies of games of chance. Since the early 1970’s, a vast literature has amassed on derived relational responding showing that when verbally able humans are taught a series of interconnected conditional discriminations involving physically dissimilar (arbitrary) stimuli, the stimuli involved often become related to each other in ways that are not explicitly trained. To illustrate, if choosing Stimulus B in the presence of Stimulus A is taught (i.e., AB), and choosing Stimulus C in the presence of Stimulus A (i.e., A-C) is also taught, it is highly likely that relations will emerge between B and A, C and A (called symmetry), B and C, and C and B (called combined symmetry and transitivity, or equivalence), in the absence of any further training. When these relations have been observed, a stimulus equivalence relation is said to have formed among the relata (Fields, Adams, Verhave, & Newman, 1990; Sidman, 1994). These untrained, but nonetheless predictable, derived stimulus relations have been the focus of concerted research attention precisely because they are not readily explained by traditional behavior-analytic principles of discrimination and stimulus generalization. Neither B nor C, for instance, have a history of differential reinforcement with regard to each other (a defining feature of discrimination learning), therefore, neither should control selection of the other. Also, the derived stimulus relations that emerge cannot be accounted for on the basis of generalization because the stimuli are all physically dissimilar and cannot be explained via simpler conditioning processes. To illustrate, it is likely that derived stimulus relations comprised of spoken words, visual stimuli and self-statements may be involved in contexts where gambling occurs. For example, the word “casino” participates in a derived relation with actual casinos. Moreover, during a visit to a casino, a gambler is likely to emit the self-discrimination that he or she “feels

41

lucky”. We may consider the spoken word, “casino” as Stimulus A, the actual casino as Stimulus B (i.e., A-B), and the selfstatement, “I feel lucky” as Stimulus C (i.e., A-C). With a relational history such as this, a gambler is likely to utter the selfstatement, “I feel lucky”, when visiting a casino (B-C) and may also, when uttering the statement in other, non-gambling contexts “see” actual casinos (C-B) “in the absence of the thing seen” (Skinner, 1974, p. 91). Derived relations such as these are likely to highly diffused, flexibly adapting and assimilating to novel environments. This means that a gambler need only think of a casino in order to spontaneously derive relations involving self-discriminations of “feeling lucky” and others. Research on derived relational responding has generated scores of basic research studies, applied extensions, and conceptual analyses. The chief reason for the burgeoning research and theoretical advances in this area is that it now appears possible to explain the emergence of untrained stimulus relations in the absence of a direct history of reinforcement. The behavioral process by which this occurs, and which is still hotly debated (Hayes et al., 2001; Hayes & Barnes-Holmes, 2004; Sidman, 1994), has the capacity to alter virtually all other operant behavior, and account for the complex behavior emitted by verbally able humans, be they in the classroom, therapy room, or casino. Thus, what is at stake in research on derived relational responding is the opportunity to develop a contemporary, functional analytic account of verbal behavior itself. Transformation of Stimulus Functions: A Functional Account of Verbal Events A central feature of derived relational responding - the transformation of stimulus functions - makes it directly relevant to an empirical analysis of gambling behavior, and with it, a contemporary approach to verbal behavior (Barnes-Holmes, Barnes-Holmes & Cullinan, 2001). Transformation of stimulus functions is said to oc-

42


cur when the psychological functions of stimuli in a derived relation are transformed based on the nature of the relation and the psychological functions of the other member(s) of that relation. For example, if A is related to B and B is related to C, and C is paired with a winning slot machine outcome that evokes arousal and approach functions, then presentations of A will also likely evoke similar conditioned arousal and approach functions by virtue of the derived C-A equivalence relation (for a review, see Dymond & Rehfeldt, 2000). In the context of a gambling, consider a gambler who plays Blackjack for the first time and enjoys it. Later, while on vacation in Germany, our gambler might learn that, in Germany, Blackjack is called “Seventeen plus four”. If she subsequently hears “Seventeen plus four”, she may show signs of approach and “see” a Blackjack table. In this way, the functions of a novel stimulus are transformed based on the functions acquired with another by virtue of the derived relation that obtains between the two. According to prevailing behavioranalytic accounts, a verbal stimulus acquires its functions based, at least in part, on participation in a derived relation or relational frame (Barnes-Holmes, Hayes, Dymond, & O'Hora, 2001; Dymond, 2008; Dymond & Rehfeldt, 2000; Dymond & Whelan, 2007; Hayes, Fox, Gifford, Wilson, Barnes-Holmes, & Healy, 2001). Approached in this way, derived relational responding and the transformation of stimulus functions represent the key behavioral processes involved in the initiation and maintenance of gambling behavior. Such processes may interact with or override programmed contingencies of reinforcement. In effect, the transformation of stimulus functions may account for the insensitivity to direct contingency control often observed in disordered gambling and may partly explain the emergence of gambling behavior that arises in the absence of a direct learning history.

Dixon, Nastally, Jackson, and Habib (2009) showed that derived equivalence relations could alter recreational gamblers’ ratings of slot machine outcomes. During a pretest phase, Dixon et al. presented participants with three graphic displays of slot machine outcomes depicting a win (i.e., three matching symbols on a payout line), a near miss (i.e., two matching symbols and one different symbol on a payout line) and a loss (i.e., three different symbols on a payout line; C1, C2 and C3, respectively), and asked them to rate how close the image was to a win. Next, participants were trained in the formation of A-B and A-C conditional discriminations, before being tested once for symmetry (B-A and C-A) and equivalence relations (B-C and C-B). The A1, A2, and A3 stimuli consisted of three abstract images, and the B1, B2, and B3 stimuli consisted of the text “win”, “loss” and “almost”, respectively. Finally, in the post-test phase participants were re-presented with the C1, C2 and C3 stimuli. Dixon et al. predicted that if derived equivalence relations were formed between the B-C and C-B stimuli, then the B3 stimulus (“almost”) should acquire some of the functions of the C3 loss image and the B2 stimulus (“loss”) should acquire some of the functions of the C2 nearmiss image (the B1 stimulus, “win”, should remain unchanged as it was related via equivalence to the C1 win image, and vice versa). Results indicated that, relative to pretest levels, the majority of participants rated the C3 “loss” stimulus as closer to a win than the C2 “near miss” stimulus. Moreover, when the requisite derived relations were not formed, the predicted performances failed to emerge. These findings demonstrate how intra-experimentally established derived verbal relations may influence recreational gamblers’ ratings of slot machine outcomes in ways that may override the contingency-relevant functions of gambling stimuli. In effect, the gamblers behaved as if the three different symbols on the payout line were closer to a win than the “almost winning” near miss

Simon Dymond and Bryan Roche display of two matching symbols (Habib & Dixon, 2010; Reid, 1986). Modeling Pregambling Behavior: Derived Transformation of Children’s Pregambling Game Playing Functions in Accordance with Equivalence Relations The findings of Dixon et al. (2009) are promising and show how an approach based on derived relational responding may be used to systematically alter recreational gamblers’ preferences for gambling relevant stimuli. Dymond, Bateman, and Dixon (2010) sought to further investigate the impact of derived, verbal relations on gambling behavior by examining whether or not a key defining feature of derived relational responding – the transformation of stimulus functions - occurs during the same type of analogue gambling tasks Johnson and Dixon (2009) used with young children. Transformation of stimulus functions is said to occur when the psychological functions of stimuli in a derived relation are transformed based on the nature of the relation and the psychological functions of the other member(s) of that relation. For example, if A is related to B and B is related to C, and C is paired with a winning slot machine outcome that evokes arousal and approach functions, then presentations of A will also likely evoke similar conditioned arousal and approach functions by virtue of the derived C-A equivalence relation (for a review, see Dymond & Rehfeldt, 2000). The transformation of stimulus functions may partly explain the emergence of gambling behavior, such as an increased preference for a novel slot machine, that arises in the absence of a direct learning history and may, ostensibly, appear to indicate control over behavior by “erroneous beliefs” (Delfabbro et al., 2009; Sevigny & Ladouceur, 2003). Dymond et al. (2010), therefore, sought to extend Johnson and Dixon’s (2009) findings by showing that children’s pregambling responses may be altered via derived relational responding and the transformation of functions. First,

43

12 children aged between 7 and 10 years old, were trained (A1-B1, A1-C1, A2-B2, & A2-C2) and tested for the formation of two, three-member equivalence relations (B1-C1, B2-C2, C1-B1, & C2-B2). Four participants failed to achieve mastery criterion after four equivalence test exposures and were removed from the study. The remaining 8 participants then proceeded to play an adapted version of Johnson and Dixon’s simulated board game. The purpose of this phase was to attach high- and low-roll functions to two dice labeled with members of the derived relations. Specifically, the die labeled B1 was programmed to always roll high numbers (4, 5 or 6), and the die labeled B2 was programmed to always roll low numbers (1, 2, or 3). This phase established a baseline of responding between the two concurrently available dice labeled B1 (“more than”) and B2 (“less than”). To complete this phase, participants were required to select the B1 (“more than”) die on at least 80% of trials. Next, the test for transformation of stimulus functions was presented with presentations of dice labeled C1 and C2. As this was a test, participants’ selections of each die were not followed by differential feedback (i.e., each trial ended with the participant’s game piece completing the racetrack). Dymond et al. (2010) found that all except one of the participants who passed the equivalence relations test selected the C1 die more often than the C2 die, despite the absence of differential feedback following each dice roll, and all except three gave C1 higher liking ratings than C2. The increased response allocation and liking ratings for C1 relative to C2 suggests that the directly trained functions of B1 and B2 were transformed in accordance with the derived equivalence relations between the B and C stimuli. These findings show, for the first time, that gambling relevant response functions may transform in accordance with derived equivalence relations. Such a demonstration greatly extends the potential utility of behavior-analytic mod-

44


els of gambling in explaining the emergence and maintenance of gambling behavior in the absence of direct reinforcement and contribute towards potential verbally based interventions to overcome disordered gambling (Petry, 2009). Undertaking this analysis with young children’s pregambling game playing is important in developing an empirical, developmental account of verbal mediation effects in terms of transformation of functions and how it may lead to disordered gambling. Modeling Gambling Behavior: Derived Transformation of Functions in Accordance with Equivalence and Nonequivalence Relations These preliminary, demonstration studies go some way towards understanding the impact of derived relational responding on gambling behavior, and lend support to the view that gambling may be considered a verbal event (Dymond, 2008). However, behavior-analytic research on gambling behavior needs to do more than demonstrate, in the basic lab, a putative role for verbal relations; it must also undertake translational analyses (Mace & Critchfield, 2010) of the behavior evoked by the situations and stimuli confronted by actual gamblers. Consider, for instance, someone entering a casino. Choice of which slot machine, or of which form of gambling (roulette, craps, poker, etc.) to play is likely to be influenced by derived, verbal functions in the form of rules and selfrules like “loosest slots in the house” and “I feel lucky”, along with formal features of the context (e.g., lights, colors, sounds, and names of slots machines). In gambling, stimulus functions such as these likely participate in multiple, contextually controlled derived relations. Gamblers’ relational histories with various stimulus functions may come to exert control over choices and override the effects of programmed contingencies. Dymond, Mills, Griffiths, Cox, and Crocker (submitted) have sought to develop a preliminary translational model of

the choices gamblers make to play differing slot machines by testing for derived transformation under various conditions of lean reinforcement and non-reinforcement. Three experiments were conducted, each involving the formation of derived equivalence relations and the training of highand low-probability payout functions for two members of the derived relations but which differed in terms of how subsequent transformation was tested. In Experiment 1, thirty participants (3 of whom scored 1 on the South Oaks Gambling Screen (SOGS; Leiseur & Blume, 1987), first formed two, three-member equivalence relations (A1-B1-C1 & A2-B2-C2). Next, they were given successive simulated slot machine exposure training with two machines labeled B1 and B2, respectively. Slot machine B1 was programmed to payout (i.e., three matching symbols on the payout line, and the addition of 1 credit to an accumulating total display) on 5 out of 25 trials (i.e., 0.2 probability), while slot machine B2 was programmed to payout on 20 out of 25 trials (i.e., 0.8 probability). Following a ratings phase in which participants rated the likelihood of winning on the two slot machines, they proceeded to the test for transformation of stimulus functions. During this phase, participants were given concurrent presentations of slot machines labeled C1 and C2, and were required to select which one they wished to play. Participants did not directly experience playing the slot machines; instead, their forced choices were recorded. The findings of Experiment 1 showed that participants rated slot machine B1 as significantly closer to a win than slot machine B2, and during transformation of functions testing they chose slot machine C1 significantly more often that slot machine C2. This derived transformation occurred under forced choice conditions in which participants were not exposed to the underlying reinforcement schedules operating with the slot machines. Such a demonstration provides evidence of ‘proof of concept’ and illustrates the impact of de-

