tematic error in self-report measures resulting from the desire of respondents to avoid embarrassment and project a favorable image to others. Yet little is known.
Social Desirability Indirect Questioning
and the Validity of
ROBERT J. FISHER Indirect (i.e., structured projective) questioning has been employed frequently in marketing and other social sciences to reduce social desirability bias, that is, systematic error in self-report measures resulting from the desire of respondents to avoid embarrassment and project a favorable image to others. Yet little is known about the validity of indirect questioning in reducing social desirability bias. This article reports on three studies that examine indirect questioning as a technique to reduce social desirability bias on self-report measures. The effects of asking indirect (i.e., structured, projective) questions were compared with direct (i.e., structured, personal) questions. The pattern of results indicates that indirect questioning reduces social desirability bias on variables subject to social influence and has no significant effect on socially neutral variables. The social nature of the differences between direct and indirect questioning groups, and the attribution of an undesirable trait to an out-group but not an in-group target, supports the view that subjects projected their beliefs and evaluations in the indirect response situation. These results are consistent across several product categories and indirect question wordings.
M
affect variable means (Peterson and Kerin 1981), Research that does not recognize and compensate for social desirability bias may lead to unwarranted theoretical or practical conclusions about consumers' psychological traits (e.g.. Campbell 1950; Peltier and Walsh 1990); purchase motivations (e.g.. Levy 1981); and attitudes, intentions, and behaviors (e.g., Mensch and Kandel 1988). An important technique used by researchers to mitigate the effects of social desirability bias is indirect (i.e., structured, projective) questioning.' Indirect questioning is a projective technique that asks subjects or respondents to answer structured questions from the perspective of another person or group (Anderson 1978; Calder and Burnkrant 1977; Robertson and Joselyn 1974). For example, marketing researchers have asked subjects to predict the types of people most likely to eat hot cereal (Westfall. Boyd, and Campbell 1957); to indicate the extent to which "a consumer" would react to reference group influence (Bearden and Etzel 1982; Brinberg and Plimpton 1986; Park and Lessig 1977); to identify the emotional responses of the anonymous author of a story describing a past consumption experience (Havlena and Holbrook 1986); and to state the
uch of what we think we know about human behavior comes from self-report measures (Peterson and Kerin 1981). Unfortunately, the basic human tendency to present oneself in the best possible light can significantly distort the information gained from self-reports. Respondents are often unwilling or unable to report accurately on sensitive topics for egodefensive or impression management reasons. The result is data that are systematically biased toward respondents' perceptions of what is "correct" or socially acceptable (Maecoby and Maccoby 1954). This phenomenon is called social desirability bias and has been found to occur in virtually all types of self-report measures and across nearly all social sciences literatures (e.g.. Levy 1981; Peltier and Walsh 1990; Robinette 1991; Simon and Simon 1975; Zerbe and Paulhus 1987). Not only is social desirability bias pervasive, but it can lead to the reporting of spurious or misleading research results. Prior studies have found that social desirability bias can attenuate, inflate, or moderate variable relationships (Zerbe and Paulhus 1987); increase measurement error (cf. Cote and Buckley 1988); and *Robert J. Fisher is assistant professor of marketing. School of Business Administration. University of Southern California. Los Angeles. CA. 90089-1421. The author would like to thank Valerie S. Folkes. Jill A. Grace. J. Jeffrey Inman. Kent Nakamoto. Linda L. Price. Dennis W. Rook, and David W. Stewart for comments on earlier versions of this article. The author also benefited from discussions with Stephen J. Hoch and comments by the editor and three reviewers.
