There are differing social cognitive models describing the attitude-behavior relations in health and social psychology, the Theory of Planned Behavior (TPB) being by far the most inclusive and successful one in the prediction of intentions and behavior (e.g., Conner & Armitage, 1998). Despite its success in the prediction arena, however, previous researchers doubted the sufficiency of the normative component of the model (e.g., Terry, Hogg, & White, 1999). Realizing its limitations, recent applications of the TPB have suggested possible avenues for improving its methodological adequacy. One avenue suggested was the inclusion of a self-identity perspective in the TPB, which has improved the predictability of intentions (e.g., Sparks & Shepherd, 1992). In spite of empirical evidence ascertaining the predictive validity of self-identity, use of the TPB with a self-identity perspective should not represent a theoretical advance. Moreover, they contend that since self-identity is implied in the measurement of attitudes and values of the Theories of Reasoned Action and Planned Behavior (TRA/TPB), its inclusion adds nothing in principle, and hence, a causal link between self-identity and intention, independent of the effects of attitudinal evaluations may not be expected. More importantly, Sparks and Shepherd stress that even if self-identity is found to be important, its validity could be viewed as an “outcome” of the behavior, performed in the past, not as an “antecedent” to it, thus insinuating a basic hypothesis onto an interaction effect. According to Sparks and Shepherd, self-identity is not expected to have an independent main effect on intention. Contrary to their assumptions, they reported a substantial main effect for self-identity even after controlling for past behavior. Recent reports also provide empirical support for the predictability of intentions from the self-identity measure, within the TPB paradigm (e.g., Conner & McMillan, 1999; Sparks & Shepherd, 1992; Terry et al., 1999; Åstrom & Rise, in press). Given that self-identity has a main effect in the prediction of intentions, teasing out its specific roles in the TPB represents an empirical challenge, thus constituting the main object of investigation in the present study.

Rooted in sociology, identity theory describes variations in behavior across the social structure in a manner that is different from the TRA’s perspective (e.g., Charng, Pivliavin, & Callero, 1988). Identity theory suggests that a person’s self-concept is organized into a hierarchy of role identities that correspond to one’s positions in the social structure, such as parent, spouse, or employee, for instance. As a psychological entity, self-identity may be defined as the salient part of a person’s self-concept which relates to behavior (e.g., Conner & Armitage, 1998; Conner & McMillan, 1999). Hence, it is a set of expectations derived from the person’s social position, referred to as a role-person merger (Charng et al., 1988), thus implying the extent to which a role is internalized as part of the self. On the other hand, as a distinct social construct, identity (social identity) may also represent the different roles that a person occupies in the social structure (e.g., Terry et al., 1999), which often expresses itself in the form of “group belonging and identifications”.

Regarding the two forms of “self-identities”, Åstrom and Rise (2001), for instance, stress that there is confusion between (self) role-identities and social identities. Following Thoits and Virshup (1997), they describe self-identity as an individual level identity composed of information on self-understanding of “ME’s”, and social identity as the reflections on the identifications of the self with a social group or category, that is, the self as an interchangeable group member (“WE’s”). However, in the present study we focus on self-identity, which is to be inferred from a set of cognitive information linking the self with a particular role category, such as “Me” as a person who takes contraceptives, or “Me” as a person who does not take contraceptives, for that matter.