Simon Dymond and Bryan Roche rived relational responding on gambling behavior, but may be considered as lacking ecological validity. For instance, it is rare for slot machine gamblers to be exposed to choices of different machines, but not to actually experience outcomes on those machines. A more realistic model of slotmachine gambling would be to expose participants to a test under conditions of extinction (i.e., in the absence of feedback following each reel spin), since this more accurately reflects how random ratio schedules are arranged on slot machines. Experiment 2 addressed this issue. In Experiment 2, the same procedural format as Experiment 1 was adopted except for the following important difference. During the test for transformation of stimulus functions, participants could actually play the slot machines labeled C1 and C2. However, when a slot machine was selected it spun as before but each reel successively stopped on a blank display. This manipulation ensured that participants were not provided with any feedback on the outcomes of the slot machines trials (i.e., extinction). Twenty-eight out of thirty participants (4 who scored 1, 3 who scored 3, and 1 who scored 6 on the SOGS) passed tests to demonstrate the formation of equivalence relations. The findings from the test for transformation of stimulus functions indicated that participants chose slot machine C1 significantly more often that slot machine C2, and rated slot machine C1 as significantly closer to a win than slot machine C2, despite never winning on either slot machine. The findings of Experiment 2 demonstrate the derived persistence of slot-machine gambling under conditions of extinction, and lend further support to the behavioral model of gambling as derived relational responding. The findings indicate that high probability winning outcome functions established during direct exposure training may come to participate in derived relations and influence the persistence of slot machine selections under conditions of extinction.

45

In seeking to further develop a behavioral model of slot-machine gambling initiation and maintenance, however, it is necessary to test derived transformation under conditions that approximate realworld reinforcement schedules. Previous research on concurrent slot machine schedules has shown that participants’ responding generally conforms to the predictions of the matching law (MacLin, Dixon, Daugherty, & Small, 2007): that is, relative response rates match relative reinforcement rates. Experiment 3 addressed the role of matched reinforcement schedules during transformation of functions testing. During this phase, participants could actually play the slot machines labeled C1 and C2. However, when a slot machine was selected it spun as before and displayed the outcome of each trial. Each slot machine was programmed to payout at a probability of 0.2 (i.e., on 4 of 20 trials). Forty out of forty-three participants (11 who scored 1 and 2 who scored 2 on the SOGS) passed tests to demonstrate the formation of equivalence relations. The findings from the test for transformation of stimulus functions indicated that participants chose slot machine C1 significantly more often that slot machine C2, and rated slot machine C1 as significantly closer to a win than slot machine C2, despite the matched probabilities. The final studies to be described here sought to investigate how slot machine response functions could come to participate in, and transform, multiple, contextually controlled derived comparative relations of more than and less than (Hoon & Dymond, 2010). Following nonarbitrary relational training and testing designed to establish contextual functions of more than and less than for two arbitrary cues, participants were tested for the formation of a contextually controlled relational network, E>D>C>B>A. All mutual (e.g., DE) tasks were tested. Next, a slot machine labeled with the middle-ranking stimulus, ‘C’, was presented that had a low payout probabil-

46


ity (0.2). During the test for transformation of functions, pairs of slot machines labeled with members of the relational network were presented. It was predicted that participants would choose the higher-ranking stimulus of each pair in accordance with the relational network. Findings supported this prediction, with participants selecting the slot machine labeled ‘A’ least often and the slot machine labeled ‘E’ most often. Responding during the test for transformation of functions showed a graded trend in accordance with the derived comparative relations, in the absence of any differential feedback. In a further experiment, Hoon and Dymond (2010) altered the direction of the trained relational network, A>B>C>D>E. Again, a low payout probability (0.2) function was attached to the slot machine labeled ‘C’. As predicted, participants’ choices on the tests for transformation of functions were in accordance with the comparative ranking of the relational network. That is, participants selected the slot machine labeled ‘E’ least often and the slot machine labeled ‘A’ most often. Selections occurred in the absence of differential feedback, and without actually playing the slot machines, yet a consistent pattern of transformation of functions was observed in accordance with the ranking of the relational network. These findings may help to understand those occasions when a gambler selects to play on a slot machine that he or she “likes more” than another, despite never had won before on the machine. The findings of Dymond et al. (2010) and Hoon and Dymond (2010) illustrate how gambling relevant stimulus functions may come to participate in, and transform, contextually controlled derived relations. Taken together, these promising findings attest to the fact that because not all objects and events in a derived relation need to be directly experienced, the potential for gambling to be controlled by increasingly complex and ever more remote contingencies is both tremendous and far-reaching.

These and other studies go some way towards providing an experimental analysis of verbal mediation effects seen in cognitive approaches to gambling behavior. Moreover, the findings demonstrate the utility of the experimental and translational analysis of gambling behavior. The observation that gamblers and nongamblers’ behavior came under the control of comparable contingencies indicates that common processes may underlie the transition from orderly, non-problem gambling to disordered or pathological gambling. Implications of a Contemporary Approach to Verbal Events for the Treatment of Disordered Gambling Derived relational responding and the transformation of stimulus functions are likely processes through which gambling behavior that at first appears to be insensitive to underlying reinforcement contingencies may, subsequently, come to be firmly established in a gambler’s repertoire. Indeed, a respectable body of experimental data now exists to support the idea that much complex human behavior, including features of gambling behavior (see above), can emerge from verbal contingencies alone. However, many research challenges lie ahead if the import of such research is to be fully realized. In particular, it is not known how immediate nonverbal and more indirect verbal contingencies controlling gambling behavior interact with each other. It may well be the case that gambling behaviors arising from verbal contingencies quickly come under the direct control of the immediate consequences of gambling (i.e., winning or losing). In other words, the direct reinforcing and punishing monetary consequences of gambling may over-ride the effects of derived relational responding contingencies. Thus, while verbal processes may initiate gambling behavior or a particular risky decision on a given occasion, such processes may not be solely responsible for the maintenance of such behavior.

Simon Dymond and Bryan Roche Alternatively, however, it may be that the ability of derived relational processes to transform the functions of reinforcing consequences (e.g., “losing means that I must be about to win soon”, otherwise known as the gambler’s fallacy) may be sufficient to maintain high rate gambling behavior even in the presence of direct punishing consequences (i.e., losing). In effect, the degree to which specific types of verbal contingencies might render gambling behavior insensitive to nonverbal contingencies is not well understood, and likely varies with the strength of each contingency, the verbal fluency of the individual and salience of behavioral consequences (i.e., the response cost, or level of gambling loss), and other factors. A complete analysis of gambling behavior that is applicable to real-world gambling will need to address this complex empirical issue. ‘Third Wave’ Behavior Therapies Identifying core processes at work in the emergence and maintenance of gambling behavior also bears on the issue of treatment. Indeed, the idea that many instances of problem behavior emerge from verbal contingencies has led in recent years to the development of a range of modern behavioral therapies such as Acceptance and Commitment Therapy (ACT; Hayes, Strosahl, & Wilson, 1999), Functional Analytic Psychotherapy (Kohlenberg & Tsai, 1991), and Dialectical Behavior Therapy (Linehan, 1993). These approaches to behavior therapy employ experiential exercises and rich talk-therapy metaphors of a type more commonly associated with the humanistic tradition (McCurry & Hayes, 1992). However, these approaches (known collectively as “third wave” approaches; see Vilardaga, Hayes, Levin & Muto, 2009) have in common an emphasis on verbal behavior, and in the case of ACT, a particular emphasis on derived relational responding processes. We will briefly consider here how insights gained by ACT researchers

47

may prove helpful in harnessing the concept of derived stimulus relations and the transformation of stimulus functions in the treatment of problematic gambling behavior. ACT is a modern behavior therapy strongly associated with relational frame theory (Hayes et al., 2001) and the mindfulness movement (Baer, 2005). This approach to therapy assumes that human suffering is ubiquitous largely due to our ability to derive bidirectional relations between words and other stimuli, and for the functions of stimuli to transform in accordance with those relations (e.g., for the thought “I feel lucky today” to transform the functions of multiple events that occur today such that everything from gambling to meeting an old friend to catching a bus, all produce the positive feelings of “luck”). According to ACT, the problem for many clients displaying addictions and other forms of psychopathology is that they employ coping methods that are inherently self-defeating. More specifically, a typical adult in our culture will deal with unpleasant private thoughts and feelings by using methods of distraction or avoidance, which ACT therapists refer to as experiential avoidance. For many clients, these methods may be effective in cases where the experiences being avoided are not sufficiently aversive to interfere with the individual’s general social functioning. However, in some cases, the very effort to avoid uncomfortable private experiences can itself be self-destructive (e.g., excessive drinking or gambling). Alternatively, the avoidance method may fail due to the inescapable fact that avoidance efforts become related to the very experiences being avoided. Thus, avoidance itself becomes one of the experiences that produces the experiences to be avoided (Blackledge & Hayes, 2001). Put simply, the more one attempts to avoid an experience the more that effort produces the very experience being avoided (see Hayes, Luoma, Bond, Masuda & Lillis, 2006). In fact, several studies have now produced empirical evi-

48


dence of this effect (e.g., Clark, Ball, & Pape, 1991; Feldner, Zvolensky, Eifert, & Spira, 2003). ACT refers to “psychological flexibility” as the preferable alternative to psychological inflexibility or experiential avoidance. It uses nonlinear language, paradox and metaphor to undermine the literality of language, which in technical terms we may now describe as derived relational responding and the transformation of stimulus functions. Hayes et al., (1999) provided extensive examples of metaphors and experiential exercises that can be used to undermine literal language, and these have been built upon in several subsequent publications (e.g., Hayes & Stroshal, 2004). The most relevant of these exercises in the current context are known as defusion or deliteralization exercises. Defusion exercises are designed to alter the context for the derived transformation of functions. Importantly, the defusion technique aims to leave all psychological functions of stimuli, even the aversive ones, fully in tact (see Blackledge, 2007). In general terms, the technique teaches clients to see their thoughts of escape from feelings (e.g., an urge to gamble), overt actions (e.g., actual gambling) and further thoughts (e.g., intentions to gamble) as just thoughts, and to be aware of and psychologically present to the alternatives to avoidance. Such a technique seems counter intuitive insofar as it expressly involves teaching a gambling addict, for example, not to avoid the feelings and thoughts associated with gambling (e.g., “I am a failure”, “Here I go again”, “I have got nothing more to loose anyway”, etc.). This approach, however, is based on a solid understanding of derived relational processes and laboratory analog studies (e.g., Healy, Barnes-Holmes, Barnes-Holmes, Keogh, Luciano & Wilson, 2008; Masuda, Hayes, Sackett, & Twohig, 2004; see also, Hesser, Westin, Hayes & Andersson, 2009; Masuda, Hayes, Twohig, Drossel, Lillis & Washio, 2009; Masuda, Twohig, Stormo, Feinstein, Chou, & Wendell, 2010) which

have shown this technique to be effective in reducing the intensity of urges and the distress with which they are associated. In a defusion exercise, the therapist socially reinforces non-avoidance and nonescape responses from aversive thoughts and feelings and thereby establishes a genuinely novel context in which the client responds to feelings of distress associated with their compulsion. More specifically, a therapeutic context is created in which escape and avoidance are not the only possibilities. Moreover, a history of nonavoidance is established in the presence of former discriminative stimuli for avoidance (i.e., thoughts of avoidance, responded to literally). In this way, the dominance of avoidance functions produced by thoughts of gambling are reduced relative to other more recently established non-avoidance functions. Avoidance may of course still occur, but not only is its probability reduced, but the distress arising from the former certainty that gambling always follows thoughts of gambling has been directly undermined. This in turn further reduces the intensity of future avoidance responses, and therefore a further weakening of the control of gambling thoughts over behavior. Paradoxically then, a reduction in gambling urges and activity can arise when the client is taught to be willing to experience those very thoughts and feelings they fear most (hence the emphasis on “acceptance” in ACT). As a concrete illustration, defusion exercises often involve asking a client to repeat aloud a word or phrase that is discriminative for problematic urges or behavior and which are typically avoided. In the case of a compulsive gambler, they may be asked to repeat the phrase; “I am going to gamble today” repeatedly. Through such vocal repetition, the auditory functions of the phrase often come to dominate over the emotional functions of the words, such that the client begins to notice more and more the sounds of each word with each repetition. They may

