did not examine the potential moderating role of the stability of PBC which is the other proximal .... Questionnaire measures of the following constructs were taken at both time points. ... The mean of these items produced a composite scale.
British Journal of Social Psychology (2000), 39, 469±493 Printed in Great Britain # 2000 The British Psychological Society
469
Temporal stability as a moderator of relationships in the Theory of Planned Behaviour Mark Conner* School of Psychology, University of Leeds, UK
Paschal Sheeran, Paul Norman and Christopher J. Armitage Department of Psychology, University of She¬eld, UK
Temporal stability of behavioural intentions and perceived behavioural control (PBC) within the Theory of Planned Behaviour were examined as moderators of the cognition±behaviour relationships in two studies. Study 1 (N ¯ 201) examined attendance at health screening appointment (infrequently performed behaviour) using an objective measure of attendance. The impact of intentions and past behaviour on behaviour was moderated by intention stability. In addition, stable intentions were more strongly related to attitudes and past behaviour. Study 2 (N ¯ 407) examined eating a low-fat diet (frequently performed behaviour) using a self-report measure of behaviour. The impact of intentions on behaviour was moderated by intention stability, while the impact of PBC and past behaviour on behaviour was moderated by PBC stability. Stable intentions were more strongly related to attitudes, subjective norms and PBC. Stable PBC was more strongly related to attitudes and past behaviour. The discussion considers the implications of the ®ndings for furthering the understanding of the proximal determinants of behaviour and emphasizes the importance of temporal stability as a key measure of the strength of intentions and PBC.
The Theory of Planned Behaviour (TPB) has been applied to the understanding of numerous behaviours (Ajzen, 1988, 1991 ; Armitage & Conner, in press). Despite this widespread application the predictive power of the TPB is far from perfect. This study aims to further the understanding of the determinants of social behaviour and improve the TPB’s predictive power by examining the moderating role of temporal stability of behavioural intentions and perceived behavioural control (PBC). The research examined the ability of these two variables to moderate the relationship between behaviour and intentions, attitudes, subjective norms, PBC and past behaviour. In addition, the determinants of stable vs. unstable intentions and PBC are examined. * Requests for reprints should be addressed to: Dr Mark Conner, School of Psychology, University of Leeds, Leeds LS2 9JT, UK (e-mail : mc!psychology.leeds.ac.uk).
470
Mark Conner et al.
Theory of Planned Behaviour The TPB describes the proximal in¯uences upon a person’s decision to engage in a particular behaviour. The theory is an extension of the widely applied Theory of Reasoned Action (TRA ; Ajzen & Fishbein, 1980 ; Fishbein & Ajzen, 1975). The TPB proposed that behaviour is determined by intentions to engage in that behaviour and PBC. Intentions here represent a person’s motivation, conscious plan or decision to exert eåort to perform the behaviour. PBC is a person’s perception of the extent to which performance of the behaviour is within his or her control or is easy or di¬cult. The concept is similar to Bandura’s (1982) concept of self-e¬cacy. Intentions are determined by three variables : attitudes, subjective norms and PBC. Attitude here refers to the overall evaluation of the behaviour by the person. Subjective norms are beliefs about whether signi®cant others think he or she should engage in the behaviour. PBC determines both intention and behaviour because if intention remains constant, the eåort expended to bring a behaviour to a successful conclusion is likely to increase with greater PBC. In addition, to the extent that PBC accurately re¯ects actual control, it is predicted to directly in¯uence behaviour. Meta-analytic reviews of the TPB provide impressive support for the predictive power of the TPB in terms of the percentage of variance explained in behaviour and intentions (see Armitage & Conner, in press ; Godin & Kok, 1996 ; Sheeran & Orbell, 1998 ; Sheeran & Taylor, 1999 ; Sutton, 1998). For example, Godin and Kok (1996) reported that attitude, subjective norm and PBC accounted for 41% of the variance in intentions in 76 applications, while intentions and PBC accounted for 34 % of the variance in behaviours in 35 applications. While the predictive power of the TPB is impressive, meta-analytic reviews such as that by Godin and Kok indicate that approximately two thirds of the variance is unaccounted for by intentions and PBC. Some of this is likely attributable to measurement problems and other factors linked to the nature of the data (see Sutton, 1998) ; nevertheless, a signi®cant proportion of `explainable ’ variance remains unaccounted for. Two ways of accounting for this unexplained variance have been oåered in the literature : additional variables and moderator variables. The TPB is thought to be complete in that any other factors are believed to have their impact via their in¯uence on components of the TPB (Ajzen, 1991). However, several studies have demonstrated an impact of additional variables upon behaviour, even after the TPB variables have been taken into account (see Conner & Armitage, 1998, for a review). While several such variables have been examined, the strongest evidence is associated with the role of past behaviour. The in¯uence of past behaviour on current behaviours is an issue that has attracted a great deal of attention (see Eagly & Chaiken, 1993, for a review). It is argued that many behaviours are determined by one’s past behaviour rather than cognitions such as described in the TPB (Sutton, 1994). This argument is based on the results of a number of studies showing that past behaviour is the best predictor of future behaviour (Bagozzi & Kimmel, 1995 ; Conner, Warren, Close, & Sparks, 1999 ; Godin, Valois, & Lepage, 1993 ; Norman & Smith, 1995 ; van Ryn, Lyttle, & Kirscht, 1996). The present study examined the role of past behaviour as an additional variable in the TPB.
Temporal stability and the TPB
471
Another approach to improving the predictive power of the TPB has been to examine moderator variables. A moderator variable in¯uences relationships between other variables (Baron & Kenny, 1986). A small number of moderators of the intention±behaviour relationship have been examined in the context of the TRA}TPB. For example, Ajzen (1985) predicted that PBC should moderate the impact of intentions on behaviour such that intentions will have a weaker impact on behaviour when PBC is low than when PBC is high ; several studies have found support for these eåects (see Armitage & Conner, in press, for a review ; see also Conner & McMillan, 1999, for an opposite ®nding for a negatively evaluated behaviour). Verplanken, Aarts, van Knippenberg, and Moonen (1998) examined the moderating eåect of habit strength on intention±behaviour relationships. They found that as habit strength increases in predictive power, the power of intention in predicting behaviour decreases, and vice versa (see, however, Kashima, Gallois, & McCamish, 1993, for a contradictory ®nding). Finally, Sheeran and Orbell (1999) found that anticipated regret moderated the impact of intentions to play the lottery such that play was most likely when participants both intended to play and anticipated regretting it if they did not play. The present research examined the role of temporal stability of components of the TPB as potential moderators. In particular, the temporal stability of intentions and PBC (as the immediate determinants of behaviour) in moderating the impact of intentions and PBC on behaviour is examined. The next section describes the theoretical background to considering temporal stability as a moderator in the TPB. Temporal stability of cognitive variables Sheeran, Orbell, and Tra®mow (1999b) de®ne temporal stability as `the extent to which an attitude remains unchanged over time regardless of whether or not it is challenged ’ (p. 722). As Ajzen (1996) argues, `¼ to obtain accurate prediction of behavior, intentions and perceptions of behavioral control must remain reasonably stable over time until the behavior is performed ’ (p. 389) ; this is because (weak) intentions or PBC measured prior to performance of a behaviour may change as a result of new information or unforeseen obstacles to action, and consequently the original measures may not accurately predict behaviour. In addition, several authors have argued that temporal stability is the de®ning property of a strong and consequential attitude (Ajzen, 1996 ; Bassili, 1996 ; Doll & Ajzen, 1992 ; Erber, Hodges, & Wilson, 1995 ; Krosnick, Boninger, Chuang, Berent, & Carnot, 1993 ; Prislin, 1996). The importance of temporal stability of constructs to their power to explain behaviour has also begun to be recognized in other domains (Biesanz, West, & Graziano, 1998). Further, Sheeran et al. (1999b) argue that temporal stability represents an operative measure of (intention) strength (Bassili, 1996). Operative measures provide indirect measures of underlying cognitive processes (Greenwald & Banaji, 1995) and as such are less likely to be in¯uenced by self-presentation processes." As such, operative measures tend to be better predictors of criterion " Clearly self-presentation can always in¯uence self-report measures. However, as temporal stability is here based on a computation rather than a direct self-report, it seems less likely that such measures would be in¯uenced by selfpresentation processes.
