Insufficient physical activity is known to cause non-communicable diseases such as hypertension, obesity, cancer, type 2 diabetes, and cardiovascular diseases [1–3]. Consequently, the need to promote physical activity (PA) has become an important public health goal [4]. Yet, the recommended level for PA – i.e. to be at least moderately physically active for 150 min per week [5] - is still not met by 31% of the world’s population [6]. To help develop more effective interventions, it is necessary to gain deeper insight into the determinants that predict PA. There are two paradigms that can be applied to explain health behaviors. The first focuses on identifying explicit beliefs of people concerning a behavior, and is inspired by a set of complementary social cognitive and ecological models which summarize multiple levels of influences on behavior [7–9]. Explicit beliefs are determinants which people are aware of, can express consciously, and are measured by self-reported questionnaires. For instance, the explicit attitude towards a behavior (e.g. ‘Being physically active is very good for my health’) or a person’s reported ability to engage in a behavior when being confronted with challenging situations, called self-efficacy (e.g. ‘I find it hard to be sufficiently physically active when I am stressed’ or ‘I find it hard to be sufficiently physically active when I dislike the activity’). The second paradigm focuses on unconscious processes which persons may not be aware of but which still influence their behavior, called implicit processes [10, 11]. Implicit attitudes are one type of implicit process. They are automatically occurring attitudes of which people are less aware and to which people do not initially have conscious access [12]. To assess implicit attitudes, computerized reaction time tasks are used, i.e. the Implicit Association Test (IAT) [13]. While several studies have applied both the explicit and the implicit paradigms, only a few focus on how to combine these approaches. The present study attempts to integrate them.

An example of the explicit paradigm is reflected by the I-Change Model [14] which has also been used to assess and change PA-related cognitions and behaviors [15–17]. According to the I-Change Model – which integrates aspects from socio-cognitive models, i.e. the Theory of Planned Behavior [9], the Trans Theoretical Model [18], Social Cognitive Theory [8] and Goalsetting Theory [19] – intention is one of the most proximal conscious determinant for behavior. Intention in turn is determined by the attitude to the behavior (comprised of perceived pros and perceived cons regarding a behavior, e.g. ‘When I am sufficiently active I have more energy’ or ‘Being sufficiently active costs me a lot of effort’), social influence (the perception of the norms and behavior of people in the social environment as well as the perceived social support, e.g. ‘Most of my friends think that I need to be sufficiently active’ or ‘Most of my friends are sufficiently active’) and self-efficacy (whether a person perceives him or herself as capable of performing a behavior when confronted with obstacles). Individuals with high levels of self-efficacy are more likely to exert effort to perform a behavior and are therefore more likely to succeed, whereas people with low levels are more likely to fail [20]. PA behavior indeed has a reliable correlation with intention, and intention in turn acts as a mediator between the explicit cognitions such as attitude, knowledge, self-efficacy, social norms and behavior and self-efficacy also has a direct effect on PA behavior [21–25]. In most of the publications on PA, this paradigm is the most dominant one and most interventions aim to increase PA levels by changing explicit cognitions. A review concludes, however, that this approach is only modestly effective [26], and the contemporary idea is that implicit cognitions need to be taken into account, in addition to explicit cognitions.

The relatively new concept of combining implicit and explicit cognitions is reflected in dual process models [10, 11, 27, 28]. According to the Reflexive-Impulsive Model (RIM) [10], an example of a dual-process model, an impulsive and a reflective system exist, both of which guide behavior. Whereas the reflective system is composed of reasoned, deliberate, and conscious motives, the impulsive system is a composition of affective responses and automatically associated behavioral tendencies. According to the RIM, the reflexive and impulsive systems can influence behavior in different ways. One way is the double dissociation pattern [29], according to which spontaneous behavior is predicted best by the impulsive system, and deliberate behavior by the reflexive system [30–33]. Another potential way of how the two types operate is referred to as the additive pattern [29], meaning that both systems explain unique variance in one behavior. This pattern has indeed been shown for purchasing healthy food [34], dental flossing [35] and also with regard to PA. Concerning the latter behavior, it has been demonstrated that automatic, less conscious processes play a unique role alongside explicit cognitions in explaining past [36, 37] and future PA behavior [38] as well as the maintenance of PA [39, 40]. From this perspective, it follows that PA is regulated by both impulsive (or implicit) and reflective (or explicit) cognitions. This conclusion was indeed reached in a recent review [41].

Although explicit and implicit constructs have been shown to play a role in determining PA, it is not clear which of the above-stated patterns can be applied to PA. Conroy and colleagues [38] showed that implicit and explicit cognitions explain unique variance in PA behavior, i.e. favoring the additive pattern. Berry and colleagues [42], however, challenged this approach and concluded from their study that implicit and explicit cognitions are not only directly related to PA behavior, but that implicit attitudes interact with the intention to be active. This is in line with a third way of operating, namely the interactive pattern, meaning that the reflective and impulsive systems interact synergistically to predict behavior [29]. Also, Cheval [43] and colleagues found that impulsive processes interacted with PA intentions. More precisely, PA intentions predicted PA behavior when the impulsive approach tendencies toward the opposite behavior of PA, namely sedentary behavior, were low or moderate. By contrast, strong impulsive approach tendencies toward sedentary behavior blocked the effect of intention on behavior. These findings suggest that the way implicit and explicit processes jointly influence PA might be more complex than so far assumed.