A close inspection of the findings from earlier reports on the main effects of self-identity, provides justification for its inclusion in the TPB. Conner and Armitage (1998) argue that the TPB and identity theory are comparable with regard to their assumption on the determination of behavior from intention. Thus, behavior as a result of some rational decision-making process (e.g., Charng et al., 1988; Conner & Armitage, 1998), is implied in both theoretical perspectives. Biddle, Bank, and Slavings (1987) describe self-identity both as a product of social interaction and as a cause of subsequent behaviors, hence suggesting the nature and significance of self-identity perspective in the pre-diction of behaviors. Nevertheless, the two theoretical perspectives differ with regard to their specific focuses. Identity theory is concerned more with the larger sociological implications of behavior, in particular, the consistent patterns of behavior linking the “actor” to the socially identifiable collective roles (identities) defining the social structural context (e.g., Charng et al., 1988). By contrast, the TPB, being more psychological in perspective, is less concerned about the social implications of behavior and hence is more focused on the implications of social context on the behavior. Perhaps an outstanding distinction between the two perspectives is their views on the determinants of intentions and behavior. Accordingly, identity theory assumes that intentions are likely to be influenced more by salient “role identities”, while the TRA/TPB assumes that they are influenced more by attitudes and norms (e.g., Charng et al., 1988; Conner & Armitage, 1998). Moreover, the TRA/TPB views behavior as a result of discrete personal decisions, while identity theory views behavior as a product of an interactive process in which the roles of the self and of others are defined within the limits of the social structure (e.g., Charng et al., 1988).

Moreover, the two perspectives differ on the role of past behavior in the prediction of intentions. While there is no sound theoretical basis for the moderation effects of past behavior in the TPB (Ajzen, 1991), identity theory assumes the presence of such moderation effects for past behavior. Thus, the effect of self-identity on intentions is not primarily additive but interactive, varying with previous experience of role-related behavior (e.g., Charng et al., 1988). The longer people have occupied a role, the more likely it is that intention to engage in these behaviors is based on the salience of role-identity for that particular behavior and less upon the TPB variables. Thus, the presence of “interaction” between self-identity and habit implicitly suggests the distinction between identity theory and the TPB.

Empirical studies confirmed the independent effect of self-identity in the prediction of intention within the TRA/TPB perspectives (e.g., Åstrom & Rise, in press; Biddle et al., 1987; Charng et al., 1988; Terry et al., 1999). The validity of self-identity was ascertained, for instance, in the prediction of voting behavior (Granberg & Holmberg, 1990), consumption of organically grown vegetables (Sparks & Shepherd, 1992), cannabis use (Conner & McMillan, 1999), consumption of a low-fat diet (Armitage & Conner, 1999), and consumption of alcohol (Conner, Warren, Close, & Sparks, 1999).

On top of its independent effect, past behavior was reported to have a significant moderation effect on the self-identity-intention relationships in the TRA (e.g., Charng et al., 1988), and in the TPB (e.g., Conner & McMillan, 1999; Thompson & Rise, 2000). The latter two TPB applications reported an interaction effect in the opposite direction of what would have been expected. Furthermore, Armitage and Conner’s (1999) findings supported the hypothesized moderation effect, which suggests that the strength of the self-identity measure might hinge on the strength of past behavior. Based on their data, Conner and McMillan concluded that, although the TPB is an additive model, past behavior sets the boundary conditions for the effects of self-identity on intention in the extended TPB model. In contrast, more recently Terry et al., (1999) and Åstrom and Rise (2001) found no evidence to support the hypothesized moderation effects. One possible explanation for the mixed results on the validity of the hypothesized moderation effect was the dependency of such interactions on the nature of the behavior under review, implicitly suggesting the difficulties involved in the interpretation of interactions under different sets of behaviors. Although the above studies suggest the boundary conditions under which the effects of self-identity on intention increase or decrease, it is not yet clear how this effect might be interpreted.

Furthermore, the concept of self-identity is somewhat diffuse (Armitage & Conner, 1999), and hence may capture a number of constructs not currently included within the TPB (e.g., Sparks & Shepherd, 1992). Given the empirical evidence ascertaining the validity of self-identity and past behavior in the TPB, investigating the moderating effect of past behavior is justifiable from a behaviorist perspective. According to this perspective, behavior is more readily influenced by habitual factors (e.g., Conner & McMillan, 1999; Sutton, McVey, & Glanz, 1999), rather than by cognitive factors included in both the TPB and identity perspectives. After all, it has been argued that given the repeated performance of a behavior, intentions are less likely to be influenced by the controlled cognitive processes described in the TPB or identity theory than by the automatic processes (e.g., Åstrom & Rise, in press; Conner & McMillan, 1999; Sutton et al., 1999). In the light of the above arguments, investigating the interactions between past behavior and self-identity constitutes an empirical question that may clear the misgivings surrounding the question on whether self-identity is, indeed, only an antecedent to behavior, or is also a result of the behavior in the past (cf. Sparks & Shepherd, 1992).