'In marketing research the emphasis has been on »/istructured projective methods such as thematic apperception tests, sentence completion, and word association (e.g., Levy 1981; Rook 1985; for a review see Kassarjian [1974]). Unlike the structured approach identified here, unstructured metJiods allow subjects much greater freedom in organizing and responding to the projective stimulus. Consequently, unstructured projective questioning is particularly valuable in helping subjects express latent motivations. 303 1993 by JOURNAL OF CONSUMER RESEARCH, Inc. • Vol. 20 • September 1993 All rights reserved. 0093-5301/94/2002-0010$2.00
304
JOURNAL OF CONSUMER RESEARCH
importance of various attributes in determining the choices of "most people" (Alpert 1971). Indirect questioning is thought to reduce the distortion of "private" opinions that are revealed to the researcher by asking respondents to "report on the nature of the external world" rather than about themselves (Westfall et al. 1957. p. 138). It is expected that respondents project their unconscious biases into ambiguous response situations and reveal their own attitudes (Campbell 1950; Holmes 1968; Sherwood 1981). For example, if a respondent predicts that "other people" drink and drive under certain circumstances the answer would be interpreted to mean that the respondent is likely to behave in this manner. Indirect questioning allows respondents to "describe their own feelings behind a facade of impersonality" (Simon and Simon 1975. p. 586). Despite good theoretical reasons to believe that indirect questioning compensates for social desirability bias, empirical research on the subject is limited and inconclusive. Only one prior study in marketing has empirically examined the validity of indirect questioning as a technique to reduce social desirability bias. Unfortunately, this study used unstructured indirect measures and structured direct measures, making interpretation ofthe results problematic (see Steele 1964). Other researchers have focused on topics such as the predictive validity of direct versus indirect methods (Alpert 1971) or methodological studies related to the design of projective stimuli (Anderson 1978; Robertson and Joselyn 1974). Beyond the lack of empirical verification to support the use of indirect questioning as a means of reducing social desirability bias, the extent to which indirect questioning provides information about the respondent is unclear. Rather than provide insights into the self, indirect questions may actually reveal what respondents predict a "typical other" might do or think (Maccoby and Maccoby 1954). For example, an adolescent might predict that his or her peers will shoplift under certain circumstances on the basis of knowledge of peer shoplifting and not because s/he is likely to shoplift him- or herself. The adolescent's response is an objective judgment that is independent of his or her own intentions and behaviors. The objectives ofthe three studies reported here are to evaluate indirect questioning as a means of reducing social desirability bias, to explore the potential for this method to add other forms of systematic error to selfreport measures, and to gain insights into the extent to which indirect questioning reveals information about the self. The following sections develop the research hypotheses, specify the research designs, present the findings, and discuss the results and implications of these experiments.
STUDY 1 The effects of questioning method on social desirability bias depend on the constructs under investigation
and the social expectations perceived by the respondent (Paulhus 1984). On this basis, hypotheses must be developed within a conceptual context that specifies the likely nature and direction of these effects. The present study employs the Miniard and Cohen (1983) model of behavioral intentions for this purpose because it partitions the determinants of behavioral intentions into variables that are likely to be differentially affected by social desirability bias. The Miniard and Cohen model specifies that behavioral intentions are a function of individuals' expectations about the personal and normative consequences of undertaking a behavior (see Miniard and Cohen [1983] for more details on this model). The personal component ofthe model is composed of an individual's beliefs about the likelihood that a behavior leads to salient personal outcomes and an evaluation of these outcomes. Salient personal outcomes may be positive or negative and are valued as ends in themselves, that is, they have intrinsic worth to the individual. These consequences are perceived without reference to the visibility of the considered behavior, the subsequent approval or disapproval of referent others, or the extent to which the behavior serves to affiliate one with a social group. Social desirability bias is unlikely to distort self-report measures of personal outcomes because this type of outcome is based on internalized values that are independent of social considerations. For example, the purchase of a new compact-disc player might be motivated by personal outcomes related to the durability of the compact disc's recording medium. This product characteristic provides both financial and psychological benefits or outcomes because valued recordings deteriorate less quickly than is the case with other media. The likelihood and importance of these outcomes do not depend on what others might think or do. Accordingly, self-reported personal outcomes should not be affected by social desirability bias because respondents are unlikely to feel threatened or embarrassed about accurately reporting their views. If personal outcomes are independent of social desirability bias, significant differences should not exist between direct and indirect measures of this variable. A test ofthe independence of personal outcomes to social expectations is achieved by an examination of the variable's behavior under anonymous and unanonymous response conditions. Removing anonymity increases the pressure on subjects to respond in a socially acceptable manner (cf. Kelman 1961). Given that personal outcomes are socially neutral, the anonymity of the research setting should not affect responses under either direct or indirect questioning. In both cases, subjects are reporting beliefs and evaluations that are independent of social influence. The normative component ofthe model is composed of individuals' beliefs about the likelihood a behavior leads to salient social outcomes and an evaluation of
SOCIAL DESIRABILITY BIAS .
