The mediating role of constructs representing reasoned-action and automatic processes on the past behavior-future behavior relationship

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY-NC-ND 4.0 https://creativecommons.org/licenses/by-nc-nd/4.0/ The mediating role of constructs representing reasoned-action and automatic processes on the past behavior-future behavior relationship © 2020 Elsevier Accepted version (Final draft) Brown, D. J.; Hagger, M. S.; Hamilton, K. Brown, D. J., Hagger, M. S., & Hamilton, K. (2020). The mediating role of constructs representing reasoned-action and automatic processes on the past behavior-future behavior relationship. Social Science and Medicine, 258, Article 113085. https://doi.org/10.1016/j.socscimed.2020.113085 2020 Journal Pre-proof The Mediating Role of Reasoned-Action and Automatic Processes from Past-toFuture Behavior Daniel J. Brown, Martin S. Hagger, Kyra Hamilton PII: S0277-9536(20)30304-X DOI: https://doi.org/10.1016/j.socscimed.2020.113085 Reference: SSM 113085 To appear in: Social Science & Medicine Revised Date: 18 February 2020 Accepted Date: 22 May 2020 Please cite this article as: Brown, D.J., Hagger, M.S., Hamilton, K., The Mediating Role of ReasonedAction and Automatic Processes from Past-to-Future Behavior, Social Science & Medicine, https:// doi.org/10.1016/j.socscimed.2020.113085. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier Ltd. The Mediating Role of Reasoned-Action and Automatic Processes from Past-toFuture Behavior Daniel J. Brown Griffith University Martin S. Hagger University of California, Merced and University of Jyväskylä Kyra Hamilton Griffith University Author Note Daniel J. Brown, School of Applied Psychology, Menzies Health Institute Queensland, Griffith University, Brisbane, Queensland, Australia; Martin S. Hagger, Psychological Sciences, University of California, Merced, USA and Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland; Kyra Hamilton, School of Applied Psychology, Menzies Health Institute Queensland, Griffith University, Brisbane, Queensland, Australia. Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. This project was supported by the Australian Government Research Training Program. Martin S. Hagger’s contribution was supported by a Finland Distinguished Professor (FiDiPro) fellowship from Tekes, the Finnish funding agency for innovation. *Correspondence concerning this article should be addressed to Daniel Brown, Health and Psychology Innovaitons (HaPI) Research Lab, School of Applied Psychology, Griffith University, 176 Messines Ridge Road, Mt Gravatt, QLD 4122. Email: [email protected] Figure 1. Relations among proposed model constructs including standardized parameter estimates. Estimates on the upper, center, and lower lines are estimates for the binge drinking, flossing, and sun safety behaviors, respectively. Effects of age, gender, and education on each model construct have been omitted for clarity. Intention T2 Automaticity Behavior T1 Automaticity Perceived Behavioral Control Attitude Subjective Norms Frequency .344 *** .426*** .719*** .105 .263*** .135 .111 .094 -.164* Recency Routine Past Behavior .601 *** .530*** .364*** .391 *** .249*** -.050 .203 ** .081 -.015 .917 .967 .919 .328 *** .345*** .363*** .670 *** .899*** .710*** .420 *** .424*** .466*** .454 *** .301*** .369*** .104 .147* .089 .158 * .205** -.087 .366 *** .526*** .428*** .204 ** .514*** .383*** .902 .941 .933 .884 .897 .938 Table 1 Model Fit and Quality Indices for Structural Equation Models for Binge Drinking, Flossing, and Sun Safety Index Binge drinking Flossing Sun safety GoF .575 .678 .592 AR 2 .382 *** .518 *** .382 *** APC .164 ** .187 ** .201 ** AVIF 1.272 1.601 1.342 Note. * p < .05 ** p < .01 *** p < .001 1 Introduction To develop effective interventions in order to modify people’s behavior one needs to first isolate the mechanisms that guide the behavior and then test the extent to which the mechanisms magnify or diminish behavioral engagement. Previous research has often turned to theories of social cognition to guide investigations aimed at identifying the determinants for health behaviors and, importantly, the processes by which these determinants relate to each other and the behavior. A close examination of the major theories that have been applied to the understanding of health behavior assumes that behavior is determined by a reasoned, intentional process in which an individual invests effort in order to pursue an action. The theory of planned behavior (TPB: Ajzen, 1991), given intention is central to the model, is perhaps the most widely used social–cognitive theory of behavior. According to the model, intention is the most proximal predictor of behavior, with intention determined by three social–cognitive variables: attitude (overall evaluations of performing the behavior), subjective norm (social pressure from important others to perform the behavior), and perceived behavioral control (perceived amount of control over behavioral performance; also more recently theorized as moderating the intention-behavior relationship; Ajzen, 1991). Meta-analytic studies support the use of the TPB in predicting health-related behaviors (e.g., McEachan et al., 2011; Rich et al., 2015) Social cognition theories such as the TPB tend to focus on a relatively narrow set of determinants that include constructs that represent reasoned, intentional determinants of