The theory of planned behaviour:
The theory of planned behaviour: Self-identity, social identity and group norms
Terry, D. J., Hogg, M. A., & White, K. M. (1999). The theory of planned behaviour: Self-identity, social identity and group norms. The British Journal of Social Psychology, 38, 225-44.
Only note please reference to source:
The aim of the present study was to examine further the role that self-identity plays in the theory of planned behaviour and, more specifically, to: (1) examine the combined effects of self-identity and social identityconstructs on intention and behaviour, and (2) examine the effects of self-identity as a function of past experience of performing the behaviour. The study was concerned with the prediction of intention to engage in household recycling and reported recycling behaviour. A sample of 143 community residents participated in the study. It was prospective in design: measures of the predictors and intention were obtained at the first wave of data collection, whereas behaviour was assessed two weeks later. Selfidentity significantly predicted behavioural intention, a relationship that was not dependent on the extent to which the behaviour had been performed in the past. As expected, there was also evidence that the perceived norm of a behaviourally relevant reference group was related to behavioural intention, but only for participants who identified strongly with the group, whereas the relationship between perceived behavioural control (a personal factor) and intention was strongest for low identifiers.
Contrary to expectations, social psychological research has consistently found that attitudes do not have a strong impact on people's behaviour (Wicker, 1969). On the basis of such findings, several researchers have challenged the view that there is a simple attitude-behaviour relationship, arguing instead that we need to employ more complex models if we are to predict people's behaviour from their attitudes. Influential in this respect have been Fishbein & Ajzen's (1975) theory of reasoned action, and its recent extension, the theory of planned behaviour (Ajzen, 1985, 1987; Ajzen & Madden,1986). Although there is general support for the theory of planned behaviour (the most complex of the two models), the sufficiency of the model has been questioned (see Eagly & Chaiken, 1993). In particular, recent research has, in accord with the predictions ofidentity theory (Stryker, 1968,1980,1987; also Burke, 1980; McCall & Simmons,1978; Turner,1978), demonstrated a role for self-identity. In the present paper, we report the results of a study of recycling intentions and behaviour that examined further the role that self-identity may play in the theory of planned behaviour and, more specifically: (1) examined the combined effects of selfidentity and social identityconstructs-derived from social identity theory (Tajfel & Turner, 1979, 1986; see also Hogg & Abrams, 1988) and self-categorization theory (Turner, 1985; Turner, Hogg, Oakes, Reicher & Wetherell, 1987); and (2) examined the effects of self-identity as a function of past experience of performing the behaviour.
Theories of reasoned action and planned behaviour
According to the theory of reasoned action (Fishbein & Ajzen, 1975), behaviour can best be predicted from a person's intention, or willingness, to perform the behaviour. Intention is, in turn, proposed to be a function of two conceptually independent components: (1) an attitudinal component, and (2) a normative component (subjective norm). The attitudinal component reflects the favourableness of people's evaluation of the behaviour, whereas the subjective norm refers to people's perception of the extent to which others who are important to them think that they should perform the behaviour. In an extension to the theory of reasoned action, the theory of planned behaviour, Ajzen (1985, 1987; Ajzen & Madden, 1986) proposed that perceived behavioural control will emerge as an additional predictor of intentions and actual behaviour-the latter effect evident only in relation to behaviours that cannot be performed at will.
Two decades of research have revealed, across a range of behaviours, general support for the theories of reasoned action and planned behaviour (Ajzen, 1988, 1991; Sheppard, Hartwick & Warshaw, 1988; Terry, Gallois & McCamish, 1993). However, support for the role of subjective norm in both theories has been relatively weak (see Ajzen, 1991; Terry & Hogg, 1996; Trafimow & Finlay, 1996). Recent research has also examined whether the theory of planned behaviour incorporates all the major predictors of intention and behaviour. One variable that has consistently been found to emerge as an additional distinctive predictor of intentions is selfidentity-that is, the extent to which performing the behaviour is an important component of a person's self-concept (Eagly & Chaiken, 1993; Sparks & Shepherd, 1992).
Self-identity and the theory of planned behaviour
Theoretically, the link between self-identity and behavioural intentions is predicated on the basis of identitytheory (Stryker, 1968, 1980, 1987; also Burke, 1980; McCall & Simmons, 1978; Turner, 1978), which conceives of the self not as a distinct psychological entity, but as a social construct (see also Mead, 1934). More specifically, Stryker (1968, 1980; Stryker & Serpe, 1982) proposed that we have distinct components of self for each of the role positions that we occupy. The self is, therefore, conceived as a collection of identitiesthat reflects the roles that a person occupies in the social structure. Central to identity theory is the view that to understand action-or in more psychological terms, to understand and predict behaviour-it is necessary to conceive of the self and the wider social structure as being inextricably linked. As well as being influenced by the wider social structure, the self is conceived `as an active creator of social behaviour ' (Stryker, 1968, p. 385).
As noted by Callero, `Role identities, by definition, imply action' (1985, p. 205). In more specific terms, a role can be defined as a set of expectations as to what constitutes role-appropriate behaviour (Simon, 1992). To engage in role-congruent behaviour serves to validate a person's status as a role member (Callero, 1985).