these outcomes. This component reflects the extent to which individuals are motivated by the expectations of another person or group. The individual, by fulfilling (or acting contrary to) the expectations of others, anticipates socially mediated rewards (or punishments). The individual does what a referent wants him or her to do because it is instrumental in achieving a social purpose. Social desirability bias is likely to affect self-reported normative outcomes because the pursuit of social approval as a purchase motivation is inconsistent with social norms and expectations. For example, "shopping list" research by Haire (1950) and replications or extensions by others (e.g., Robertson and Joselyn 1974) suggest that consumers are unable or unwilling to express the extent to which they are motivated by social factors. When asked directly, consumption motivations tend to focus on the functional aspects of a product such as taste or gas mileage (Haire 1950). Consumption motivations such as social positioning and the approval of others are revealed only with indirect questioning. Research on American values also indicates that the overt pursuit of social approval through consumption is undesirable. Americans consistently assign a low rank to social recognition (the respect and admiration of others) and a high rank to independence as cultural values (Rokeach 1979, p. 133: Rokeach and Ball-Rokeach 1989, p. 778). Of 18 terminal values, social recognition was ranked 17th in 1971 and 18th in 1974 and 1981, while independence was ranked third in all three studies. Given that self-reported values represent culturally desirable ideals (Rokeach 1979), these rankings imply that expressing one's desire for social approval is inappropriate in American culture. Accordingly, socially desirable responding behavior should cause normative outcomes to be understated with direct questioning because subjects are inclined to present themselves in a way that is consistent with social expectations. Under indirect questioning, however, subjects are able to disengage themselves from the social implications of their responses, leading to larger estimates of normative beliefs and evaluations. Subjects should most actively engage in impression management when they believe their responses will be used as evidence about themselves and not as objective statements about the "real world." The anonymity of the research setting should interact with questioning method to have a differential effect on self-reported normative outcomes. Removal of anonymity should cause subjects answering direct questions to feel greater pressure to make their responses consistent with social expectations. These responses are most likely to have symbolic or communicational properties because they represent the personal thoughts and feelings of the subject on a socially sensitive topic and the responses are visible to others. However, eliminating anonymity is not expected to have the same effect on subjects answering indirect questions because they are
305 distanced from their responses by the projective method. A summary of study hypotheses related to the effects of questioning method and anonymity on mean scores follows: Hla: Method of questioning has no effect on mean personal outcomes because of the social neutrality of this variable. , Hlb: Indirect questioning reduces social desirability bias, resulting in higher mean normative outcomes. H2a: Anonymity has no effect on mean personal outcomes because of the social neutrality of this variable. H2b: Removing anonymity increases social desirability bias, resulting in lower mean normative outcomes under direct questioning. Method A 2 (questioning method) X 2 (anonymity) betweensubjects experiment was performed to test the hypotheses. A convenience sample of 184 male and female undergraduate students was selected and randomly assigned to the conditions (46 per cell). A student sample was selected because of the widespread study of this population in consumer research on normative influence and other socially sensitive topics. Subjects were contacted within a normal classroom situation and told that the purpose of the study was to generate information to be used in the advertising campaign for a new product targeted at college students. A fictional new product was selected to avoid the influence of prior beliefs on subjects' responses. A pretest indicated the new product, an innovative stereo headphone designed for use with Walkman-type cassette or compact-disc players, was believable and of interest to those in the target market." Moreover, the pretest indicated that students perceived that adoption of the new product had significant social implications because of its visibility and innovative features. To enhance involvement, subjects were shown a mock-up of the new product, a professionally designed brochure layout, and entered in a drawing for three $50 prizes for participating in the study. The experimental manipulations took the form of instructions to subjects and question wording.-* In the direct questioning condition, subjects were asked to respond to a series of items in terms of their own beliefs and evaluations. In the indirect questioning condition, subjects were asked to predict the likely responses of "a typical college student." Item wording differed be^The new headphone design was described as similar in size and weight lo those currently used with portable stereo systems, except that the new design does not require wires connecting the headphone set to the player unit. 'All manipulations and measures are available from the author.