action. Such approaches tend not to explicitly account for the pervasive effects of past behavior on key constructs of psychological theories and their relations with health behaviors (e.g., Albarracín et al., 2001; Conner et al., 1999; Hagger et al., 2018). However, research has demonstrated that including past behavior in these theories accounts for substantive additional variance in subsequently measured (future) behavior through the residual effect of 2 past behavior on future behavior. Inclusion of past behavior has also been shown to attenuate the size of the effects of intention and other social-cognition constructs on future behavior. Some argue that these residual effects are likely an artifact of assessing past behavior and future behavior using the same measure, leading to shared method variance (Ajzen, 1991, 2002). However, research using different measures of past and future behavior has revealed residual effects (Brown et al., 2018; Verplanken, 2006). Others, therefore, argue that the residual effects afforded by past behavior may model non-conscious processes including habits and decisions based on implicit cognition or behavioral ‘scripts’ (Hagger, 2020; Hagger & Chatzisarantis, 2014; Strack & Deutsch, 2004; Triandis, 1977). In seeking to understand the effects of past behavior in social-cognitive theories, Ouellette and Wood (1998) proposed two pathways by which past behavior relates to future behavior: a direct pathway and an indirect pathway. The direct effect is said to model an automatic process, similar to that proposed in dual-process theories (Strack & Deutsch, 2004). The second, indirect pathway, in which past behavior affects future behavior via conscious, intentional processes, such as those characterized in social-cognition theories (Ajzen, 1991). Ouellette and Wood (1998) hypothesized that well-practiced behaviors, occurring in consistent contexts likely reflect habitual patterns that are expected to be automatically repeated in the future. These automatic behaviors can be intentional and goaldependent or be perceived as non-volitional and counter-intentional (i.e., outside of an individual’s awareness/control or in opposition to an individual’s intention; Gardner et al., 2015; Ji & Wood, 2007). Conversely, the authors describe how novel or less practiced behaviors are likely governed by conscious, intentional processes. While past behavior has been shown to attenuate the effects of social cognitive variables (Brown et al., 2018), studies suggest that their constructs still account for unique variance in behavior and mediate, at least partially, the effects of past behavior on future behavior (Hagger et al., 2018). 3 Although previous research has demonstrated support for the direct and indirect effects of past behavior on future behavior in social cognition theories models (Ouellette & Wood, 1998), few studies have tested the effects of other constructs that reflect nonconscious decision making, such as habits, on behavior alongside past behavior (Bamberg et al., 2003; van Bree et al., 2015; Verplanken, 2006). If indirect effects of past behavior on future behavior model automatic, spontaneous processes to behavioral enactment as proposed in dual process theories (Strack & Deutsch, 2004) and, thus, capture the automaticity component of habit they would be expected to mediate the effect of past behavior on subsequent behavior. We aimed to examine these propositions in the present study by testing the TPB in a range of health behaviors and included measures of past behavior and behavioral automaticity across two time-points. An additional issue with research examining the role of past behavior in social cognition theories is the large variation in arbitrary time frames used to measure past behavior. For example studies have measured past behavior over one week (Mullan et al., 2015), four weeks (Caudwell et al., 2019), and six months (Luszczynska & Cieslak, 2009). Measuring past behavior over arbitrary timeframes may miss patterns of behavioral engagement that could potentially be informative on the mechanisms and processes that determine subsequent health behavior. For example, engagement in sun safety behaviors is likely to be highly seasonal in that they are performed frequently in the summer months and seldom in winter months (Sun et al., 2014; Xiang et al., 2015). A measure of past behavior that refers to sun safety behaviors in close proximity to subsequent behavior is likely to lead to stronger past behavior effects than if the time frame of the past behavior measure extended beyond the current season. Furthermore, Ouellette and Wood (1998) suggested that a different pattern of past behavior effects will occur for infrequently compared to regularly performed behaviors. It would seem reasonable to account for both long-term (i.e., distal) and 4 short-term (i.e., recent) enactments of past behavior as well as the frequency with which the behavior is measured. Perugini and Bagozzi (2001) proposed differentiating between longterm (up to 1 year) and more recent (within the last month) measures of past behavior. Similarly, others have used items relating the extent to which the behavior is routinized to further capture the regularity of daily behavioral performance (Verplanken & Orbell, 2003). Extending this research, the present study tested effects of past behavior using multiple methods that account for the