Identity theory, thus, provides a clear justification for the inclusion of self-identity as a predictor of intention, given that, in both the theories of reasoned action and planned behaviour, intention is regarded as the most proximal predictor of behaviour. In support of this proposal, Biddle, Bank & Slavings (1987) found that school continuation was significantly predicted by intention to stay at school which was, in turn, predicted by attitude, norms and identification as a person who would continue at school. Similarly, Charng, Piliavin & Callero (1988) found that people were more likely to intend to give blood if donating blood was an important part of their self-identity. Comparable results have been found in relation to voting intention and behaviour (Granberg & Holmberg, 1990), adherence to an exercise regime (Theodorakis, 1994), and intention to consume organically grown vegetables (Sparks & Shepherd, 1992).
Taken together, previous tests of the theories of reasoned action and planned behaviour have provided convincing support linking self-identity to behaviour. However, the results are unclear in relation to two points: First, it is not clear whether the effect of self-identity on behaviour is only indirect (through behavioural intention, e.g. Charng et al., 1988), only direct (Theodorakis, 1994), or both indirect and direct (Granberg & Holmberg, 1990)-other studies have found a link between self-identity and intention but have not assessed behaviour (Sparks & Shepherd, 1992). Second, only two studies have examined the role of self-identity in the theory of planned behaviour (Sparks & Shepherd, 1992; Theodorakis, 1994), and only one of these assessed both intention and behaviour (Theodorakis, 1994). Thus, the present research was designed to examine the effects of self-identity-both indirect (through intention) and direct-in the theory of planned behaviour. The research was also designed to examine the extent to which a focus on social identity and group membership, in addition to self-identity, can improve our understanding of the influences on behavioural decision making.
Self-identity and social identity
As noted in two recent theoretical reviews (Hogg, Terry & White, 1995; Thoits & Virshup,1997), identitytheory-the theoretical basis for the inclusion of self-identity in the theories of reasoned action and planned behaviour-has many similarities to social identity theory (Tajfel & Turner, 1986; see also Hogg & Abrams, 1998) and its extension, self-categorization theory (Turner, 1985; Turner et al., 1987). According to social identity theory, an important component of the self-concept is derived from memberships in social groups and categories. When people define and evaluate themselves in terms of a self-inclusive social category (e.g. a sex, class, team) two processes come into play: (1) categorization, which perceptually accentuates differences between in-group and out-group, and similarities among in-group members (including self) on sterotypical dimensions; and (2) self-enhancement which, because the self-concept is defined in terms of group membership, seeks behaviourally and perceptually to favour the in-group over the out-group. Social identities are cognitively represented as group prototypes that describe and prescribe beliefs, attitudes, feelings and behaviours that optimize a balance between minimization of in-group differences and maximization of intergroup differences.
More specifically, according to social identity theory, there is a continuum between personal and social identity-shifts along this continuum determine the extent to which group-related or personal characteristics influence a person's feelings and actions (Tajfel & Turner, 1979). If a particular social identity is a salient basis for selfconception, then the self is assimilated to the perceived in-group prototype-which can be thought of as a set of perceived in-group norms-such that self-perception, beliefs, attitudes, feelings and behaviours are defined in terms of the group prototype. Thus, social identities should influence behaviour through the mediating role of group norms-people will be more likely to engage in a particular behaviour if it is in accord with the norms of a behaviourally relevant group membership, particularly if the identity is a salient basis for self-definition (Terry & Hogg, 1996; White, Terry & Hogg, 1994). If the group membership is not salient, then people's behaviour and feelings should be in accord with their own personal and idiosyncratic characteristics rather than group norms.
On this basis, the present research was designed to test the proposal that group norms at least for people who identify strongly with the group-will influence behavioural intentions. In the theory of reasoned action, Fishbein & Ajzen (1975) included subjective norm, defined as the extent of perceived pressure from others to perform the behaviour under consideration, as a predictor of behavioural intention. As noted, tests of the theory (and its extension, the theory of planned behaviour, which also incorporates the subjective norm-intention link) have failed to find strong support for the proposed role of subjective norm (see Ajzen, 1991). Recently, Terry & Hogg (1996) showed that, in accord with predictions derived from social identity theory, the perceived norms of a specific and behaviourally relevant reference group were related to students' intentions to engage in health behaviours (regular exercise and sun-protective behaviour), but only for students who identified strongly with the group. Also consistent with social identity theory was the finding that the relationship between perceived behavioural control and intention to engage in regular exercise was stronger for participants who did not identify strongly with the reference group than for the high identifiers.1 From a social identity perspective the extent to which participants perceive that they are able to perform the behaviour at ease perceived behavioural control, a personal characteristic of central relevance to behavioural choice (Ajzen, 1991; Bandura, 1977)-should be related to intention more strongly for the low identifiers than for the high identifiers.