306
Iween conditions only to the extent that third-person wording was substituted for first-person wording in the indirect condition. For example, the words "the typical college student will . . ." were substituted for "I will . . ." in the indirect condition. The items measuring the constructs of interest were randomly mixed with other items unrelated to this study. Anonymity was manipulated with information about the likelihood subjects would be asked to discuss their responses with a researcher after completing the instrument. Subjects in the unanonymous condition were told, "You may be asked to discuss your responses with a researcher when you have finished." In addition, subjects in this condition were asked to include their student identification number on the first page of the instrument. Subjects in the anonymous condition were told. "Your responses are completely anonymous," and they were not asked to include their student identification number.
Measurement After a review of the relevant literatures, 10 items were generated to measure the personal and normative outcome constructs. The items, worded in the first person {i.e., directly), were administered as part of a larger data collection efi'ort to 90 graduate and undergraduate students at another university. Two items were deleted by corrected item-to-total correlations and principal axis factor analysis with varimax rotation. The analyses resulted in eight items measuring the two constructs of interest. Measures of the constructs were designed specifically for the student sample and new headphone product. For personal outcomes, a pretest on an unrelated sample indicated that one of the new product's key advantages was the increased freedom of movement it afforded. This attribute was identified by pretest respondents as being intrinsically valued, that is. desirable regardless of the social aspects of consumption. Consequently, beliefs and evaluations of personal outcomes were measured with items such as "the new headphones provide more freedom of movement than other headphones." For normative outcomes, subjects were asked to indicate their beliefs and evaluations of statements such as "students I know would have a favorable reaction if I bought one of the new products." The belief components of the outcome variables were measured with seven-point "highly likely" to "not at all likely" scales. The evaluation components were measured with sevenpoint "very important" to "very unimportant" scales. For hypothesis testing, the multi-item scales were summated to form a single indicator of each construct.
Results Measurement. Confirmatory factor analyses were conducted for the individual two-factor models within
JOURNAL OF CONSUMER RESEARCH TABLE 1 SCALE CHARACTERISTICS: STUDY 1
Group measure Direct {n = 92): Personal outcomes Beliefs only Evaluations only Normative outcomes Beliefs only Evaluations only Indirect {n ^ 92): Personal outcomes Beliefs only Evaluations only Normative outcomes Beliefs orly Evaluations only
No, of items
Actual scale range
Scale mean
SD
Alpha
4 4 4 4 4 4
47-144 12-24 15-24 0-105 0-22 0-24
109,4 20,4 21.3 36,4 12.7 92
23.2 2.9 2.6 29.6 5.8 6.4
79 .68 .73 .91 .91 .92
4 4 4 4 4 4
42-144 9-24 11-24 0-132 5-22 0-24
111.8 20.6 21.4
27.1 3.3 2.9 31.6 4.0 5.2
,81 .75 .79 .89 .82
65.8 15.9 15.3
,88
NOTE,—Personal outcomes and normative outcomes were measured as the sum of the product of beliefs (i.e.. "not at all likely" to 'highly likely") and evaluations (i,e., "very unimportant" to "very important'). All scales were scored from zero to six. Higher numbers indicate higher levels of each variable.