Simon Dymond and Bryan Roche even begin to laugh at the amusing sounds of some of the words. Eventually, they may realize, with the help of the therapist, that the avoidance or approach functions of the word “gamble” have diminished, whist other stimulus functions have increased (see Masuda et al., 2004). This exercise can then serve as a metaphor for all disturbing thoughts and feelings in future therapy sessions. In this way, the ability of such thoughts to control (or be “fused with”) overt action decreases. The outcomes of these treatment approaches may then be measured using objective indices and self-report instruments, a range of which is now available within research on gambling. Conclusions & An Agenda for the Future Derived relational responding research has tremendous implications for the analysis of gambling behavior and for the development of empirically supported verbally based interventions, but behavior analysts can learn from interventions in other, related domains. Indeed, it is likely that the future development of effective techniques to treat disordered gambling will involve a range of components from other, related behavioral interventions. For instance, contingency management methods that are effective in treating substance abuse disorders (Silverman, Roll, & Higgins, 2008) may be capable of being adapted to treat disordered gambling, which is often highly comorbid with substance abuse (Petry, 2005, 2009). It is likely that contingency management procedures could target the ‘altered schedule of social reinforcement’ that gamblers undergoing treatment must contact, and in conjunction with pharmacotherapeutic procedures could prove useful in maintaining the effects of treatment contingencies (Grant, Kim, & Hartman, 2008; Petry, Wienstock, Ledgerwood, & Morasco, 2008). However, before further combinations of interventions to treat disordered gambling are proposed, a great deal more empirical work remains to be conducted,

49

both on the components of effective treatment interventions and the impact of derived relational responding on gambling behavior. Great strides have been made in the experimental analysis of gambling behavior. As the research momentum grows, it becomes critical to explain the role of verbal behavior on gambling behavior. The present article has highlighted how research on derived relational responding and the transformation of stimulus functions offers a potential analytic strategy with which to approach the behavior of verbally able humans engaging in games of chance. The potential of such an approach has yet to fully realized. Yet, only further research and the continued development of empirically supported treatment interventions will help to determine whether or not the research efforts currently underway will pay off in the future. REFERENCES Baer, R. A. (2005). Mindfulness-based treatment approaches: Clinician's guide to evidence base and applications. New York: Academic Press. Barnes-Holmes, D., Barnes-Holmes, Y., & Cullinan, V. (2000). Relational frame theory and Skinner’s Verbal Behavior: A possible synthesis. The Behavior Analyst, 23, 69-84. Barnes-Holmes, D., Hayes, S. C., Dymond, S., & O'Hora, D. (2001). Multiple stimulus relations and the transformation of stimulus functions. In S. C. Hayes, D. Barnes-Holmes, and B. Roche, (Eds.), Relational frame theory: A post-Skinnerian account of human language and cognition (pp. 51-71). New York: Kluwer Academic/Plenum. Baum, W. M. (1979). Matching, undermatching, and overmatching in studies of choice. Journal of the Experimental Analysis of Behavior, 32, 269-281.

50


Blackledge, J. T. (2007). Disrupting verbal processes: Cognitive defusion in acceptance and commitment therapy and other mindfulness-based psychotherapies. The Psychological Record, 57, 555-576. Blackledge, J. T., & Hayes, S. C. (2001). Emotion regulation in Acceptance and Commitment Therapy. JCLP/In session: Psychotherapy in Practice, 57, 243–255. Clark, D. M., Ball, S., & Pape, D. (1991). An experimental investigation of thought suppression. Behaviour Research and Therapy, 29, 253-257. Delfabbro, P., Lambos, C., King, D., & Puglies, S. (2009). Knowledge and beliefs about gambling in Australian secondary school students and their implications for education strategies. Journal of Gambling Studies, 25, 523539. Dixon, M. R., & Delaney, J. (2006). The impact of verbal behavior on gambling behavior. In P. M. Ghezzi, C. A. Lyons, M. R. Dixon, and G. R. Wilson (Eds.), Gambling: Behavior theory, research and application (pp. 171-190). Reno, NV: Context Press. Dixon, M. R., Nastally, B. L., Jackson, J. E., & Habib, R. (2009). Altering the near-miss effect in slot machine gamblers. Journal of Applied Behavior Analysis, 42, 913-918. Dymond, S. (2008). Approaching gambling as a verbal event: Commentary on Fantino & Stolarz-Fantino (2008). Analysis of Gambling Behavior, 2, 7981. Dymond, S., & Barnes, D. (1995). A transformation of self-discrimination response functions in accordance with the arbitrarily applicable relations of sameness, more-than, and less-than. Journal of the Experimental Analysis of Behavior, 64, 163-184.

Dymond, S., Bateman, H., & Dixon, M. R. (2010). Derived transformation of children’s pregambling game playing. Journal of the Experimental Analysis of Behavior, 94, 353-363. Dymond, S., Mills, K., Griffiths, J., Cox, A., & Crocker, V. Derived transfer of response allocation and outcome ratings in a simulated slot machine task. Manuscript in preparation. Dymond, S., & Rehfeldt, R. A. (2000). Understanding complex behavior: The transformation of stimulus functions. The Behavior Analyst, 23, 239-254. Dymond, S., & Whelan, R. (2007). Verbal relations and the behavior analysis of gambling. Analysis of Gambling Behavior, 1, 19-20. Fantino, E. (2008). Behavior analysis: Thriving, but how about its future? Journal of the Experimental Analysis of Behavior, 89, 125-127. Feldner, M. T., Zvolensky, M. J., Eifert, G. H., & Spira, A. P. (2003). Emotional avoidance: An experimental test of individual differences and response suppression using biological challenge. Behaviour Research and Therapy, 41, 403-411. Fields, L., Adams, B. J., Verhave, T., & Newman, S. (1990). The effects of nodality on the formation of equivalence classes. Journal of the Experimental Analysis of Behavior, 53, 345– 358. Grant, J. E., Kim, S. W., & Hartman, B. K. (2008). A double-blind, placebocontrolled study of the opiate antagonist naltrexone in the treatment of pathological gambling urges. Journal of Clinical Psychiatry, 69, 783–789. Habib, R., & Dixon, M. R. (2010). Neurobehavioral evidence for the “near miss” effect in pathological gamblers. Journal of the Experimental Analysis of Behavior, 93, 313-328.

Simon Dymond and Bryan Roche Hayes, S. C., & Barnes-Holmes, D. (2004). Relational operants: Processes and implications: A response to Palmer’s review of Relational frame theory. Journal of the Experimental Analysis of Behavior, 82, 213-224. Hayes, S. C., Barnes-Holmes, D., & Roche, B. (2001), Relational frame theory: A post-Skinnerian account of human language and cognition. New York: Kluwer Academic/Plenum. Hayes, S. C., Fox, E., Gifford, E. V., Wilson, K. G., Barnes-Holmes, D., & Healy, O. (2001). Derived relational responding as learned behavior. In S. C., Hayes, D. Barnes-Holmes, & B. Roche (Eds.), Relational frame theory: A post-Skinnerian account of language and cognition (pp. 21-49). New York: Kluwer Academic/Plenum Publishers. Hayes, S. C., & Stroshal, K. D. (2004). A practical guide to Acceptance and Commitment Therapy. New York: Springer-Verlag. Hayes, S. C., Strosahl, K. D., & Wilson, K. G. (1999). Acceptance and commitment therapy: An experiential approach to behavior change. New York: Guilford Press. Healy, H. A., Barnes-Holmes, Y., BarnesHolmes, D., Keogh, C., Luciano, C. & Wilson, K. (2008). An experimental test of a cognitive defusion exercise: Coping with negative and positive self-statements. The Psychological Record, 58, 623-640. Hesser, H., Westin, V., Hayes, S. C., & Andersson, G. (2009). Clients' insession acceptance and cognitive defusion behaviors in acceptance-based treatment of tinnitus distress. Behaviour Research and Therapy, 47, 523528. Hoon, A. E., & Dymond, S. (2010). Transformation of slot machine response functions in accordance with derived comparative relations. Manuscript in preparation.

51

Hoon, A., Dymond, S., Jackson, J., & Dixon, M. R. (2007). Manipulating contextual control over simulated slot machine gambling. Analysis of Gambling Behavior, 1, 109-122. Hoon, A., Dymond, S., Jackson, J., & Dixon, M. R. (2008). Contextual control of slot machine gambling: Replication and extension. Journal of Applied Behavior Analysis, 41, 467-470. Johnson, T. E., & Dixon, M. R. (2009). Influencing children’s pregambling game playing via conditional discrimination training. Journal of Applied Behavior Analysis, 42, 73-81. Kallick, M., Suits, D., Dielman, T., & Hybels, J. (1976). Survey of American Gambling Attitudes and Behaviors (USGPO Stock No. 052-003-00254). Washington, DC: U.S. Government Printing Office. Kohlenberg, R. J., & Tsai, M. (1991). Functional analytic psychotherapy: Creating intense and curative therapeutic relationships. New York: Plenum Press. Lesieur, H. R., & Blume, S. B. (1987). The South Oaks Gambling Screen (The SOGS): A new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144, 1184–1188. Linehan, M. M. (1993). Cognitivebehavioral treatment of borderline personality disorder. New York: Guilford Press. Mace, F. C., & Critchfield, T. S. (2010). Translational research in behavior analysis: Historical traditions and imperative for the future. Journal of the Experimental Analysis of Behavior, 93, 293-312. MacLin, O. H., Dixon, M. R., Daugherty, D., & Small, S. L. (2007). Using a computer simulation of three slot machines to investigate a gambler’s preference among varying densities of near-miss alternatives. Behavior Research Methods, Instruments, and Computers, 39, 237–241.