Although different patterns of influence have been demonstrated, we argue that the two patterns are not necessarily mutually exclusive. Implicit attitudes and explicit determinants could both have a direct effect on behavior (additive pattern) and also interact with each other (interactive pattern). Until now, the two operating models have not been tested in a single study. Furthermore, former studies, such as the one by Cheval and colleagues [43], investigated the interactive pattern only between impulsive tendencies and the explicit construct intention. We aim to take this research approach one step further and raise the question whether implicit processes might also interact with the above-mentioned explicit cognitions that predict intention (perceived pros, perceived cons, social norms, social modeling, self-efficacy) and intention itself. Just as impulsive tendencies in the study by Cheval et al. [43] either reinforced or disinhibited the relationship between intention and behavior, we assume that implicit attitudes could have a reinforcing or inhibiting effect on the relationship between explicit cognitions and intention. For instance, it is conceivable that a person who perceives many pros regarding PA has an even stronger intention to become active when he or she unconsciously evaluates the behavior as positive. If, however, the same person evaluates PA unconsciously as negative, we expect this negative implicit attitude to inhibit the effect of perceived pros on intention. The similar pattern of reasoning could be applied to the other predictors of intention. Although intention does not necessarily lead to behavior, it still accounts for 23% of the variance in PA [44] and is regarded as an important step in the adoption and maintenance of behavior and as a good predictor in the context of protective behaviors such as PA [45]. Shedding light on the joint role that implicit attitudes and explicit cognitions play in intention formation could help to further elucidate this process.

The aim of the present study was three-fold. First, we investigated the direct effects of implicit attitudes and explicit cognitions on PA behavior (Fig. 1). As found in the former two studies [38, 43], we expect both implicit attitudes and explicit cognitions to predict unique variance in PA behavior (H1). Second, interactions between implicit attitudes and intention and implicit attitudes and self-efficacy were examined (Fig. 2). Just like Cheval [43] and in line with an interactive pattern of behavior prediction, we assume implicit attitudes also moderate the relationship between intention and PA and self-efficacy and PA (H2). Third, interactions between explicit cognitions and implicit attitudes were assessed (Fig. 3). We expect that the positive influence of the explicit cognitions (perceived pros, social norms, social modeling and self-efficacy) on intention is strengthened by positive implicit attitudes. The negative effect of perceived cons on intention is expected to be weakened by positive implicit attitudes but strengthened by negative implicit attitudes (H3).

The present study aimed to shed light on the question how implicit attitudes influence PA intention and behavior together with well-known explicit predictors of PA. Direct effects of these variables as well as interactions between them were examined. Results showed that implicit attitudes did not have a direct effect on PA behavior albeit via other explicit cognitions. The fact that implicit attitudes did not have a direct effect on PA behavior at any measuring point is in contrast to our hypothesis as well as to earlier results of Conroy and colleagues [38] and Cheval and colleagues [43]. Both authors found that, after controlling for explicit motivational predictors, implicit processes significantly contributed to PA prediction and hence support for the additive pattern. Whereas above authors assessed PA behavior using pedometers, we assessed PA levels by means of a self-report questionnaire, which, despite its shown validity [54], is less accurate than direct measurements [58, 59]; this could be a reason for the non-significant findings. Follow-up studies using accelerometers may be needed to obtain further insight into whether or not implicit processes influence actual PA behavior directly.

Although we did not find any direct effects, moderating effects were demonstrated: i.e. positive implicit attitudes strengthened the positive relationship between self-efficacy and PA behavior at the first follow-up. Negative implicit attitudes were found to weaken this relationship. In addition, and similar to Cheval et al. [43], we found that neutral but not positive implicit attitudes strengthened the positive relationship between intention and PA at the second follow-up. It seems surprising that positive implicit attitudes did not strengthen the relationship between intention and PA, but this could be explained by a ceiling effect as the intention of participants to be active was already very strong. Nonetheless, the findings support the idea of an interactive pattern of influencing PA behavior which is also in line with the findings of Cheval and colleagues [43]. If the intention to be active is already strong, positive implicit attitudes do not seem to support the effect on behavior, whereas neutral implicit attitudes do. In order to strengthen the likelihood that intention translates into behavior, our results suggest that one should at least aim to diminish a negative implicit attitude and create a neutral implicit attitude.