In sum, the first aim in the present study was to provide empirical data on the role of self-identity in the TPB. Based on the assumption that self-identity provides additional information about the respondent’s self-concept in relation to contraception, the authors hypothesized that its inclusion in the TPB would have a significant main effect in the prediction of intention (H1). The second aim in the study was to investigate the moderating effects of past behavior on the self-identity-intention relations, and tease out the boundary conditions for such effects. Accordingly, the authors hypothesized that past behavior would have a significant moderation effect on the self-identity-intention relationship (H2), in addition to its main effect.

Intention (I): Intention has been assessed in several ways. For instance, Lugoe and Rise (1999) assessed it as the respondents’ decision to adopt the behavior (i.e., condom use) in the future. Maintaining correspondence with ‘past behavior’, in the present study intention was measured as a likelihood of behavioral frequency on a likelihood scale using three modal contraceptive behaviors among the target population (cf. Fekadu & Kraft, 2001).

Attitude (A): Following TPB’s suggestion (Ajzen & Fishbein, 1980), and empirical results from earlier reports (e.g., Armitage, Conner, Loach, & Willetts, 1999; Lugoe & Rise, 1999), an indirect estimate of attitude was determined from 11 behavioral outcomes weighted by the evaluation of these outcomes, and then summing the results across all outcomes. Since the “expectancy-value formulation” (i.e., multiplicative assumption) was not a valid procedure (e.g., Fekadu & Kraft, 2001), an “indirect measure of attitudes” was computed by adding the 11 beliefs and the corresponding evaluations (cf. Evans, 1991; Fekadu & Kraft, 2001, for the statistical procedures).

Subjective norm (SN): Following similar procedures used for the attitude measure, an indirect measure for subjective norms was computed by adding the scores on the five normative beliefs and the corresponding scores on the motivation to comply items in a unipolar fashion (cf. Fekadu & Kraft, 2001).

Perceived behavioral control (PBC): Following similar procedures as in A and SN above, an “indirect” estimate for PBC was computed by taking the summative term on the seven control beliefs. The strength of the control factor was not found to be significant when examined using the hierarchical regression model (cf. Fekadu & Kraft, 2001).

Past behavior (PB): Respondents were asked: “In the last four months, have you used: (1) “oral contraceptive pills?” (2) “condoms?” (3) “behavioral methods?” Then, a summative score was computed by adding the responses on the above three items (cf. Fekadu & Kraft, 2001).

Self-identity (SI): Following the procedure used by Terry et al. (1999), two items were constructed to assess the extent to which the performance or nonperformance of “contraception” was an important part of the respondents’ self-identity. The items were (1) “Contraception is an important part of who I am” and (2) “I am not a type of person oriented to engaging in contraception”. Responses for both items were coded on a 4-point scale with values ranging from (1) strongly disagree to (4) strongly agree. The response to the latter question was recoded, and then a summative score was computed to represent the self-identity measure, the higher scores indicating a greater self-identity as a person who uses contraceptives.

Table 1 presents both the univariate and bivariate results among the model variables. Although the majority of the respondents had a favorable contraceptive attitude, the social climate was not in favor of contraception (with more than 52% of the respondents reporting an unfavorable social climate for contraception). Respondents further revealed a poor perception of behavioral control over contraception. All variables had significant bivariate association with intention, with SN (Pearson’s r = .53, p < .01) having the strongest bivariate association. Self-identity’s correlation with intention (Pearson’s r = .31, p < .01) is comparable with that of the TPB variables.

