Abstract
This study investigated whether gender-role related traits agency and communion contribute to successful health behavior change, in an interplay with domain-specific psychosocial factors, namely, agency, mediated by health-related self-efficacy, and communion, moderated by social support. Data from women (N = 282) participating in the GOAL Lifestyle Implementation Trial were analyzed using structural equation modeling. Agency and increase in self-efficacy both independently predicted waist circumference reduction in the 1-year follow-up. Individuals high in communion succeeded in waist reduction only if they received social support. Initial self-efficacy increase predicted 3-year waist reduction. Gender-role orientation, together with social environment, influences behavior change intervention outcomes.
Introduction
The prevalence of type 2 diabetes (T2D) is on the rise worldwide, but T2D could be prevented with lifestyle changes; for example, a weight loss of even 5 percent can delay or prevent the onset of the disease (Paulweber et al., 2010). Weight loss is also related to improvements in quality of life (Wright et al., 2013). However, very few of those pursuing weight loss achieve long-term results (Wadden et al., 2002). Various psychosocial factors predict health behavior change. This study investigates the relevance of gender-role orientation—agency and communion—in a T2D prevention intervention with a special interest on mediation and moderation of the effects by self-efficacy and social support.
Gender-role orientation refers to personality characteristics that are traditionally associated with either men or women, referred to as psychological masculinity and femininity, or agency and communion (Bakan, 1966; Bem, 1974; Helgeson, 1994; Spence and Helmreich, 1978). Agency is characterized by a focus on the self and autonomy and demonstrated in instrumental traits such as decisiveness, ambition, and assertiveness; communion is characterized by a focus on other people and relationships and demonstrated in expressive traits such as gentleness, compassion, and helpfulness. As a result of gender-role socialization, men are more likely to develop agentic, self-assertive (“masculine”) qualities, while women develop communal, other-oriented (“feminine”) qualities (Helgeson, 1994; Spence and Helmreich, 1978).
Both high agency and high communion are related to adaptive functioning in many areas, including healthy psychological adjustment and emotional well-being, among both men and women (Helgeson, 1994; Huselid and Cooper, 1994; Saragovi et al., 2002). Agency consistently predicts better psychological and even physical health (Annandale and Hunt, 1990; Bassoff and Glass, 1982; Taylor and Hall, 1982). It is also positively associated with health interest, physical activity, and body weight maintenance (Danoff-Burg et al., 2006; Robbins et al., 1991). Among cardiac patients, it predicts increases in well-being and mental but not physical functioning (Fritz, 2000). Communion associates with better relationships (Helgeson, 1994) and more received and provided social support (Ghaed and Gallo, 2006; Reevy and Maslach, 2001), but its associations with eating are mixed (Hepp et al., 2005; Mosher and Danoff-Burg, 2008). Agency and communion are not opposite ends of a continuum: scoring high on both have been suggested to relate to optimal outcomes (Bem, 1974).
More proximal, psychosocial factors might mediate or interact with the more general personality attributes (McEachan et al., 2010) in producing behavior change outcomes. Self-efficacy reflects an individual’s beliefs about his or her capabilities to execute a sequence of actions (Bandura, 1986, 1997). Strong self-efficacy facilitates selecting more challenging tasks and fosters sustained effort in them, also in the face of obstacles (Bandura, 1997). Increase in self-efficacy in interventions is associated with improvements in health behaviors and their health outcomes (e.g. Laatikainen et al., 2012). Both agency and self-efficacy share the notion of the self as an active agent. However, whereas agency is a general disposition, demonstrated across diverse behaviors, self-efficacy can vary across contexts and domains. Given the theoretical and empirical associations between the two (Choi, 2004; Hermann and Betz, 2004), high agency may influence weight loss through health-related self-efficacy. Agentic individuals might be more prone to hold more self-efficacious beliefs across different life domains, including health behavior. They might also more quickly build self-confidence in a novel behavioral domain, due to generally ambitious tendencies and overall high self-confidence. In a health behavior intervention, this would show as a steeper increase of self-efficacy in executing health-enhancing behaviors (“dynamic mediation”, see Hankonen et al., 2010).
In addition, environmental factors such as social support might moderate the personality–outcome relationship. Social support has been theorized to facilitate health behavior change (Schwarzer and Fuchs, 1996). However, empirical research on the role of social support in weight loss and its maintenance has produced mixed findings (Dombrowski et al., 2012; Elfhag and Rössner, 2005), suggesting indeed a possibility of interaction effects. Moreover, studies investigating whether social support is equally important in predicting physical activity among men and women are inconsistent (Molloy et al., 2010; Sallis et al., 1999), and as one explanation, an interaction effect between the gender-related trait communion and social support has been suggested (Molloy et al., 2010).