472
Mark Conner et al.
variables (Bassili, 1996). Hence, one might expect the temporal stability of intentions and PBC within the TPB to be important moderators of the relationship between these variables and behaviour. Relatively few studies have examined how properties of intentions might moderate relationships with behaviour and, as far as the present authors are aware, only one study has examined how properties of PBC might determine its relationship to behaviour. For example, Bagozzi and Yi (1989) found degree of intention formation moderated the intention±behaviour and attitude±behaviour relationships. Similar results have been reported by other authors for certainty or con®dence in one’s intentions (e.g. Bassili, 1993 ; Pieters & Verplanken, 1993). Temporal stability has been addressed in just three studies. Bagozzi and Yi (1989) examined intention stability but failed to ®nd any evidence to support a moderating eåect on intention±behaviour or attitude±behaviour relationships. Doll and Ajzen (1992) compared the relative ability of temporal stability vs. accessibility to moderate the relationship between behaviour and attitude, PBC and intentions. There was evidence of stability but not accessibility acting as a moderator. In each case the eåect was such that greater stability was associated with stronger relationships between attitude, PBC, intentions and subsequent behaviour, although the simultaneous eåects were not tested. In addition to their contradictory ®ndings, both studies can be criticized for not assessing temporal stability prior to the participants actually performing the behaviour. This means that causal ordering has not been eliminated as a potential alternative explanation of the ®ndings in these studies. Sheeran et al. (1999b) overcame this di¬culty by obtaining measures of intentions at two time-points prior to the participants’ performances of the behaviour. Their study demonstrated that intentions with greater stability were more predictive of subsequent behaviour (studying over the winter vacation). However, Sheeran et al. did not examine the potential moderating role of the stability of PBC which is the other proximal predictor of behaviour in the TPB. In summary, while moderating eåects of stability have been demonstrated for both intentions and PBC, no study has satisfactorily examined their simultaneous eåects. Such examination would be valuable in order to determine which is the more powerful moderator and to discover how the relationship between past and future behaviour might be attenuated (see below). Understanding the role of past behaviour in the TPB As noted earlier, measures of past behaviour frequently oåer superior predictions of behaviour than behavioural intentions (e.g. Bagozzi & Kimmel, 1995 ; Bentler & Speckart, 1979 ; see Ouellette & Wood, 1998, for a review). These ®ndings are problematic for models such as the TPB because they suggest that the cognitive variables speci®ed in the model are not su¬cient to explain behaviours which have been performed frequently in the past (Ajzen, 1991). Relatively little research has examined the factors that might attenuate the relationship between past behaviour and behaviour (but see Orbell, Hodgkins, & Sheeran, 1997 ; Verplanken et al., 1998, for exceptions). Ajzen (1991) regards the role of past behaviour as a test of su¬ciency of the TPB and suggests that its eåects should be mediated by PBC in particular
Temporal stability and the TPB
473
because repetition of behaviour should lead to enhanced perceptions of control. This might be expected to be especially likely when PBC also possesses temporal stability. Sheeran et al. (1999b) demonstrated that intention stability moderated the past behaviour±behaviour relationship such that past behaviour was unrelated to behaviour when intentions were stable. In contrast, when intentions were unstable, past behaviour was the best predictor of subsequent behaviour. Sheeran et al. (1999b) argue that this is because changing an habitual behaviour requires strong intentions (see Ouellette & Wood, 1998) and temporal stability is a de®ning property of strong intentions. The present study examined the relative power of intention stability and PBC stability to moderate past behaviour±behaviour relationships. Ouellette and Wood (1998) distinguished behaviours on the basis of their frequency of performance in their review of the relative impact of past behaviour and intentions on behaviour. They argue that for well practised (frequently performed) behaviours in constant contexts, past behaviour will guide behaviour because the processes that initiate and control performance become automatic (` habituallycontrolled ’ behaviours). In contrast, for behaviours which are not well learned (infrequently performed) or are performed in unstable or di¬cult contexts, the impact of past behaviour on behaviour is assumed to be mediated by intentions (` consciously controlled ’ behaviours). While Ouellette and Wood’s meta-analytic review supports this view, in the present research it was anticipated that intention stability and PBC stability might moderate past behaviour±behaviour relationships for both infrequently and frequently performed behaviours. This was because it was considered that stability would be a stronger moderator of this relationship than simple frequency of past performance of the behaviour. Sheeran et al. (1999b) report data relevant to this issue. In their study, the temporal stability of intentions moderated past behaviour±behaviour relationships such that the signi®cant impact of past behaviour at low levels of intention stability disappeared at high levels of intention stability. However, Sheeran et al. examined only one combination from Ouellette and Wood’s categorization of behaviours in terms of frequency of performance, namely a frequently performed behaviour (studying over the winter vacation). The present research was designed to assess the moderation of past behaviour in a more systematic fashion. In particular, the present research examined two behaviours that diåer in their frequency of performance. Study 1 examined a behaviour that is performed annually (attending for a health screen), while Study 2 examined a behaviour that can be performed daily (eating a low-fat diet).#
Hypotheses The present study examines two extensions to the TPB. First, the role of temporal stabilities of both intentions and PBC were examined as potential moderators of the relationship between intentions, PBC and subsequent behaviour. It was predicted # Ouellette and Wood (1998) also specify the stability of context in which the behaviour is performed and the di¬culty of performing the behaviour as further moderators (p. 58). Neither of these variables were speci®cally addressed in the present studies, but see general discussion.
474
Mark Conner et al.
that intentions and PBC which possess greater temporal stability would be more predictive of behaviour compared to less stable intentions and PBC. Secondly, the role of past behaviour within the TPB was examined. Following Ouellette and Wood (1998), it was predicted that past behaviour would have a relatively weak impact on behaviour and intentions would have a relatively strong eåect for the infrequently performed behaviour (Study 1). Conversely, for the frequently performed behaviour (Study 2), it was expected that past behaviour would have a relatively strong eåect and intentions would have a relatively weak eåect on behaviour. However, it was also predicted that intention stability and}or PBC stability would moderate past behaviour±behaviour relationships such that stable intentions and}or stable PBC would attenuate the impact of past behaviour on future behaviour in both studies. Finally, the correlates of stable and unstable intentions and PBC were examined. STUDY 1 Study 1 examined attendance at health screening. This is an infrequently performed behaviour (annually) that one might expect to be under conscious control. Selfreport measures of the TPB variables and objective measures of past behaviour and behaviour were employed. By employing objective measures of behaviour, it was possible to test whether the moderating eåects of intention and PBC stability could be observed in data that were not contaminated by self-report biases. Method Participants and design The study design was longitudinal with data collected at four time points. The TPB data were collected by postal questionnaire at two time points separated by 12 months. Approximately one month after receiving each postal questionnaire, participants received an invitation to attend a health screening appointment at their local general practice in the next month. Patient records were then used to reliably determine who attended for a screening appointment in response to each invitation. A total of 201 patients from a single general practice completed and returned both questionnaires (out of 749 who received questionnaires; 27 % response rate) and were included in the study (mean age ¯ 35 years; range ¯ 31±42). These 201 did not diåer on the TPB variables from the other 132 respondents who only completed the ®rst questionnaire (F(4,328) ¯ 1± 93, p " ± 10). There were roughly equal numbers of male and female participants. Responses from the initial questionnaire were employed only to compute measures of intention and PBC stability. Attendance in response to the initial invitation for screening was used as a measure of past behaviour. TPB measures collected in the second questionnaire were used to predict behaviour. Attendance in response to the second invitation for screening was used as the measure of behaviour.