The second regression analysis (Table 3, Part B) was run for the high level of past behavior. The results in model 1 reveal that the TPB variables significantly predicted intention (R² = 30%, p < .001), with all the variables as significant predictors. In model 2, the inclusion of self-identity increased the multiple R2 significantly (R² = 3%, p < .015). Unlike in the lower group, its inclusion did not diminish the effects of attitudes. Besides, self-identity (β = - .18, p < .015) explained 3.6% of the variance in intention, independently. However, as indicated by the negative beta, the more an individual reported an identity as one who does use contraceptives, the less one intended to do so in the future. Thus, as the strength of past behavior increases, the predictive power of self-identity decreases, although it remains significant.

The results of this study indicated that self-identity makes a significant contribution to the prediction of intention, thus adding further to the growing body of literature on self-identity as an independent predictor of intention in the TPB (e.g., Åstrom & Rise, 2001; Biddle et al., 1987; Sparks & Guthrie, 1998; Terry et al., 1999). Given the findings suggesting the predictability of intention from self-identity, independent of past behavior, the present authors follow others (e.g., Åstrom & Rise; Conner & McMillan, 1999; Sparks & Guthrie) in concluding that self-identity represents a more active and deliberative cognitive process that is distinguishable from the TPB variables, thus favoring their initial hypothesis, which predicted an independent effect for self-identity. Nevertheless, Sparks and Shepherd (1992) argue that an independent effect for self-identity in the TPB should be examined carefully in terms of the inadequacy of the attitude measurement. In fact, the results in Table 3 affirm the above suspicion, and hence, provide an indication of the intricate nature of the relations between self-identity and attitudes (discussed later), when the level of past behavior has been manipulated. Arguably, the lack of a measure of “anticipated affect reaction” (e.g., Richard & Van der Pligt, 1995) in the present study, which was believed to add significantly to the prediction of intention (e.g., Sutton et al., 1999), might have contributed to the observed main effect for self-identity. The results in the augmented model further revealed that “past behavior” alone accounted for an additional 7.8% of the variance explained in intention, hence, favorably comparing to earlier reports (e.g., Lugoe & Rise, 1999; Sutton et al., 1999).

In addition to its main effect on intention, past behavior had a significant moderating effect on self-identity-intention relations, although not in the hypothesized direction. The presence of interaction between self-identity and past behavior was consistent with the authors’ expectations and in keeping with the hypothesis of the self-identity theory, which predicted that identity labels based on previous behaviors have strong effects on intentions and subsequent behaviors (e.g., Biddle et al., 1987; Charng et al., 1988; Conner & McMillan, 1999). In contrast, some recent applications (e.g., Åstrom & Rise, in press; Terry et al., 1999) did not find empirical evidence for the hypothesized moderation effects.

At the lower level of past behavior (Table 3, Part A), self-identity was a significant predictor of intention, consistent with the hypothesized direction for such moderation effects. However, at the high level of past behavior (Table 3, Part B), its contribution was significant, but negative (a result that was not consistent with the hypothesized interaction). The unexpected negative beta might be explained in terms of the prevailing noncontraceptive social climate, which could be inferred from the strength of the “subjective norm”. Thus, when the strength of past behavior increases from “lower” to “higher” levels, the predictive power of self-identity decreases, although it remains significant at both levels. It may thus be concluded that self-identity was more predictive of contraceptive intentions at “lower”, than at “higher” levels of past behavior. These findings are in tune with some previous reports. For instance, Conner and McMillan (1999) reported a negative interaction between self-identity and habit strength in the prediction of intention to use cannabis. Similarly, Charng et al. (1988) reported that self-identity was a stronger predictor of intention at the “moderate” than at the “higher” level of habit strength, further adding to the relative weakness of self-identity-intention relationships at higher levels of habit strength, a result apparent in the present study as well. Even the study by Charng et al. (though reporting a positive overall interaction between habit and self-identity) did not provide full support for the “hypothesized moderation effect”.