Social identity theory, like identity theory, conceives of the self as being socially defined: Social identity is a construct that mediates the relationship between the self and the broader social structure of groups and categories. Moreover, both theories acknowledge the fact that self- or role identities (identity theory) andsocial identities (social identity theory) vary in relative importance to a person's self-concept although social identity theory considers the contextual salience of a particular identity to be more responsive to immediate situational cues than does identity theory. Because of these broad metatheoretical similarities betweenidentity theory and social identity theory, the two perspectives are difficult to compare empirically (see Hogg et al., 1995; Thoits & Virshup, 1997). For instance, in the context of attitude-behaviour relations, both perspectives would agree that people are motivated to engage in identity-related behaviours-to do so, serves to validate an important component of the self-concept.
Although similar, there is an important point of difference between identity theory and social identity theory (Hogg et al., 1995; Thoits & Virshup, 1997). Identity theory focuses on role identities-such as mother, academic and blood-donorwhereas social identity theory focuses on identities that emanate from group memberships. Although role identities can be conceived of as broad social aggregates, to the extent that others perform the same roles, they can be contrasted with social identities, which are fundamentally group memberships. Rather than simply being aggregated entities, social identities define the self explicitly in terms of membership in a self-inclusive group, thus intergroup perceptions and behaviours and intragroup influence are important areas of interest for social identity theorists.
The central issue for the present research was the question of how self-identity and group-related constructs jointly influence behavioural decisions. One perspective is that self- or role identities, to the extent that roles can be conceived of as shared social positions, are social categories and, hence, can be regarded as social identities (see Turner, Oakes, Haslam & McGarty, 1994). From this perspective, the importance of the behavioural role for self-definition (measure of self-identity) should not be a significant predictor of intentions once group norms and group identification are taken into account. In other words, a behavioural role should become an important component of self only to the extent that the norms of a self-relevant and salient reference group support the behaviour. An alternative perspective is that because most people perform a variety of different social roles, role identities are personal characteristics; thus, self-identity should predict intentions only for those people who do not strongly identify with the reference group. The view of roleidentities as personal identities is problematic because role identities are essentially social to the extent that role positions involve aggregates of people who perform the same role; however, in most instances, the aggregate is a relatively disembodied group that is unlikely to engender the same intragroup conformity and intergroup responses as other group memberships, thus the view of role identities as social identities is also problematic.
It may be that self- (or role) identities can be distinguished from both personal and social identities. Indeed, Tajfel (1981) suggested that role identities fall somewhere between the personal or group pole of the interpersonal-intergroup continuum. Moreover, Thoits & Virshup (1997) drew clear distinctions among personalidentities, role identities and social identities. They conceptualized role identities as individual-levelidentities or 'me's' because they reflect the definition of self as a person who performs a particular socialrole, whereas group-level identities are conceptualized as 'we's' because they reflect identifications of the self with a social group or category. As collective identities, both role and social identities are contrasted with personal identities, where the self is defined in terms of unique and idiosyncratic characteristics. On the basis of this framework, it was proposed that independently of the effects of personal characteristics and group norms (both of which were proposed to be dependent on strength of group identification), the extent to which the self is defined in terms of the role (self-identity) would predict intentions. This is because performing the behaviour helps to validate that part of the self-concept that emanates from role identities and, as a consequence, provides the person with positive and meaningful self-evaluations (Thoits & Virshup, 1997).
The proposal that strength of identification with a behavioural role will predict intentions independently of the effects of social identity-related effects is inconsistent with self-categorization theory (Turner et al., 1994), but not necessarily inconsistent with Tajfel's (1981) conceptualization of role-related identities. However, consistent with both perspectives, is the proposal that, although self-identity may have an independent influence on intentions, perceived group norms and self-identity should be correlated to the extent that the group membership is an important basis for selfconception and the group norm is perceived to be generally in support of the behaviour. Thus, when the group norm supports the behaviour, the relationship between self-identity and the strength of the norm should be stronger for high than for low identifiers.
Self-identity and the prediction of repeated behaviours
The present study also examined whether the effects of self-identity vary as a function of repeated experience of performing the relevant behaviour. Charng et al. (1988) reasoned that if a behaviour has been performed repeatedly in the past, and thus under habitual control, then decisions to engage in it in the future should depend more on the importance of the behaviour for the person's self-identity than on judgments and feelings about the behaviour (attitude and perceived control) or the perceived expectations of others (subjective norm). When a behaviour becomes a relatively automatic response, the role of cognitive determinants of both intention and actual behaviour should diminish (Triandis, 1979; cf. Kashima, Gallois & McCamish, 1993), whereas the effect of self-identity should strengthen because repeated performance of a behaviour increases both the likelihood that the behaviour is an important component of the self-identity and the person's motivation to validate his or her status as a role member.
As predicted, Charng et al. (1988) found that the relationship between attitude and intention to donate blood was stronger for first-time donors than for other donors (see also Bagozzi,1981); however, there was only weak evidence that the relationship between self-identity and intention varied as a function of extent of past behaviour. The moderating effect of past behaviour on the self-identity-intention relationship may be weaker in relation to infrequently performed behaviours (e.g. blood donation) than for behaviours that have been performed frequently in the recent past (e.g. adherence to an exercise program, regular recycling etc.). The latter type of behavioural roles is more likely to be internalized as a salient component of the selfconcept, and hence more likely to influence behavioural intentions than less frequently performed behaviours-a proposal that was tested in the present research.