Moreover, we found implicit attitudes moderated the relationship between several explicit cognitions and intention. Firstly, implicit attitudes moderated the relationship between perceived cons and intention as well as between self-efficacy and intention at baseline. In line with our hypothesis, negative implicit attitudes strengthened and positive implicit attitudes weakened the negative relationship between perceived cons and intention. It seems that for those participants who reported exercise not to be beneficial or pleasant (as measured by the explicitly perceived cons), the positive implicit associations with PA acted as a buffer between perceived cons and intention. Moreover, the positive relationship between self-efficacy and intention was strengthened by neutral and positive implicit attitudes. Regarding self-efficacy, it seems conceivable that the effect of intention on PA behavior is stronger when a person does not only perceive him or herself as being capable of performing the behavior, but also has a positive, or at least a neutral, unconscious attitude towards the behavior. Thus, when intending to increase PA intention, positive implicit attitudes appear to be more beneficial. The interactions were not significant at one and 3 months’ follow-up, which could either be due to the weaker power of the sample, or to the assumption that implicit attitudes only have a short-term influence on the effect of perceived cons and intention and self-efficacy and intention.

Secondly, at one and 3 months’ follow-up, implicit attitudes moderated the relationship between social modeling and intention. The impact of other people’s behavior on the intention to become physically active was significantly greater when the implicit attitude was negative compared to when it was positive. One explanation for this finding could be derived from Festinger’s cognitive dissonance theory [60], according to which, individuals seek consistency among their cognitions. When an inconsistency between attitudes or behaviors occurs, the individual is motivated to resolve it as it is accompanied by negative feelings [61]. Feeling implicitly negative about being physically active while at the same time perceiving important people in one’s environment as being physically active, might create dissonance. In order to resolve this, individuals might reduce the importance of the implicit attitude and follow the behavior of others. In this case, the explicitly perceived modeling behavior might override the implicitly perceived negative implicit association. In the present study, the negative implicit attitude had a positive effect on the relationship between social modeling and intention. However, when there is no dissonance, i.e. when a person holds a negative implicit attitude and is surrounded by people who are not sufficiently active, negative implicit attitudes might strengthen the negative relationship between social modeling and intention, as was also the case for the relationship between perceived cons and intention. As interventions may not be able to change or control behavior or the perception of peer or parent behavior, they might rather attempt to reduce the impact of these perceptions on intention by creating a positive implicit attitude. Training or changing implicit associations has been applied to reduce social anxiety [62], alcohol consumption [63], to increase implicit self-esteem [64, 65] and only recently to increase PA levels [66, 67]. While Berry and colleagues and Markland and colleagues demonstrated short-term changes in implicit attitudes via exercise imagery or the provision of (counter attitudinal) information, computerized tasks have not yet been used in this context, but might offer a fruitful alternative. More research is, therefore, needed to understand how stable and changeable implicit attitudes actually are, especially over time. Moreover, in order to understand conditions under which dissonant and congruent implicit and explicit attitudes are beneficial or detrimental for PA behavior, further research is required.

When interpreting our findings, the following possible limitations need to be taken into account. First, the study sample was quite homogenous as far as age, education, and socio-economic status were concerned and had, on average, a very positive explicit attitude and a strong intention to be physically activity, which is not representative of the general public [68]. Second, for practical reasons, PA levels were measured by self-report. It is not clear to what extent participants were explicitly aware of activities which occurred spontaneously and excluded planned, structured exercise (e.g. using the stairs), and whether they were able to report them. Despite the satisfactory validity of the SQUASH [54], supplementing it with more objective measures, such as accelerometers or pedometers, could provide a more adequate report about activity levels as suggested by other studies [58, 59]. Third, we had a high drop-out rate at our follow-up measures (29% at T1 and 62% at T2) which could be due to an absence of commitment to participate in all three measures, a panel fatigue [69] or simply due to time constraints of the student sample as the last measure was conducted shortly before the exam period. As a consequence, our sample suffered from low power after one and 3 months which makes the interpretation of (non-)findings challenging. Fourth, stratified analyses for people with a negative, neutral, or positive implicit attitude were conducted using small sub-samples; these are also likely to have suffered from low power. Fifth, although neutral or positive implicit attitudes might help to increase PA intention, the intention behavior gap still remains. Research aimed at reducing this gap should be further stimulated.

Summarizing, one can conclude that the present findings challenge the dual-process approach which, until now, only assumed a direct influence of implicit attitudes on behavior, not via other explicit constructs. Although different modes of influence were suggested by Perugini [29], including an interactive pattern of influence between implicit and explicit attitudes, a thorough examination or integration of further determinants, such as intention, has not yet been carried out. Both approaches have in fact developed in isolation. We argue that this division needs to be reconsidered as our findings and those of Cheval and colleagues [43] demonstrate that unconscious processes are indeed associated with more conscious processes. A unique contribution of the present research is the examination of interactions between implicit attitudes, which are part of dual-process models, and explicit cognitions, which are summarized in socio-cognitive models. Potential improvements for interventions are thus provided. Future research needs to build on these findings by testing whether interventions which target both implicit attitudes and explicit cognitions result in greater activity intention and actual behavior change. Another avenue for future work, especially for the area of model testing and model improvement, is to investigate whether the relationships found in the present study are rather unique to PA or valid across diverse health-related behaviors. Shedding light on these issues may not only aid the development of even more successful interventions to promote physical activity, but also the aspiration to improve global health.