In the light of the above discussions, the present authors follow Conner and McMillan (1999) in concluding that the stronger impacts of self-identity on intention over the first few performances of the behavior may reflect the role of the initial experiences in strengthening the relevance of self-identity to intentions. Likewise, the subsequent weakening of self-identity-intention relations (negative interaction) at higher levels of habit strength implies that intentions become less under the control of the cognitive factors (e.g., self-identity), and become more automatic in nature (see e.g., Conner & McMillan).

An outstanding result in both cases (low and high past behaviors) is the observed “stability” in subjective norm and the “variability” in attitude measures across the models, which provide support for the intricacies involved in attitude and self-identity relations (e.g., Sparks & Shepherd, 1992). The reduction in the statistical significance of the attitude component might be attributed to the reduction of the power in such analysis that involved lower sample size. As can be seen in Table 3 (Part A), the role of attitudes (TPB variable) was reduced to an insignificant level, probably due to the reliance on the belief-based measure of the attitude component that is based on direct experience. Comparatively speaking, the strength of the attitude component among the sample with “high PB” may be explained in terms of the automatic retrieval of attitudes that are based on direct experience. Regarding the strength of experience-based attitudes, Abrahams, Sheeran, and Orbell (1998), quoted Fazio (1990) who found that direct experience with behavior strengthens attitudes by making them more accessible in memory and, therefore, more responsive to relevant contextual cues – such as inquiries or opportunities to act in an attitude-relevant manner, further strengthening the validity of the results. In earlier reports a similar diminishing effect was also documented on the predictive validity of attitudes (e.g., Charng et al., 1988), the subjective norms and perceived behavioral control (e.g., Armitage & Conner, 1999), and for the subjective norms only (Terry & Hogg, 1996), when the TPB was augmented with a self-identity measure, hence providing evidence of the complex nature of the relationships between self-identity and the TPB variables. As a result of this, the present authors speculate that the above effects may be attributed to the dichotomization of past behavior, which was otherwise a continuous variable.

There are some methodological limitations inherent in the present study. The first is the reliance on self-reports for measuring past behavior and intentions. However, relying on such indirect measures is imperative, given the difficulty of taking direct account of sexual and contraceptive behaviors among Ethiopian female adolescents. Second, the authors relied only on belief-based measures of the TPB variables and did not take direct measures of these same variables, with a presumption on the preponderance of such belief-based measures over direct measures. As a result, there is a lack of comparative information to test the validity of the measurement procedures which were employed in this study. However, the standard measurement procedures were employed to indirectly assess those variables. Besides, the study was not a methodological piece, but rather a comparative analysis examining the predictive validity of a Western model in an African setting, where one cannot find a valid instrument to assess a wider range of subjective feelings. Hence, its contributions must be examined in terms of providing information on the cultural relativity of these models in the collectivist cultures of Africa.

Third, the study did not have the actual measure of future behavior, which may call for a longitudinal design. In connection with this point, Åstrom & Rise (2001) argue that given the difficulty of collecting data in a community-based survey, relying on intention alone would suffice for the major purpose of the TPB applications. They argue further that verifying the predictability of the model’s components constitutes the major purpose of TPB applications. Besides, intentions accounted for 20-40% of the variance in behavior (Åstrom & Rise), thus adding further to the meager contributions of prospective studies as far as prediction of future behavior is concerned. In addition, given that intentions are significantly influenced by past behavior, a sound correlation between past behavior and intention points to a reasonably reliable measure of intention (Åstrom & Rise), as was the case in the present study (r = 0.40, p < .01), thus, fairly predicting the likelihood of performing the behavior in the future. Arguably, therefore, it may be concluded that the exclusion of a behavior measure from the model should not represent a major methodological concern.

Abraham, C., Sheeran, P., & Orbell, S. (1998). Can social cognitive models contribute to the effectiveness of HIV-preventive behavioural interventions? A brief review of the literature and a reply to Joffe (1996, 1997) and Fife-Schaw (1997). British Journal of Medical Psychology, 17, 297-310.

Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park: Sage.

Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Process, 50, 179-211.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice-Hall.

Armitage, C., & Conner, M. (1999). Predicting validity of the theory of planned behavior: The role of questionnaire format and social desirability. Journal of Community and Applied Social Psychology, 9, 261-272.

Armitage, C. J., Conner, M., Loach, J., & Willetts, D. (1999). Different perception of control: Applying an extended theory of planned behavior to legal and illegal drug use. Basic and Applied Social Psychology, 21(4), 301-316.

Åstrom, A., & Rise, J. (2001). Young adults’ intention to eat healthy food: Extending the theory of planned behavior. Psychology & Health, 16(2), 223-237.

Biddle, B., Bank, B. J., & Slavings, R. L. (1987). Norms, preferences, identities and retention decisions. Social Psychology Quarterly, 50, 322-337.

Charng, H-W, Pivliavin, J. A., & Callero, P. L. (1988). Role identity and reasoned action in the prediction of repeated behavior. Social Psychology Quarterly, 51, 303-317.

Cohen, J. (1978). Partialed products are interactions; partialed powers are curve components. Psychological Bulletin, 85(4), 858-866.

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, 15, 1429-1464.

Conner, M., & McMillan, M. (1999). Interaction effects in the theory of planned behaviour: Studying cannabis use. British Journal of Social Psychology, 38, 195-222.

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.

Evans, M. G. (1991). The problem of analysing multiplicative composites: Interactions revisited. American Psychologist, 46(1), 6-15.

Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model as an integrative framework. In M. P. Zana (Ed.), Advances in Experimental Social Psychology, 23, 75-109.

Fekadu, Z., & Kraft, P. (2001). Predicting intended contraception in a sample of Ethiopian female adolescents: The validity of the theory of planned behavior. Psychology & Health, 16(2), 207-222.

Granberg, D., & Holmberg, S. (1990). The intention-behavior relationship among U.S. and Swedish voters. Social Psychology Quarterly, 53, 44-54.

Lugoe, W., & Rise, J. (1999). Predicting intended condom use among Tanzanian students using the theory of planned behavior. Journal of Health Psychology, 4(4), 497-506.

Pedhazur, E. J., & Schmelkin, L. P. (1991). Measurement, design, and analysis: An integrated approach. Mahwah, NJ: Erlbaum.

Richard, R., & Van der Pligt, J. (1995). Anticipated affect reactions and prevention of AIDS. British Journal of Social Psychology, 34, 9-21.

Sparks, P., & Guthrie, C. A. (1998). Self-identity and the theory of planned behavior: A useful addition or an unhelpful artifice? Journal of Applied Social Psychology, 28, 1393-1410.

Sparks, P., & Shepherd, R. (1992). Self-identity and the theory of planned behavior: Assessing the role of identification with “green consumerism”. Social Psychology Quarterly, 55, 388-399.

Sutton, S., McVey, D., & Glanz, A. (1999). A comparative test on the theory of reasoned action and the theory of planned behavior in the prediction of condom use intentions in a national sample of English young people. Health Psychology, 18(2), 72-81.

Terry, D. J., & Hogg, M. A. (1996). Group norms and the attitude-behaviour relationship: A role of group identification. Personality and Social Psychology Bulletin, 22, 776-793.

Terry, D. J., Hogg, M. A., & White, K. M. (1999). The theory of planned behavior: Self-identity, social identity, and group norms. British Journal of Social Psychology, 38, 225-244.

Thoits, P. A., & Virshup, L. K. (1997). Me’s and we’s: Forms and functions of social identity. In R. Ashmore, & L. Jussim (Eds.), Self and identity: Fundamental issues, (Vol. 1, pp. 106-133). New York: Oxford University Press.

Thompson, M., & Rise, J. (2000). Extending the theory of planned behavior: The role of past behavior and social influence. (In press).