Measures Questionnaire measures of the following constructs were taken at both time points. These items were part of a larger set not reported here (see Conner & Norman, 1994 ; Norman & Conner, 1993, 1996). Intention. Intention to attend a health check was assessed using three items, each on a 7-point bipolar (® 3 to 3) scale. These were; `I intend to attend a health check if oåered the opportunity (extremely unlikely±extremely likely) ’, ` If you had the opportunity, how likely is it that you would attend a health
Temporal stability and the TPB
475
check (extremely unlikely±extremely likely) ?’, and ` If I was oåered a health check I would try to attend (extremely unlikely±extremely likely) ’. The mean of these items produced a composite scale. Cronbach’s a were ± 93 and ± 92 for Times 1 and 2, respectively; test±retest reliability was r ¯ ± 45 (p ! ± 001). Attitude. Attitude was assessed using a semantic diåerential scale. Respondents were presented with the sentence, ` Attending a health check would be ¼’. Eight pairs of adjectives were rated, each on a 7-point bipolar (® 3 to 3) scale : bad±good, harmful±bene®cial, worrying±reassuring, unpleasant±pleasant, unsatisfying±satisfying, negative±rewarding, and foolish±wise. The mean of the eight scales was used as a composite score (Cronbach’s a Time 1 ¯ ± 86, Time 2 ¯ ± 88; test±retest r ¯ ± 57 (p ! ± 001)). Subjective norm. The subjective norm measure was composed of two items each scored on a 7-point bipolar (® 3 to 3) scale : ` People who are important to me would (de®nitely advise me not to attend±de®nitely advise me to attend a health check) ’, `People who are important to me would (de®nitely not attend a health check±de®nitely attend a health check)’. Cronbach’s a were ± 76 and ± 79 for Times 1 and 2, respectively ; test±retest reliability was r ¯ ± 45 ( p ! ± 001). Perceived behavioural control. PBC was assessed using three 7-point unipolar ( 1 to 7) items: ` I could easily attend a health check if I wanted to (extremely unlikely±extremely likely) ’, `How much control have you over whether you attend a health check (very little control±complete control)’, `Attending a health check would be (di¬cult±easy)’. The mean of the three items was used in analyses. Cronbach’s a were ± 61 and ± 74 for Times 1 and 2, respectively; test±retest reliability was r ¯ ± 45 (p ! ± 001). Behaviour. Behaviour was determined reliably from patient records. Attendance to the two health check invitations was coded into attend (scored as 1ÐTime 1 : N ¯ 157; Time 2 : N ¯ 129) or did not attend (scored as 0ÐTime 1 : N ¯ 44; Time 2: N ¯ 72). Time 1 measures tap past behaviour, while Time 2 measures tap future behaviour. Intention stability. A number of indices of temporal stability has been reported in the literature. Traditionally, stability is evaluated based on the correlation coe¬cient between answers to repeated questions (Batista-Foguet & Saris, 1997), in other words test±retest correlations. In order to compute an individual diåerence measure this correlation needs to be computed for each individual. For example, Sheeran et al. (1999b) employed a within-participant Pearson correlation between repeated items employed at the two time points to tap the construct of interest. Campbell (1990) presents three alternative measures of temporal stability: the sum of absolute diåerences between the repeated items at the two time points (sum of absolute diåerences) ; the average absolute diåerence adjusted for maximum possible change (mean adjusted diåerence) ; and the number of items exhibiting any change (number items changing). A further possibility is the absolute diåerence between the sum of items used to tap a construct at each time point$ (absolute diåerence). The within-person correlation is probably the measure to be preferred because of its similarity to the traditional between-persons test±retest correlation used to tap stability. However, this measure is potentially problematic with a limited number of data points (fewer than 5) because of the tendency for all items within a time point to receive the same value, making the computation of a correlation impossible. This would result in the loss of approximately 50 % of respondents in the present data. In its place an index of stability was employed which was the mean of these other four measures of temporal stability (after standardizing each measure). This measure proved to be internally reliable (a ¯ ± 96) and signi®cantly correlated with the within-subjects correlation (r ¯ ± 48, p ! ± 001). In order to ensure that high scores on this stability measure represented higher levels of stability, the scores were reversed by subtracting the mean stability score (for an individual) from zero. PBC stability. PBC stability was assessed as 0 minus the mean stability score (based upon standardized values of the sum of absolute diåerences ; the mean adjusted diåerence ; the number items changing ; and the absolute diåerence). The four items were internally reliable (a ¯ ± 94) and this stability measure correlated signi®cantly with the within-participants correlation (r ¯ ± 48, p ! ± 001). $ The authors thank a reviewer for this suggestion.
476
Mark Conner et al. Results
TPB analyses Table 1 presents the means, standard deviations and intercorrelations for the Time 2 TPB measures, intention stability, PBC stability, past behaviour, and behaviour. Behaviour and intentions are signi®cantly associated with all other variables, with intentions being the strongest predictor of behaviour and subjective norms being the strongest predictor of intentions. Logistic regression of behaviour onto TPB variables (Table 2) indicated that intentions (Table 2, col. 1) and PBC (Table 2, col. 2) were the only signi®cant predictions.% This equation correctly classi®ed 79 ± 7 % of respondents into the attend}did not attend groups. The addition of past behaviour to the equation (Table 2, col. 3) indicated that a modest, but signi®cant additional percentage of respondents were correctly classi®ed. Examination of the unstandardized beta weights (B) indicated intention, PBC and past behaviour were all signi®cantly, and independently, associated with behaviour (Table 2, col. 3). Temporal stability as a moderator Table 1 indicates that the temporal stability of intentions and PBC were both signi®cantly positively correlated with behaviours ( ps ! ± 01), but were not correlated with past behaviour. Both stability measures showed signi®cant positive correlations with intentions ( ps ! ± 001), with the correlation between intention stability and intentions being substantial (r ¯ ± 62, p ! ± 001). This indicates that respondents with positive intentions to attend screening had more stable views compared to those with negative intentions. Moderated logistic regression analysis was employed to test the moderation hypothesis (Baron & Kenny, 1986) and mean-centred variables were used to minimize problems of multicollinearity (Aiken & West, 1991). Table 2 (col. 4) shows the results of these analyses. Intentions, past behaviour and the interactions between intentions stability and intentions (b ¯ ± 60, p ! ± 01) and between intention stability and past behaviour (b ¯ ® ± 108, p ! ± 05) were signi®cant predictors. The ®nal equation correctly classi®ed 82 ± 4 % of respondents.& Neither intention stability nor PBC stability were signi®cant direct predictors of behaviour. The nature of the signi®cant interactions were probed using simple slope analyses (Aiken & West, 1991) by examining the regression lines at three levels of the hypothesized moderator (intention stability), in other words the mean level and at one standard deviation above and below the mean (Fig. 1). For the intention ¬ intention stability interaction, simple slope analyses demonstrated how % Because the behaviour measure was dichotomous, logistic regression was deemed the most appropriate analysis. & There was concern that multicollinearity might still constitute a problem in the data despite mean-centring the data. It was therefore decided to examine the variance in¯ation factor (VIF) scores for each variable in the equation. In no case did the scores show evidence of multicollinearity problems. In addition, the analyses were rerun, entering one interaction term at a time after entering all other variables. This produced essentially identical results, with only the intention ¬ intention stability (B ¯ ± 45, p ! ± 001) and past behaviour ¬ intention stability (B ¯ ® ± 81, p ! ± 01) interactions being signi®cant.
* p!
±
±
01 ; *** p !