52


McCurry, S. M., & Hayes, S. C. (1992). Clinical and experimental perspectives on metaphorical talk. Clinical Psychology Review, 12, 763–785. Madden, G. J., Ewan, E. E., & Lagorio, C. H. (2007). Toward an animal model of gambling: Delay discounting and the allure of unpredictable outcomes. Journal of Gambling Studies, 23, 6383. Masuda, A., Hayes, S. C., Sackett, S. F., & Twohig, M. P. (2004). Cognitive defusion and self-relevant negative thoughts: Examining the impact of a ninety-year-old technique. Behaviour Research and Therapy, 42, 477-485. Masuda, A., Hayes, S. C., Twohig, M. P., Drossel, C., Lillis, J., & Washio, Y. (2009). A parametric study of cognitive defusion and the believability and discomfort of negative self-relevant thoughts. Behavior Modification, 33, 250-262. Masuda, A. Twohig, M. P., Stormo, A. R., Feinstein, A. B., Chou, Y, & Wendell, J. W. (2010). The effects of cognitive defusion and thought distraction on emotional discomfort and believability of negative self-referential thoughts. Journal of Behavior Therapy & Experimental Psychiatry, 4, 11-17. Nastally, B. L., Dixon, M. R., & Jackson, J. W. (2010). Manipulating slot machine preference in problem gamblers through contextual control. Journal of Applied Behavior Analysis, 43, 125129. Petry, N. M. (2005). Pathological gambling: Etiology, comorbidity, and treatment. Washington, DC: American Psychological Association. Petry, N. M. (2009). Disordered gambling and its treatment. Cognitive and Behavioral Practice, 16, 457-467. Petry, N. M., Wienstock, J., Ledgerwood, D., & Morasco, B. (2008). A randomized trial of brief interventions for problem and pathological gamblers. Journal of Consulting and Clinical Psychology, 76, 318–328.

Reid, R. L. (1986). The psychology of the near miss. Journal of Gambling Behavior, 2, 32–39. Roche, B., Kanter, J. W., Brown, K. R., Dymond, S., & Fogarty, C. C. (2008). A comparison of “direct” versus “derived” extinction of avoidance. The Psychological Record, 58, 443-464. Sevigny, S., & Ladouceur, R. (2003). Gamblers’ irrational thinking about chance events: The ‘double switching’ concept. International Gambling Studies, 3, 162–170. Skinner, B. F. (1974). About behaviorism. London: Penguin. Sidman, M. (1994). Equivalence relations and behavior: A research story. Boston, MA: Authors’ Cooperative. Silverman, K., Roll, J. M., & Higgins, S. T. (2008). Introduction to the special issue on the behavior analysis and treatment of drug addiction. Journal of Applied Behavior Analysis, 41, 471-480. Stewart, I., & McElwee, J. (2009). Relational responding and conditional discrimination procedures: An apparent inconsistency and clarification. The Behavior Analyst, 32, 309-317. Vilardaga, R. M., Hayes, S. C., Levin, M. E., & Muto, T. (2009). Creating a strategy for progress: A contextual behavioral science approach. The Behavior Analyst, 32, 105-133. Volberg, R. A. (1994). The prevalence and demographics of pathological gamblers: Implications for public health. American Journal of Public Health, 84, 237–241. Volberg, R. A. (1996). Prevalence studies of problem gambling in the United States. Journal of Gambling Studies, 12, 111–128. Wardle, H., Sproston, K., Orford, J., Erens, B., Griffiths, M., Constantine, R., & Pigott, S. (2007). British gambling prevalence survey. Retrieved from http://www.gamblingcommission.gov. uk/research__consultations/research/b gps.aspx

Simon Dymond and Bryan Roche Weatherly, J. N., & Dixon, M. R. (2007). Toward an integrative behavioral model of gambling. Analysis of Gambling Behavior, 1, 4-18. Whelan, R., Barnes-Holmes, D., & Dymond, S. (2006). The transformation of consequential functions in accordance with the relational frames of more-than and less-than. Journal of the Experimental Analysis of Behavior, 86, 317-335. Zlomke, K. R., & Dixon, M. R. (2006). Modification of slot-machine preferences through the use of a conditional discrimination paradigm. Journal of Applied Behavior Analysis, 39, 351– 361. Action Editor: Jeffrey N. Weatherly

53


2010, 4, 54–60


The Roulette Near-Miss Effect Mark R. Dixon Southern Illinois University The near-miss effect has been repeated documented in the published literature as a variable that impacts gambling behavior. The effect, however, has been almost exclusively studied using slot machines. The present investigation sought to explore the effect of almost winning while playing roulette. When 28 participants were given the opportunity to play roulette and rate the closeness to wins after every trial, ratings varied as a function of numerical value between number bet and number won for most players. These results extend the findings that almost winning (e.g., a near-miss effect) is present for the game of roulette and defines the parameters of such an effect. Implications for the treatment of pathological gamblers are presented. Keywords: Near-miss, Roulette, Gambling, Addiction, Risk-taking --------------------------------------------------

When partaking in a game of chance, many players will find themselves becoming quite pleased upon producing a winning outcome. The unexpected, probabilistic, reinforcement schedule maintains behavior for great periods of time. However, when that same player loses, yet his/her loss looks "close" to a win, a paradox appears to occur. Close-win outcomes are often rated by players as being somehow closer to wins than other types of losses (Dixon, Nastally, Jackson, & Habib, 2009; Dixon & Schreiber, 2004). What is interesting is that all losses are just that - losses, and none are more predictive of a win right around the corner. This tendency to categorize certain losses as more valuable or predictive of a win has been termed the near-miss effect (Reid, 1986). Previous research has noted the potential for these types of outcomes to promote problem gambling (Griffiths, 1991), produce preferences for such outcomes over total losses (MacLin, Dixon,

Daugherty, & Small, 2007), as well as generate specific neurological activity usually only occurring during wins for pathological gamblers (Habib & Dixon, 2010). While there has been considerable growth in the exploration of the near-miss effect by persons who gamble, what has not been seen is much extension beyond a simple slot-machine preparation. With slot machines, the near miss is clearly defined as 2 winning symbols on the payoff line, and the remaining 1 winning symbol somewhere right above or below the payoff line. A notable exception to the exclusive study of near misses via slot machines was conducted by Dixon, Nastally, Hahs, Horner-King, and Jackson (2009) using the casino game of Blackjack. Using a single-deck game, participants rated how close 50 consecutive hands of blackjack were to a win. Losses were latter categorized as either "busts" or "non-busts" with the former being defined as a loss to the dealer where the player did obtain card values of over 21, and the latter being card values of less than 21. A nearmiss effect was shown only for the non-bust losses and it decreased in strength as the difference between dealer and player hand

Address all correspondence to: Mark R. Dixon Behavior Analysis and Therapy Program Rehabilitation Institute Southern Illinois University Carbondale, IL 62901 Email:mdixon@siu.edu

54

Mark R. Dixon

55

card values increased (Dixon et al., 2009). Near misses thus appear to be present on more than just slot machines, and their presence is not simply a factor of formal similarity to a win. In the above blackjack study, the additional feature of "bust" or "non-bust" modulated the effect, suggesting a role for psychological function to impact participant outcomes (i.e., ratings). Additional research has demonstrated the flexibility of the near-miss effect based on conditioning history of the participants (Dixon et al., 2009). To date, the published gambling literature has yet to aggressively explore the presence of near misses in other casino games. Such losses might be present at the craps table when rolling dice "close" to those that are designated as winning combinations or at the video-poker machine where a royal flush is missed by just one card. Expanding the scope of the investigation of the near miss to other casino games is critical to determine if the effect is a feature of actual slot machine or as a psychological process that leads to faulty decision making. Behavior analytically, one might consider this an issue of a structural characteristic of one game or a frequently occurring discriminative stimulus present across many casino-type games that sets the occasion for players to respond by wagering even when a win is not reliability predictable. Others might also consider the near miss to be a contextual stimulus, setting event, or motivational operation that participates in a field of interaction to increase the probability of a gambler responding (gambling), under a certain set of stimulus conditions. However, regardless of definition, as researchers we must continue to explore the robustness of the near miss across games, and determine the conditions under which it is demonstrated.

As a result, the present investigation attempted to explore the potential for the game of roulette to produce near-miss outcomes from gamblers. Using self-reports of "closeness to a win" this study examined a series of wins and losses from individuals gambling at roulette. METHOD Participants Twenty-eight college students with no self-reported history of gambling disorders participated in the present study for course extra-credit and a chance to win a $50 gift card to a local retail establishment. Additional contingencies were imposed such that the first 5 students that selected a winning number were given 10 extra-credit points towards their final course grade, the next 5 students to select a winning number received 5 extra-points, and the remainder of the students received 1 extra-credit point. While not available during the current study, additional opportunities for course extracredit were afforded to the students that may have wanted more. Participants recruited for this study were informed that, if they had a gambling problem, they would be allowed to obtain identical compensation for a nongambling study. Setting / Apparatus The experimental procedure took place in a single-room stadium style classroom following the completion of an undergradgraduate class lecture. The room cpntained a large number of chairs, desks, and speaker's podium with two large screens for presentation displays which faced the class attendees. A computerized roulette interface was presented on one of the large screens and can be seen in Figure 1.

56

ROULETTE PLAYERS

Figure 1. Display of the experimental interface.

Roulette

All programmed contingencies were at fair odds. In other words, each location on the roulette reel had an equal probability of being displayed on the board (p = 1/38). Participants were also provided with a data sheet which included space to write trial number, number on the board which they wished to wager upon, and the number on the board which was eventually displayed as the winning number. They were also provided with a Likert-type scale from 1 to 10 in which they were to circle the number that they believed was "how close their outcome was to a win." Procedure At the onset of the experiment, all participants completed an informed-consent document, and were instructed that they

were able to win additional course extra credit, and an opportunity for a cash prize for completion of the study. Course credit was administered as described above, and the potential cash prize consisted of their name being entered into a lottery for the drawing of a single 50 USD gift card to a local retail establishment. Participants were instructed the following verbally by the experimenter: "You will be asked tonight to wager on a single number that you see displayed on this computerized roulette board. You will complete a data sheet whereby you will indicate the number of trial that you are on, the wager you make, and the resulting outcome of each spin. After you select a number you think will win, place your pen on the desk, and your hands in your lap. Once everyone's hands are down, I will

Mark R. Dixon

57

click on the spin button and we will all watch the wheel spin. If the winning number that is displayed on the board is the number you bet on, please raise your hand for me and let me know that you won. If you did not win, please keep your hands down until we have checked all 'winners.' At that time you will be allowed to rate how close your wager was to the winning wager. At this time you can also bet on the next game. We will be watching you to ensure you are performing this task correctly, and if not, you will be dismissed from the study."

To ensure that participants did not "cheat" and wait until the winning number on the roulette board was displayed to write down their own wagered number, each participant was instructed to put their pen on the table, and their hands in their lap until the spin of the wheel occurred, and the winning number revealed. Any participant that had wagered on that winning number was instructed to raise their hand to allow one of three researchers in the room to check their data sheet for correspondence between numbers. If correspondence was observed, contingencies as previously specified were delivered. This procedure produced 100% correspondence between self-reported wins

and observed data sheets. No participant was ever observed attempting to alter a data sheet mid or post spin of the reel. Correspondence between all participants' recording of the obtained winning number and the experimenter's recording of the number matched 100% of the trials conducted. Once the participant wagered on the winning number, they were dismissed from the remainder of the study. This resulted in varying lengths of exposure to the experimental procedure, which increased based on repeated losses. For example, one participant may have won after 3 trials and another after 34 trials, thus the procedure would have resulted in more obtained data from the second participant. The procedures continued until all participants had wagered on a winning number (85 trials). RESULTS All recruited participants completed the experimental procedures. Wins were obtained by participants on as few trials as 3 and as many trials as 85. Figure 2 displays the average subjective rating for all participants on losing trials.

Figure 2. Average rating of closeness to a win as a function of numerical value difference between number bet upon by the participant and the obtain winning number displayed on the roulette board.

10

Avg Subjective Rating

9 8 7 6 5 4 3 2 1 0 1

6

11

16

21

‐ numerical value of bet and win

26

31

58

ROULETTE PLAYERS

Figure 3. Average rating of closeness to a win as a function of spatial difference between number bet upon by the participant and the obtained winning number displayed on the roulette board.