In summary, the present study was designed to examine further the effects of selfidentity in the context of attitude--behaviour relations. The behavioural focus for the research was recycling behaviour: recent research has highlighted the importance of attitudinal variables in the prediction of pro-environmental actions (Oskamp, Harrington, Edwards, Sherwood, Okuda & Swanson, 1991; Smith, Haugtvedt & Petty, 1994). The first aim of the research was to examine the role that self-identity plays in the theory of planned behaviour. In accord with the view that intention is the most proximal determinant of behaviour, it was proposed that participants who regarded the role of recycling as an important component of their self-identity would be more likely to intend to perform the behaviour than those who did not (HI). The second aim of the research was to consider the combined role that self-identity and social identity play in the theory of planned behaviour. On the basis of social identity theory, it was proposed that, independent of the effect of self-identity, the norms of a behaviourally relevant reference group (friends and peers in the community) would be related to intentions, but only for people who identified strongly with the group (H2). Furthermore, it was proposed that perceived behavioural control would be more strongly related to the intentions of the low identifiers than the high identifiers (H3), and that if the group norm was perceived to be generally in support of the behaviour, then the correlation between the strength of the group norm and selfidentity would be stronger for the high identifiers than for the low identifiers (H4).
The third aim of the research was to examine the moderating influence of past behaviour on the predictors of intention. Specifically, it was proposed that the relationship between self-identity and intention would be strongest for people who had performed the behaviour frequently in the past (H5). At the same time, it was expected that the relationships between intention and both attitude and perceived behavioural control would be weaker as a function of past experience of performing the behaviour (H6), as would the relationships between behaviour and both intention and perceived behavioural control (H7).
On the basis of the theory of planned behaviour, additional hypotheses were that attitude towards the behaviour (H8) and perceived behavioural control would predict intention (H9); people who intended to perform the behaviour would be more likely than others to actually do so (H10); and perceived behavioural control would predict behaviour (H11).
Participants and design
The target population was members of households with access to recycling bins provided by the local city council. A sample of 63 males and 80 females (N = 143; age range 17 to 59 years; M = 32.66, SD = 12.57) was recruited to participate in the study using a 'snowballing' technique (with the assistance of advancedsocial psychology students who recruited up to two participants each). The characteristics of the sample (age, occupational status and marital status) were compared to the 1991 Australian census data for the city of Brisbane, which indicated that the sample was broadly representative of the population from which it was drawn, although there was some over-representation of younger, unmarried residents in the present sample.
The study was prospective (longitudinal) in design, with the first questionnaire (Time 1) assessing participants' intention to engage in the target behaviour-that is, to `put out for recycling all newspaper and glass, aluminium/tin products, and plastic products that can be recycled during the next fortnight', as well as the proposed predictors of household recycling. Two weeks later (Time 2), 114 (80%) participants reported their recycling behaviour during the past two weeks. Preliminary analyses revealed that the participants who failed to provide data at Time 2 (N = 29) did not differ from those who did on age, gender, marital status, occupational status, intention to engage in the target behaviour, or on any of the proposed predictor variables.
Table 1 presents the means, standard deviations and the Cronbach's alpha coefficients for each of the predictors. As shown in the table, all the scales were reliable.
Two items were used to assess strength of intention: 'I intend to engage in household recycling (the target behaviour was specified in each of the relevant items; see above) during the next fortnight', 1 = extremely unlikely to 7 = extremely likely; and 'I (1 = do not intend to 7 = do intend) to engage in household recycling during the next fortnight'. As recommended by Ajzen & Fishbein (1980), a measure of attitude towards the target behaviour was obtained using four evaluative semantic differential scales (e.g. unpleasant to pleasant, good to bad, favourable to unfavourable). Two items were used to assess the perceived social pressure-subjective norm `If I engaged in household recycling during the next fortnight, most people who are important to me would...' (1 = approve to 7 = disapprove) and `Most people who are important to me think that I (1 = shouldn't to 7 = should) engage in household recycling during the next fortnight'. One item was reverse scored. Based on the items used by Ajzen & Madden (1986), four items assessed perceived behavioural control. Participants indicated the extent to which they perceived that they could perform the target behaviour at will (e.g. `If I wanted to, it would be easy for me to engage in household recycling during the next fortnight' (1 = strongly disagree to 7 = strongly agree) and `How much control do you have over whether you engage in household recycling during the next fortnight?' (I = absolutely no control to 7 = complete control).
Three items assessed the extent to which engaging in household recycling was an important component of the person's self-concept--self-identity. These items were adapted from those used by Charng et al. (1988) and Sparks & Shepherd (1992). These items were: To engage in household recycling is an important part of who I am (1 = no, definitely not to 7 = yes, definitely; I am not the type of person oriented to engage in household recycling (1 = completely false to 7 = completely true); and I would feel at a loss if were forced to give up household recycling (1 = strongly disagree to 7 = strongly agree). To assess perceived group norm, participants responded to four items (see Terry & Hogg, 1996; White et al., 1994) assessing their perceptions of the reference group norms for performing the target behaviour-.g. How many of your friends and peers would engage in household recycling?' (1 = none to 7 = all) and Most of my friends and peers think that me engaging in household recycling during the next fortnight would be...' (1 = undesirable to 7 = desirable). A measure of group identification was obtained with four items (based on those used by Brown, Condor, Mathews, Wade & Williams, 1986; and Hogg, Cooper-Shaw & Holzworth, 1993) designed to assess strength of identification with the reference group (e.g. How much do you identify with your group of friends and peers?'; 1 = not very much to 7 = very much), as well as feelings of belongingness to the group (e.g. In general, how well do you feel you fit into your group of friends and peers?' ; I = not very much to 7 = very much).
