The influence of self-efficacy and past behaviour on the physical activity intentions of young people

Transcription

1 Journal of Sports Sciences, 2001, 19, The influence of self-efficacy and past behaviour on the physical activity intentions of young people MARTIN S. HAGGER, * NIKOS CHATZISARANTIS and STUART J.H. BIDDLE epartment of Psychology, University of Shefield, Western Bank, Sheffffield S10 2TN, Department of Sport Sciences, Brunel University, Osterley Campus, Isleworth, Middlesex TW7 5DU and Department of Physical Education, Sports Science and Recreation Management, Loughborough University, Ashby Road, Loughborough LE11 3 TU Accepted 26 April 2001 The aim of the present study was to identify the influence of self-efficacy and past behaviour on young people s physical activity intentions using an augmented version of Ajzen s theory of planned behaviour. We hypothesized that self-efficacy would exhibit discriminant validity with perceived behavioural control and explain unique variance in young people s intentions to participate in physical activity. We also expected that past physical activity behaviour would attenuate the influence of attitude, subjective norms, perceived behavioural control and self-efficacy on intention. The sample comprised 1152 young people aged 13.5± 0.6 years (mean ± s) who completed inventories assessing their physical activity intentions, attitudes, subjective norms, perceived behavioural control, self-efficacy and past physical activity behaviour. A confirmatory factor analysis demonstrated that the constructs of the Theory of Planned Behaviour achieved discriminant validity. Furthermore, the measures of attitudes, subjective norms, perceievd behavioural control and self-efficacy were significantly related to their respective belief-based measures, supporting the concurrent validity of the measures of the Theory of Planned Behaviour. A non-standard structural equation model demonstrated that attitude and self-efficacy were strong predictors of physical activity intention, but perceived behavioural control and subjective norms were not. Self-efficacy attenuated the influence of attitudes and perceived behavioural control on intention. Past behaviour predicted intention directly and indirectly through self-efficacy and attitude. The present findings demonstrate that young people with positive attitudes and high self-efficacy are more likely to form intentions to participate in physical activity. Furthermore, controlling for past physical activity behaviour revealed that the unique effects of self-efficacy and attitudes on young people s physical activity intentions were unaltered. Keywords: attitudes, perceived behavioural control, Theory of Planned Behaviour. Introduction Recent reviews have acknowledged the role of regular physical activity in young people s physical and mental health (Biddle et al., 1998). However, increasing research evidence indicates that young people do not participate in physical activity of sufficient intensity, duration and frequency for health benefits (Cale and Almond, 1992; Sleap and Warburton, 1996). The onus is, therefore, on exercise psychologists to study the key antecedents of physical activity participation in * Address all correspondence to Martin Hagger, Department of Psychology, University of Essex, Wivenhoe Park, Colchester C04 3SQ, UK. hagger@essex.ac.uk young people to identify targets for intervention (De Bourdeaudhuij, 1998). Importantly, intentional models of social cognition have been adopted to identify and examine the pattern of influence of the psychological antecedents of physical activity behaviour in young people (Atsalakis and Sleap, 1996; Craig et al., 1996). Many social cognitive approaches to the study of intentional behaviour have used the framework proposed by Ajzen (1985), known as the Theory of Planned Behaviour. The model suggests that the proximal predictor of behaviour is an individual s stated intention to perform the target behaviour in a given context and at a given time. Intention is considered to be a motivational variable and is a context-specific representation of goal-directed behaviour (Bloom, 2000). Journal ofsporw Sciences ISSN printassn X online Taylor & Francis Ltd

2 2 Intention is a function of a person s attitudes, subjective norms and perceived behavioural control over the target behaviour. Attitudes represent an individual s predisposition towards engaging in the behaviour and are underpinned by the beliefs that participation in the behaviour will result in certain outcomes and the evaluation of these outcomes as having positive or negative benefits. The subjective norm variable assesses a person s subjective estimate of the extent that important others want them to participate in the target behaviour. It is a function of the beliefs that salient others want the individual to engage in the behaviour and the willingness of the person to comply with these significant others. Finally, perceived behavioural control reflects the person s assessment of the capacities (e.g. skills and abilities) and the limiting or facilitating factors (e.g. barriers and access to facilities) regarding behavioural engagement. The Theory of Planned Behaviour has been successful in the prediction of physical activity behaviour in adults (for reviews, see Godin, 1994; Hausenblas et al., 1997) and young people (Atsalakis and Sleap, 1996; Craig et al., 1996). These studies have revealed that attitudes and perceived behavioural control tend to have the most pervasive influence on physical activity intention in adults, with subjective norms having less of a role. However, research has shown that subjective norm tends to have a stronger influence on physical activity intention in young people than adults (Godin and Shephard, 1986; Shephard and Godin, 1986). This indicates that there may be developmental differences in the social cognitive antecedents of physical activity behaviour. Therefore, further evidence is required with representative samples of young people in relation to social cognitive influences on physical activity behaviour. Recent research suggests that the perceived behavioural control construct in the Theory of Planned Behaviour should be separated into internal and external aspects to allow an examination of the differential effects of these control-related perceptions on physical activity behaviour (Terry and O Leary, 1995). Furthermore, evidence suggests that regular past engagement in physical activity behaviour tends to