10

Avg Subjective Rating

9 8 7 6 5 4 3 2 1 0 1

2

3

4

5

6

7

8

9

10

‐ between Location of Bet and Win on Board

Data are plotted as a function of the numerical value difference between the wagered and the winning number. Subjective ratings were extremely high (close to a win) when the difference in value was small, and decreased considerably as the differences in value increased. A Pearson Product-Moment Correlation was calculated to support the visual analysis of the data. This analysis revealed a significant correlation between the difference in numerical value of wager and win and subjective rating, r(30) = -.90, p < .05). Figure 3 displays a similar trend when the subjective ratings are plotted against the difference between location on the roulette board between the wagered and the winning number. Here, however, the negative data trend is less pronounced. A Pearson Product-Moment Correlation was also calculated to support the visual analysis of the data. This analysis also revealed a significant correlation between the proximal distance on the roulette board between

wagered and winning number and subjective rating, r(8) = -.88, p < .05). DISCUSSION The near-miss effect occurs when a gambler believes that certain losing outcomes are closer to wins than other losing outcomes. In an objective reality, all losses are just that - losses. None are more predictive of a win that is bound to occur. Furthermore, no loss, no matter how much it might "look like a win", is indicative of being close to a win. Each outcome from typical casino games such as slot machines, craps, or roulette is independent of the next. The game knows no history of prior outcomes. Instead, the gambler incorrectly assumes history or certain losses reveal information about the future. The present findings extend the published research that has documented a near-miss effect in slotmachine players (Dixon & Schreiber, 2004), and blackjack players (Dixon et al., 2009) to the game of roulette, as well as support the conceptualizations that near misses do in

Mark R. Dixon

59

fact occur when people are gambling (Reid, 1986). In the present study, twenty eight roulette players rated losing outcomes closer to winning when that losing outcome was a) close in numerical value to the winning number, and b) close in proximity on the actual roulette table to the winning number. The two factors noted here are to a fair degree autocorrelated with each other, and future research is warranted which could determine if one of these factors is more responsible for producing inflated subjective ratings than the other. With hopes of maintaining a fair degree of external validity, the present study utilized a roulette simulation that closely resembled those found in casinos as well as mirrored that found in online gaming environments. The makeup of the roulette game, and the board specifically, grouped numbers close together proximally that were close together numerically. As a result, it is possible that the current participants' rating behavior was a product of one or a combination of both of these factors. A future study might consider altering the roulette board display to examine if nonnumerically close numbers, if grouped together proximally on the board, would alter the correlational relationship that was observed in Figure 3. A similar examination could be made by examining the distance on the actual roulette wheel between winning number and wagered number. The reel itself does not have numerical numbers placed proximal. In this study, it was not possible, given the setting configuration. Additional data capturing along these lines would thus allow for more sophisticated data analyses to be conducted such as a regression model which contained distance on board, distance on reel, and distance in numerical value. The obtaining of a near-miss effect by roulette players was not entirely surprising given that an increasing body of literature is emerging on the presence of this effect by

gamblers. What remains to be answered at this time however, is what this "effect" really is. From a behavioral perspective, such an "effect" beckons notions of internal casual states of the organism, that are somehow flaws of rationality or decision making abilities. Behavior analysts are not content with such an explanation, and need to examine the factors in the environment that may be producing the near-miss response. In the present study's case, the inflated self-reported ratings of certain losses compared to other losses. With respect to slot machines, the near miss might be defined as a result of stimulus generalization, given that two winning symbols on the payoff line and a third right above or below the payoff line looks physically similar to an actual win. The data of the current study tend to weaken this explanation. The numerical value of a 7 is not typographically similar to an 8, nor is a 9 to a 10. However, an 8 is very similar to an 18, and a 1 to an 11. Yet, it is the numerical value that was correlated to the near miss, not the physical similarity of the stimuli. Previous research by Dixon et al. (2009) has suggested that the near-miss effect is a product of verbal behavior, or a verbal construction, rather than a product of stimulus generalization. The current data support this assertion, given that a participant's pre-experimental history is probably very rich for relating the number 8 as being a little less than 9, and 1 being far from 11. Such a history comes to bear in an experimental, or gambling, experience and alters responding accordingly. Working on the assumption that the near-miss is verbally constructed, therapists have a fair amount of resources by which they may be able to minimize or reverse the effect in their patients. Behavioral techniques such as deliteralization, transformation of stimulus functions, and/or altering relational networks may have

60

ROULETTE PLAYERS

promise. These techniques have been successful at altering response allocations of gamblers to game alternatives when contingencies have remained in tack (e.g., Zlomke & Dixon, 2006), and it is thus possible that they may be able to alter nearmiss ratings and control by near-miss stimuli as well. Preliminary findings support this notion (Nastally & Dixon, this issue). In summary, the near-miss effect occurs in multiple casino games. The present data add the game of roulette to the list made up previously of only slot machines and blackjack. The means by which the "effect" is produced has yet to be comprehensively explained but the current data weaken the notion that physical characteristics of the stimuli are solely responsible for the exhibiting of a near-miss response. If pathological gamblers have developed deep rooted notions of what are considered wins and almost wins, and those gamblers alter subsequent gambling because of such nearmiss outcomes, the present data suggest that care providers need to be very careful in treatment. Near misses are not just a part of a slot machine, and they are not just part of the internal workings of an illogical gambler. Instead the near-miss effect appears to be the outcome of certain environmental arrangements, and given that position, altering such arrangements either via differential reinforcement or through verbally constructed means, should be able to produce change in the pathological gambler. With the lives of thousands of pathological gamblers in need of treatment, researchers should extend investigations of the near-miss forward with the hope being that caregivers will not need to do rely on unfounded treatment approaches that may at best lead to "almost" a success in therapy.

REFERENCES Dixon, M. R., & Schreiber, J. (2004). Nearmiss effects on response latencies and probability estimations of slot machine players. The Psychological Record, 54, 335–348. Dixon, M. R., Nastally, B. L., Jackson, J. E., & Habib, R. (2009) Altering the nearmiss effect in slot machine gamblers. Journal of Applied Behavior Analysis, 42, 913-918. Dixon, M.R., Nastally, B.A., Hahs, A. D., Homer-King, M., & Jackson, J.W. Blackjack players demonstrate the near miss effect. (2009). Analysis of Gambling Behavior. 2, 56-61. Griffiths, M. (1991). Psychobiology of the near-miss in fruit machine gambling. Journal of Psychology, 125, 347–357. Habib, R. & Dixon, M.R. (2010). Neurobehavioral evidence for the `nearmiss` effect in pathological gamblers. Journal of the Experimental Analysis of Behavior, 93, 313-328. MacLin, O. H., Dixon, M. R., Daugherty, D., & Small, S. (2007). Using a computer simulation of three slot machines to investigate a gambler’s preference among varying densities of near-miss alternatives. Behavior Research Methods, Instruments, and Computers, 39, 237–241. Reid, R. L. (1986). The psychology of the near miss. Journal of Gambling Studies, 2, 32–39. Zlomke, K. R., & Dixon, M. R. (2006) Modification of slot-machine preferences through the use of a conditional discrimination paradigm. Journal of Applied Behavior Analysis, 39, 351-361. Action Editor: Jeffrey N. Weatherly


2010, 4, 61–75


Concurrent Validity of the Gambling Functional Assessment (GFA): Correlations with the South Oaks Gambling Screen (SOGS) and Indicators of Diagnostic Efficiency Joseph C. Miller, Mark R. Dixon, Amanda Parker, Ashley M. Kulland, & Jeffrey N. Weatherly* University of North Dakota, Southern Illinois University, Southern Illinois University, University of North Dakota, & University of North Dakota Concurrent validity of the recently introduced Gambling Functional Assessment (GFA) was assessed by comparison with the long-used South Oaks Gambling Screen (SOGS) in two nonclinical adult samples (N = 201, 49% female; N=101, 74% female). Correlations between GFA total scores and its four content scores with SOGS scores were promising (r = .04 to .61), with the content score relating to Escape yielding the highest correlations (.45, .61) and the score relating to Attention yielding the lowest. Performance in the second sample, where the SOGS-defined base rate of pathological gambling (28.7%) was high, was best for Escape scores, which efficiently categorized SOGS-defined cases. The present data suggest that the GFA content area of Escape shows promise at classifying pathological versus nonpathological gambling, while the GFA as a whole may be a useful treatment tool, allowing clinicians to identify the mechanisms that may be maintaining gambling in their patients seeking treatment for pathological gambling. Keywords: Concurrent validity, Gambling Functional Assessment, Escape, South Oaks Gambling Screen, Adults ____________________________________________________________

The current Diagnostic and Statistical Manual (DSM-IV; American Psychiatric Association, 1994) defines pathological gambling as “persistent and recurrent maladaptive gambling behavior” (p. 618). As with most DSM-defined disorders, the diagnostic criteria are a la carte, with the individual needing to display at least five of 10 potential symptoms to be given the diagnosis. Not all symptoms are directly linked to the behavior itself, however. For example, the first criterion, preoccupation with gambling, refers to planning and mental rehearsal for future gambling and rumination

about past gambling experiences. Several subsequent criteria refer to the negative life consequences of the behavior, its practical maintenance, or concealment. Apart from apparent “withdrawal” symptoms reflected in the criterion “is restless or irritable when attempting to cut down or stop gambling,” only one criterion, “gambles as a way of escaping from problems or of relieving a dysphoric mood,” refers to maintenance mechanisms—in this case, negative reinforcement. Thus, the current diagnostic criteria emphasize pathological outcomes and deemphasize the means of behavioral maintenance. In contrast, the Gambling Functional Assessment (GFA; Dixon & Johnson, 2007) was designed to determine the consequences that might be maintaining the individual’s gambling. It was designed around the as-

*Address correspondence to Jeffrey N. Weatherly, Ph.D. Department of Psychology University of North Dakota Grand Forks, ND 58202-8380 Phone: (701) 777-3470 Fax: (701) 777-3454 Email: jeffrey.weatherly@und.edu

61

62

CONCURRENT VALIDITY OF THE GFA

sumption that different individuals may gamble for different reasons, and thus need different styles of treatment to successfully overcome excessive gambling. For example, one person may gamble to try and avoid the pain of a dysfunctional marriage, while another may gamble for the physiological rush or sensory experience it gives him/her. While the severity of the disorder for these two individuals could be very similar, the cause, and thus the required treatment, could be much different. This type of “functionbased” assessment approach has been utilized for a number of clinical disorders from self-injury and aggression (e.g., Iwata, Dorsey, Slifer, Bauman, & Richman (1994) to eating disorders (e.g., Piazza et al., 2003). The reasons for gambling assessed by the GFA are not necessarily pathological in and of themselves, though there are theoretical reasons to suspect that different maintenance mechanisms may be more or less likely to result in pathological gambling in some individuals (e.g., see Weatherly & Dixon, 2007). The GFA is a 20-item, Likert-type, selfreport instrument designed to identify four possible maintaining functional consequences of gambling (i.e., reinforcement contingencies): Sensory, Attention, Tangible, and Escape (see also Durand & Crimmins, 1988). Sensory functions might include the lights, sounds, or physical bodily sensations associate with gambling. Attention functions may include the social enjoyment of being with friends while gambling, or the emotional embraces of a loved one who provides compassion to the gambler upon returning from the casino. Tangible functions might include gambling to acquire casino “points” or “comps,” as well as the possibility of gaining sums of money. Finally, the escape functions might include gambling to numb oneself from certain life pains or stressors, or to replace dealing with difficult psychological issues.