Past recycling behaviour was assessed at Time 1 with a single item-`During the past three months, how much of your household garbage that can be recycled (i.e. newspaper and glass, aluminium/tin products and certain plastic products) have you put out for recycling?' (1 = none at all to 7 = everything). At Time 2, a five-item measure of reported behaviour was obtained. Using the same scale that assessed past behaviour, participants indicated how much of their recyclable garbage that they had put out for recycling during the past fortnight. They also indicated, on separate items, how much glass, newspaper, aluminium/tin products and plastic products (which can be recycled) they had put out for recycling during the past fortnight.
Data analysis procedure
Three sets of analyses were performed.2 First, hierarchical regression analysis were conducted to examine the effects of self-identity in the theory of planned behaviour. The second and third sets of analyses tested the significance of interactions involving group identification and past behaviour, respectively. Because both sets of interactions involved continuous variables, it was appropriate to use moderated regression analyses for this purpose (see Cohen & Cohen, 1983). Prior to conducting the regression analyses, correlations among the predictors were computed. As shown in Table 1, the variables were empirically distinguishable. In no instance did the correlation between any two scales approach the mean scale reliability (reliabilities shown in main diagonal; see Campbell & Fiske, 1959). Principal component analyses (with varimax rotation) also supported the empirical distinction among the variables. In the first analysis (accounting for 72.9 % of the variance), the items assessing self-identity and each of the predictors of intention outlined by the theory of planned behaviour (attitude, subjective norm and perceived behavioural control) loaded on separate factors. The second analysis was conducted to examine the empirical distinction between the subjective norm and group norm measures--the analysis accounted for 70% of the variance, and the respective sets of items loaded on different factors. A final analysis was conducted on the identity-related items (selfidentity, group norm and group identification). The analysis yielded a three-factor solution (accounting for 66% of the variance)- each set of items loaded on a sepearate factor.
Self identity and the theory of planned behaviour
Analysis predicting behavioural intention. A hierarchical regression analysis predicting intention was conducted to determine if self-identity emerged as a significant predictor after control of the components of the theory of planned behaviour and past behaviour (the effects of past behaviour were controlled in all analyses because of consistent evidence linking past behaviour to intention and reported behaviour; see Bentler & Speckart, 1979; Fredericks & Dossett, 1983). Variables were entered into the analysis in the following order: (1) past behaviour; (2) attitude, subjective norm and perceived behavioural control; and (3) self-identity.
As shown in Table 2, there was some support for the theory of planned behaviour. The components of the model-as a block accounted for a significant proportion of variance in intention, after control of the effect of past behaviour. Past behaviour, attitude and perceived behavioural control emerged as significant distinctive predictors of intention-participants had a stronger intention to engage in household recycling if they had engaged in the behaviour regularly in the past, if they had a positive attitude towards doing so (H7), and if they perceived a high level of perceived behavioural control (H8). In line with previous literature, subjective norm failed to emerge as a significant predictor of intention (see Terry & Hogg, 1996).
After control of the effects of past behaviour and the components of the theory of planned behaviour, entry of self-identity into the regression equation accounted for a significant increment of variance in intentions. As predicted under HI, participants were most motivated to engage in the behaviour if the behavioural role was an important component of their self-identity.
Analysis predicting reported behaviour. To examine the predictors of reported behaviour, a three-step hierarchical regression analysis was performed. The hypothesized predictors of behaviour-intention and perceived control-were entered in the second step of the analysis (after past behaviour), whereas the measures not proposed to influence behaviour directly (attitude, subjective norm, self-identity) were entered on the third step. As shown in Table 2, inclusion of the measures of intention and perceived behavioural control (as a block) explained a significant proportion of variance in self-reported recycling behaviour, as did past behaviour. As expected, entry of attitude, subjective norm, and self-identity in the third step of the analysis did not account for a significant proportion of variance in reported behaviour. When all the variables were in the equation, there was evidence that people were more likely to engage in the behaviour if they had done so in the past and if they indicated at Time 1 that they intended to do so (H9). Contrary to H10, perceived behavioural control did not significantly predict reported behaviour.
Self-identity, group norms and group identification
The second set of analyses tested H2 and H3 which proposed that, independent of the effects of self-identity, level of reference group identification would interact with group norm and perceived behavioural control to predict intention. To test these predictions, multiplicative terms between group identification, and both group norm and perceived behavioural control were computed. The interaction terms were entered in the final step of the analysis after control of past behaviour (step 1), the components of the theory of planned behaviour (step 2) and the identity-related variables (self-identity, group identification and group norm, step 3). Group norm and group identification were entered into the equation because the significance of a multiplicative term cannot be tested without controlling for the component main effects (Cohen & Cohen, 1983). To ensure that multicollinearity between the predictors and the interaction terms did not distort the results of the analyses, the interactive terms were based on deviation scores, i.e. scores deviated from their means (see Aiken & West, 1991). Although not predicted effects, a second analysis examined whether the relationships between intention and attitude or self-identity varied as a function of group identification.
