Exploring the transfer of self-efficacy: academic self-efficacy predicts exercise and nutrition self-efficacy / Una exploración de la transferencia de la autoeficacia: la autoeficacia académica predice la autoeficacia nutritiva y del ejercicio físico

Abstract In this research, we investigated the transfer of self-efficacy (SE) beliefs from one domain to another. Specifically, we explored whether academic SE beliefs might be able to predict exercise (Study 1) and nutrition (Study 2) SE. Additionally, to explore the underlying mechanisms, participants reported on domain confidence, perceived similarity and domain importance. Specifically, in Study 1, undergraduate students reported on their academic SE beliefs, confidence in the academic and exercise domains and their exercise SE beliefs (n = 50). In Study 2, participants reported on their academic SE beliefs, confidence in the academic and nutrition domains and their nutrition SE beliefs (n = 71). Moreover, in both studies, the perceived importance and the perceived similarity between domains were measured. Multiple regression analyses revealed that academic SE beliefs, academic confidence and exercise confidence were significant predictors of exercise SE beliefs (Study 1); for Study 2 nutrition SE was predicted by academic SE, academic confidence, nutrition confidence and nutrition importance diet. Additionally, we found a partial suppression effect in both studies: academic confidence strengthened the relationship between academic SE and exercise SE in Study 1, and between academic SE and nutrition SE in Study 2. Discussion focuses on the similarities and differences between the studies’ results and the implications for the literature.

Palabras clave: confianza; metacognición; autoeficacia; transferencia Having a strong sense of control over one's behavioursi.e., being highly selfefficaciousenables individuals to set higher goals for themselves, invest more energy in attaining these goals and to recover more quickly from setbacks. Indeed, self-efficacy (SE) 1 beliefs play a central role in predicting a number of behaviours, ranging from academic achievements (e.g., Richardson, Abraham, & Bond, 2012) to physical activity (e.g., Luszczynska, Schwarzer, Lippke, & Mazurkiewicz, 2011;Rovniak, Anderson, Winett, & Stephens, 2002) and dietary behaviour (Kreausukon, Gellert, Lippke, & Schwarzer, 2012). The majority of researchers view SE beliefs as being domain-specific, and believe the predictive power of SE to be mainly limited to one domain or task (e.g., Zimmerman, 1995).
However, a number of researchers have argued for a generalized sense of SE, or a belief in one's competence to cope with a wide range of demanding or novel tasks, and to deal effectively with stressful situations (Sherer et al., 1982). For example, Sherer et al. (1982) hypothesized that individuals develop a general sense of SE over time due to a build-up of experiences, which allows them to approach novel situations with renewed expectations and vision. Moreover, there are some indications that in addition to the existence of a general sense of SE, SE beliefs from one domain may 'transfer' to another domain. Bandura (1986) already discussed these transference effects and stated that SE in one domain 'tends to generalize to other situations . . . as a result, behavioural functioning may improve across a wide range of activities ' (p. 399). Thus, this implies that even though one does not know how self-efficacious one will be in a certain domain, generalized experiences from another domain may still provide individuals with a general sense of control and competence. In line with this, Jackson and Dimmock (2012) have shown that students' academic SEtheir perceived ability to manage their study commitmentspredicted their exercise intentions and engagement, providing evidence for transfer of SE between conceptually different domains.