attenuate the social cognitive influences on intention and prospective behaviour (Yordy and Lent, 1993; Bagozzi and Kimmel, 1995). The main aim of the present study was to examine the differential effects of two control-related constructs - perceived behavioural control and self-efficacy - on the physical activity intentions of young people. In addition, we wished to determine whether the influences of the control-related variables, attitudes and subjective norms on physical activity intentions were attenuated with the inclusion of past behaviour. Hagger et al. Distinguishing perceived behavioural control and selfeficacy It has been proposed that perceived behavioural control accounts, in part, for an individual s confidence in a given set of circumstances or self-efficacy (e.g. DuCharme et al., in press), as advocated by social learning theory (Bandura, 1977,1997). This notion was adopted by Ajzen (1991), who overtly aligned the perceived behavioural control construct with self-efficacy on many occasions. However, Estabrooks and Carron (1998) argued that the interchangeable use of the terms self-efficacy and perceived behavioural control is inconsistent with theory. Indeed, there is increasing empirical and conceptual support for a distinction between perceived behavioural control and self-efficacy (Terry and O Leary, 1995; Sparks et al., 1997; Manstead and Van Eekelen, 1998; Armitage and Conner, 1999; Armitage et al., 1999). Empirically, researchers have extended the study of relations between self-efficacy and intentions by successfully incorporating the construct into or alongside variables from the Theory of Planned Behaviour. Studies have demonstrated that self-efficacy predicts intentions alone (McCaul et al., 1993; Terry, 1993; Yordy and Lent, 1993; Morrison et al., 1995; Richard et al., 1995; Terry and O Leary, 1995; Buunk et al., 1998), behaviour alone (Van Ryn and Vinokur, 1992; Conner et al., 1998) or both intention and behaviour (Dzewaltowski et al., 1990; Biddle et al., 1994; DuCharme and Brawley, 1995; Van Ryn et al., 1996) in addition to the independent effects of perceived behavioural control on intention. Furthermore, factor analyses of the items traditionally used to operationalize the perceived behavioural control construct have extracted two clear factors indicating that individuals tend to treat the items differently (Armitage and Conner, 1999). Conceptually, Terry and O Leary (1 995) made this distinction by proposing that self-efficacy reflects a person s abilities (internal aspects of control) with respect to performing physical activity behaviour and that perceived behavioural control reflects barriers (external aspects of control) towards performing physical activity behaviour. An example of an internal factor is an individual s perceived confidence in engaging in physical activity and an example of an influential external factor is a barrier like bad weather. Terry and O Leary demonstrated that internal aspects of control, labelled self-efficacy, and external aspects of control, labelled perceived control over the behaviour, achieved discriminant and predictive validity in the Theory of Planned Behaviour. Armitage and Conner (1999) further supported this internalexternal distinction by providing a tighter definition of

3 Young people s physical activity intentions 3 self-efficacy and perceived behavioural control within the framework of the Theory of Planned Behaviour. This tighter definition saw perceived behavioural control defined solely by items referring to control over difficult situations and barriers, whereas self-efficacy was defined by internal evaluations of competence and ability to perform the behaviour. The concurrent validity of these constructs was supported with an associated set of internal and external control beliefs. The predictive validity of the distinction between selfefficacy and perceived behavioural control has also been supported in studies of health behaviour (Sparks et al., 1997; Armitage and Conner, 1999; Armitage et al., 1999). The present study aims to augment this research by differentiating between perceived behavioural control and self-efficacy in a study of young people s physical activity intentions. Self-efficacy towards physical activity has been conceptualized and measured in several ways. Terry and O Leary (1995) and Armitage and Conner (1999) both adopted a battery of items to measure self-efficacy that achieved high construct validity and reliability. The measures determined overall estimates of selfefficacy from the individual s assessment of all the possible barriers and facilitating conditions that came to mind and their evaluation of their abilities, This is advantageous because these perceptions have behavioural utility, as shown by predictions of behaviour in other studies (e.g. Povey et al., 2000). However, the disadvantage of this method is that the individual confidence estimates, in the face of certain barriers and facilitating conditions, are not assessed explicitly. Bandura (1977), on the other hand, believed that self-efficacy should reflect a person s evaluation of their confidence in performing a given behaviour in the face of salient barriers and facilitating conditions (Bandura and Cervone, 1983). Indeed, this has been shown to be an important predictor of exercise adherence (Dzewaltowski, 1989). Measures of this conceptualization of self-efficacy provide a more explicit means of eliciting a person s overall judgement of selfefficacy compared with the items referring to general perceptions of confidence and ability used in other studies. Such a conceptualization of self-efficacy enables the researcher to identify the relative contribution that each confidence estimate from each barrier makes to the overall self-efficacy construct. This permits the identification of specific aspects of control that may be of interest to the practitioner for intervention targets, but also provides an overall picture of the influence of self-efficacy on physical activity intentions. Bandura and Cervone (1983) used a free-response format to elicit salient barriers and facilitating conditions from the target population, which were used to formulate items to measure self-efficacy. If the items reliably define a latent self-efficacy factor, the researcher can be confident that the self-efficacy construct is an adequate reflection of self-efficacy and can identify those items that contribute most to self-efficacy estimates. The present study aimed to measure