the direct and indirect questioning groups. Both models have acceptable internal and external consistency given an overall X" of 34.25 (df^ 19, p = .017) with a goodness-of-fit index (GFI) ^ .92 for the direct group and a X^ of 33.45 {df= 19. p = .021) with a GFI = .92 for the indirect group. The normed-fit index (Bentler and Bonett 1980) indicates that 91 percent of the observedmeasure covariation is explained in each of the measurement models. The internal consistency of the measures is supported by three additional tests. First, the scales exhibited good internal consistency with coefficient alphas for personal outcomes of .79 (direct) and .81 (indirect) and fornormative outcomes of .91 (direct) and .89 (indirect). Second, variance-extracted estimates reflect the amount of variance captured by a measure relative to random measurement error (Fornell and Larcker 1981). Ail measures had variance-extracted estimates that approached or exceeded the .50 level with values of .48 (direct) and .53 (indirect) for personal outcomes and .71 (direct) and .66 (indirect) for normative outcomes. Finally, all indicator /-values in the confirmatory factor analysis exceeded 5.0 (p < .001) in both groups. See Table 1 for a summary of scale characteristics. Manipulation Checks. Data from the experiment were analyzed with an ANOVA. First, the anonymity manipulation was evaluated via a summated four-item manipulation check on perceived anonymity. Items included "My responses on this survey can be traced back to me" (reverse scored). The manipulation was successful given a significant main effect of the manipulation on perceptions of anonymity (/^( 1,180) ^ 90.42, p< .001) with the anonymous mean {X = 17.17) higher
307
SOCIAL DESIRABILITY BIAS
than the unanonymous mean (X = 10.96). No unintended main or interaction effects were found. Second, the anonymity manipulation may have caused subjects to feel more accountable for their responses than students in the unanonymous condition, thus causing subjects in this group to respond more carefully (cf. Tetlock and Kim 1987). To evaluate this possible confound, a summated four-item manipulation check was used to measure subjects' perceptions about the care with which they responded. Items included "1 was careful when I answered the questions on this survey." No significant effects (/? > .10) were found for anonymity, questioning method, or anonymity X questioning method on response care. Tests of Hypotheses. Hypothesis la was supported with no main effect of questioning method on personal outcomes (F(l,180) ^ .43, NS), with similar means in both the direct {X = 109.4) and indirect (A' = 111.8) conditions. Hypothesis lb was supported with a main effect of questioning method on normative outcomes (F{ 1.180) = 43.15, /J < .01), with the indirect questioning mean (A* = 65.8) higher than the direct questioning mean {X = 36.4). These two results indicate that subjects made the same evaluations for a socially neutral variable across questioning conditions but reported significantly lower evaluations for a socially sensitive (undesirable) purchase motivation when asked directly. The lack of a main effect for anonymity on personal outcomes with indirect questioning predicted in Hypothesis 2a was supported (F(M80) = .02, NS), with similar mean scores across the anonymous {X = 110.3) and unanonymous (A' = 110.8) conditions. This result suggests that personal outcomes are independent of social influence because subjects made the same evaluations for themselves as for others, regardless of the anonymity of their responses. Hypothesis 2b stated that a reduction in anonymity will lower mean normative outcomes only in the direct questioning condition. A simple-effects test revealed a significant reduction (F( I. i 80) = 3.97, p < .05) in mean normative outcome scores in the direct questioning condition (.^anonymous - 42.7 and .V^n^inonymoas ^ 30.1), but not in the indirect questioning condition (JC^^^nymous ^ 61.1 and Vunanonymous = 70.5). These differences were also reflected in a significant anonymity X questioning method interaction (/•( 1,180) ^ 6.04, p < .05). There was no main effect of anonymity on normative outcomes (F( 1,180) = . 13, NS), with similar means across the anonymous {X = 51.9) and unanonymous (A" = 50.3) conditions. These results indicate the social sensitivity of normative outcomes given the lower mean scores that occurred in the unanonymous, direct cell. The results are summarized in Table 2 and Figure 1. Table 2 also includes the effects of the manipulations on the individual beWei' and evaluation components of the outcome variables. In terms of the individual components. Table 2 reveals significant effects for questioning method on normative
beliefs and evaluations and the anonymity X questioning method interaction only for evaluations. The larger effect for the evaluations component suggests it is more acceptable to predict that others are likely to approve of a consumption behavior than it is to indicate that this outcome is important.