Communion by definition entails high concern for social environment. It might, on the one hand, enable asking for social support (e.g. Hirokawa and Dohi, 2007; Reevy and Maslach, 2001) and help to mobilize one’s social resources to facilitate goal achievement. On the other hand, it might also lead to more dependence on support when facing obstacles. Individuals high in communion and concerned about social relationships in general may especially suffer from low social support and benefit from high social support. For them, the reciprocity of support can be expected to be relevant. In other words, communion should be associated with better weight loss but only among those with high social support. Social support would “fuel” the weight-loss efforts for those high in communion. Two prior studies lend support to this moderation hypothesis: Cheng (1999) found that communion was related to depression but only among those whose social support decreased, indicating that social support acts as a buffer for negative mental health especially for communal individuals. Alcohol treatment utilizing social support yielded better outcomes for those high in communion (John et al., 2008).
The present study
The purpose of this study was to investigate how agency, communion, and more proximal psychosocial factors (health-related self-efficacy and social support) were related to 1-year changes and 3-year changes in waist circumference among women in a lifestyle intervention. We expected to find that (1) high agency was associated with reductions in waist circumference, (2) the effects of agency were mediated by larger increases in self-efficacy during the intervention (“dynamic mediation model”), and (3) communion predicted reductions in waist circumference but only among those with high social support. No prior study has investigated the mediation between agency and self-efficacy or the interaction between communion and social support in the context of physical health goals, such as obesity reduction, in a long-term follow-up.
Method
Study setting and sample
Participants were middle-aged (50–65 years) women (N = 282) who took part in the GOAL Lifestyle Implementation Trial, a group intervention to prevent T2D. Intervention objective was to facilitate the adoption of physical activity and healthy nutrition objectives to reduce the risk for T2D (Tuomilehto et al., 2001). (For a detailed description of objectives, sample, and design, see Absetz et al., 2007; Uutela et al., 2004.) The ethical commission in Päijät-Häme hospital district and the Ethical Committee of the National Public Health Institute gave their approval for the project. Participants provided a written informed consent, and they were treated according to the American Psychological Association (APA) ethical standards.
Measures
Measurements were conducted at baseline (T1), after the intensive phase of the intervention at 3 months (T2), at 1 year (T3), and at 3 years (T4).
Gender-role orientation was measured with Personal Attributes Questionnaire (PAQ; Spence et al., 1973) at T4, a semantic differential of attributes that are stereotypically viewed typical of the male role, female role, or both. The agency scale taps instrumental, self-assertive qualities (e.g. competitive, independent, and stands up well under pressure), and the communion scale taps expressive, interpersonal-oriented qualities (e.g. warm, friendly, and aware of others’ feelings). The participants rated themselves on these qualities on a 5-point scale. In middle-aged adults, personality traits are relatively stable (Caspi et al., 2005).
Using exploratory and confirmatory factor analyses, the measurement model of gender-related traits agency and communion was specified, and some conceptually and theoretically justified respecifications were done. The analyses suggested a 2-factor solution with separate factors for agency and communion, confirming the theory. The final measurement model had an adequate fit (χ2(49) = 70.43, p = .024, comparative fit index (CFI) = .950, root mean square error of approximation (RMSEA) = .046), retaining 6 items for agency (independent, competitive, never gives up, self-confident, feels very superior, and stands up well under pressure) and 5 items for communion (easy to devote self completely to others, helpful to others, aware of others’ feelings, understanding of others, and warm in relations with others). The standardized loadings ranged from .42 to .73. Despite elimination of two agency and three communion items, the brief scales and the original scales correlated very highly (agency: r = .95; communion: r = .93, p < .001). These more parsimonious measures also enabled a better sample size to parameter ratio for the structural equation modeling (SEM). The measurement model specified two correlated residuals for agency (“competitive” with “independent” and “feels very superior” with “independent”) and one for communion (“understanding of others” and “warm in relations with others”). The factors agency and communion were positively correlated (r = .38, p < .001).
Health related self-efficacy was measured at T1 and T2, with 6 items referring to one’s confidence in dealing with the difficulties, temptations, and barriers of health-related lifestyle (T1/T2 α = .78). The items were formulated as “I can”-statements (e.g. “I can resist temptations when I know they are bad for my health”). Possible responses ranged from completely disagree (1) to completely agree (4). The measurement model for health-related self-efficacy (Hankonen et al., 2010) included three parcels to create more reliable indicators (Bandalos and Finney, 2001), and factor loading invariance across time was enforced.