selfefficacy by eliciting salient barriers and facilitating conditions by free-response and constructing a selfefficacy variable using items derived from these responses. This self-efficacy measure will be used in the prediction of physical activity intentions of young people alongside a traditional measure of perceived behavioural control. Past physical activity behaviour Although reviews of the literature support the significant prediction of physical activity intentions from attitudes, subjective norms and perceived behavioural control (Blue, 1995; Godin, 1994), some evidence suggests that the inclusion of past physical activity behaviour as a predictor of intention tends to attenuate these influences. Studies have shown that including past behaviour in regression analyses to predict physical activity intentions resulted in the extinction of the influences of attitude, subjective norm and perceived behavioural control on intention (Godin et al., 1991; Norman et al., 1999). However, including past behaviour as a predictor of intentions does not always completely extinguish the effect of the social cognitive influences of the Theory of Planned Behaviour on intentions, particularly where perceived behavioural control and self-efficacy are concerned. Norman et al. (2000) demonstrated that, among patients attending health promotion clinics, the inclusion of past behaviour extinguished the influence of attitudes and subjective norms on intentions such that perceived behavioural control and past behaviour were the only significant predictors of intention. Terry and O Leary (1995) entered past behaviour as a covariate in the regression of intention on the variables of the Theory of Planned Behaviour. They showed that the influence of self-efficacy on intention was unchanged. Other studies examining the influence of past behaviour have also indicated that attitudes and perceived behavioural control explain unique variance in intention after controlling for past behaviour (Godin et al., 1993; Yordy and Lent, 1993; Norman and Smith, 1995). It would appear, therefore, that controlling for past behaviour in the Theory of Planned Behaviour results in the influence of attitudes, subjective norms, perceived behavioural control and self-efficacy on intentions being attenuated, but not completely extinguished.

4 4 Considering the mechanisms responsible for the attenuating influence of past behaviour in models of social cognition, Bagozzi and Kimmel (1 995) claimed that past behavioural influences tend to control for the previous decision-making process towards the target behaviour. A person who has participated in the behaviour has already made the decision to engage in the behaviour. That person is, therefore, more likely to form an intention to engage in the behaviour without deliberating over their attitudes, subjective norms and perceived control provided that the conditions associated with the behaviour remain the same. In this case, the prediction of intention by past behaviour will reduce the impact of the other social cognitive variables, namely attitude, subjective norms and perceived control. Effectively, the past behaviour-intention path bypasses the indirect effects of past behaviour on intention via the attitude, subjective norms and perceived control variables. Conversely, if some effects of attitudes, perceived behavioural control and subjective norms on intentions remain after controlling for past behaviour, it indicates that the behavioural intention is influenced by situation-specific evaluations regarding the behaviour as well as evaluations about the behaviour executed in the past. Bagozzi (1981) also remarked that testing such hypotheses may reveal the true explanatory power of constructs like attitudes in models of social cognition. He cited the theory of Doob (1947), who raised doubts as to whether attitudes have utility not already subsumed under theories of human learning and motivation (p. 6 13). Therefore, if attitudes did not predict intentions when past behaviour was accounted for, their influence was likely to be spurious and only an artefact of learned behavioural dispositions. A test of whether unique effects of attitudes on intentions remain after controlling for past behaviour would be a partial test of the value of attitudes in models of social cognition and physical activity. The present study aimed to examine the social cognitive influences on the physical activity intentions of young people when accounting for past physical activity behaviour. We expected that controlling for past behaviour by including past behaviour as a predictor of all the model variables would result in the attenuation of the influence of attitudes, subjective norms, perceived behavioural control and self-efficacy on intentions, but unique influences would also remain. Study hypotheses The present study hypothesized that young people s physical activity intentions would primarily be a function of attitudes, perceived behavioural control and Hagger et al. self-efficacy, with subjective norms having a peripheral role as evidenced in meta-analytic studies (Hausenblas et al., 1997). We also expected, as outlined by Ajzen (1 985), that the belief-based measures of attitude, subjective norms and perceived behavioural control would be associated with their respective underlying sets of beliefs. In addition, we hypothesized that the discriminant validity of the intention, attitude, perceived behavioural control, self-efficacy and subjective norm variables would be supported and that self-efficacy would have a unique influence on physical activity intention. Finally, we anticipated that including the frequency of past physical activity behaviour as a predictor of all the Theory of Planned Behaviour variables and self-efficacy would result in the attenuation of the influence of attitudes, subjective norms, perceived behavioural control and self-efficacy on physical activity intentions. Methods Participants Twenty government-run schools were contacted by telephone from the local education authority lists covering the counties of Cheshire, Leicestershire, Nottinghamshire and Staffordshire. The school headteachers were asked if they would allow their pupils to participate in a study of young people s attitudes towards physical activities. Eleven schools confirmed their participation and provided written consent. The participants in the study were school pupils (555 boys, 597 girls) aged 13.5 f 0.6 years (mean f s). They were all pupils in classes covering school years 8 and 9 (age range years). Before the study began, the school pupils were given a letter to take home to their parents informing them that the study was being conducted and that if they objected they should return the accompanying proforma to the school. No proformas were returned. Measures Procedures for the development of the Theory of Planned Behaviour questionnaire published by Ajzen and Fishbein (1980) were followed. Measures of the meory of Planned Behaviour Intention to participate in the target behaviour, leisuretime physical activity, was assessed from responses to three items (e.g. I plan to do physical activities that make me out of breath at least three or more times during my free time in the next week ). Responses were