Five of the 20 total items are dedicated to each of the four functional consequences. Scores for each item range from 0 to 6, resulting in a possible maximum score of 30 in each content area (i.e., type of consequence) and a maximum raw score of 120 for the entire instrument. Reliability of the GFA has been measured in a large (N = 949) nonclinical college sample (Miller, Meier, & Weatherly, 2009). Internal consistency (Crombach’s α) was quite good for the total GFA score (.92) and for the four content scores (.80 to .84). Test-retest reliability for the total GFA score was adequate (.75) after 12 weeks. Temporal stability for three of the four content areas was likewise adequate (.69 to .71). The consequence of Escape, however, evidenced lower test-retest reliability (.40) than the other consequences, which is indicative of variability over time. The Escape content area also proved unique with respect to construct validity (Miller, Meier, Muehlenkamp, & Weatherly, 2009). Factor analysis (N=308) suggested that the GFA measured two broad constructs, interpreted as positive reinforcement and negative reinforcement, in a young-adult non-clinical sample. While strong positive correlations were observed between the positive reinforcement factor and the GFA scores for Attention (r = .84), Sensory (r = .79), and Tangible (r = .85), only the Escape scores correlated highly (r = .95) with the negative reinforcement factor. It was further observed that Escape scores were highly positively skewed; only a small minority of respondents in the upper 50th percentile of total GFA scores endorsed any items related to Escape. Miller et al. posited that the Escape score might thus be a better indicator of pathogenic, per se, behavioral maintenance function for gambling than the other three GFA content areas, as scores in these other areas were relatively normally distributed in the non-clinical sample. However, there is to date, no independent empirical evidence to

Joseph C. Miller et al. support this assertion. Likewise, there is no empirical support for the external validity of the GFA as a measure of pathological gambling. One means of establishing this criterion validity (Groth-Marnat, 2003) is direct comparison with other established measures of the same construct(s), applied at the same point in time (i.e., concurrent validity; e.g., Anastasi & Urbina, 1997; Sattler, 2001). One Criterion Measure of Pathological Gambling The South Oaks gambling Screen (SOGS; Lesieur & Blume, 1987) is a brief instrument intended to measure probable pathological gambling by sampling clinically relevant outcomes (e.g., difficulty controlling the amount of gambling, guilt about gambling, lying about or hiding gambling behavior, low efficacy for quitting despite a desire, negative interpersonal and occupational consequences, and means used or sources tapped for securing the money necessary to continue gambling). Thus, the SOGS, having been developed using prior DSM criteria (Lesieur & Blume, 1987), is similar to the DSM-IV clinical criteria, in that pathological outcomes are emphasized. The SOGS’ authors recommend a raw score of five or more as an indicator of potential pathological gambling. Reliability statistics for the measure are uniformly adequate. For internal consistency, Stinchfield (2003) found α = .81 for a large non-clinical Midwestern sample (N = 803). While Lesiuer and Blume (1987) reported α = .97 for the original norming sample, Stinchfield (2002) pointed out that this coefficient was derived using a large mixed clinical/non-clinical sample. In actual use, where reference is made to a single population, testing of a more homogeneous sample should result in less score variance and lower internal consistency, such as that reported by Stinchfield (2003). Test-retest reliability for the SOGS with a mixed clinical/non-clinical sample (N

63

= 112) was rtt = .71 with test administrations “30 or more days apart” (Lesiuer & Blume , 1987; p. 1186). The SOGS is a thoroughly researched instrument and its validity is well-accepted, despite some critiques (see Gambino & Lesieur, 2006). Thus, with respect to the identification of likely pathological gamblers, the SOGS is a legitimate criterion measure for assessment of the GFA’s validity as a screen for probable pathological gambling. Diagnostic Efficiency Relative to SOGSDefined Populations Using the SOGS’ cutoff score as criterion, it should be possible to estimate the diagnostic efficiency of various GFA cutoff scores. In other words, probable pathological gamblers and non-pathological respondents may be identified by their SOGS raw score (pathological ≥5); various GFA cutoff scores could be used to identify these same cases, and the accuracy of categorization by the GFA assessed. Indicators of diagnostic accuracy derived from this analysis would not represent the GFA’s diagnostic accuracy or efficiency per se (i.e., no diagnoses are rendered, and there is no independent confirmation of the categories defined by the SOGS cutoff score). However, classification of cases similar to that accomplished by SOGS scores would support concurrent validity of the GFA, by supporting its convergence with the SOGS categorization of cases. Hypotheses The current study used scores from the SOGS as a means of assessing the concurrent validity of the GFA scores as indicators of probable pathological gambling in two ways. First, we determined the degree of correlation between scores from the two tests—the more traditional method of demonstrating this form of criterion validity (Anastasi & Urbina, 1997; Groth-Marnat, 2003’ Nunnally & Bernstein, 1994). We hy-

64


pothesized that GFA scores would correlate highly and significantly with SOGS scores. Because the statistical significance of a correlation is relative to sample size, the magnitude of the correlation is more salient. Anastasi and Urbina (1997) suggest that such convergent correlations should be “moderately high, but not too high” (p. 127), as very high correlations may suggest that the new measure is redundant. Groth-Marnat points out that there is no universally accepted minimal correlation sufficient to support convergent validity; rather, a criterion should be set logically, following the purpose and assumptions of the tests involved, and, where possible, comparison with known correlations among tests of the same construct. Stinchfield (2002) found high correlations between SOGS scores and DSM-IV diagnostic criteria in a large Minnesota community sample, surveyed by telephone (r = .77; N = 803), and a large sample of clients seeking treatment for gambling problems at state clinics (r = .83; N = 400). Recently, four pathological gambling measures were intercorrelated in a large study of university students (N = 197) in Singapore (Arthur, Tong, Chen, Hing, Sagara-Rosemeyer, Kua, & Ignacio, 2008). Correlations between the SOGS and the Gamblers Anonymous 20, the Canadian Problem Gambling Index, and the DSM-IV diagnostic criteria for pathological gambling ranged from .60 to .79. Jimenez-Murcia et al. (2009) considered correlations with SOGS of ≥ .30 evidence of convergent validity in their evaluation of a Spanish translation of a DSM-IV based pathological gambling measure. Based on these precedents, we anticipated that correlations between SOGS and GFA scores would exceed .30. Correlations in the range of .60 or above would be considered more satisfactory, since the correlation between SOGS and the current “gold standard”

DSM-IV criteria falls at or above .60 (Arthur et al., 2008; Stinchfield, 2002). Second, we explored the GFA’s accuracy and efficiency in predicting categories (i.e., pathological versus non-pathological) as defined by the SOGS cutoff score for probable gambling pathology. This methodology is less traditional, but has several advantages. Correlational analyses reveal little about the relative diagnostic efficiency of a test, and tests that correlate may not necessarily distinguish groups with similar accuracy. Some researchers have suggested that a test's ability to classify relevant cases is a better indicator of its validity than its correlations with related measures, since such classification more closely matches realworld application. The notion of validity is tied to the application of the testing method (Cronbach, 1988). Thus, because the GFA was originally designed for clinical applications, a diagnostic approach that more closely parallels its eventual application, rather than a correlational method, would seem warranted. Moreover, the second approach allows for the identification of optimal cutoff scores for such applications, which are not produced by the correlational analysis. Sensitivity, specificity, and other indicators of diagnostic accuracy may be evaluated in the context of diagnostic efficiency relative to the base rate of pathology as indicated by the criterion measure. We therefore hypothesized that, as a valid measure of gambling pathology, the GFA would be diagnostically efficient (Meehl & Rosen, 1955) relative to the “base rate” established empirically by SOGS ≥ 5. Based on the unusual performance of the GFA Escape score seen previously (Miller et al., 2009), we further hypothesized that GFA Escape scores would evidence the greatest diagnostic accuracy relative to the SOGS-defined categories (i.e., these previous data suggest that negative reinforcement contingencies are the most

Joseph C. Miller et al. pathogenic in the context of gambling; cf., Weatherly & Dixon, 2007). METHOD Participants Data were collected from two locations in the United States: One in Nevada and one in Illinois. Demographic data are displayed in Table 1 for each sample, including gender Table 1. Demographic Variables for Participants in the Nevada and Illinois Samples. Nevada

Illinois

201 (49%)

101 (74%)

Median Mean SD

45 45.7 14.3

32 35.8 12.0

White Asian African American Hispanic Native American Other

171 6 11 9 1 3

85 3 8 1 1 3

$0-5,000 $5,000-10,000 $10,000-20,000 $20,000-30,000 $30,000-50,000 $50,000-70,000 >$70,000

2 4 13 20 34 50 74

0 1 14 24 44 15 3

Education High School / GED Associates Degree Bachelors Degree Graduate Degree

93 34 43 31

45 26 26 4

History of Treatment None Drugs Gambling Alcohol

195 4 4 5

84 3 2 15

N (% Female) Age, in Years

Race

Income

65

distribution, median, and mean age (and SD) of participants, self-identified race, annual income, and history of treatment for drug abuse, alcohol abuse, or gambling problems. Data were collected from 204 participants (49% female) in Las Vegas and Wendover, Nevada. Three of these cases were removed due to missing data. One hundred-one participants (74% female) were sampled in Rockford, Illinois. Materials and Procedure Human subjects approval was obtained from Southern Illinois University’s Human Subjects Committee prior to the sampling of participants. All participants were given a copy of an informed consent page which described the research and its purpose, the risk to the participant, as well as information on the human subjects committee’s approval and contact information if the participant had any questions regarding the research. The materials were stapled packets containing the informed consent (described above), a demographics questionnaire, and two surveys/assessments on gambling behavior- the SOGS (Lesieur & Blume, 1987) and the GFA (Dixon & Johnson, 2007). People above the age of 18 were approached by one of three researchers and asked if they would participate in a research study on gambling behavior. Individuals who agreed to participate were given the packet or the packet was read to them (depending upon their reading ability or request). Participants responded to the survey, which took an estimated 5 - 10 min to complete. Once the participant was finished, the researcher collected the survey. Participants were not given anything of material value for their participation. All participants in the Nevada sample were approached by one of three researchers in locations including, but not limited to, restaurants, outside streets, public transportation systems (e.g., the airport, trolley, and

66


bus), Laundromats, grocery stores, private transportation service (i.e., hotel van transportation), parking lots, convenience stores, pawn shops, and liquor stores—all of which were within 100 yards of a gambling establishment. Data from the Illinois sample were collected in two sports bars in Rockford. Scores for the SOGS and GFA were calculated for each participant, according to the appropriate scoring guidelines (Dixon & Johnson, 2007; Lesieur & Blume, 1987). Indicators of diagnostic accuracy. Overall accuracy of GFA categorization was tabulated, along with sensitivity, specificity, positive predictive power, and negative predictive power for a range of GFA Overall and content cutoff scores (see Results). The method and rationale follow. All calculations are predicated on the SOGS score of ≥ 5 being a valid positive indicator of probable pathological gambling. The ability of various GFA cutoff scores to accurately reproduce the SOGS-based categories was assessed. Four outcomes are possible when predicting dichotomous group membership (e.g., identifying likely pathological versus likely non-pathological respondents): true positive, false positive, true negative, and false negative. If we identify cases as probably pathological (i.e., a “positive” prediction), based on some GFA cutoff score (e.g., Escape ≥ 10), then we are correct for people who scored ≥ 5 on the SOGS (true positives) and incorrect (false positives) for those who scored < 5 on the SOGS. If the GFA cutoff score identifies pathology as being absent (a “negative” prediction, e.g., Escape < 10), then we are correct (true negatives) for cases where SOGS < 5 and incorrect (false negatives) where SOGS ≥ 5. Only two of these outcomes are correct: true positives and true negatives. Together, cases with these frequencies are used to calculate the overall

accuracy of classification (Kamphuis & Finn, 2002) using Equation 1:

% Correct Classification =

True Positives + True Negatives

(Equation 1)

N

It should be noted that accurate prediction of a low base-rate phenomena is notoriously difficult (Meehl & Rosen, 1955). For example, the prevalence of pathological gambling in the general population has been estimated at 1-3%, a low base rate occurrence (e.g., see Petry, 2005). By simply predicting that no one in a random sample of the general population gambles pathologically, we would be correct in 97%-99% of cases, despite having made no true positive predictions. Meehl and Rosen (1955) derived Equation 2 as a criterion to determine when a cutoff score is efficient (i.e., when the predictions based on the cutoff yield greater overall accuracy than use of the base rate alone):