Social support during the intervention over the past 3 months was measured at T2 with a shortened version of a scale for participation in physical activity (Sallis et al., 1987) (α = .86) (e.g. “During the past three months, how often people close to you, family, friends or relatives gave you helpful reminders to exercise”). Responses ranged from never (1) to often (3). To generate a more parsimonious measure, again, only highly redundant items from the original scale were excluded (the brief and original scale r = .96 at T2, p < .001).
Waist circumference was measured by study nurses at T1, T3, and T4. Abdominal obesity is a major predictor of T2D (Janiszewski et al., 2007) and a reliable indicator of participants’ health behavior change (both physical activity and diet). Importantly, physical activity reduces abdominal obesity also in the absence of weight change (e.g. Ross et al., 2000).
Study drop-out and intervention exposure
Out of 282 women at baseline, 207 (73.4%) participated at the 3-year follow-up. Exposure to intervention sessions was not related to agency, communion, self-efficacy, social support, or T1 waist. Study drop-out was not related to social support, but participants with lower self-efficacy at T2 were less likely (p < .01) to participate at T3 and T4, and those with larger T1 waist circumference were less likely (p < .05) to participate at T2 and T3. As these variables are included in the model, the full information maximum likelihood (FIML) estimation used reduces bias in parameter estimates for all parameters of the model (Graham, 2009).
Statistical analysis
To conduct SEM analyses, we used Mplus Version 6.0 with maximum likelihood estimation (maximum likelihood robust (MLR) estimation for models testing hypothesis 3). PASW Statistics 18.0 was used to obtain descriptive statistics and to conduct simple slope analysis for interpreting the interaction between communion and social support. FIML estimation was used to handle missing data in SEM. FIML allows using all available information efficiently without biasing estimates compared to other available missing data techniques (Graham, 2009).
The CFI and RMSEA were used to assess goodness of fit. Considering the sample size and model complexity of the present study, CFI > .92 and RMSEA < .07 were interpreted as an indication of satisfactory model fit (Hair et al., 2006). To test the specific hypotheses, the χ2 difference test was used. To compare the fit between alternative nested models, a series of nested models were run. If the χ2 difference test resulted in a nonsignificant difference, the more parsimonious (constrained) model was preferred. For testing hypothesis 3, χ2 difference test using the log-likelihood values was employed. Bayesian Information Criterion (BIC) and Sample Size Adjusted BIC were investigated, for which lower values indicate better fit. As the study hypotheses were directional in nature, one-tailed testing was performed in evaluating the significance of the parameter estimates.
Changes in self-efficacy and waist circumference were modeled with latent change regression models (McArdle, 2009), where the variable measured at follow-up (e.g. T2) is regressed with a weight of 1 on the T1 variable, and a latent change score is estimated similarly with a weight of 1 on T2 variable. Change-regression models are suitable for analyzing the processes in an intervention study (McArdle, 2009).
To test the three hypotheses, a series of structural models were calculated. First, we specified a model with only agency and communion predicting waist circumference change directly (see Figure 1, Model 1). In Model 2, we added T1–T2 change in self-efficacy as a predictor and a potential mediator and T2 social support as a predictor (see Figure 1, Model 2). Within Model 2, we also tested hypothesis 2 (whether self-efficacy increase acted as a mediator of agency on waist circumference decrease, by examining indirect effects of agency through self-efficacy on waist circumference change) as well as hypothesis 3 (whether T2 social support interacted with communion in predicting waist circumference change, using the Latent Moderated Structural Equations (LMS) method). The set of separate models containing the interaction is not shown. All sets of models were also conducted for waist circumference T1–T4.

Model 1: Direct effects. Model 2: Dynamic mediation model.
The communion × social support interaction hypothesis was also tested by calculating a separate regression analysis containing the main effects for communion and social support, and the interaction term of these. Communion and social support scales were both mean-centered prior to calculating the interaction term (Aiken and West, 1991). In order to interpret this interaction effect, the association of communion with waist circumference change was analyzed at ±1 standard deviation (SD) from the mean of social support (Aiken and West, 1991). While latent change regression models were used in other analyses, we used a simple difference score (T2 value − T1 value) in the simple slope analyses as well as in the bivariate correlations (Table 1) obtained in PASW.