5 -I - Young people s physical activity intentions 5 given on 7-point semantic differential scales anchored by the word-pair likely-unlikely. This scale exhibited adequate internal consistency as indicated by the Cronbach alpha reliability coefficient (a = 0.77). Attitude was measured using three semantic differential scales in response to the item: My doing physical activities that make me out of breath at least three or more times in the next week is.... The three scales were good-bad, exciting-boring and fun-unpleasant. The alpha reliability for this scale was adequate (a = 0.85). Subjective norm was measured from a single statement: Most people important to me think I should do physical activities that make me out of breath at least three or more times in the next week. This was evaluated on a scale using the likely-unlikely wordpair. Although the use of a single item to measure subjective norm is quite common and consistent with the TPB [Theory of Planned Behaviour] (Courneya and Friedenreich, 1999, p. 116), it is important to recognize that single-item statements are likely to be influenced by measurement error that cannot be corrected for statistically (Martin, 1982). Perceived behavioural control was measured directly using three items. The item statements were: Whether or not I participate in physical activities that will make me out of breath at least three or more times in the next week is entirely up to me, measured with strongly agreestrongly disagree scale end-points; It is mostly up to me whether I do physical activities that make me out of breath at least three or more times in the next week, measured with true-false scale end-points; and How much personal control do you feel you have over participating in physical activities that will make you out of breath at least three or more times in the next week?, measured with very little control-complete control scale end-points. The alpha reliability for this scale was modest (a = 0.68), but not inconsistent with previous measures of perceived behavioural control that have used these items (Sheeran et al., 2001). However, it must be made clear that this is a limitation of the current measure. Measures of beliefs Behavioural beliefs and outcome expectations. In a pilot study, one class of 37 pupils (16 boys, 21 girls) aged years, who were not participants in the main study, completed a free-response format questionnaire to elicit relevant beliefs for the belief-based measures of the attitude, subjective norm and control variables. The pupils were asked to list all the advantages and disadvantages of participating in regular physical activity. A content analysis of the responses revealed that the modal advantages were helps you get fitlstay in shape, is good fun, helps you improve your skills ; the modal disadvantages were might get injured and makes you hot and sweaty. These were used to formulate items for an expectancy-value measure of behavioural beliefs. Behavioural beliefs were measured by responses to the statement: Doing regular physical activities will [salient belief]. Outcome evaluations were measured with responses to the statement: Doing physical activities to [salient belief] is followed by a 7-point scale anchored by the good-bad word-pair. The belief-based measure of attitude was defined from the items comprising the product of the behavioural belief score and the outcome evaluation score for each salient belief. Normative beliefs. A second question in the pilot study asked respondents to list the people important to them. Responses gave the following salient others: mother, father, grandmother, grandfather, sister, brother, friends and teacher. To make the questionnaire generic to the sample, these salient others were reduced to the following categories: parents, grandparents, other family members, fiiends and teacher. These were used to formulate items to assess normative beliefs and motivation to comply for an expectancy-value measure of subjective norms. Items assessing normative beliefs were formed using the following statement: My [salient referent] think(s) that 1 should participate in physical activities that will make me out of breath at least three or more times in the next week. This was measured on a 7-point scale with the likely-unlikely end-points. Motivation to comply items were assessed using the following statement: Generally speaking, I want to do what my [salient referent] think(s). This was assessed using a 7-point scale anchored by the very much-not at all end-points. The belief-based measure of subjective norms was produced by multiplying the normative belief component by the motivation to comply component for each referent. Control beliefs. In another question on the freeresponse format questionnaire, the pupils were asked: Can you list all the things that might get in your way of you doing physical activities in the next week? Responses reflected both internal and external aspects of control: going out with friends, bad weather, doing homework, having other hobbies and being no good at it. The responses were used to formulate the control beliefs items, which asked respondents how much the given barriers would be likely to interfere with their physical activity participation. The items were anchored by the almost never-frequently word-pair. Selj-eficacy. Previous studies examining the influence of self-efficacy on physical activity behaviour have measured self-efficacy as confidence in the ability &; -