There are a number of variables which influence the transfer of SE beliefs, or which might explain the underlying process. One such explanation for the transfer of SE beliefs is offered by a metacognitive perspective (e.g., Petty, Briñol, & Tormala, 2002;Petty, Briñol, Tormala, & Wegener, 2007). In this perspective, the focus is on the individual's perception of her or his thoughts and, relevant for the current research, the (metacognitive) confidence with which they are held. Specifically, in addition to having primary cognitions, consisting for example of one's SE beliefs in a certain domain, individuals differ in the certainty with which they hold these beliefs, their perception of the validity of these beliefs and a self-understanding of one's weak and strong points (Petty et al., 2007;Schraw, 1998). One's level of metacognitive confidence can thus undermine or enhance one's primary cognitions, by casting doubt on or boosting them (e.g., Wichman et al., 2010), and influences behaviour accordingly (Petty et al., 2002). For example, although two individuals might have similar SE beliefs about their ability to exercise regularly, one of these individuals might be very confident about these beliefs, while the other might have considerable doubts about the validity of these expectations; it follows that one might expect the first individual to be more successful in maintaining an exercise regime than the second individual. With respect to the transfer of SE from one domain to another, one might thus assume that the (metacognitive) confidence of an individual in these beliefsin both the domains under investigationinfluences this transfer. Thus, in the current research, to investigate the influence of confidence on the SE transfer process, we will add confidence in the domains under investigation as mediating variables in our analyses.
Additionally, the perceived similarity between domains will be considered in the current studies, since it has been found that perceived similarity between tasks aids the transfer of SE from one domain to another (e.g., Bong, 1997), revealing an increase in generality, or transfer, of academic SE when perceived similarity between tasks increases. Moreover, research on judgments-of-learning (e.g., Koriat, 1997;Koriat & Bjork, 2006;Metcalfe, 2009) indicates that the degree to which individuals are able to make accurate predictions of their future performance in a certain domain is to a large extent an inferential process. That is, individuals rely on certain beliefs or theories they have about their own abilities or about the difficulty of the tasks at handor, a 'feeling of knowing' (Koriat, 1997). These metacognitive beliefs, paired with certain cues of the tasks at handsuch as the degree of familiarity with a certain task or the similarity ('fit') between tasks in different domainsmake it possible to predict future performance. Extended to SE beliefs, this reasoning would lead to the prediction that the degree of similarity between domains should help in the transfer of SE beliefs from one domain to another. Specifically, having confidence in one's skills in the academic domain should produce stronger SE beliefs in another domain if the domains are perceived to overlap. In the current research we will therefore ask about participants' perception of the similarity of the domains under investigation.
A final factor that will be investigated in the current research, because it is expected to play a role in the process underlying SE transference effects, is the importance assigned to the domains in which the transfer effect is expected to occur. The importance of a goal is strongly related to the effort individuals put into achieving it (e.g., Seijts, Meertens, & Kok, 1997). Moreover, there is research which indicates that academic SE moderates the relationship between domain importance and test anxiety, such that pupils with high academic SE were less anxious about a test in a highly important domain than pupils with lower academic SE (Nie, Lau, & Liau, 2011). In the study on SE transfer by Jackson and Dimmock (2012) mentioned above, differences in domain importance were found to moderate the relationship between academic SE and exercise SE.
To sum up, this research is aimed at exploring whether a transfer effect exists between different SE domains, and which variables can help predict the transfer of SE across these domains. More specifically, we investigate if, and to what extent, academic SE beliefs predict exercise SE beliefs (Study 1) and nutrition SE beliefs (Study 2). We additionally include confidence in the domains under investigation, perceived similarity of the domains and importance of the domains in our analyses to investigate whether they explain additional variance in SE beliefs in these 'new' domains. Moreover, we investigate the mediating effect of confidence.