±
001.
Behaviour Intention Attitude Subjective norm Perceived behavioural control Intention stability PBC stability Past behaviour
05; ** p !
1. 2. 3. 4. 5. 6. 7. 8.
TPB variables 1± 00
1 ± 34*** 1± 00
2 ±
22** 60*** 1± 00 ±
3 ±
17* 67*** ± 58*** 1 ± 00 ±
4 ±
26*** 33*** ± 37*** ± 26*** 1 ± 00 ±
5 ±
7 ±
18** 26*** ± 29*** ± 24*** ± 61*** ± 34*** 1± 00 ±
201)
17* 62*** ± 41*** ± 42*** ± 25** 1 ± 00 ±
6
Table 1. Means, standard deviations and intercorrelations for Study 1 variables (N ¯
±
23** 19** ± 08 ± 15* ± 14* ± 09 ± 12 1± 00 ±
8
0 ± 78 2 ± 30 1 ± 86 2 ± 23 6 ± 08 0 ± 00 0 ± 00 0 ± 64
M
0 ± 41 1 ± 26 0 ± 97 1 ± 03 1 ± 18 0 ± 95 0 ± 92 0 ± 48
SD
Temporal stability and the TPB 477
* p!
±
Intention Attitude Subjective norm Perceived behavioural control Past behaviour BI stability PBC stability Intention ¬ BI stability Attitude ¬ BI stability Subjective norm ¬ BI stability PBC ¬ BI stability Past behaviour ¬ BI stability Intention ¬ PBC stability Attitude ¬ PBC stability Subjective norm ¬ PBC stability PBC ¬ PBC stability Past behaviour ¬ PBC stability BI stability ¬ PBC stability
1. 2.
05; ** p !
±
01 ; *** p !
±
19 ± 2*** 79 ± 4
0 ± 55***
B
®
0± 59** 0± 08 0± 32 0± 32*
B
6± 35 79± 7
001. Data were mean centred to aid interpretation of interaction terms.
Improvement in model v # Percent correctly classi®ed
3. 4.
Predictors
Step
®
5± 70* 79± 9
0± 56** 0± 11 0± 35 0± 29* 0± 91*
B
®
®
®
®
®
®
®
®
®
23± 3* 82± 4
1± 56*** 0± 00 0± 58 0± 27 0± 93* 1± 28 0± 30 0± 60** 0± 02 0± 06 0± 20 1± 08* 0± 06 0± 41 0± 37 0± 07 0± 42 0± 25
B
Table 2. Logistic regressions of behaviour onto TPB variables, past behaviour, intention stability, PBC stability and stability interaction terms in Study 1 (N ¯ 201)
478 Mark Conner et al.
Temporal stability and the TPB
479
the relationship between intention and behaviour varies as a function of intention stability (Fig. 1, left panel). When intention stability is low intentions positively predict behaviour (B ¯ ± 99, p ! ± 01). However, as intention stability increases from moderate to strong, the predictive power of intentions increases (B ¯ 1 ± 56, p ! ± 001, and B ¯ 2 ± 13, p ! ± 001, respectively). Thus, intentions were more predictive of screening attendance when intentions were stable. For the past behaviour ¬ intention stability interaction, simple slope analyses demonstrated (Fig. 1, right panel) that when intention stability is low or moderate, past behaviour strongly predicts behaviour (B ¯ 1 ± 96, p ! ± 001; B ¯ ± 93, p ! ± 01, respectively). However, when intention stability is high, past behaviour is unrelated to behaviour (B ¯ ® ± 10, n.s.). Thus, past behaviour is unrelated to screening attendance when intentions showed high levels of stability, while for less stable intentions past behaviour shows a signi®cant positive relationship to behaviour. These eåects are consistent with the hypotheses.
Figure 1. Simple regression slopes for Study 1. Left panel shows regression of behaviour on intentions for low intention stability (mean® 1 standard deviation), moderate intention stability (mean) and high intention stability (mean 1 standard deviation) participants. Right panel shows regression of behaviour on past behaviour for low intention stability (mean® 1 standard deviation), moderate intention stability (mean) and high intention stability (mean 1 standard deviation) participants.
Correlates of stable vs. unstable intentions and PBC The next set of analyses explored whether stable vs. unstable intentions or PBC might be diåerentially correlated with TPB variables and past behaviour. Table 3 shows the correlates of intentions and PBC for both groups and a Z test of the
480
Mark Conner et al.
diåerence between correlations from independent samples. All variables were signi®cantly correlated with intentions for both groups except for past behaviour in the unstable intention group. Stable intentions were signi®cantly more strongly correlated with attitudes than were unstable intentions (the diåerence for past behaviour was marginally signi®cant, p ! ± 10). Examination of the correlations for stable and unstable PBC (Table 3) revealed few diåerences. PBC was signi®cantly correlated with intentions and attitudes in both groups. Subjective norms were signi®cantly correlated with PBC only in the stable PBC group, while past behaviour was signi®cantly correlated with PBC only in the unstable PBC group. None of these diåerences were signi®cant. Table 3. Correlates of intention for groups with stable and unstable intentions and correlates of PBC for groups with stable and unstable PBC in Study 1 Intention or PBC stabilitya Correlated variables
Stable (N ¯
100)
Unstable (N ¯
Attitude±intention Subjective norm±intention PCB±intention Past behaviour±intention ±
61*** 67*** ± 25* ± 21* ±
±
±
Intention±PBC Attitude±PBC Subjective norm±PBC Past behaviour±PBC
±
23* 36*** ± 30*** ± 01 ±
±
±
®
Zb
101)
42*** 62*** ± 26* ± 01 29*** 30** ± 18 ± 21*
®
®
1 ± 82* 0 ± 60 0 ± 08 1 ± 42 0 ± 45 0 ± 47 0 ± 90 1 ± 56
* p ! ± 05; ** p ! ± 01 ; *** p ! ± 001. a Top half of table: the two groups are based on intention stability; bottom half of table : groups are based on PBC stability. b These diåerences were unaåected by range restrictions in the variables.
Discussion Study 1 supported the hypotheses. The temporal stability of intentions moderated the impact of intentions on behaviour such that stable intentions are signi®cantly more predictive of subsequent behaviour. In addition, the temporal stability of intentions moderated the eåects of past behaviour on behaviour such that a signi®cant eåect of past behaviour when intentions were unstable was reduced to non-signi®cance when intentions were stable. These ®ndings replicate those of Sheeran et al. (1999b). However, the present study also employed objective measures of past behaviour and behaviour demonstrating that such eåects are not merely attributable to consistency among self-report measures. In addition, the present study demonstrated that the eåects attributable to intention stability occur in the presence of PBC stability measures, but that PBC stability did not moderate relationships with behaviour.
Temporal stability and the TPB
481
The analyses of the correlates of stable and unstable intentions and PBC also revealed a number of interesting ®ndings. Attitude and past behaviour were more strongly correlated with stable intentions than unstable intentions. These ®ndings also broadly support the ®ndings of Sheeran et al. (1999b). There were no signi®cant diåerences in the correlates of stable vs. unstable PBC. The general discussion considers these eåects in the context of the Study 2 ®ndings. STUDY 2 Study 2 examined eating a low-fat diet. In this study, self-report measures of all variables were employed. Since eating a low-fat diet is a frequently performed behaviour, a strong past behaviour±behaviour relationship was anticipated. It would be especially impressive, therefore, if temporal stabilities could moderate relationships in the present study. Method Participants and design Study 2 also employed a longitudinal design with data collected by postal questionnaire at three time points. TPB data were collected at the ®rst two time points (separated by ®ve months) and a self-report measure of behaviour was taken three months after the Time 2 data collection. A total of 407 hospital workers completed all three questionnaires (out of 1155 who received all questionnaires; 35 % response rate) and were included in the study (mean age ¯ 37 ± 4 years ; range ¯ 20±64 years). These 407 respondents did not diåer on the TPB variables from the other 392 respondents who only completed the ®rst questionnaire (F(4,794) ¯ 1 ± 99, p " ± 10). Responses from the ®rst questionnaire were employed to compute stability measures of intentions and PBC. TPB measures and a measure of past behaviour in the second questionnaire were used to predict self-reported behaviour in the ®nal questionnaire.