6 6 Hagger et al. to engage in the behaviour in the face of barriers (DuCharme and Brawley, 1995). Consequently, researchers have reported that, when presented with barriers, self-efficacy perceptions are a powerful predictor of exercise intention (Dzewaltowski, 1989). As a result, the three most frequent responses to the barriers question on the free-response questionnaire administered in the pilot study were used as the salient barriers for the self-efficacy items. These were going out with friends, bad weather and doing homework. Items to measure self-efficacy used the following statement: How confident are you in doing physical activities at least three times in the next week when (salient barrier)? ; responses were given on a 10-point scale representing 0 to 100%. The internal consistency of this scale was satisfactory (a = 0.81). Past behaviour. Frequency of past physical activity behaviour was assessed using the following questionnaire item: How often have you participated in physical activities that make you out of breath in the past 6 months. This was measured on a scale using the hardly ever-uery ofcen end-points. Although single-item selfreports of physical activity behaviour are not as comprehensive or detailed as the many self-report measures of physical activity behaviour available, single-item measures have demonstrated validity and reliability with objective measures of physical activity behaviour (Godin and Shephard, 1985). In the main study, the questionnaires were administered to groups of manageable size (maximum 100 participants per group) in a quiet classroom. Physical activities were defined for the participants as activities that you do in your spare time that make you out of breath or breathe very fast. The questionnaires were completed together as a class with the researcher reading each question aloud and then providing adequate time for the class to ask questions and make their responses. Results Behauioural beliefs and attitudes The concurrent validity of the measures of attitude, subjective norms, perceived behavioural control and selfefficacy was assessed by association with their respective belief-based measures as recommended by others (Armitage and Conner, 1999). Semi-partial correlations between the expectancy-value, belief-based measures of attitude and the attitude measure are presented in Table 1. It can be seen that doing physical activity for fun accounts for the largest proportion of variance in attitude (37.2%) when the remaining beliefs were partialled out. This suggests that young people s attitudes are dominated by beliefs about physical activity for providing enjoyment. Administration The adequacy of the item statements in the questionnaire for use with young people was tested in a pilot study on a small sample of 4 school pupils (25 boys, 19 girls) who were not participants in the main study (Hagger et al., 1997). The aim of this pilot study was to ensure that the language used and the scales adopted were appropriate. Feedback from the participants in this pilot study and from the class teachers suggested that young people were able to understand the questionnaire items. Normative beliefs and subjective norm Table 2 presents the semi-partial correlation coefficients between expectancy-value, belief-based measures of normative beliefs and the subjective norms measure. The beliefs of parents (2.25%) and friends (1.44%) accounted for the greatest proportion of the variance in subjective norms. The large proportion of unexplained variance suggests that there may be other aspects tapped by the social norms variable other than the beliefs of significant others. These other aspects may be represented by constructs like social support and descriptive Table 1. Semi-partial correlations (SP) between attitude and expectancy-value, belief-based measure of attitude, and individual beliefs and evaluations (mean +_ s) Belief-based attitude SP Behavioural belief Evaluation Doing physical activity helps me to get fit and to stay in shape 0.08* 4.53 rf: f 0.82 Doing physical activity is good fun 0.61* 6.90f S 0.84 Doing physical activity helps improve my skills 0.08* 4.98 S I 1.61 I might get injured doing physical activity k k 1.03 Doing physical activity might make me hot and sweaty 0.12* 5.75 f f 1.94 * F < 0.01, ** P< Job:

7 Young people s physical activity intentions 7 Table 2. Semi-partial correlations (SP) between subjective norm and expectancy-value, belief-based measure of subjective norm, and individual normative beliefs and motivation to comply (mean _+ S) Belief-based subjective norm SP Normative belief Motivation to comply My parent(s) think(s) * 5.94 k _ My grandparents think * 5.14 k f 0.84 My other family members think ** 5.35 f i 1.61 My friends think * 5.27 f k 1.03 My teacher thinks f f 1.94 * P< 0.01, ** P< Table 3. Semi-partial correlations (SP) between perceived behavioural control (PBC) and self-efficacy and control belief components (mean k s) Control belief PBC Self-efficacy Control beliefs Going out with friends * 5.01 k 2.14 Bad weather 0.15* 0.18* 4.49 f 2.30 Doing homework -0.09* -0.07** 3.39? 2.39 Having other hobbies 0.14* 0.08* 3.90 f 2.23 Being no good at it * 2.75 k 2.05 * P< 0.01, ** P< SP norms that have been recently included in the Theory of Planned Behaviour (Courneya et al., 2000). Control beliefs, perceived behavioural control and self-eficacy The partial correlations between control beliefs and the self-efficacy and perceived behavioural control variables are shown in Table 3. It can be seen that perceived behavioural control is significantly related to the external beliefs of bad weather, having other hobbies and doing homework. Self-efficacy is also associated with these external beliefs but also with the belief related to lack of competence, being no good at physical activity. This suggests that, although perceived behavioural control appears to be related to barriers alone, self-efficacy is also characterized by more internal, competence-related beliefs. It is possible that self-efficacy does not isolate the external aspects of control but that the perceived behavioural control variable does. Discriminant validity A confirmatory factor analysis using the maximum likellhood method (Bentler, 1989) was conducted to assess the construct and discriminant validity of the study variables. Since the statistics produced by the maximum likelihood estimation of covariance structures is likely to be sensitive to departures from multivariate normality, a robust method was used to estimate the model chi-square and fit indices according to the formulae set out by Satorra and Bentler (1988). Latent constructs were formed for the intention) attitude) perceived behavioural control and self-efficacy variables. Each latent variable was