Study 1
In this first study, we investigate the transfer of academic SE beliefs to the exercise domain, and explore whether academic SE predicts exercise SE. For young adults, the transition to university often means a reduction in time spent on physical exercise, and exercise levels are in many cases below recommended levels (e.g., Haase, Steptoe, Sallis, & Wardle, 2004). Given that insufficient physical activity among students is associated with lower mental and physical well-being (Bray & Kwan, 2006), it is important to know how to increase physical exercise among this group.

Participants
The survey sample consisted of 74 undergraduate students (age M = 20.12, SD = 1.65) who voluntarily participated in the study and received partial credit towards course fulfillment. Given that we were interested in exercise SE, students were asked about the amount of time they spent exercising on a weekly basis. Students who spent less than 105 minutes per week (or 15 minutes per day) exercising were left out of analyses, so that the final sample was n = 50 (14 males; exercise time per week M = 274.90, SD = 192.66 minutes). All procedures and materials were approved by the local Ethics Committee of Psychology.

Measures & procedure
For an overview of all means, SDs and reliability coefficients, see Table 1. Upon entering the online survey environment, participants received information about the study's purpose, i.e., to investigate the associations between academic performance and a number of personality variables. After having provided informed consent, participants provided demographic information such as their gender, age, GPA and current study year, and they completed questions about the type and amount of exercise they performed on a weekly basis.
Next, they completed the academic and exercise SE measurements, which were adapted from Bandura's SE scale (2006). Subjects were asked to rate their belief of how likely they thought they were to perform a certain behaviour successfully, even when they would run into obstacles (0 = 'extremely unlikely' and 100 = 'extremely likely'). Participants completed this measure twice, once for academic SE and once for exercise SE. An example statement used to measure academic SE was: 'I can study efficiently when I am disappointed by the grades I received in a course'. An example of exercise SE was: 'I can perform my exercise routine regularly even when I do not reach my exercise goals'.
Participants then completed the domain importance measure (based on Jackson & Dimmock, 2012). Four questions probed the importance of both domains for participants, in the present and in the future, and for themselves as well as for their parents (1 = 'extremely unimportant' and 7 = 'extremely important'), e.g., 'How important is exercising for you at the moment?' Next, participants were asked to rate how similar they found a behaviour in the academic domain to a comparable behaviour in the exercise domain, in a manner adapted from Solomon (1977). For example, participants rated how similar they considered 'scheduling time to work out' to 'scheduling time to study' (0 = 'completely dissimilar', to 100 = 'completely similar'). In total, they had to rate 12 of these instances on similarity, four of which were filler items.
Lastly, domain confidence was examined by a single question for each domain under investigation (0 = 'no confidence at all' and 100 = 'full confidence'): 'How confident are you that you can study efficiently?' and 'How confident are you that you can exercise regularly?' After completing all measures, participants were thanked for their participation, debriefed about the aim of the study and received course credit.

Results & discussion
First, correlations between the variables were calculated. As can be seen in Table 1, academic SE and exercise confidence were both strongly associated with exercise SE, whereas the other variables (academic confidence, academic domain importance, exercise domain importance and perceived similarity) were not significantly associated with exercise SE. However, since these variables were intercorrelated, all variables were included in the regression analyses. Next, hierarchical multiple regression analysis was conducted. Weekly exercise (in minutes) and self-reported GPA were entered in the first step, academic SE was entered in the second step, academic and exercise confidence were entered in step 3, exercise and academic importance were entered in step 4, and in the final step perceived similarity between domains was entered (see Table 2 for an overview of all results). This analysis indicated that academic SE by itself explained 20.4% of the variance found in exercise SE scores, and that confidence in the academic and exercise domains accounted for an additional 34%. In the final model GPA, academic SE, academic confidence and exercise confidence were unique predictors of exercise SE.
Next, we conducted process bootstrapped mediation analyses (5,000 bootstrap samples; all variables standardized). First we investigated academic confidence as a mediator, including exercise confidence as a covariate in the model. This analysis revealed partial negative mediation (or suppression) of academic confidence (indirect effect β = −.14, 95% CI [−.403, −.006]; R 2 = .35, F(2, 47) = 12.85, p < .001),indicating that the total effect increased when the mediator was added (total effect β = .39, 95% CI [.143, .630], direct effect β = .52, 95% CI [.287, .756]). Next, we conducted an analysis including exercise confidence as a mediator and academic confidence as a covariate, but this analysis showed no These results suggest that, as expected, academic SE predicts exercise SE, suggestive of transfer or generalization. Additionally, we found that both academic and exercise confidence predict exercise SE beliefsalthough the association was positive for exercise confidence and negative for academic confidence. Interestingly, when adding academic confidence as a mediator to explain the relationship between academic SE and exercise SE, we found evidence for partial suppressionsuggesting that the relationship between academic SE and exercise SE becomes stronger when this variable is taken into account. Although this result should be interpreted with caution since these analyses are exploratory in nature, it is in line with a metacognitive confidence perspective (Petty et al., 2002(Petty et al., , 2007 which holds that the confidence with which one holds one's primary cognitionssuch as SE beliefsmight enhance (or undermine) these cognitions.
We expected that part of the variance in exercise SE would be explained by the importance participants ascribed to both the academic and the exercise domain (see, e.g., Jackson & Dimmock, 2012), but we found no effects for these variables. Similarly, we hypothesized that when domains share certain characteristics, transfer of SE should be facilitated due to one's familiarity with the task at hand (e.g., Koriat, 1997). Study and exercise routines are similar in the sense that both require conscious planning and monitoring, which should make it easier for skills acquired in one of these domains to transfer to the other (see, e.g., Jackson & Dimmock, 2012). However, in the current study perceived similarity did not predict exercise SE. One could speculate that rather than being seen as similar due to the self-regulatory demands placed on the individual, studying and exercising might have been perceived as interfering with each otherin the sense that time spent exercising cannot be spent studying. In the next study, we therefore focus on a behaviour that should not be seen as interfering with each other when factors like time management are taken into account: maintaining a healthy diet, or nutrition SE. Specifically, in Study 2 we will investigate whether nutrition SE is predicted by academic SE, and again, domain importance, domain similarity and domain confidence are included as possible predictors.

