E10 An example to help you report studies using Theory of Planned Behavior (TPB) measures

I have added this section following several requests from students, particularly those studying on health psychology masters courses in the UK, to provide some guidance about how to report studies based on social cognition models of behaviour. This particular example uses the Theory of Planned Behavior (TPB), but it could be used with little modification to report any social cognition model. As well as the METHOD and RESULTS provided in this section, there are now many published papers using such models and therefore no shortage of studies on which to base your write up if you have conducted such a study. The principal issues (and solutions) are the same as those described in Section 3.7 of the book and throughout Section E of this Web site. In that sense there is nothing distinctive about reporting a study using the TPB or similar models, although you may find yourself using reasonably advanced methods of statistical analysis, such as hierarchical multiple linear regression.

Below are some extracts from a paper by Paul Norman and Mark Conner. I am grateful to them for allowing me to use their work in this way. This provides a useful framework on which to model your own reports. Norman and Conner's RESULTS section provides some nice examples of how to report and interpret different forms of linear regression and illustrates the use of tables to present regression results (see also Section B7.7). Note, however, that strictly speaking differences between beta values should be established using an appropriate test, such as a Sobel test. Note, also, that they combine the PARTICIPANTS (which they label RESPONDENTS) and PROCEDURE sections and these sections are best kept separate. They also cite significance levels instead of actual probabilities associated with their obtained statistics - you are advised wherever possible to cite the actual probability (see Section 4.2 of the book). Note also that they do not cite the degrees of freedom and N for their chi square statistics in the later sections of the RESULTS. Again, you are advised to do this (for more on citing chi square, see Section 4.6.1 of the book and Section B2.1 of this Web site).

The full reference for this paper is:

Norman , P., & Conner, M. (2006). The theory of planned behaviour and binge drinking: Assessing the moderating role of past behaviour within the theory of planned behaviour. British Journal of Health Psychology, 11, 55-70.

Method

Three hundred and ninety-eight undergraduate psychology students completed TPB questionnaires in relation to binge drinking ( M age = 20.26 years, SD = 4.37; 92 male, 305 female, 1 missing). One week later the respondents completed a second questionnaire on their binge drinking behaviour over the previous week. Completed questionnaires were obtained from 273 respondents who had previously completed the time 1 questionnaire (68.6% response rate). In order to assess any attrition biases, the time 1 questionnaire responses of those respondents who completed both questionnaires were compared with those respondents who had only completed the time 1 questionnaire. No significant differences were found between the two groups on any of the TPB or binge drinking measures, although females were more likely to have completed the time 2 questionnaire than males (71.1% vs. 59.8%), χ² (1, N = 273) = 4.23 p < .05.

Measures

The time 1 questionnaire included direct measures of the main constructs of the TPB constructed in line with recommendations (Ajzen & Fishbein, 1980; Conner & Sparks, 1996). All items were measured on 7-point response scales and coded so that high values indicated high levels on the variable of interest. Frequency of binge drinking was also assessed in the time 1 questionnaire, and binge drinking in the previous week was assessed at one-week follow-up. Binge drinking was defined in both questionnaires as consuming at least five pints of beer (or ten shorts/glasses of wine) in a single session for men, or consuming at least three and a half pints of beer (or seven shorts/glasses of wine) in a single session for women (see Bennett, Smith & Nugent, 1994).

Attitude towards engaging in a binge drinking session during the next week was measured using five semantic differential scales (i.e., engaging in a binge drinking session over the next week would be . bad-good, foolish-wise, harmful-beneficial, pleasant-unpleasant, enjoyable-unenjoyable; Cronbach's alpha = .90).

Subjective norm was measured using two items (i.e., people who are important to me would approve/disapprove of me engaging in a binge drinking session over the next week; people who are important to me think I should/should not engage in a binge drinking session over the next week; Cronbach's alpha = .85).

Four items were used to assess self-efficacy (i.e., for me, engaging in a binge drinking session over the next week would be . easy/difficult, how certain are you that you could engage in a binge drinking session over the next week?, how confident are you that you could engage in a binge drinking session over the next week? if I wanted to, I could easily engage in a binge drinking session over the next week; Cronbach's alpha = .87).

Perceived control was measured using four items (i.e., how much control do you have over whether or not you engage in a binge drinking session over the next week?, I feel in complete control over whether or not I engage in a binge drinking session over the next week, how much do you feel that whether or not you engage in a binge drinking session over the next week is beyond your control?, how much will factors outside your control influence whether or not you engage in a binge drinking session over the next week?; Cronbach's alpha = .75).

Behavioural intention was assessed using four items (i.e., do you intend to engage in a binge drinking session over the next week?, I will engage in a binge drinking session over the next week, how likely is it that you will engage in a binge drinking session over the next week?, I intend to engage in a binge drinking session over the next week; Cronbach's alpha = .97).

Past binge drinking at time 1 was assessed by asking respondents to indicate, on average, how often they engaged in a binge drinking session. From this, the frequency of binge drinking sessions per week was calculated and entered into the analysis.

Binge drinking at time 2 was assessed by asking respondents to indicate whether or not they had engaged in a binge drinking session over the previous week.

Results

Descriptive Findings

The means, standard deviations and intercorrelations between the variables under consideration are presented in Table 1. Respondents reported engaging in binge drinking an average of 1.51 times per week and 66.7% engaged in a binge drinking session during the one-week follow-up period. As shown in Table 1, intentions to engage in binge drinking were correlated with each of the TPB constructs such that strong intentions were associated with positive attitudes, strong perceptions of social pressure, strong feelings of self-efficacy and weak perceptions of control. Intention was also positively correlated with past binge drinking and negatively correlated with age. Binge drinking at time 2 was correlated with all the TPB constructs, with the exception of perceived control. Thus, binge drinking at time 2 was associated with positive attitudes, strong perceptions of social pressure, strong feelings of self-efficacy and strong intentions. Binge drinking at time 2 was also positively correlated with past binge drinking and negatively correlated with age.