set a priori to explain the covariances between the items pertaining to each scale. One item loading for each factor was arbitrarily set to unity to ensure the model was identified. In addition, the non-latent, single-item measures of subjective norm and past behaviour were also included in the model. All of the latent and non-latent variables were made to correlate with each other, as is typical in confirmatory factor analysis models (Bentler, 1989). Model goodness-of-fit was assessed with the normed fit index (Bentler and Bonett, 1980), the comparative fit index (Bentler, 1990), the non-normed fit index (Bentler, 1990) and the standardized root mean square of the model residuals. The three fit indices should equal or exceed 0.95 for adequate fit of model to the data and the standardized root mean square of the model residuals should be 0.08 or less to represent acceptable goodness-of-fit (Hu and Bentler, 1999). The analysis resulted in a model that exhibited a very large and significant X2-value relative to available

8 8 degrees of freedom. While large and significant x2- values are indicative of poor fit, Byrne (1994) and Bentler (1989) suggested that the goodness-of-fit x2- values in covariance structure analyses are very sensitive to sample size and are therefore often significant. It is important, therefore, to assess model fit using criteria that account for sample size, such as the normed fit index, comparative fit index and non-normed fit index. Hagger et al. According to these criteria, the model demonstrated satisfactory goodness-of-fit (Table 4, Model 1). Most items demonstrated adequate factor loadings approaching 0.70, suggesting that the latent factor accounted for approximately half of the explained variance in the items. Factor correlations, factor loadings and item error terms from the confirmatory factor analysis are shown in Fig. 1. Discriminant validity Table 4. Goodness-of-fit statistics for the estimated models Model SB-x" df NFI CFJ NNFI SRMSR Model 1: CFA Model 2: TPB Model 3: TPB* Model 4: TPB* + past behaviour Note: Model 1 = confirmatory factor analysis (CFA), Model 2 =test of the Theory of Planned Behaviour (TPB), Model 3 = augmented version of the Theory of Planned Behaviour (TPB*) including self-efficacy, Model 4 = augmented version of the Theory of Planned Behaviour (TPB*) including self-efficacy and past behamour. Abbreviations: SB-x2 = Sattora-Bentler scaled chi-square value, NFI = normed fit index, CFI = comparative fit index, NNFI = non-normed fit index, SRMSR = standardized root mean square of the model residuals. Efficacy Homework 0.68' Weather Intend 0.62: Intention 0.74 Plan 0.67* 0.56* 0.68* Good-Bad 0.56' Attitude \ \\-: Entirely Fun-Unpleasant I h \/ Exciting-Boring I I \ \/ I up to me 0.53" \ \ \0.20* - Personal control \ \ \ I 0.41* 0.60* 0.75' 0.67' 0.85' Subjective Fig. 1. Confirmatory factor analysis of the Theory of Planned Behaviour variables and self-efficacy. * P <.01

9 Young people s physical activity intentions 9 of the factors is supported if the correlation between the two factors is less than unity by an amount greater than 1.96 times the standard error, assuming alpha is set at 5% (Bagozzi and Kimmel, 1995). According to this criterion, all the constructs in the confirmatory factor analysis achieved discriminant validity. (The covariance and intercorrelation matrices for the items from the present study are available from the first author on request.) Testing study hypotheses The hypotheses relating to the Theory of Planned Behaviour variables were tested using a series of nonstandard structural equation models (Bentler, 1989). The models were estimated using the robust maximum likelihood method, and the normed fit index, comparative fit index, non-normed fit index and standardized root mean square of the model residuals were used to estimate the overall goodness-of-fit of the models. Initially, the predictive validity of the traditional Theory of Planned Behaviour model stipulated by Ajzen (1985) was tested. This model hypothesised that physical activity intentions would be a function of attitudes, subjective norms and perceived behavioural control. The model displayed an adequate fit of the hypothesized model with the data (Table 4, Model 2). Intentions were significantly predicted by attitudes (standardized coefficient = 0.57, P < 0.01) and perceived behavioural control (standardized coefficient = 0.20, P < O.Ol), and attitudes and perceived behavioural control accounted for 48.2% of the variance in intentions. Contrary to the findings of other studies, subjective norms had no influence in the model. This model is shown in Fig. 2. An augmented version of the Theory of Planned Behaviour was tested to determine the influence of self-efficacy in the prediction of young people s physical activity intentions. Self-efficacy was included as an independent predictor of physical activity intentions. This model demonstrated adequate fit with the data (Table 4, Model 3). Self-efficacy was a strong predictor of intentions (standardized coefficient = 0.58, P C 0.01). The 95% confidence intervals (CI,,5) were calculated for the standardized structural coefficients in Model 2 and Model 3. No overlap occurred for the confidence intervals of the attitude4ntention standardized coefficients in Model 2 (CIo.95 = ) and Model 3 (CIo,g5 = ). This provides evidence to support the prediction that the inclusion of self-efficacy significantly attenuated this path. (The authors are grateful to David Markland for suggesting the use of confidence intervals to determine the significance of the attenuation effects in the structural Fig. 2. Structural diagram for the non-standard structural equation model (Model 2) representing the Theory of Planned Behaviour for young people s physical activity. D, = structural disturbance term for latent intention construct. * P < Subjective No; i\ / / I \ Efficacy } Fig. 3. Structural diagram for the non-standard structural equation model (Model 3) representing the Theory of Planned Behaviour for young people s physical activity augmented to include self-efficacy. D, = Structural disturbance term for latent intention construct. * P c equation models.) The inclusion of self-efficacy also completely swamped the effect of perceived behavioural control on intentions, reducing it to zero. This model accounted for a greater amount of variance in physical activity intentions (66.4%) than Model 2. A structural equation model diagram of this model is shown in Fig. 3.