Discussion The results of this experiment suggest that indirect questioning operates to mitigate social desirability bias and does not systematically affect the means of variables that are independent of social influence. Overall, the pattern of effects implies that subjects projected their beliefs and evaluations when responding to indirect questions. The specific findings of study 1 are discussed below. The lack of questioning method and anonymity effects on self-reported personal outcomes is evidence that personal outcomes are independent of social influence and that indirect questioning does not systematically bias mean scores of variables of this type. First, the lack of an anonymity effect indicates that subjects* self-reports of personal outcomes were not influenced by the threat of social pressure. This finding is consistent with the asocial nature of personal outcomes and demonstrates that subjects were not concerned about the symbolic or communicational aspects of their responses (cf. Miniard and Cohen 1983). Second, subjects made very similar predictions in the direct and indirect questioning groups for personal outcomes. The lack of a main effect for questioning method on personal outcomes suggests that subjects projected their own beliefs and evaluations into the ambiguous response situation. That is, no systematic upward or downward bias resulted from asking subjects to make predictions about typical others as op' posed to making evaluations in the first person. Indirect questioning, by contrast, produces a very different pattern of responses for self- and typical-other questions. Mean estimates of normative outcomes varied across questioning-method conditions, and removing response anonymity lowered mean normative outcomes with direct questioning. The main effect of questioning method on normative outcomes combined with the absence of a main eflect for personal outcomes implies that social desirability bias is operating on selfreported normative outcomes. The sociai nature of the effect is evidence that the results are not due to global variations in how subjects evaluated these outcomes for themselves and others. If subjects simply make different self- versus other predictions, mean differences should result for both personal and normative outcomes across questioning methods. The significantly lower normative outcome mean in the unanonymous. direct questioning cell supports this interpretation because removing anonymity only affected the mean of the socially sensitive variable. Subjects evidently felt the need
308
JOURNAL OF CONSUMER RESEARCH TABLE 2 ANQVA RESULTS: STUDY 1 Means
Measure and source Personal outcome: Beliefs X evaluations Anon Oues Anon X ques Beliefs only: Anon Ques Anon X ques Evaluations only: Anon Ques Anon X ques Normative outcomes: Beliefs X evaluations Anon Ques Anon X ques Beliefs only: Anon Ques Anon X ques Evaluations only: Anon Ques Anon X ques
df
1 1
Sum of squares
F
Anonymous and direct
Unanonymous and indirect
13.6 272.7 869.6
.02 .43 1.37
110.3 109.4
110.8 111.8
1
.2
1
2.2
20.5 20,4
20.4 20.6
1
26.6
.02 .23 2.81
1
1.4 .2 .2
.18 .03 .03
21.3 21.3
21.4 21.4
1
1 1
' •" •
1 t 1
115.8 39.619.6 5,544,0
.13 43.15" 6.04-
1 1 1
1.6 453.9 35.7
.06 18.30" 1.44
.09
1
52 2 1,704.4 244.3
1.62 52.777.56"
.21
1 1
.18 .02
'I
51.9 36.4
50.3 65,8
14.2 12.7
14.4 15.9
12.8
11.7 15.3
9,2
.03
NOTE.—n = 184 for Study 1. The independent variables are the djchotomous manipulations while the dependent variables are the continuous manipulatjon checks. Anon, manipulated anonymity: ques, manipulated questioning method. Although the hypotheses relate to the overall personal outcome and normative outcome variables, the effects of the manipulations on the individual belief and evaluation components are included to provide additional detail.
•p < .05. "p < .01.
'
to manage their responses when answering directly worded questions on a socially sensitive variable in an unanonymous situation.
STUDY 2 A second experiment was designed to investigate the effects of indirect questioning on the relationships between personal and normative outcomes and intentions. This study was undertaken for two reasons. First, many consumer researchers on socially sensitive topics are interested in structural orassociational evidence rather than mean scores {e.g., Alpert 1971; Fishbein and Ajzen 1975). This investigation could not be carried out in study 1 because a Box's M-test revealed significant differences in the covariance matrices across anonymity conditions within the indirect group (X" ^ 13.67, df = 3, /? < .01). As a consequence, it would be inappropriate to pool the anonymous and unanonymous covariance matrices within each questioning group. Second, it is desirable to establish the reliability of the key findings of study 1 through a partial replication.