Study 2 Method
Participants & procedure Seventy-one undergraduate students 2 (age M = 20.25, SD = 1.70) voluntarily participated in the study and received partial credit towards course fulfillment after having completed the online questionnaire. Again, ethical approval was obtained from the local Ethics Committee of Psychology.
The procedure and all the measures used in this second study were identical to those of Study 1, with the exception that in the current study the questions were related to the nutrition domain. Moreover, to be able to statistically control for any possible influence of BMI on nutrition SE beliefs, participants reported their height and weight (BMI: M = 21.96, SD = 3.88) in the demographics section of this study. For a summary of all means, SDs and reliability coefficients of the measures used in Study 2, see Table 3.

Results & discussion
Again, first we investigated the correlations between all variables. As can be seen in Table 3, nutrition SE was significantly associated with academic SE, nutrition confidence and nutrition importance, but not with the other variables. However, since these variables were intercorrelated, all variables were included in the regression analyses.
Next, hierarchical multiple regression analysis was conducted. BMI and selfreported GPA were entered in the first step, academic SE was entered in the second step, academic and nutrition confidence were entered in step 3, academic and nutrition importance were entered in step 4, and in the final step, perceived similarity between domains was entered (see Table 4 for an overview of all results). This analysis indicated that academic SE by itself explained 27% of the variance found in nutrition SE scores, and that confidence in the academic and nutrition domains accounted for an additional 21%. In the final model, academic SE, academic confidence, nutrition confidence and nutrition importance were unique predictors of nutrition SE.
Next, we conducted process bootstrapped mediation analyses (5,000 bootstrap samples; all variables standardized). Similarly to our approach in Study 1, we first investigated academic confidence as a mediator, adding nutrition confidence as a covariate in the model. Again, this analysis revealed partial Thus, similarly to the results of Study 1, both academic SE and confidence in the domains under investigation were significant predictors of SE beliefs in the nutrition domain. And, as in Study 1, subsequent mediation analyses revealed a similar pattern of results: the effect of academic SE on nutrition SE seems to be increased when academic confidence was included in the model, but not when nutrition confidence was added. In addition, the importance an individual attaches to healthy eating also significantly predicted nutrition SE. This is in line with a study showing that female college students who valued healthy eating and saw themselves as healthy eaters were more likely to adhere to a healthy diet than students who did not identify as healthy eaters (Kendzierski & Costello, 2004). As in Study 1, no significant results were found for academic domain importance or the perceived similarity of the academic and nutrition domains.