Measures The following measures were used in the questionnaires. These items were part of a larger set not reported here (see Armitage & Conner, 1999). Intention. Intention to eat a low-fat diet was assessed using three items, each measured on a 7-point bipolar (® 3 to 3) scale. These were : ` I intend to eat a low-fat diet in the future (de®nitely do not±de®nitely do) ’, ` I plan to eat a low-fat diet in the future (de®nitely do not±de®nitely do) ’, and ` I want to eat a low-fat diet in the future (de®nitely do not±de®nitely do) ’. The mean of these items produced a composite scale. Cronbach’s a were ± 82 and ± 82 for Times 1 and 2, respectively ; test±retest reliability was r ¯ ± 61 ( p ! ± 001). Attitude. Attitude was assessed using a semantic diåerential scale. Respondents were presented with the sentence, `My eating a low-fat diet in the future is ¼’. Eight pairs of adjectives were rated, each on a 7-point bipolar (® 3 to 3) scale: bad±good, unfavourable±favourable, harmful±bene®cial, unpleasant± pleasant, negative±positive, unenjoyable±enjoyable, unsatisfactory±satisfactory, and useless±useful. The mean of the eight scales was taken as a composite score. Cronbach’s a were ± 89 and ± 90 for Times 1 and 2, respectively ; test±retest reliability was r ¯ ± 64 ( p ! ± 001). Subjective norm. A global measure of subjective norm was measured with a composite score derived from four items: `People who are important to me think I (should not eat a low-fat diet±should eat a lowfat diet) ’, ` People who are important to me would (disapprove of my eating a low-fat diet±approve of my eating a low-fat diet) ’, ` People who are important to me want me to eat a low-fat diet (strongly
482
Mark Conner et al.
disagree±strongly agree) ’, and `I feel under social pressure to eat a low-fat diet (strongly disagree±strongly agree) ’. All were measured on 7-point bipolar (® 3 to 3) scales and a mean score created. Cronbach’s a were ± 66 and ± 65 for Times 1 and 2, respectively; test±retest reliability was r ¯ ± 56 ( p ! ± 001). Perceived behavioural control (PBC). PBC was assessed using ®ve 7-point unipolar ( 1 to 7) items (e.g. ` How much personal control do you feel you have over eating a low-fat diet? (very little control±complete control) ’). The mean of the ®ve items produced composite scores. Cronbachs a were ± 79 and ± 76 for Times 1 and 2, respectively ; test±retest reliability was r ¯ ± 58 ( p ! ± 001). Behaviour. Behaviour was measured by two 7-point self-perception measures of behaviour (scored 1 to 7): ` I ate a low-fat diet in the last 3 months (strongly disagree±strongly agree)’ and ` How often did you eat a low-fat diet in the last 3 months (never±frequently)’. The two items correlated ± 73 at Time 2 and ± 77 at Time 3. The mean of the two items at each time point produced composite scores. The measure in the second questionnaire constitutes the measure of past behaviour and that in the ®nal questionnaire is the measure of behaviour. Intention stability. Intention stability was assessed as 0 minus the mean stability score (based upon standardized values of the sum of absolute diåerences ; the mean adjusted diåerence; the number items changing ; and the absolute diåerence). The four items were internally reliable (a ¯ ± 93) and this stability measure correlated signi®cantly with the within-participant correlation (r ¯ ± 53, p ! ± 001). PBC stability. PBC stability was assessed as 0 minus the mean stability score (based upon standardized values of the sum of absolute diåerences ; the mean adjusted diåerence ; the number items changing ; and the absolute diåerence). The four items were internally reliable (a ¯ ± 91) and this stability measure correlated signi®cantly with the within-participant correlation (r ¯ ± 36, p ! ± 001).
Results TPB analyses Table 4 reports the intercorrelations, means and standard deviations for the Time 2 TPB measures, intention stability, PBC stability, past behaviour and behaviour. Behaviour and intentions are signi®cantly associated with all other variables, with past behaviour being the strongest predictor of behaviour and attitude being the strongest predictor of intentions. Multiple regression of behaviour onto TPB variables indicated that intentions and PBC (Table 5, cols 1 and 2) were the only signi®cant predictors and explained 40 ± 9 % of the variance. The addition of past behaviour to the equation (Table 5, col. 3) explained a large and signi®cant additional percentage of the variance in behaviour (15 ± 1 %). Examination of the beta weights indicated intention, and past behaviour were each signi®cantly independently associated with behaviour (Table 5). Temporal stability as a moderator Table 4 indicates that the temporal stability of intentions and PBC were both positively correlated with behaviour, intentions and past behaviour ( ps ! ± 01). This indicates that respondents with stable intentions and stable PBC were more likely to have eaten a low-fat diet in the past, intended to do so in the future and reported eating a low-fat diet.
** p !
±
±
001.
Behaviour Intention Attitude Subjective norm Perceived behavioural control BI stability PBC stability Past behaviour
01; *** p !
1. 2. 3. 4. 5. 6. 7. 8.
TPB variables 1 ± 00
1 ± 61*** 1± 00
2 ±
50*** 66*** 1 ± 00 ±
3 ±
39*** 50*** ± 56*** 1 ± 00 ±
4 ±
52*** 64*** ± 61*** ± 36*** 1 ± 00 ±
5 ±
7 ±
29*** 30*** ± 25*** ± 13** ± 46*** ± 44*** 1± 00 ±
407)
18** 42*** ± 22*** ± 16** ± 34*** 1± 00 ±
6
Table 4. Means, standard deviations and intercorrelations for Study 2 variables (N ¯
±
72*** 65*** ± 50*** ± 37*** ± 55*** ± 21*** ± 29*** 1 ± 00 ±
8
5 ± 18 1 ± 48 1 ± 89 1 ± 17 5 ± 60 0 ± 00 0 ± 00 5 ± 09
M
1± 61 1± 34 0± 95 1± 27 0± 98 0± 91 0± 89 1± 67
SD
Temporal stability and the TPB 483
* p!
±
05; ** p !
±
Intention Attitude Subjective norm Perceived behavioural control Past behaviour BI stability PBC stability Intention ¬ BI stability Attitude ¬ BI stability Subjective norm ¬ BI stability PBC ¬ BI stability Past behaviour ¬ BI stability Intention ¬ PBC stabiltiy Attitude ¬ PBC stability Subjective norm ¬ PBC stability PBC ¬ PBC stability Past behaviour ¬ PBC stability BI stability ¬ PBC stability
1. 2.
01 ; *** p !
±
001.
R# R# change Model F
3. 4.