'
Consistent with study I, social desirability bias should differentially affect variables that are neutral and sensitive to social influence. Again using the Miniard and Cohen (1983) model of behavioral intentions, the relationship between personal outcomes and intentions is expected to be unaffected by questioning method. Subjects should not be motivated to "manage" the relationship between persona! outcomes and intentions because their responses are independent of social expectations. In the present context, there appear to be no relevant norms governing beliefs and evaluations of the freedom of movement offered by the new headphones. No differences should exist in the relationship between personal outcomes and intentions across questioning methods. The estimated effect of normative outcomes on intentions, however, should be attenuated by social desirability bias when direct questioning is used. Subjects may consciously or unconsciously engage in image management by distorting the association between their normative outcome and intention scores. For example,
SOCIAL DESIRABILITY BIAS FIGURE 1 PERSONAL AND NORMATIVE OUTCOME BY OUESTIONING METHOD AND ANONYMITY
150140130'ersonai Outcomes n o -
Indirect
•
0
— = = 2Direct
10090BO7060-
I
No
Yes
Anonymity
Anonymity
subjects may lower normative outcome scores when intentions are high to avoid giving the impression they are motivated by social approval. The relationship between normative outcomes and intentions is suppressed because social desirability bias adds systematic variance to the normative outcome variable that is unrelated to behavioral intentions (cf. Conger 1974). Formally, H3a: The association between persona! outcomes and intentions is the same with direct and indirect questioning because the variance common to these variables is not subject to social desirability bias. H3b: Indirect questioning removes social desirability variance from normative outcomes, resulting in a stronger association between normative outcomes and intentions under this form of questioning.
Method A two-group experiment was conducted in which the single factor was questioning method. Data collection procedures, the new product example, and the manipulation of questioning method remained the same as in study 1. Male and female undergraduate students were randomly assigned to either the direct {u = 170) or indirect (n = 182) groups. This sample was drawn from a different university than that of study 1.
Measurement Measures of persona! and normative outcomes were the same as those used in study 1. Intentions to adopt
early were measured with one indicator composed of the summation of four self-report items. The four-item directly worded scale included items such as "I will probably purchase one ofthe new products soon after they are on the market," measured on a seven-point "strongly agree" to "strongly disagree" scale. The indirect scale asked subjects to respond in terms of the typical student. The coefficient alphas for this scale were .93 (direct) and .81 (indirect). The means were 13.5 (direct) and 14.9 (indirect).
Results Before testing the formal hypotheses, an analysis of personal and normative outcome mean scores reveals the same pattern of differences between groups found in study 1. Specifically, the means for personal outcomes are not significantly different across the direct and indirect questioning groups (A'dircci = 96.7, .findireci = 101.2, t = 1.5. NS), while the mean for normative outcomes is significantly higher in the indirect group (Xa.tcci = 39.4, X.^direa ^ 63.2, / - 8.72. /J < .001, one-tailed test). As in study I. these results suggest a systematic social desirability effect on normative outcome mean scores with direct questioning. Tests of differences in the estimated effects of personal and normative outcomes on intentions in study 2 were undertaken in two ways. First, given differences in the ratio of the standard deviations for the predictor and criterion variables across questioning groups, the appropriate test of differences in the relationship between outcomes and intentions is a comparison of productmoment correlations across subgroups (Arnold 1982). On the basis of Fisher's Z-transformation, Hypothesis
310
JOURNAL OF CONSUMER RESEARCH
3a is supported with no significant difference in the correlation between personal outcomes and intentions across groups (rawca = -432, r.ndirect = -493, z = .77, NS). Hypothesis 3b is supported with a significantly higher correlation between normative outcomes and intentions in the indirect group (rdirea = -269. rindireci ^ -442, z = 1.81, ;? < .05. one-tailed test). Second, an ordinary least squares path analysis model was run for each questioning-method group in which intention to early adopt was regressed on personal and normative outcomes. This analysis was performed to compare the parameter estimates that might result from path-analytic studies under different questioning methods (see Table 3). Congruent with the study's hypotheses, the standardized beta weight for normative outcomes on intentions is higher in the indirect (/? = .332, / = 5.19, p < .001) compared to the direct (^ = .205, t - 2.94, p