As shown in Table 3, entry of the identity-related variables in the third step of the regression analysis accounted for a significant increment of variance in intentions. When all the variables were in the equation, self-identity had a significant main effect on intention. Entry of the identification x group norm and the identification x perceived behavioural control interactions into the analysis explained a significant increment of variance in intention. Both the group norm and the perceived behavioural control interactions were significant. In accord with H2, there was a relationship between group norm and intention, but only for those participants who identified strongly with the reference group (see Fig. 1). Simple slope analyses (analogous to performing simple main effects in ANOVA designs; see Aiken & West, 1991) confirmed this pattern of results: The relationship between group norm and intention was significant at one SD above the mean on the measure of identification (f = .40, t = 3.83, p < .001), but not at one SD below the mean (# = .10, t = 1.05, n.s.). The identification by perceived behavioural control interaction revealed an opposite pattern of results (see Fig. 2). Perceived behavioural control was related to intention for low identifiers (p = .73, t = 6.64, p < .001), but not for high identifiers (S = .11, t < 1, n.s.) thus supporting H3 and qualifying the main effect for perceived behavioural control. Entry of the second set of interaction terms (identification x attitude, identification x self-identity) failed to account for a significant increment of variance in intentions. Additional analyses revealed that neither set of interaction terms explained a significant increment of variance in reported behaviour.
It was predicted under H4 that when the group norm was perceived to be prorecycling (above the mean), the correlation between perceived group norm and selfidentity would be stronger for high group identifiers than for low identifiers. There was support for this hypothesis: the correlation between the two variables was significantly stronger for the high identifiers (r = .42), than for the low identifiers (r = -.07, q = 2.17,p < .05; correlation significant atp < .01 for the high identifiers and non-significant for the low identifiers).
Moderating effect of past behaviour
The third set of analyses tested whether past behaviour moderated the effects of selfidentity or the effects of any of the components of the theory of planned behaviour on intention or reported behaviour. To test the significance of these interactions, multiplicative terms between past behaviour and each of the predictors were computed (based on centred scores to avoid multicollinearity between the main effects and the interactions). For intention, the interactions were entered into the regression equation (after the component main effects) in two sets. The interactions involving self-identity and perceived control were entered in the first set, whereas those involving attitude and subjective norm were entered in the second. In both analyses, the set of interaction terms failed to account for a significant increment of variance in intentions (contrary to H5 and H6). However, subsequent analyses-in which the interaction terms were entered separately-revealed a weak past behaviour x attitude interaction (,3 = -.12, p < .07). Consistent with H6, simple slope analyses revealed that the relationship between attitude and intention was stronger for participants who had not performed the behaviour regularly in the past (fl = .46, t = 5.70, p < .001) than it was for those who had done so (fl = .23, t = 2.18, p<.05). For reported behaviour, entry of interaction terms involving past behaviour with intention, perceived behaviour control and self-identity faild to account for a significant increment of variance.
The aim of the present study was to examine further the role that self-identity plays in attitude-behaviour relations and, more specifically, in the theory of planned behaviour. As expected, self-identity had an indirect relationship with reported behaviour through behavioural intention, a relationship that was not dependent on the extent to which the behaviour had been performed in the past. Additional results of the study demonstrated that identity-related influences on intention should be broadened to encompass role identities, as well as a focus on that part of the selfconcept that derives from group membership. Specifically, the perceived norm of a behaviourally relevant reference group was related to intentions for people who strongly identified with the group (that is, for whom the group membership was an important component of their self-concept), but not for those who did not.
After control of the components of the theory of planned behaviour (attitude, subjective norm and perceived behavioural control), self-identity emerged as an independent predictor of intention-participants who regarded the role of recycling as an important component of their self-identity were more motivated to engage in the behaviour than those who did not. This result is consistent with previous research (e.g. Biddle et al., 1987; Charng et al., 1988; Sparks & Shepherd, 1992) and helps to clarify the precise nature of the role of self-identity in the theory of planned behaviour. The current results indicate that the relationship between self-identity and reported behaviour is indirect, through behavioural intention, rather than direct, a pattern of results that accords with the supposition that intention is the most proximal determinant of actual behaviour (Ajzen, 1991; Fishbein & Ajzen, 1975).