Predicting Binge Drinking Intentions

A hierarchical linear regression analysis was used to predict intention to engage in a binge drinking session over the next week (see Table 2). The independent variables were entered in three blocks: (i) age and sex, (ii) attitude, subjective norm, self-efficacy and perceived control, and (iii) past binge drinking. In this way it was possible to examine the predictive utility of the TPB variables, controlling for the effects of age and sex, as well as the additional predictive utility of past behaviour. Age and sex explained 8% of the variance in binge drinking intentions, R² = .08,
adj. R² = .07, F (2, 386) = 16.63, p < .001, with both variables emerging as significant independent predictors. The addition of the TPB variables led to a substantial increase in the amount of variance explained in binge drinking intentions, ? R² = .66, ? F = 246.74, p < .001. The addition of the TPB variables resulted in the beta weights for age and sex becoming non-significant, indicating that their effects on intention were fully mediated by the TPB variables (Baron & Kenny, 1986). The variables in the regression equation at step 2 explained 74% of the variance in binge drinking intentions, R² = .74, adj. R² = .74, F (6, 382) = 184.16, p < .001, with attitude, self-efficacy and perceived control (negative relationship) having significant beta weights. The addition of past binge drinking produced a further small increase in the amount of variance explained in binge drinking intentions, R² = .03, F = 42.92, p < .001. Attitude, self-efficacy and perceived control, along with past binge drinking, all made significant contributions to the regression equation at step 3. Together the variables under consideration were able to explain 77% of the variance in binge drinking intentions, R² = .77, adj. R² = .77, F (7, 381) = 182.41, p < .001.

In order to assess the moderating role of past behaviour on TPB-intention relations, interaction terms were constructed between each of the TPB constructs and past behaviour and entered into the regression analysis in order to predict binge drinking intentions (Baron & Kenny, 1986). The variables were mean centred prior to the construction of the interaction terms in order to minimise any problems of multi-collinearity and to aid the interpretation of the results (Aiken & West, 1991). A number of researchers have commented on the low power of moderated regression analyses to detect moderation effects, which may stem from a range of factors such as sample size, intercorrelations between the predictors and measurement error (see Anguinis & Stone-Romero, 1997). As a result, the interaction terms were entered at the fourth step in the regression analysis reported above. A significant increment in the amount of variance explained in binge drinking intentions would indicate that past behaviour moderates the impact of the TPB variable on intention. Only the attitude x past behaviour interaction term produced a significant increment in the amount of variance explained in intention scores, R² = .01, F = 173.43, p <.001. The direction of the beta weight, ß = -.14, p <.001, indicates that the attitude-intention relationship became weaker as the frequency of past behaviour increased. The nature of the interaction between attitude and past behaviour was explored in greater detail using simple slopes analysis (Aiken & West, 1991). Regression lines were examined at three levels of the hypothesized moderator (i.e., the mean level and one standard deviation above and below the mean). Attitude was found to be a stronger predictor of intention under low, B = 1.13, p < .001, than under moderate, B = .88, p < .001, and high, B =.63, p < .001, levels of past behaviour. Nevertheless, it should be noted that attitude was a significant predictor of intention at all three levels of past behaviour.

Predicting Binge Drinking at Time 2

Given that binge drinking was a dichotomous variable, a hierarchical logistic regression analysis was performed to predict binge drinking at time 2 (see Table 3). The independent variables were entered in four blocks: (i) age and sex, (ii) intention, self-efficacy and perceived control, (iii) attitude and subjective norm, and (iv) frequency of binge drinking at time 1. In this way it was possible to examine the predictive utility of the TPB variables, controlling for the effects of age and sex, as well as the sufficiency of the TPB. The addition of age and sex produced a significant improvement on the constant-only model, χ² = 11.36, p < .01, although only age emerged as a significant independent predictor. The addition of intention, self-efficacy and perceived control at step 2 produced a further significant improvement in the model, χ² = 57.78, p < .001, with intention and self-efficacy emerging as significant independent predictors. The impact of age became non-significant at this step, indicating that its effect on behaviour was fully mediated by the TPB. The addition of attitude and subjective norm at step 3 failed to improve the model, χ² = 2.32, ns . The addition of past binge drinking at step 4 produced a further small, but significant, improvement in the model, χ² = 11.36, p < .01. Intention, self-efficacy and past binge drinking emerged as significant independent predictors in the final model.

The moderating role of past behaviour on the intention-behaviour relationship was assessed by constructing an interaction term between intention and past behaviour and entering it into the regression analysis at a fifth step [1]. A significant improvement in the model would indicate that past behaviour moderates the impact of intention on behaviour. In line with expectations, the addition of the intention x past behaviour interaction term produced a significant improvement in the model, χ² = 3.93, p <.05. The direction of the regression coefficient, B = -.20, p < .05, indicates that the intention-behaviour relationship became weaker as the frequency of past behaviour increased. Simple slopes analysis revealed that intention was a significant predictor of binge drinking behaviour under low, B = .10, p < .001, and moderate, B = .06, p < .01, levels of past behaviour. However, as the frequency of past behaviour increased, the predictive power of intention decreased and became non-significant under high levels of past behaviour, B = .003, ns .

References

Table 1
Descriptive Statistics and Intercorrelations Between the Measures

Table 2
Predicting Binge Drinking Intentions: Hierarchical Regression Analysis (Betas)

Table 3
Predicting Binge Drinking at Time 2: Hierarchical Logistic Regression Analysis (Unstandardised Regression Coefficients)