False Positives, using the Procedure Base Rate of Event > Base Rate of No Event True Positives, using the Procedure (Equation 2)

Using SOGS-defined groups, the Base Rate of Event is the percentage of respondents with SOGS ≥ 5, the Base Rate of No Event is 1- (Base Rate of Event), and “the Procedure” is the identification of likely pathological and non-pathological respondents using the GFA cutoff score of interest. Efficiency, as defined by Meehl and Rosen (1955), is one important criterion used to identify optimal cutoff scores for a test. However, in clinical use, “optimal” is variously defined (Groth-Marnat, 2003; Kamphuis & Finn, 2002), depending mostly on the importance assigned to avoiding false positives versus false negatives. For example, false positives might be more acceptable

Joseph C. Miller et al.

test as lacking the trait. Specificity is calculated using Equation 4. In the current study, Specificity is defined as the number of identified likely non-pathological gamblers, as determined by the SOGS (true negatives), divided by the total number of likely nonpathological gamblers (true negatives + false positives). Specificity reflects how well the test discounts cases that are likely not pathological.

than false negatives for a test of suicidality, because failing to detect suicidal intent may have far more dire consequences than mislabeling an individual as potentially suicidal. A practitioner might retain an inefficient test, because it produces few false negatives and identifies all or nearly all of the suicidal respondents (true positives). Therefore, other indicators of diagnostic accuracy, such as sensitivity, specificity, positive predictive power, and negative predictive power, are often of interest. Sensitivity is the proportion of cases in which a trait (present) is identified by the test (true positives) relative to the total number of cases where the trait is present. Sensitivity is calculated using Equation 3. In the current case, the trait is probable pathological gambling (operationalized as SOGS ≥ 5), true positives would be those likely pathological gamblers identified as such by GFA data, false negatives would be likely pathological gamblers not identified by GFA data, and the sensitivity of the GFA score would be equal to the number of SOGS-defined probable pathological gamblers identified by GFA (true positives) divided by the total number of SOGS-identified probable pathological gamblers (true positives + false negatives).

Sensitivity =

67

Specificity =

True Negatives True Negatives + False Positives (Equation 4)

Positive predictive power (PPP) is the proportion of cases predicted to have the trait that indeed have the trait. PPP can be calculated using Equation 5. PPP is, in the current case, the proportion of respondents identified as likely pathological by GFA data who earned a SOGS score of five or more. True Positives True Positives + False Positives (Equation 5)

PPP =

Negative predictive power (NPP) is the proportion of cases predicted to lack the target trait that indeed lack it. NPP can be calculated using Equation 6. Here, NPP is the proportion of respondents identified by the GFA as probably non-pathological who score less than five on the SOGS.

True Positives True Positives + False Negatives (Equation 3)

NPP =

Specificity is the proportion of cases without the trait correctly identified by the

True Negatives True Negatives + False Negatives (Equation 6)

Table 2. Correlations with SOGS Total Score for Two Samples. Attention

Escape

Tangible

Sensory

GFA Total

Nevada Sample (N=201; BR=7.5%)

.24

.45

.44

.42

.49

Illinois Sample (N=101; BR=28.7%)

.04

.61

.24

.38

.44

BR = Base Rate (SOGS≥5)

68


RESULTS Correlations Table 2 displays correlations between the SOGS score and the total GFA score and each of the four GFA content scores for both the Nevada and Illinois samples. Nevada sample (N = 201). The correlation between the SOGS and total GFA score was significant at the = .01 level, though the correlation was modest (r = .49). Similarly, significant correlations were found between the SOGS and GFA scores for Escape (r = .45), Sensory (r = .42), and Tangible (r = .44). The correlation between SOGS and GFA Attention scores appeared smaller than for the other GFA content areas (r = .243; p < .01). Illinois sample (N = 101). Correlations were more variable for the Illinois respondents, with coefficients for GFA scores on Attention (r = .04) and Tangible (r = .24) failing even to meet the significance criterion of α = .01. GFA Total (r = .44) and Sensory (r = .38) score correlations with the SOGS were both significant (p < .01). Correlations between the SOGS and GFA Escape scores yielded the largest coefficient (r = .61; p < .01) for either sample. Diagnostic Efficiency with Respect to SOGS-Defined Categories Tables 3, 4, and 5 display the sensitivity, specificity, positive predictive power (PPP), and negative predictive power (NPP) across a range of cutoffs for the four content and the total GFA scores in the Illinois and Nevada samples. Data are bolded where the cutoff score yielded efficient overall prediction (using criterion in Eq. 2) relative to the base rate, which was 7.5% for the Nevada sample, and 28.7% for the Illinois sample. Due to its unique factor loadings and distri-

bution (Miller, Meier, Muehlenkamp, & Weatherly, 2009), and its moderate to high correlations with SOGS (Table 2), the Escape score is of particular interest. Illinois Sample. The Escape scores performed best in the Illinois sample, consistent with the pattern of correlations displayed in Table 2. The efficiency criterion was met when Escape ≥ 11. At this cutoff, sensitivity was 38% and specificity was 94%, reflecting the relative importance of minimizing false positives when base rates are less then 50%. This cutoff score correctly classified 78% of the sample. Nevada sample. Both sensitivity and specificity were uniformly lower over the same range of Escape cutting scores in this sample. The maximum Escape sensitivity was 80%, versus 90% in the Illinois sample. DISCUSSION In terms of convergence with the SOGS, the GFA appeared to perform somewhat differently in the two samples, and across content scores. One reason may be the differences in the two samples. In the Nevada sample, 7.5% of respondents scored ≥5 on the SOGS—the instrument’s criterion for probable pathological gambling (Lesieur & Blume, 1987). The frequency of scoring 5 or more on the SOGS for the Illinois sample (28.7%) was nearly four times as high. The Nevada sample appeared to be somewhat wealthier and better educated overall. Only bar goers were sampled in Illinois, while Nevada respondents came from a variety of locations near gambling establishments. It should also be remembered that the GFA and SOGS are intended to measure two different, though related, constructs. The SOGS measures range and frequency of gambling behaviors, as well as behaviors—legal or illegal—serving to facilitate or obfuscate the


69

Table 3. Diagnostic Accuracy of GFA Total Score Cutoffs for the Illinois & Nevada Samples). SOGS≥5 is the criterion. Illinois Sample N = 101; BR = 28.7%

Nevada Sample N = 201; BR = 7.5%

Cut

Sens

Spec

PPP

NPP

%C

Sens

Spec

PPP

NPP

%C

≥50 ≥48 ≥46 ≥44 ≥42 ≥40 ≥38 ≥36 ≥34 ≥32 ≥30 ≥28

0.52 0.52 0.69 0.76 0.76 0.79 0.79 0.83 0.86 0.86 0.93 0.97

0.75 0.67 0.65 0.64 0.60 0.57 0.44 0.38 0.32 0.26 0.19 0.13

0.46 0.39 0.44 0.46 0.43 0.43 0.37 0.35 0.34 0.32 0.32 0.31

0.79 0.77 0.84 0.87 0.86 0.87 0.84 0.84 0.85 0.83 0.88 0.90

0.68 0.62 0.66 0.67 0.64 0.63 0.55 0.51 0.48 0.44 0.41 0.37

0.47 0.53 0.67 0.67 0.67 0.67 0.73 0.80 0.80 0.80 0.80 0.80

0.96 0.95 0.94 0.93 0.89 0.86 0.81 0.78 0.72 0.69 0.67 0.62

0.50 0.44 0.48 0.42 0.32 0.27 0.24 0.23 0.19 0.17 0.16 0.15

0.96 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.98 0.98

0.93 0.92 0.92 0.91 0.87 0.84 0.81 0.78 0.72 0.70 0.68 0.63

BR = Base Rate, i.e., % of N for whom SOGS≥5 Sens = Sensitivity = True Positives / (True Positives +False Negatives) Spec = Specificity = True Negatives / (True Negatives + False Positives) PPP = Positive Predictive Power = True Positives / (True Positives + False Positives) NPP = Negative Predictive Power = True Negatives / (True Negatives + False Negatives) %C = Percent Correct Overall = (True Positives +True Negatives) / N

gambling (i.e., the quantity of gambling and maladaptive outcomes). In contrast, the GFA assesses reasons for gambling in general, with no reference to maladaptive consequences; the only consequences assessed are those that maintain the behavior. The distributions of scores may reflect the differences between the tests. SOGS scores are highly positively skewed, with 92.5% of the Nevada respondents and 71.3% of the Illinois respondents falling below the cutoff score of five. GFA Total scores are more normally distributed, reflecting a range of functions maintaining gambling behavior among those who gamble, though, not necessarily, pathologically. Given that the two instruments measure different constructs, the more modest of the correlations might be expected. However, for the GFA to be useful (valid) as a diagnostic instrument, it

should be able to discriminate the same populations as the SOGS. That is, it should be able to discriminate between pathological and nonpathological respondents. The current clinical definition of pathological gambling (i.e., “persistent and recurrent maladaptive gambling behavior”) suggests many possible assessment approaches. One, a purely clinical and empirical approach, focuses on the maladaptive outcomes of the problem behavior. Such an approach, exemplified by the SOGS (Lesieur & Blume, 1987), catalogs negative consequences in close relationships, financial problems, time investment, etc. but does not address the reasons for the behavior’s persistence and recurrence. This emphasis ties the test closely to DSM diagnostic criteria, which often avoid defining disorders using any single theoretical model (i.e., the SOGS is atheoretical, consistent with its

70



71

72


origin in the pointedly atheoretical criteria of the DSM). Another, theoretically based, approach emphasizes the proposed underlying causes of the behavior and the mechanisms of maintenance. Use of the GFA (Dixon & Johnson, 2007), and its underlying behavioranalytic theoretical perspective, emphasizes reinforcing consequences. This theory-based approach has added value for clinicians, as the diagnostic indicators suggest theoretically relevant and practical targets for intervention. In other words, identification of the mechanisms maintaining a behavior is also, by definition, identification of the means for changing it. By drawing distinctions between the descriptive and theoretically driven assessment approaches, we do not mean to suggest that the two are somehow contrary or incompatible. Any such suggestion would be moot, given the need for diagnostic schemes that may be applied irrespective of theoretical orientation, and the universal acceptance of the DSM system for classifying pathology. Theoretically-based methods, such as the GFA, may serve as a means of bridging the gap between diagnosis and treatment, clarifying the intervention targets by exposing the means of maintenance. Further research will be needed to explore the utility of the GFA as a treatment-planning tool. A useful first step would be to correlate GFA scores with various outcomes in treatment for gambling addictions, such as indicators of treatment compliance, symptom reduction or remission, and post-treatment relapse. Data from the current study support the concurrent validity of only one GFA component, Escape, relative to the SOGS, i.e., as a diagnostic indicator. Performance differences across the two samples are enlightening. In the Illinois sample, the base rate of gambling pathology, as measured by the SOGS, was much higher than in the Nevada sample, and much higher than estimates for

the general population (APA, 1994; Petry, 2005). In this way, the Illinois sample was the closer of the two to a ‘clinical” sample, where the base rate of pathology would be expected to be higher than in a general, nonclinical group. In this sample, the GFA Escape score performed better than other GFA content scores. SOGS and GFA Escape scores shared about 37% of variance (r = .61, the highest overall). Correlations of this magnitude are not uncommon for measures of similar, though distinct, constructs like those measured by the SOGS and GFA. For example, Verbal and Performance IQ scores of the Wechsler Adult intelligence Scales, 3rd Edition, correlate at .68 to .80, depending on the age of the subject (Tulsky, Zhu, & Ledbetter, 2002). Indicators of substance abuse from the Minnesota Multiphasic Personality Inventory, 2nd Edition (Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989), the MacAndrew Alcoholism Scale— Revised and the Addiction Admission Scale, correlate at r = .48 (Greene, 1999). GFA Escape and SOGS scores were distributed similarly, with most respondents in the ostensibly non-clinical sample endorsing few items, if any, on either. This similarity in distribution contributed to the comparatively good sensitivity and specificity (in the Nevada sample) of the Escape cutoffs scores. While the higher base rate in the Illinois sample, relative to the Nevada sample, would be expected to contribute as well, performance did not improve for all of the GFA content scores. Analysis of GFA diagnostic efficiency using SOGS ≥ 5 as criterion (Tables 3, 4, 5) indicated that the Escape subscale most accurately replicated SOGS-based classification. Escape was the only GFA score to meet Meehl and Rosen's (1955) criterion for efficiency (Table 4). That is, it was the only score to predict SOGS-based categories better than prediction by the base rate alone.