By demonstrating further support for the role of self-identity in the prediction of behaviour, the present results underpin the importance of incorporating identityrelated constructs into the theory of planned behaviour (see Eagly & Chaiken, 1993). They also provide support for the view that such influences on behaviour should include not only the extent to which the behavioural role is a salient component of the self-concept, but also constructs that can be derived from a focus on the selfconceptions that emanate from group membership, viz. social identities. As predicted under H2, there was evidence that, for people who strongly identified with the reference group, intention to perform the behaviour was influenced by perceived group norms. These results are consistent with those reported by Terry & Hogg (1996), and extend this previous work by showing that the relationship between group norms and intention (for the high identifiers) is independent of the extent to which performing the behavioural role is a central component of the person's selfconception. In addition to the finding that group identification moderated the relationship between group norm and behavioural intention, there was evidence that the relationship between perceived behavioural control-a personal rather than a group-based construct-and intention was stronger for participants who did not identify strongly with the reference group of friends and peers than for the high identifiers. This finding is consistent with results reported by Terry & Hogg (1996, Study 1) and accords with a social identity/self-categorization perspective, which would predict that when one's identity as a unique person is salient, personal beliefs and feelings are likely to form the most cognitively accessible basis for behavioural choice (see Fazio.1990).
At a theoretical level, the present results provide empirical data on which to compare identity theory andsocial identity/self-categorization theory, a need identified by recent theoretical comparisons of the two approaches (Hogg et al., 1995; Thoits & Virshup, 1997). Taken together, the effects observed for self-identityand group identification underscore the importance of taking into account both selfidentity and social identityconstructs in attitude-behaviour relations. In other words, they demonstrate quite clearly that in the attitude-behaviour context, at least, the two perspectives are complementary, a possibility implied by Hogg et al. (1995) and discussed more explicitly by Thoits & Virshup (1997). In the present study, the self-relevance of the behavioural role and the perceived group norm--for the high group identifiers-were related, independently, to people's motivation to engage in the behaviour of interest.
More specifically, the present results are useful in clarifying the relationship among personal identities, roleidentities (or self-identities) and social identities. As noted, Thoits & Virshup (1997) conceptualized roleidentities and social identities as collective identities (cf. personal identities), but argued that, in most instances, the two types of identities can be distinguished, a distinction that accords with the fact that Tajfel (1981) did not regard role identities as being synonymous with either personal or group identities. The present results provide some support for the view that role and social identities can be distinguished. Even after the effects of group norms and group identification were controlled, self-identity was an independent predictor of intentions. Moreover, the effects of self-identity were constant, irrespective of levels of group identification, suggesting that a role or self-identity is not a personal identity, but rather a distinct form ofsocial identity that can simultaneously influence, along with group-level constructs, intentions. The latter point is important because, from a self-categorization perspective, aspects of personal identity should be responsive to variation in strength of group identification in the same way as was observed for perceived behavioural control (a personal variable). It should, however, be acknowledged that inspection of the self-identity items (based on those that have been used in previous research) reveals that they were strongly focused on self-definition, and hence could be regarded as indicators of personal identity. Future research should seek to develop measures of self-identity that reflect, more closely, the socially constructed nature of role identities. However, the fact that the effects of self-identity were not dependent on group identification is inconsistent with the view that role identities are a form of personal identity.
The present results point to a complex model of behavioural prediction that needs to take into account the relevance of the behavioural role for self-definition in addition to, and in a more dynamic way, the salience and norms of behaviourally relevant social identities. Nevertheless, in accord with the prediction derived fromsocial identity/self-categorization theory, there was evidence that in the face of a prorecycling group norm, the strength of the group norm and self-identity were correlated for high identifiers but not for low identifiers. These results suggest that definitions of self in terms of behavioural roles are not entirely independent from definitions of self in terms of social identities, and that at least for people who identify strongly with the group membership, strength of self-identification in terms of the behavioural role is related to the perceived norms of the group, a pattern of results that is consistent with self-categorization theory (see Turner et al., 1994).
In addition to examining the roles that both self-identity and social identity may play in the prediction of intentions, the present research examined the effects of selfidentity as a function of the extent to which the behaviour had been performed in the past. Contrary to the reasoning of Charng et al. (1988), there was no evidence that past behaviour moderated the relationship between self-identity and either intention or reported behaviour. These results are in line with those reported by Charng et al and suggest that the importance of validating the person's status as a role member by performing the behaviour impacts on intentions, even for people who have not performed the behaviour repeatedly in the past. Consistent with Charng et al., there was some evidence that the relationship between attitude and intention was moderated by past behaviour; attitude had a stronger relationship with intention for participants who had not performed the behaviour often in the past than for those who had repeatedly performed the behaviour. This pattern of results supports the contention that the impact of cognitive determinants on intentions lessens as a function of past experience of performing the behaviour (Bagozzi, 1981; Triandis, 1979).
The findings observed for the components of the theory of planned behaviour were largely in accord with predictions-intention predicted actual behaviour, whereas attitude and perceived behavioural control predicted intention. Consistent with previous research (see Ajzen, 1991; Terry & Hogg, 1996), behavioural intention was not related to subjective norm; however, the predicted effects for group norm were observed (see above). The fact that perceived behavioural control did not emerge as a significant predictor of actual behaviour was contrary to expectations, but not necessarily inconsistent with the theory of planned behaviour. According to this model (see Ajzen, 1991), perceived behavioural control will influence actual behaviour only if the behaviour is not completely under the person's volitional control. Whether or not people behave in accord with their recycling intentions-particularly in a city where recycling bins and services are provided-is unlikely to be influenced by external factors. Thus, it can be regarded as a relatively controllable behaviour, a supposition that is supported by the relatively high mean score on the measure of perceived behavioural control (cf. Madden, Ellen & Ajzen, 1992).