Correlations between study variables.
p < .001; **p < .01; *p < .05; †p < .10.
Calculated as difference scores: T1 value subtracted from T2 value.
Results
Descriptive statistics
Means of agency and communion were 3.43 (SD = .52) and 3.96 (SD = .51), and they were normally distributed. Mean of social support was 1.69 (SD = .69). The majority of the sample was abdominally obese: Mean of waist circumference was 102.9 cm (SD = 11.1 cm) at T1 and 101.6 cm (SD = 11.6 cm) at T3. Changes in waist circumference ranged from −16.0 cm to 13.2 cm (1-year) and −20.0 cm to 21.0 cm (3-year). Changes were distributed normally, showing no substantial skew or kurtosis. Communion was positively correlated with social support and self-efficacy, and agency was positively correlated with self-efficacy (Table 1).
Hypothesis 1: agency predicts waist circumference change
To explore the structural relationships between the variables, we first specified a model with agency and communion predicting 1-year (T1–T3) change in waist circumference, without any control variables. The model (χ2(67) = 118.63, p < .001, CFI = .939, RMSEA = .052) suggested a statistically significant direct effect for agency (standardized regression effect: −0.17, p = .04) but none for communion (.07, p = .26) on the outcome. However, the χ2 difference test indicated that the fit of an alternative model that constrained the effect of agency to be zero was worse at the level of p = .087 (Δχ2 = 2.93, Δdf = 1, p = .087). The χ2 difference test indicated an equally good fit, and hence, preference for an alternative model constraining the effect of communion to zero (Δχ2 = 0.44, Δdf = 1, p = .509) (model fit indices and details of the χ2 difference test available from first the author on request). The interaction between agency and communion was also tested, but the interaction term did not yield significance (p = .735).
Hypothesis 2: agency–self-efficacy mediation
Model 2 tested whether the change in self-efficacy during the intervention predicted waist circumference change (χ2(246) = 400.22, p < .001, CFI = .923, RMSEA = .047). Larger increases in self-efficacy and higher agency were related to reduction in waist circumference by T3, confirmed also by χ2 difference tests (p < .05). Communion and social support were not significant predictors.
Dynamic mediation was specified in Model 2. Here, agency did not predict change in self-efficacy. Furthermore, the indirect effect of agency through self-efficacy change was negligible (−0.017, p = .438), suggesting no mediation.
Hypothesis 3: communion × social support interaction
The interaction between communion and social support was tested within Model 2 (the models including the interaction term not shown as figures). As expected, communion was positively (but only moderately) correlated with social support (.19, p = .037). The interaction term between communion and social support was significant in predicting waist circumference change in Model 2 (−4.79, p = .032). In a separate model without any covariates, the interaction term was also significant (−4.29, p = .033). A χ2 difference test comparing Model 2 to a nested model that constrained the regression effect between the interaction term and waist circumference change to be zero also suggested that the interaction was significant (p < .05).
This interaction effect was interpreted using simple slope analysis (Aiken and West, 1991) (Figure 2): Those scoring high on both social support and communion lost the most weight, but those with low social support and high communion lost the least weight (communion β for those with high social support: −1.46, p = .037; communion β for those with low social support: .64, p = .23).

The interaction between communion and social support.
The same models were tested with the 3-year (T1–T4) waist circumference change. Model 1 (χ2(67) = 97.381, p = .009, CFI = .955, RMSEA = .040) indicated that neither agency (−.07, p = .188) nor communion (.09, p = .175) had significant direct effects. Model 2 (χ2(247) = 375.277, CFI = .930, RMSEA = .043) indicated that while other predictors were unrelated, the T1–T2 change in health-related self-efficacy significantly predicted the T1–T4 waist change (−.19, p = .019). Finally, the interaction hypothesis did not receive support in models predicting T1–T4 waist change (p = 0.192).
Discussion
This study aimed to examine the dynamics of gender-role orientation and more proximal and specific psychosocial factors in reducing abdominal obesity. As expected, (1) higher agency was associated with a greater 1-year waist circumference reduction. Contrary to our hypothesis, (2) agency’s effects were not mediated by increase in self-efficacy. Instead, both factors were independently related to 1-year waist circumference reduction. Consistent with our hypothesis, (3) communion had no direct association with waist circumference reduction, but was associated with reduced waist circumference when high social support was reported. Testing the same hypotheses with the 3-year waist circumference change as the outcome, the only significant predictor was the initial self-efficacy increase.