10 10 Hagger et al. Fig. 4. Structural diagram for the non-standard structural equation model (Model 4) representing the Theory of Planned Behaviour for young people s physical activity augmented to include self-efficacy and controlling for past behaviour. D, = structural disturbance term for latent intention construct. * P < The role of past physical activity behaviour in the model was examined by re-estimating the model in Fig. 2, but with structural paths freed from past behaviour to all of the model variables. This effectively controlled all the social cognitive variables in the augmented Theory of Planned Behaviour for the influence of past physical activity behaviour. The model fit statistics approached but did not satisfy Hu and Bentler s (1999) cut-off values for indices of good fit (Table 4, Model 4). Therefore, some caution should be used when interpreting this model as an adequate explanation of the data from the present study. Past behaviour significantly predicted all the model variables, with the strongest influences on self-efficacy (standardized coefficient = 0.63, P < 0.01) and attitude (standardized coefficient = 0.50, P < 0.01). Examination of the confidence intervals for the structural coefficients in Model 4, and a comparison of them with those from Model 3, showed that no significant attenuation of the model relationships occurred as a result of the inclusion of past behaviour. The hypothesis that past behaviour would attenuate the model relationships was therefore rejected on the basis of this model. Overall, the influence of past behaviour on intention was both direct and indirect via attitude and selfefficacy. This evidence suggests that the Theory of Planned Behaviour as specified by Model 3 may be the most appropriate model for describing these data. The structural coefficients for Model 4 are shown in Fig. 4. Discussion The aim of the present study was to examine the influence of self-efficacy and past behaviour on young people s physical activity intentions in an augmented version of Ajzen s (1985) Theory of Planned Behaviour. The latent variables of intention, attitude, perceived behavioural control and self-efficacy achieved concurrent and discriminant validity as shown by the semi-partial correlations with the belief-based measures and the confirmatory factor analysis, respectively. As expected, attitudes and perceived behavioural control were significant predictors of young people s physical activity intentions, whereas subjective norms were not. Subsequent models showed that self-efficacy had a strong influence on physical activity intentions as hypothesized and accounted for additional variance in intentions. Self-efficacy also significantly reduced the influence of attitudes on intentions and reduced the influence of perceived behavioural control on intentions to zero. The inclusion of past behaviour in the augmented model revealed that past behaviour significantly predicted all the Theory of Planned Behaviour variables and self-efficacy but, contrary to our hypotheses, did

11 Young people s physical activity intentions 11 not attenuate the influence of attitude, subjective norm, perceived behavioural control and self-efficacy on intention. The Theory of Planned Behaviour and young people s physical activity intentions The strong effect of attitudes in the present study has been corroborated in several other studies on physical activity and young people (e.g. Godin and Shephard, 1986; Atsalakis and Sleap, 1996). Reviews of the Theory of Planned Behaviour have concluded that the dominant role of attitudes in the prediction of intention is supported in the domain of physical activity (Godin, 1994; Blue, 1995; Hausenblas et al., 1997), as well as across many other behaviours (Sheppard et al., 1988). These results suggest that, for young people and physical activity, the formation of behavioural plans or intentions is dominated by attitudes towards that behaviour. The dominant role of personal attitude on physical activity intentions is also accompanied by a lesser role for social pressures to engage in such behaviours. The non-significant influence of subjective norms on intention in the present study supports the contention that subjective norm has a peripheral role in the Theory of Planned Behaviour. Although contrary to the operationalization of the Theory of Planned Behaviour by Ajzen (1985), this is not a surprising finding, given current empirical tests of the model. Many studies have shown a greatly reduced or non-significant effect of subjective norms on physical activity intentions among young people (Shephard and Godin, 1986) and adults (Hausenblas et al., 1997). Conceptually, the lack of contribution by subjective norms in the model is counterintuitive, since it might be surmised that young people usually have to comply with the wishes of parents, which is less likely to be a strong influence among adults. This has been shown to be the case in studies of the social cognitive predictors of the physical activity intentions of young people (Shephard and Godin, 1986; Theodorakis et al., 1991). However, as the participants in the present study were entering their teenage years, it may be that the new-found independence associated with social development in early adolescence compelled them to hold parental and peer pressure in lesser regard than their personal volition. The non-significance of social influences on physical activity intentions may also be a result of the measures used to quantifi the construct. Povey et al. (2000) stressed the need to move away from single-item measures and focus on a more diverse set of measures for the subjective norms variable. Future tests of the Theory of Planned Behaviour for the physical activity behaviour of young people may, therefore, need to account for other social influences on physical activity intentions such as peer approval and descriptive norms. The role of perceived behavioural control and self-eficacy The significant influence of perceived behavioural control on physical activity intentions corroborates the results of several studies that have demonstrated a significant effect of this variable in a physical activity context (Godin and Shephard, 1986; Atsalakis and Sleap, 1996). The strength of this association in the present study is slightly lower than in other studies and in meta-analytic reviews of the Theory of Planned Behaviour for physical activity behaviour (Hausenblas et al., 1997). Indeed, some reviewers have proposed that the influence of perceived behavioural control on intention is at least as strong as that of attitude in a physical activity context (Blue, 1995; Hausenblas et al., 1997). However, this influence may be inflated because the reviews did not account for the inclusion of other confounding variables such as