Predictors
Step 61***
367 367*** 234± 0*** ±
±
±
b
409 042*** 68 ± 4*** ±
±
±
38*** 09 ± 08 ± 19*** ±
b
560 151*** 101 ± 8*** ±
±
±
14** 06 ± 07 ± 08 ± 53*** ±
b ±
±
29*** 01 ± 04 ± 11* ± 42*** ® ± 08 ± 07 ± 17** ® ± 11 ® ± 02 ® ± 02 ® ± 02 ® ± 06 ± 03 ± 04 ± 12* ® ± 13** ± 02 ± 594 ± 034*** 31 ± 4*** ®
b
Table 5. Hierarchical regressions of behaviour onto TPB variables, past behaviour, intention stability, PBC stability and stability interaction terms for Study 2 (N ¯ 407)
484 Mark Conner et al.
Temporal stability and the TPB
485
Moderated (linear) regression analysis was employed to test the moderation hypothesis and mean-centred variables were used to minimize problems of multicollinearity. Table 5 (col. 4) shows the results of these analyses. Intentions, PBC, past behaviour, and the interactions between intention stability and intentions (b ¯ ± 17, p ! ± 01), between PBC stability and PBC (b ¯ ± 12, p ! ± 05), and between PBC stability and past behaviour (b ¯ ® ± 13, p ! ± 01) were signi®cant, with the ®nal equation explaining 59 ± 4 % of the variance in behaviour.’ Neither intention stability nor PBC stability were signi®cant predictors of behaviour. The nature of the signi®cant interactions was probed using simple slope analyses in the same manner as Study 1 (Fig. 2). For the intention ¬ intention stability interaction, simple slope analyses demonstrated how the relationship between intention and behaviour varies as a function of intention stability (Fig. 2, top left panel). When intention stability is low, intentions positively predict behaviour (b ¯ ± 20, p ! ± 05). However, as intention stability increases from moderate to strong, the predictive power of intentions increases (b ¯ ± 35, p ! ± 01; b ¯ ± 50, p ! ± 001, respectively). Thus, intentions were more predictive of eating a low-fat diet when intentions were stable. For the PBC ¬ PBC stability interaction, simple slope analysis (Fig. 2, top right panel) demonstrated that when PBC stability is low, PBC is unrelated to behaviour (b ¯ ± 02, n.s.). However, as PBC stability increases from moderate to strong, the predictive power of PBC increases (b ¯ ± 19, p ! ± 05; b ¯ ± 36, p ! ± 01, respectively). Thus, PBC was more predictive of eating a low-fat diet when PBC was stable. For the past behaviour ¬ PBC stability interaction, simple slope analyses demonstrated how the relationship between past behaviour and behaviour varies as a function of PBC stability (Fig. 2, bottom panel). At moderate and low levels of PBC stability, past behaviour is positively related to behaviour (b ¯ ± 41, p ! ± 001; b ¯ ± 63, p ! ± 001, respectively) ; however, at high levels of PBC stability, past behaviour is unrelated to behaviour (b ¯ ± 19, n.s.). These eåects are consistent with the hypotheses.
Correlates of stable vs. unstable intentions and PBC The ®nal set of analyses explored whether stable vs. unstable intentions or PBC might be diåerentially correlated with TPB variables and past behaviour. Table 6 shows the correlates of intentions and PBC for both groups and a Z test of the diåerence between correlations for independent samples. All variables were signi®cantly correlated with intentions for both groups. However, several of these diåerences in correlation between the two groups were signi®cant. In particular, stable intentions were more strongly correlated with attitudes, subjective norms and PBC than were unstable intentions. There were no diåerences for past behaviour. Examination of the correlation for stable and unstable PBC (Table 6) revealed PBC to be signi®cantly correlated with all variables. However, stable PBC was more ’ Examination of VIF scores did not show evidence of multicollinearit y problems in the data. Rerunning the analyses entering one interaction term at a time after entering all other variables produced essentially identical results, except in the case of the past behaviour ¬ intention stability interaction which was marginally signi®cant in these analyses (b ¯ ® ± 08, p ¯ ± 09).
486
Mark Conner et al.
Figure 2. Simple regression slopes for Study 2. Top left panel shows regression of behaviour on intentions for low intention stability (mean® 1 standard deviation), moderate intention stability (mean) and high intention stability (mean 1 standard deviation) participants. Top right panel shows regression of behaviour on PBC for low PBC stability (mean® 1 standard deviation), moderate PBC stability (mean) and high PBC stability (mean 1 standard deviation) participants. Bottom panel shows regression of behaviour on past behaviour for low PBC stability (mean® 1 standard deviation), moderate PBC (mean) and high PBC stability (mean 1 standard deviation) participants.
Temporal stability and the TPB
487
Table 6. Correlates of intention for groups with stable and unstable intentions and correlates of PBC for groups with stable and unstable PBC in Study 2 Intention or PBC stabilitya Stable (N ¯
Correlated variables
202)
Unstable (N ¯
Attitude±intention Subjective norm±intention PCB±intention Past behaviour±intention ±
80*** 60*** ± 66*** ± 63*** ±
±
±
Intention±PBC Attitude±PBC Subjective norm±PBC Past behaviour±PBC ±
64*** 67*** ± 41*** ± 65*** ±
±
±
49*** 38*** ± 54*** ± 66*** 58*** 52*** ± 31*** ± 42***
Zb
205)
®
5 ± 63*** 2 ± 93** 1 ± 89* 0 ± 52 0 ± 96 2 ± 35** 1 ± 15 3 ± 27***
* p ! ± 05; ** p ! ± 01 ; *** p ! ± 001. a Top half of table: the two groups are based on intention stability; bottom half of table : groups are based on PBC stability. b These diåerences were unaåected by range restrictions in the variables.
strongly correlated with attitudes and past behaviour than were unstable PBC. Neither of the other diåerences were signi®cant. Discussion Study 2 broadly replicated the ®ndings of Study 1, but with a behaviour that is performed much more frequently. The temporal stability of intentions moderated the impact of intentions on behaviour such that stable intentions are signi®cantly more predictive of subsequent behaviour. In addition, the temporal stability of PBC moderated the eåects of PBC and past behaviour on behaviour. The present study demonstrates that PBC stability, rather than intention stability, moderated the impact of past behaviour, at least for this frequently performed behaviour. The analyses of the correlates of stable and unstable intentions and PBC also revealed a number of interesting ®ndings. Attitude, subjective norm and PBC were each more strongly correlated with stable intentions than with unstable intentions. For PBC, attitudes and past behaviour were more strongly correlated with stable PBC than unstable PBC. The general discussion returns to these eåects. GENERAL DISCUSSION The two studies reported above contribute to the debate on the factors that determine the strength of intention±behaviour and past behaviour±behaviour relationships (Conner & Armitage, 1998 ; Orbell et al., 1997 ; Ouellette & Wood, 1998 ; Sheeran et al., 1999b ; Verplanken et al., 1998). Both of the present studies
488
Mark Conner et al.