The findings on agency imply that it is adaptive and beneficial in the short term. The items in the agency scale tap personality attributes relevant in individual goal pursuit (e.g. Baum and Locke, 2004), such as independence, competitiveness, and the tendency of not giving up. Personality traits, such as self-control, have also been linked with better weight-loss outcomes (Will Crescioni et al., 2011). The mediation hypothesis did not receive support, suggesting that the effect was not facilitated either by initially higher health-related self-efficacy or by larger increases in it. Other aspects of agency, beyond those related to self-confidence, facilitate the change process. Although at the baseline these factors were positively and significantly related, the findings suggest that domain-specific self-efficacy expectancies are modifiable regardless of agency.
Fritz (2000) failed to find associations between agency and improvements in physical functioning among cardiac patients, but the data were not analyzed separately for genders. In fact, the association between agency and health behaviors has been reported to be weaker among men than among women (Robbins et al., 1991). Hence, whether the role of agency in obesity reduction applies to men remains to be explored in future studies.
Consistent with our moderation hypothesis, women high in communion benefited from high and suffered from low social support in their weight-loss pursuit. Communion is associated with received as well as provided social support (Fritz and Helgeson, 1998; Reevy and Maslach, 2001). Being on the giving end may drain one’s energy and lead to negligence of one’s own needs, whereas reciprocal support results in optimal outcomes. In line with other findings (Cheng, 1999) and a recent hypothesis (Molloy et al., 2010), the combination of high communion and high social support had the most beneficial effects. 1
Finally, all the effects of gender-role orientation had ceased by the 3-year follow-up. The intervention may have increased goal homogeneity only temporarily, so that in a long-term follow-up, the participants differentiate: some might have given up the behavior change and the weight-loss goals, and other external determinants (not only one’s personality traits) might have kicked in. Furthermore, agency might be related to a relatively short-term goal achievement only. The results differ between the follow-ups and need replication.
Results regarding the positive effects of self-efficacy are in line with extensive earlier evidence (e.g. Bandura, 1997). It is notable that the magnitude of self-efficacy change during the first 3 months had lasting effects on not only 1-year but also on the 3-year waist circumference reduction.
While personality traits are rarely seen as malleable (Loehlin et al., 2005), similar personal resources can be promoted through teaching agentic behavioral skills, and thus accentuate appropriate selfishness and an agentic “cognitive focus on getting the job done” among women (Bem, 1974: 156)—if not throughout the whole behavioral spectrum but at least regarding the specific domain of health behavior and weight loss. Competitiveness might be especially beneficial in weight-loss attempts: this item showed the strongest correlation to waist circumference reduction in our data. In fact, setting challenging goals (De Vet et al., 2013) has shown to be a potentially important contributor in weight loss due to its effects on effort. Another practical implication is to teach strategies to elicit social support for high-communion women.
A major limitation of the study is the retrospective measurement of agency and communion. We cannot rule out a reverse causal pathway, for example, the waist reduction causing more agentic self-assessments, or causation by a third, unknown factor of changes both in agency and waist circumference. However, as findings from longitudinal personality research indicate relative stability of personality traits, especially in middle-aged population (Caspi et al., 2005), they are likely to have been very similar had they been measured 3 years earlier at the baseline. As gender-role orientation was not associated with 3-year waist circumference change, common method bias due to simultaneous measurement can be ruled out. Another limitation is the social support variable, which taps social support related to physical activity only, not dietary behavior, also relevant for weight loss. However, physical activity improves clinical risk factors and reduces abdominal obesity even in the absence of body mass index (BMI) change (e.g. Ross et al., 2000), and hence, is more relevant in this study.
First, the study strengths include using an objectively measured behavior outcome—waist circumference—instead of self-reported variables. Second, our study question was novel: agency and communion in relation to dynamic psychosocial and health behavior change outcomes in a longitudinal intervention design, extending as far as 3 years. Third, we employed state-of-the-art statistical methodology to model change, to handle missing information, and to test alternative structural models.
This study, to our knowledge, was the first to explore gender-role related traits in relation to success in a lifestyle intervention, with an exceptionally long follow-up, and to demonstrate one potential pathway through which agentic and communal traits can contribute to better health. We also presented a rare prospective longitudinal finding, linking the initial 3-month increase in self-efficacy with 3-year abdominal weight loss. Gender-role orientation seems to be relevant for behavior change processes in interventions. However, in the long-term, strong self-efficacy is more significant for goal attainment.
Footnotes
Funding
This research received funding from the Finnish Social Insurance Institution (Kela) and the Academy of Finland.