self-efficacy. The inclusion of self-efficacy in the present study was, therefore, a crucial one. Its inclusion aimed to help disentangle the differential effects of perceived behavioural control and self-efficacy on young people s physical activity intentions. The attenuation to zero of the relationship between perceived behavioural control and intention, as a result of the inclusion of self-efficacy in the model, indicates that the variance that perceived behavioural control shares with intention is accounted for by self-efficacy. This supports the findings of other studies that have made the distinction between self-efficacy and perceived behavioural control. Yordy and Lent (1 993) and Armitage and Conner (1999) demonstrated that selfefficacy was an important predictor of physical activity and eating a low-fat diet intentions respectively, but perceived behavioural control was not. In these studies, self-efficacy was defined by an internal and external distinction, which is similar to distinction between perceived behavioural control and self-efficacy made in the present study. The unique contribution that the present study makes to this distinction is that it focuses on a conceptualization of internal perceptions of control through abilities (internal perceptions of efficacy) in the face of barriers, Therefore, if a young person feels that they have the capacity and faculties to participate in the given behaviour in the face of salient external constraints, then they are more likely to participate in the behaviour. This is consistent with Bandura s (1997)

12 12 conceptualization of the self-efficacy beliefs construct and its influence on motivation. Bandura suggested that, if competence in the behaviour is sufficient to overcome all barriers or external compromising influences over control, then the behaviour is likely to be enacted. Controlling for past behaviour The present study is, to our knowledge, the fist to examine the influence of past behaviour on young people s physical activity intention, attitudes, subjective norms, perceived behavioural control and self-efficacy. Controlling for past behaviour did not attenuate the influence of attitudes, subjective norms, perceived behavioural control and self-efficacy on physical activity intentions in the present study as hypothesized. This is contrary to the findings of studies on adults and physical activity (Godin et al., 1993; Yordy and Lent, 1993; Bagozzi and Kimmel, 1995) and other behaviours (Norman and Smith, 1995). Furthermore, meta-analyses (Conner and Armitage, 1998; Ouellette and Wood, 1998) and narrative reviews (Sutton, 1994) performed on several behaviours also support the attenuation of the social cognitive influences on intention by the direct relationship between past behaviour and intention. These authors argue that regular enactment of the behaviour in the past results in reduced cognitive deliberation over the upcoming behaviour, This is why past behaviour accounts for some of the cognitive influences on intention. The lack of attenuation of the social cognitive influences on intention by past behaviour in the present study suggests that past behavioural engagement does not have the same cognitive-reducing influences on young people s physical activity intentions that has been reported in adults (e.g. Yordy and Lent, 1993). Instead, young people base their decisions to participate in physical activities on their situation-specific evaluation of their future participation in physical activity. This is corroborated by the Theory of Planned Behaviour model being a better description of the data than the model that included past behaviour. A possible reason why the physical activity intentions of young people are dominated by attitudes and self-efficacy may be that they do not have such elaborate and organized past behavioural patterns as do adults. It is important to stress that past behaviour did have an influence on the social cognitive predictors of intention and on intention itself, but the effects were exclusively additive. Furthermore, there are indirect and direct effects of past behaviour on intentions. This suggests that current evaluations of attitudes, perceived behavioural control and self-efficacy are necessary to Hagger et a1. translate past action into intentions to participate in physical activity in the future. Limitations of the study and further research As with any cross-sectional study, caution must be exercised when assessing the generalizability of the results. Cross-sectional studies, particularly those with large sample sizes as in the present study, provide useful information as to the possible predictors of psychological constructs like intentions. However, causality can only be inferred based on the relationships provided at that instant in time. Further support needs to be provided through longitudinal and experimental studies. Future research that tests the influence of selfefficacy on the physical activity intentions of young people over time or manipulates self-efficacy towards physical activity in an experimental paradigm may provide additional support for the current cross-sectional findings. It should also be noted that the models in the present study had X2-values that were very high relative to the degrees of freedom (df) available. Structural equation analysts have often adopted the ratio between the goodness-of-fit xz and degrees of freedom k2/df ratio) as an indicator of good fit; values of 2.0 or less have been cited as a criterion for goodness-of-fit (Marsh et al., 1988). By this criterion, the models in the present study do not achieve good fit. However, Marsh et al. indicated that a strong, statistically significant sample size effect occurs with this stand-alone index of good fit, such that X2/df ratios tend to be inflated with increasing sample size. Given that the sample size in the present study was relatively large (n = 1152) and close to the upper end of Marsh and co-workers simulation (n = 1600), it is reasonable to assume that this lack of fit may be an artefact of sample size. Therefore, despite the lack of fit indicated by the X ldfratios for the models in the present study, incremental fit indices that are not influenced by the sample size effect, such as the non-normed fit index, may be more appropriate for determining the adequacy of the model fit. Since Models 1,2 and 3 in the present study achieve good fit on the basis of these criteria, we are confident that the hypothesized models adequately describe the present data. Conclusions To conclude, it is necessary to include self-efficacy and past behaviour in the Theory of Planned Behaviour to understand the social cognitive influences on the physical activity intentions of young people. This study has shown that it is the internal aspects of control in the -B.CL