examined the temporal stability of intentions and PBC as potential moderators of these relationships. Based on previous research it was predicted here that temporally stable intentions and PBC would moderate intention±behaviour and past behaviour± behaviour relationships. In both three-wave longitudinal studies reported this proved to be the case. In Study 1, stable intentions were strong predictors of behaviour whereas unstable intentions were only weakly related to behaviour. In addition, past behaviour had no impact on behaviour when intentions were stable, but had a signi®cant impact on behaviour when intentions were unstable. In Study 2, again it was found that stable intentions were strong predictors of behaviour while unstable intentions were only weak predictors of behaviour. In addition, stable PBC was predictive of behaviour, while unstable PBC was unrelated to behaviour. Finally, past behaviour was unrelated to behaviour when PBC was stable, but was strongly related to behaviour when PBC was unstable. These ®ndings support and extend those of Sheeran et al. (1999b) who examined a frequently performed behaviour (studying over the Christmas vacation). Like Sheeran et al. the present research demonstrates that intention stability moderates intention±behaviour relations. The present study shows that these eåects occur whether or not the behaviour is performed frequently (Ouellette & Wood, 1998). This is important because it demonstrates the power of stable intentions to determine behaviour even under conditions which promote behaviours becoming under the control of habitual processes (i.e. frequent performance of a behaviour). These ®ndings provide insight into why individuals with equivalent intentions might diåer in their capacity to enact their intentions. Stable intentions are more likely to be translated into subsequent action. This would suggest that future research might usefully examine both extremity of intentions and stability of intentions in trying to understand intention±behaviour relationships. In this view intention stability is a theoretically derived and empirically testable moderator variable which can be used to address the intention±behaviour gap (Sheeran et al., 1999b). The present data also support the idea that the stability of PBC can act as a moderator of PBC±behaviour relationships (Study 2). Stable PBC was a strong predictor of behaviour, whereas unstable PBC was unrelated to behaviour. It is not entirely clear why this eåect should only be observed for a frequently performed (` habitually controlled ’) behaviour and not for an infrequently performed (` consciously controlled ’) behaviour. However, one possibility is that experience of behaviours that can be performed frequently permits the formation of more stable and accurate PBC. This speculation is consistent with Ajzen’s (1991) analysis of the predictive validity of PBC and is supported by a highly signi®cant correlation between past behaviour and PBC stability in Study 2 that is not evident in Study 1 (but see discussion of research limitations below). The data also indicated that intention stability (Study 1) or PBC stability (Study 2) moderates the relationship between past behaviour and behaviour. Data demonstrating that past behaviour explains subsequent behaviour over and above intentions and PBC questions the su¬ciency of these accounts of human behaviour. Ouellette and Wood (1998) have suggested that one way out of this di¬culty is to divide behaviours into consciously controlled and habitually controlled. The former type of behaviours are infrequently performed and are assumed to be under the
Temporal stability and the TPB
489
control of intentions (and presumably PBC in the TPB), whereas the latter are frequently performed and are assumed to be under the control of past behaviour. Ouellette and Wood (1998) provide some support for this view. Nevertheless, this view does have its own problems since it does not explain how past behaviour in¯uences subsequent behaviour (see Verplanken & Aarts, 1999, for a discussion of possible processes) or provide insight into how past behaviour±behaviour relations can be attenuated, especially for frequently performed behaviours. The present ®ndings demonstrate the utility of examining intention and PBC stability in understanding past behaviour±behaviour relationships. They also provide support for diåerent moderators for infrequently and frequently performed behaviours. In particular, in Study 1 it was found that intention stability moderated the impact of past behaviour on behaviour for infrequently performed behaviours, while for frequently performed behaviours (Study 2) it was found that PBC stability moderated the impact of past behaviour on behaviour. These ®ndings imply that stable intentions are capable of overcoming one’s previous behavioural tendencies when the behaviour is performed infrequently. However, when a behaviour is performed frequently, stable intentions are not capable of breaking the link between past and future behaviour, although PBC stability is capable of breaking this link. The authors suspect that this might be because stable PBC implies more accurate PBC and that accurate PBC is essential in changing an ` habitual ’ behaviour. These results are interesting in relation to Sheeran et al.’s (1999b) ®ndings. In their study, stable intentions did break the link between past behaviour and behaviour for a frequently performed behaviour. However, the behaviour they examined (studying in the Christian vacation) might be characterized as being carried out in an unstable (or diåerent) context compared to where it is usually carried out. In contrast the behaviours examined in the present studies were likely to be carried out in stable contexts. Thus, it may be that stable intentions (or PBC) only completely break the link between past and future behaviour for infrequent behaviour or frequent behaviour performed in unstable contexts. Further (and ideally experimental) research is required to address relationships between temporal stability, accuracy, the frequency of performance of a behaviour, and stability of context in which behaviour is performed. The present study also examined the correlates of stable and unstable intentions and PBC. Consistent with the TPB, attitude, subjective norm and PBC were signi®cant correlates of stable and unstable intentions in both studies. However, past behaviour was only a consistent signi®cant correlate for the frequently performed behaviour (Study 2), but not for the infrequently performed behaviour (Study 1). More interestingly, attitudes were signi®cantly more strongly correlated with stable intentions than unstable intentions in both studies. These ®ndings partially support those reported by Sheeran et al. (1999b) and suggest that intentions based on attitudes may be better predictors of behaviour (Sheeran, Norman, & Orbell, 1999a). In Study 2, subjective norms and PBC were also signi®cantly stronger correlates of stable compared to unstable intentions. For PBC, it was found that there were no signi®cant diåerences in the correlates of stable PBC compared to unstable PBC in Study 1. In contrast, Study 2 demonstrated that attitudes and past behaviour were signi®cantly more strongly correlated with stable than unstable PBC. This latter
490
Mark Conner et al.
®nding supports the view that stable PBC is likely to be the result of previous experience with the behaviour. These ®ndings provide interesting insights into the determinants of temporal stability in intentions and PBC. Further research is required both to replicate the present ®ndings and to examine these eåects experimentally. In interpreting the ®ndings of the present studies, a number of potential methodological problems should be noted. A ®rst issue relates to the conceptualization and measurement of temporal stability. Stability was conceptualized as the extent to which a construct changes in an individual over time. Thus the authors were interested in measures of degree of change rather than direction of change.( The measure employed in the present studies was not the widely used within-person correlation measure (e.g. Sheeran et al., 1999b) because of the limited numbers of data points available for each individual on each construct. Nevertheless, the present measure of stability did show internal reliability and convergent validity with the within-participants correlation measure. Thus the results appear unlikely to be compromised by the particular measure of temporal stability employed. A second issue relates to a number of diåerences between the two reported studies. Emphasis is placed on the frequency of performance of the behaviour in interpreting the diåerences between studies. The behaviours likely also diåer in a number of other ways (e.g. degree of personal control, perceived importance, etc.). However, the design of the two studies also diåers in a number of other ways. The two studies diåered in whether the measures of behaviour (and past behaviour) were objective (Study 1) or self-report (Study 2). While this may have aåected the amount of variance in behaviour explained, there is reason to believe this would not explain the diåerences between studies. Armitage and Conner (in press) report that the TPB is signi®cantly more predictive of self-reported behaviour than objectively measured behaviour. Nevertheless, they did not report any signi®cant diåerences in the relative weight of the determinants of behaviour. Another problem relates to the items used in the two studies. While there were slight diåerences in the wording of the items and the number of items used to tap each construct, the diåerences in internal reliabilities were modest. Finally, there were diåerences between the two studies in the time interval between measurement of TPB variables and behaviour (Study 1 : one month ; Study 2 : three months). However, other research (Randall & Wolå, 1994) demonstrates that for intention±behaviour relationships the correlation is consistent over considerably longer periods than the variation between the present studies. Hence, while the two studies diåered in a number of ways, the similarity in the key ®ndings and overlap with previous research on other behaviours (Sheeran et al., 1999b) gives some con®dence in the generalizability of the present ®ndings. In conclusion, the present research demonstrates that the temporal stability of intentions moderates relationships between intentions and behaviour and that the stabilities of both intentions or PBC moderates the relationship between past behaviour and behaviour. Stable intentions are associated with greater intention± behaviour consistency for both types of behaviour, while PBC stability moderated PBC±behaviour consistency only for the frequent behaviour. Stable intentions are also associated with a reduction in the impact of past behaviour on behaviour for ( Various hypotheses about direction of change may be of interest. However, such measures of change do not show good convergent validity with traditional measures of stability such as test±retest correlations.