General discussion
Health goals often compete with goals in other important life domains over the extent to which people are able and willing to devote time, energy and money towards attaining them. However, research has also shown that individuals are quite capable of regulating behaviours across different domains, especially when they frequently perform certain behaviours. Moreover, efficacy beliefs may transfer, or generalize, across contexts when there are consistencies in the selfregulatory skills required in those domains, such as planning, monitoring and scheduling various activities, decreasing the perception of goal conflict and promote perceptions of goal complementarity (Jackson & Dimmock, 2012;Riediger & Freund, 2004). In both of the studies reported here, we found that academic SE explained a substantial proportion of the variance of SE in another domain, suggesting transfer, or generalization, may indeed occur.
Additionally, the domain confidence of participants to perform the desired behaviour explained a significant proportion of the variance in nutrition and exercise SE. As was mentioned above, confidence might be a general, trait-like construct underlying many other self-beliefs, including efficacy beliefs, and as such is an important predictor of achievements (Stankov, Lee, Luo, & Hogan, 2012). Indeed, (metacognitive) confidence can be seen as a secondary cognition about one's primary cognitionsin this case, one's efficacy beliefs in the academic and exercise/nutrition domainsand indicates the perceived validity or certainty of these beliefs (e.g., Petty et al., 2002Petty et al., , 2007 which in turn is likely to be influenced by one's prior experiences conducting this behaviour. One's level of SE may thus be enhanced or undermined by metacognitive confidence, and in the current studies we found some preliminary evidence for a partial suppression effect of academic confidence on the relationship between academic SE and exercise/nutrition SE, suggesting that including academic confidence increases the predictive value of academic SE. However, there are also studies which show that a general sense of confidence in one's abilities is not predictive of performance when domain and task-specific SE beliefs are controlled for (e.g., Pajares & Johnson, 1994). In future research, it should be investigated whether the inclusion of specific academic versus exercise/nutrition tasks, rather than focusing on more self-regulatory beliefs such as planning, scheduling and selfmonitoring, decreases the predictive value of academic SE, and as such limits transfer.
Interestingly, the importance attached to maintaining a healthy diet was a significant predictor of nutrition SE, but a similar effect could not be found for exercise importance as a predictor of exercise SE. One explanation for this finding might be inherent to the behaviour under investigation: the ever-present temptation to not eat healthily, made more difficult by the presence of unhealthy foods on and off campus. Successfully resisting temptations has been shown to increase one's chances of reaching a goal (e.g., Freitas, Liberman, & Higgins, 2002), and the likelihood of resisting temptations increases with increased goal importance, or 'want-to' motivations (e.g., Milyavskaya, Inzlicht, Hope, & Koestner, 2015). Our results suggest that wanting to reach the goal of eating healthily is part of one's perceptions of having the necessary skills to do so, whereas this is not the case for exercising. Rather, inspection of the regression results for both studies seems to suggest that for exercise SE, the confidence with which one holds these efficacy beliefs seems more crucial: a significant proportion (34%) of the variance in exercise SE is explained by academic and exercise confidence, exceeding the predictive value of academic SE.

Limitations and suggestions for further research
Although it has been argued, and shown in previous studies, that domain similarity is an important variable in the transfer of SE from one domain to another (e.g., Bong, 1997), it did not significantly predict exercise and nutrition SE beliefs in the current research. To us, this is suggestive of the generality of SE: feelings of competence can generalize, or transfer, from one domain to another, independent of perceived similarities between these domains. Indeed, according to Bandura (1997), individuals are likely to generalize their SE between domains when different activities require generic self-regulatory capabilities, which might apply to academic work as well as the maintenance of an exercise regime or a healthy diet. However, future research should investigate a number of variables which likely moderate the transfer of efficacy beliefs between domains, but which were left out of consideration in the current studies. Most importantly, the role of motivation should be examined, since it is also known that the motivational component of academic SE beliefs is most predictive of, for example, academic performance (e.g., Valentine, DuBois, & Cooper, 2004).
First, the origin of the motivation for the behaviour under investigation should be investigated: is it autonomousinfluenced by an inherent interest in the behaviouror controlledinfluenced by external contingencies (see e.g., Deci & Ryan, 2000). Second, the type of goals individuals pursue is important; that is, whether they are striving to be competent or to outperform others (Grant & Dweck, 2003). With respect to the current studies, one could hypothesize that the type of activityexercising versus eating healthilymay also lead to adopting different goals, which in turn influence one's SE beliefs (Komarraju & Nadler, 2013). In this context, the trans-contextual model proposed by Hagger and Chatzisarantis (2012) might be of relevance. This model focuses on the role of autonomous motivation as a mechanism through which transfer of academic activities to extra-curricular activities takes place.
A number of other limitations should also be addressed. First, our research is a first exploratory attempt at uncovering transfer effects, and its design does not allow us to draw causal conclusions from our findings. Ideally, longitudinal studies should be conducted, measuring academic SE beliefs in addition to SE in other domains, as well as tracking actual achievements. Moreover, a longitudinal or diary study can provide information about individuals' exact amount of exercise and consumed calories, as well as academic time management and performance, as well as monitoring fluctuations in SE beliefs. An additional limitation is the relatively small sample size employed in the current research. Although we do feel the samples are representative of the larger student population, larger sample sizes increase the power and thus the generalizability of these results.

Notes
1. To increase readability of the text, from this point forward we will refer to selfefficacy as SE. 2. Due to a technical error, data regarding participant gender were not collected.