Joseph C. Miller et al. This occurred in the Illinois sample, where, as stated earlier, the base rate was much higher than typically observed in nonclinical settings (Petry, 2005). "Efficiency" does not necessarily equal clinical utility, however. Clinicians may use test scores for different purposes (e.g., to "rule out" or "rule in" a diagnosis) for which different types of errors are more or less tolerable. Depending on the intended use, other accuracy indicators may be of greater interest to clinicians. In the Illinois sample, PPP at Escape≥14 was .91, meaning that, in this sample, there was a 91% chance that a positive result on GFA Escape would be confirmed by SOGS ≥ 5. At this same cutoff, there was a 79% chance that a negative finding (Escape < 14), or rule-out, would be confirmed by SOGS < 5 (NPP = .79). Specificity was excellent at this same cutoff (.99), while sensitivity was poor (.35). These data suggest that, with base rates similar to those found in clinical settings, Escape ≥ 14 is a highly conservative (resulting in an acceptably low probability of false positive results) threshold for identifying probable gambling pathology, as defined by the SOGS. These findings must be considered tentative because of the nonclinical nature of the sample and its limited size. In the Nevada sample, where the base rate was much closer to that of the general population, a curoff as low as Escape≥2 yielded acceptable sensitivity (.80) and specificity (.76) and excellent NPP (.98). PPP, however, was poor (.21), owing to the low base rate and the test's specificity. No Escape cutoff score met efficiency criteria at this lower base rate. As mentioned above, factor analysis supports Escape as the only GFA measure of negative reinforcement, and it is quite possible that negative versus positive reinforcement contingencies may be critical to the etiology of pathological gambling (Miller et al., 2009). Morasco, Weinstock, Ledgerwood, and Petry. (2007) reported that patho-

73

logical gamblers in treatment indicate negative reinforcement as an important contributor to maintenance of their gambling behavior. The Illinois data, though not a clinical sample, suggest that the GFA Escape score may be useful in identifying pathology in a clinical setting (e.g., among patients referred for gambling problems or who report distress or impairment related to their gambling behavior). A study of diagnostic efficiency within a true clinical population, where independent confirmation of diagnoses is available, will be needed to verify this possibility. In the Nevada sample, with roughly one quarter of the Illinois sample’s base rate of potential pathological gambling, performance of the GFA relative to SOGS was poorer than in the Illinois sample. While convergent correlations were less variable than in the Illinois sample, none of the coefficients matched the magnitude of the GFA Escape score. As the SOGS is a “screen,” these results may not be surprising. The SOGS has been used in large research studies to establish prevalence rates among sectors of the general population, where base rates are low (e.g., Gill, Dal Grande, & Taylor, 2006; Philippe & Vallerand, 2007), and has demonstrated its effectiveness in these contexts. The current data suggest that the GFA may not be as useful as the SOGS in this capacity. Further validation will be necessary to establish the GFA Escape score as a reliable indicator of pathology, though the data collected to date are mixed. The Escape score performed better where the base rate of SOGS-defined pathology was highest, suggesting it may not perform well as a screening for pathology in community samples. While the Sensory, Attention, and Tangible scores do not appear to measure SOGSidentified probable pathology to the extent that the Escape score does, these components of the GFA may still have some clini-

74


cal, if not diagnostic, utility. If the GFA Escape score proves to discriminate well between real pathological and nonpathological cases in future studies involving clinical populations, other GFA content scores may be useful in treatment planning by assisting in the identification of salient maintenance functions for persons whose gambling behavior has already been deemed pathological. At present, however, evidence for the diagnostic utility of the positive reinforcement functions assessed by the GFA is very limited. As with the majority of clinical disorders, the diagnosis is only a first step towards successful treatment and recovery for the person suffering from the affliction. For over 20 years, the SOGS has provided researchers and treatment providers with a means of easily assessing the severity of gambling for a given individual. However, syndromal classification is only the beginning. Afterwards, the clinician needs ways to understand, assess, and eventually treat reasons for why individuals continue to gamble when the odds of winning are surely against them. A function-based approach has yielded an effective means by which to discover the heterogeneity of specific clinical populations, and it appears promising that such an approach will yield great benefits for the field of pathological gambling treatment. The GFA is a promising assessment device, and with it, perhaps the odds of effective treatment will become just a bit more favorable. REFERENCES American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders 4th Edition (DSM-IV). Washington, D.C.: Author. Anastasi, A. & Urbina, S. (1997). Psychological Testing, 7th Ed. Upper Saddle River, NJ: Prentice Hall.

Arthur, D., Tong, W.L., Chen, C.P., Hing, A.Y., Sagara-Rosemeyer, M., Kua, E.H., Ignacio, J. (2008).The validity and reliability of four measures of gambling behavior in a sample of Singapore university students. Journal of Gambling Studies, 24, 451-462. Butcher, J.N., Dahlstrom, W.G., Graham, J.R., Tellegen, A. & Kaemmer, B. (1989). Manual for the Restandardized Minnesota Multiphasic Personality Inventory; MMPI-2. An Administrative and Interpretive guide.Minneapolis, MN: University of Minnesota Press. Cronbach, L.J. (1988). Five perspectives on the validity argument. In H. Wainer &H.I. Braun (Eds.) Test Validity (pp. 3-18). Hillsdale, NJ: Erlbaum. Dixon, M.R., & Johnson, T.E. (2007). The gambling functional assessment (GFA): An assessmentdevice for identification of the maintaining variables of pathological gambling. Analysis of Gambling Behavior, 1, 44-49. Durand, V.M. & Crimmins, D.B. (1988). Identifying the variables maintaining self injurious behavior. Journal of Autism & Developmental Disorders, 18, 99-117. Gambino, B. & Lesieur, H. (2006). The South Oaks gambling Screen (SOGS): A rebuttal to critics. Journal of gambling Issues, 17, 116. Gill, T., Dal Grande, E., & Taylor, A.W. (2006). Factors associated with gamblers: A population-based cross-sectional study of South Australian adults. Journal of Gambling Studies, 22, 143-164. Greene. R. (1999). MMPI-2: An interpretive Manual (2nd Ed.). Needham Heights, MA: Allyn & Bacon. Groth-Marnat, G. (2003). Handbook of psychological assessment. (4th Ed.). New York: John Wiley & Sons. Iwata, B. A., Dorsey, M. F., Slifer, K. J., Bauman, K. E., & Richman, G. S. (1994). Toward a functional analysis of self-injury. Journal of Applied Behavior Analysis, 27, 197-209.

Joseph C. Miller et al. Jimenez-Murcia, S., Stinchfield, R., AlvarezMoya, E., Juarrieta, N., Bueno, B., Granero, R., Aymami, M.N., Gomez-Pena, M., Martinez-Gimenez, R., Fernandez-Aranda, F., & Vallejo, J. (2009). Reliability, validity, and classification accuracy of a Spanish translation of a measure of DSM-IV criteria for pathological gambling. Journal of Gambling Studies, 25, 93-104. Kamphuis, J.H. & Finn, S.E. (2002) Incorporating base rate information into daily clinical decision-making. In J/ Butcher (Ed.) Clinical Personality Assessment: Practical Approaches (2nd Ed.) pp. 257-268. New York: Oxford University Press. Lesieur, H.R. & Blume, S.B. (1987). The South Oaks Gambling Screen (SOGS): A new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144, 1184-1188. Meehl, P.E. & Rosen, A. (1955). Antecedent probability and the efficiency of psychometric signs, patterns, or cutting scores. Psychological Bulletin, 52, 194-216. Miller, J.C., Meier, E., Muehlenkamp, J. & Weatherly, J.N. (2009). Testing the construct validity of Dixon & Johnson’s (2007) Gambling Functional Assessment. Behavior Modification, 33, 156-174. Miller, J.C., Meier, E., & Weatherly, J.N. (2009). Assessing the reliability of the Gambling Functional Assessment Screen. Journal of Gambling Studies, 25, 121-129 Morasco, B.J., Weinstock, J., Ledgerwood, D.M., & Petry, N.M. (2007). Psychological factors that promote & inhibit pathological gambling. Cognitive & Behavioral Practice, 14, 208-217. Nunnally, J. & Bernstein, I. (1994). Psychometric Theory (3rd Ed.). NY: McGraw-Hill. Petry, N.M. (2005). Pathological Gambling: Etiology, Comorbidity, and Treatment. Washington, D.C.: American Psychological Association. Philippe, F. & Vallerand, R.J. (2007). Prevalence rates of gambling problems in Montreal, Canada: A look at old adults & the role of passion. Journal of Gambling Studies, 23, 275-284.

75

Piazza, C.C., Fisher, W.W., Brown, K.A., Shore, B.A., Patel, M.R., Katz, R.M., Sevin, B.M., Gulotta, C.S., & Blakely-Smith, A. (2003). Functional analysis of inappropriate mealtime behaviors. Journal of Applied Behavior Analysis, 36, 187-204. Sattler, J. (2001). Assessment of Children: Cognitive Applications. La Mesa, CA: Jerome M. Sattler. Publisher. Stinchfield, R. (2002). Reliability, validity, and classification accuracy of the South Oaks Gambling Screen (SOGS). Addictive Behaviors, 27, 1-19. Stinchfield, R. (2003). Reliability, validity, and classification accuracy of a measure of DSM-IV diagnostic criteria for pathological gambling. American Journal of Psychiatry, 160, 180-182. Tulsky, D., Zhu, J., & Ledbetter, M. (Eds.). (2002). WAIS-III/WMS-III Technical Manual. San Antonio, TX: The Psychological Corporation. Weatherly, J.N., & Dixon, M.R. (2007). Toward an integrative behavioral model of gambling. Analysis of Gambling Behavior, 1, 4-18. Action Editor: Andrew Brandt

76

Guest Reviewers F. Richard Ferraro, University of North Dakota David M. Ledgerwood, Wayne State School of Medicine Joelle Ruthig, University of North Dakota Randy Stinchfield, University of Minnesota Medical School

This journal is Copyright © 2010 by Jeffrey N. Weatherly, publisher, Analysis of Gambling Behavior. All rights are reserved. All information contained within is provided as is. The AGB journal, its publisher, authors, and agents, cannot be held responsible for the way this information is used or applied. The AGB journal is not responsible for typographical errors. Analysis of Gambling Behavior (AGB) is published twice per year (summer and winter). AGB is a print journal and back issues are available online. AGB is an independent publication and is in no way affiliated with any other publications. The materials, articles, and information provided in this journal have been prepared by the staff of the AGB journal for informational purposes only. The information contained in this journal is not intended to create any kind of patient-therapist relationship or representation whatsoever.

Make sure to subscribe to the newest and most scientifically advanced journal for understanding gambling behavior. Back Issues are Available 1 year subscription (2 issues published per year) $40 USD Professionals and Institutions $20 USD Students (with documented student Status) Send check or money order to: Analysis of Gambling Behavior Department of Psychology University of North Dakota Grand Forks, ND 58202-8380 Phone: (701) 777-3470 Fax: (701) 777-3454