Temporal stability and the TPB
491
infrequently performed behaviours, while stable PBC is associated with a reduction in the impact of past behaviour on behaviour for frequently performed behaviours. Future research might usefully compare the predictive validity of temporal stability vs. other operative indicators of intention and PBC strength (e.g. accessibility ; Doll & Ajzen, 1992) in furthering the understanding of the determinants of behaviour. References Aiken, L. S., & West, S. G. (1991). Multiple regressio n : Testing and interpreting interactions. Newbury Park, CA: Sage. Ajzen, I. (1985). From intentions to action : A theory of planned behavior. In J. Kuhl & J. Beckman (Eds.), Action control : From cognition to behaviors (pp. 11±39). New York : Springer. Ajzen, I. (1988). Attitudes, personality and behavior. Milton Keynes : Open University Press. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179±211. Ajzen, I. (1996). The directive in¯uence of attitudes on behavior. In P. Gollwitzer & J. A. Bargh (Eds.), Psychology of action (pp. 385±403). New York : Guilford. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood-Cliås, NJ: Prentice-Hall. Armitage, C. J., & Conner, M. (1999). The theory of planned behaviour: Assessment of predictive validity and `perceived control ’. British Journal of Social Psychology, 38, 35±54. Armitage, C. J., & Conner, M. (in press). E¬cacy of the theory of planned behaviour : A meta-analytic review. British Journal of Social Psychology. Bagozzi, R. P., & Kimmel, S. K. (1995). A comparison of leading theories for the prediction of goaldirected behaviours. British Journal of Social Psychology, 34, 437±461. Bagozzi, R. P., & Yi, Y. (1989). The degree of intention formation as a moderator of the attitude±behavior relationship. Social Psychology Quarterly, 52, 266±279. Bandura, A. (1982). Self-e¬cacy mechanism in human agency. American Psychologis t, 37, 122±147. Baron, R. M., & Kenny, D. A. (1986). The moderator±mediator variable distinction in social psychological research : Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173±1182. Bassili, J. N. (1993). Response latency versus certainty as indexes of the strength of voting intentions in a CATI survey. Public Opinion Quarterly, 57, 54±61. Bassili, J. N. (1996). Meta-judgmental versus operative indexes of psychological attributes : The case of measures of attitude strength. Journal of Personality and Social Psychology, 71, 637±653. Batista-Foguet, J. M., & Saris, W. E. (1997). Tests of stability in attitude research. Quality and Quantity, 31, 269±285. Bentler, P. M., & Speckart, G. (1979). Models of attitude±behavior relations. Psychologica l Review, 86, 452±464. Biesanz, J. C., West, S. G., & Graziano, W. G. (1998). Moderators of self±other agreement: Reconsidering temporal stability in personality. Journal of Personality and Social Psychology, 75, 467±477. Campbell, J. D. (1990). Self-esteem and clarity of the self-concept. Journal of Personality and Social Psychology, 59, 538±549. Conner, M., & Armitage, C. J. (1998). Extending the theory of planned behavior: A review and avenues for further research. Journal of Applied Social Psychology, 28, 1429±1464. Conner, M., & McMillan, B. (1999). Interaction eåects in the theory of planned behaviour : Studying cannabis use. British Journal of Social Psychology, 38, 195±222. Conner, M. T., & Norman, N. (1994). Comparing the health belief model and the Theory of Planned Behaviour in health screening. In D. R. Rutter & L. Quine (Eds.), Social psychology and health: European perspectives (pp. 1±24). Aldershot : Avebury. Conner, M., Warren, R., Close, S., & Sparks, P. (1999). Alcohol consumption and the theory of planned behavior : An examination of the cognitive mediation of past behavior. Journal of Applied Social Psychology, 29, 1676±1704.
492
Mark Conner et al.
Doll, J., & Ajzen, I. (1992). Accessibility and stability of predictors in the theory of planned behavior. Journal of Personality and Social Psychology, 63, 754±765. Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes. Fort Worth, PA : Harcourt Brace Jovanovich. Erber, M. W., Hodges, S. D., & Wilson, T. D. (1995). Attitude strength, attitude stability, and the eåects of analyzing reasons. In R. E. Petty & J. A. Krosnick (Eds.), Attitude strength: Antecedents and consequences (pp. 433±454). Hillsdale, NJ : Erlbaum. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior. New York : Wiley. Godin, G., & Kok, G. (1996). The Theory of Planned Behavior: A review of its applications to healthrelated behaviors. American Journal of Health Promotion, 11, 87±98. Godin, G., Valois, P., & Lepage, L. (1993). The pattern of in¯uence of perceived behavioral control upon exercising behaviour : An application of Ajzen’s Theory of Planned Behavior. Journal of Behavioral Medicine, 16, 81±102. Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition : Attitudes, self-esteem, and stereotypes. Psychologica l Review, 102, 4±27. Kashima, Y., Gallois, C., & McCamish, M. (1993). The theory of reasoned action and co-operative behaviour : It takes two to use a condom. British Journal of Social Psychology, 32, 227±239. Krosnick, J. A., Boninger, D. S., Chuang, Y. C., Berent, M. K., & Carnot, C. G. (1993). Attitude strength : One construct or many related constructs? Journal of Personality and Social Psychology, 65, 1132±1151. Norman, P., & Conner, M. T. (1993). The role of social cognition models in predicting attendance at health checks. Psychology and Health, 8, 447±462. Norman, P., & Conner, M. T. (1996). Applying the theory of planned behavior to the prediction of reattendance at health checks : The role of past behavior. Journal of Applied Social Psychology, 26, 1010±1026. Norman, P., & Smith, L. (1995). The theory of planned behaviour and exercise : An investigation into the role of prior behaviour, behavioural intentions and attitude variability. European Journal of Social Psychology, 27, 403±415. Orbell, S., Hodgkins, S., & Sheeran, P. (1997). Implementation intentions and the theory of planned behavior. Personality and Social Psychology Bulletin, 23, 953±962. Ouellette, J. A., & Wood, W. (1998). Habit and intention in everyday life : The multiple processes by which past behavior predicts future behavior. Psychologica l Bulletin, 124, 54±74. Pieters, R. G. M., & Verplanken, B. (1993). Intention±behaviour consistency: Eåects of consideration of set size, involvement and need for cognition. European Journal of Social Psychology, 25, 531±543. Prislin, R. (1996). Attitude stability and attitude strength : One is enough to make it stable. European Journal of Social Psychology, 26, 447±477. Randall, D. M., & Wolå, J. A. (1994). The time interval in the intention±behaviour relationship : Metaanalysis. British Journal of Social Psychology, 33, 405±418. Sheeran, P., Norman, P., & Orbell, S. (1999a). Evidence that intentions based on attitudes better predict behaviour than intentions based on subjective norms. European Journal of Social Psychology, 29, 403±406. Sheeran, P., & Orbell, S. (1998). Do intentions predict condom use ? Meta-analysis and examination of six moderator variables. British Journal of Social Psychology, 37, 231±250. Sheeran, P., & Orbell, S. (1999). Augmenting the predictive validity of the theory of planned behavior: Roles for anticipated regret and descriptive norms. Journal of Applied Social Psychology, 29, 2107±2142. Sheeran, P., Orbell, S., & Tra®mow, D. (1999b). Does the temporal stability of behavioral intentions moderate intention±behavior and past behavior±future behavior relations ? Personality and Social Psychology Bulletin, 25, 721±730. Sheeran, P., & Taylor, S. (1999). Predicting intentions to use condoms : Meta-analysis and comparison of the theories of reasoned action and planned behavior. Journal of Social Psychology, 29, 1624±1675. Sutton, S. (1994). The past predicts the future : Interpreting behaviour±behaviour relationships in social psychological models of health behaviours. In D. R. Rutter & L. Quine (Eds.), Social psychology and health : European perspectives (pp. 71±88). Aldershot : Avebury.
Temporal stability and the TPB
493
Sutton, S. (1998). Predicting and explaining intentions and behavior. How well are we doing ? Journal of Applied Social Psychology, 28, 1317±1338. van Ryn, M., Lyttle, L. A., & Kirscht, J. P. (1996). A test of the theory of planned behavior for two health-related practices. Journal of Applied Social Psychology, 26, 871±883. Verplanken, B., & Aarts, H. (1999). Habit, attitude, and planned behaviour : Is habit an empty construct or an interesting case of goal-directed automaticity ? European Review of Social Psychology, 10, 101±134. Verplanken, B., Aarts, H., van Knippenberg, A., & Moonen, A. (1998). Habit versus planned behaviour : A ®eld experiment. British Journal of Social Psychology, 37, 111±128. Received 17 December 1998; revised version received 23 June 1999