The present study had a number of strengths: it was longitudinal in design and it obtained measures of both intention and reported behaviour from a sample of community residents. Nevertheless, it was not without limitations. Although the sample was broadly representative of the population from which it was drawn, it was not a large sample nor was it randomly selected. The reliance on a self-reported measure of behaviour was also a limitation of the present study: The use of such measures is likely to inflate the observed correlations between predictors and behaviour due to social desirability concerns and response consistency effects. Future research on the role of identity-related constructs in attitude-behaviour relations would benefit from cross-validation of the present results and those reported in related studies (e.g. Charng et al., 1988; Sparks & Shepherd, 1992; Terry & Hogg, 1996) with data obtained from other paradigms-such as experimental researchwhere overt behavioural measures are easier to obtain.
Overall, the results of the present study are important in that they help to clarify the role that self-identityplays in the theory of planned behaviour, as well as highlighting the need to consider the role of both self- and social identity constructs in the context of attitude-behaviour relations. As noted, the latter aspect of the research has important theoretical implications, given the recent theoretical interest in the distinction betweenidentity theory and social identity/self-categorization theory (Hogg et al., 1995; Thoits & Virshup, 1997). At a broader level, the present study underpins further the important role that social influence-conceptualized as both self-identities and perceived norms of self-relevant social identities-plays in attitude-behaviour relations, an emphasis that has not always been apparent in this area of research (cf. Fishbein & Ajzen, 1975; see Charng et al., 1988 and Terry & Hogg, 1996, for a similar point). The results of the present study also have applied implications, in that they provide suggestions as to the type of variables that should be targeted in intervention programmes designed to encourage pro-environmental behaviours.
Thanks are due to Barbara Mullin and Gloria Hynes for their assistance with data collection and data entry.
1 In the study of the determinants of sun-protective behaviour among females, Terry & Hogg (1996) found that the effects of attitude, rather than perceived behavioural control, was stronger for the low identifiers than for the high identifiers. This result is likely to be specific to appearance-related behaviours (such as sun-protective behaviours for females) and therefore did not form the basis for the hypotheses that were tested in the present study.
a Data were analysed on the full sample of males and females. Preliminary analyses revealed that the bivariate correlations between the predictors and the dependent variables (intention and reported behaviour) were similar for males and females. Gender differences were not explored further because they were not a theoretical focus of the present research.
Aiken, L. S. & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage.
Ajzen, I. (1985). From intentions to actions: A theory of planned behaviour. In J. Kuhl & J. Beckmann (Eds), Action-control: From cognition to behaviour, pp. 11-39. Heidelberg: Springer. Ajzen, I. (1987). Attitudes, traits, and actions: Dispositional prediction of behaviour in personality and social psychology. In L. Berkowitz (Ed.), Advances in experimental social psychology, vol. 20, pp. 1-64. New York: Academic Press.
Ajzen, I. (1988). Attitudes, personality, and behaviour. Milton Keynes: Open University Press. Ajzen, I. (1991). The theory of planned behaviour. Organizational behaviour and human decision processes, 50, 179-211.
Ajzen, I. & Fishbein, Ms (1980). Understanding attitudes and predicting social behaviour. Englewood Cliffs, NJ: Prentice-Hall.
Ajzen, I. & Madden, T. J. (1986). Prediction of goal-directed behaviour: Attitudes, intentions, and perceived behavioural control. Journal of Experimental Social Psychology, 22, 453-474. Bagozzi, R. P. (1981). Attitudes, intentions, and behaviour: A test of some key hypotheses. Journal of Personality and Social Psychology, 41, 607627.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behaviour change. Psychological Review, 84, 191-215.
Bentler, P. M. & Speckart, G. (1979). Models of attitude-behaviour relations. Psychological Review, 86, 452464.
Biddle, B. J., Bank, B. J. & Slavings, R. L. (1987). Norms, preferences, identities and retention decisions.Social Psychology Quarterly, 50, 322-337.
Brown, R. J., Condor, S., Mathews, A., Wade, G. & Williams, J. A. (1986). Explaining intergroup differentiation in an industrial organisation. Journal of Occupational Psychology, 59, 273-286. Burke, P. J. (1980). The self: Measurement requirements from an interactionist perspective. Social Psychology Quarterly, 43, 18-29.
Callero, P. L. (1985). Role-identity salience. Social Psychology Quarterly, 48, 203-215. Campbell, D. T. & Fiske, D. W. (1959). Convergent and discriminant validation by the multi-trait multi-method matrix. Psychological Bulletin, 56, 81-105.
Charng, H. W., Piliavin, J. A. & Callero, P. L. (1988). Role identity and reasoned action in the prediction of repeated behaviour. Social Psychology Quarterly, 51, 303 317. Cohen, J. & Cohen, P. (1983). Applied multiple regression for the behavioural sciences. New York: Erlbaum. Eagly, A. H. & Chaiken, S. (1993). The psychology of attitudes. Fort Worth, TX: Harcourt Brace Jovanovich.