frequency, recency, and routinization of behaviors. Specifically, the measure included items assessing long-term (last year), short-term (last month), and routine patterns of behavior, proposing that these would indicate a multi-component measure of past behavior that would capture the essence of individuals past performance of health behaviors. The Current Study and Hypotheses Based on Ouellette and Wood’s (1998) propositions, we propose a set of key hypotheses relating to the effects of reasoned, intentional pathways and non-conscious, automatic pathways to action for three health behaviors in three independent samples: binge drinking in university students, dental flossing in community-dwelling adults, and parental sun safety behaviors of their 2 to 5 year-old children. The proposed model is presented in Figure 1. In the present model, the reasoned, intentional processes are represented by the effects of the social-cognition constructs from the TPB. Specifically, attitude, subjective norm, and perceived behavioral control are expected to significantly and directly predict intentions and intentions and perceived behavioral control are expected to significantly and directly predict the target behaviors. We also expect the effects of the TPB constructs on behavior would be mediated by intentions (Ajzen, 1991). 11 supporting construct validity of the factors. Importantly, the second-order factor loadings for the frequency, recency, and routine indicators of the past behavior latent variable were large (>.884) and statistically significant (p < .001) for all behaviors. Composite (ρ) reliability coefficients, AVE, and intercorrelations for model variables are presented in Appendix C (supplemental materials). Reliability coefficients exceeded the .700 criterion, and AVE values exceeded the recommended .500 criterion. Correlations among the latent variables also indicated no problems with discriminant validity. Missing values analysis using Little’s (1988) missing completely at random (MCAR) test revealed a significant value for the flossing behavior sample (χ 2 = 437.599, df = 357, p = .002), but not for the binge drinking behavior (χ 2 = 431.810, df = 416, p = .286) or parent sun safety behaviors (χ 2 = 296.450, df = 385, p = 1.000) samples. Model Effects Standardized parameter estimates for the hypothesized relations among factors are presented in Figure 1 and as a table in Appendix D (supplemental materials). Overall, the model accounted for 68.8%, 65.3%, and 61.6% of the variance in intention to binge drink, floss, and adopt safe sun behaviors, respectively, and 39.4%, 78.8%, and 27.9% of the variance in the binge drinking, flossing, and sun-safe behaviors, respectively. Results revealed statistically significant effects of attitudes on intentions to engage in binge drinking and flossing, but not for sun safety behaviors. There was a statistically significant effect of subjective norms on intentions for flossing, but not for the other behaviors. Perceived behavior control significantly predicted intention for binge drinking and flossing but not for sun safety behaviors. Intention statistically predicted flossing, but no effect was found for binge drinking and sun safety behaviors. There was a statistically significant effect of perceived behavioral control on behavior for sun safety behaviors, but not for the other behaviors. Automaticity at T1 predicted automaticity 6-weeks later (T2), and automaticity at 12 T2 predicted behavior, for each of the three behaviors. Statistically significant effects were found from past behavior on attitude, subjective norms, perceived behavioral control, intentions, and automaticity at T1 and T2 for all behaviors. There was a statistically significant effect of past behavior on behavior for binge drinking, but not for flossing or sun safety behaviors. Focusing on the indirect effects, we found a statistically significant indirect effect of attitudes on behavior mediated by intention for flossing behavior, but no other indirect effects of the model variables on behavior mediated by intention. We also found indirect effects of T1 automaticity on behavior mediated by T2 automaticity for all behaviors. We found no indirect effects of past behavior on behavior through attitudes, subjective norms, perceived behavioral control, and intention in each of the models. There was, however, a statistically significant indirect effect of past behavior on behavior via T1 automaticity, and via T2 automaticity for all behaviors. Multi-group analyses identified a few differences in effects across the three samples, although the differences reflected the relative size of the effects across samples, rather than differences reflecting effects that were different from zero and those that were indistinguishable from zero. Effects of attitudes and perceived behavioral control on intentions, and intentions on behavior, were significantly smaller in the sun safety sample relative to the binge drinking and flossing samples. Effects of perceived behavioral control on intention and behavior were significantly larger in the sun safety sample relative to the binge drinking and flossing sample. The effect of past behavior on attitude was significantly larger in the binge drinking sample than the sun safety sample. Similarly, the effect of past behavior on subjective norms was significantly greater in the binge drinking sample than the flossing sample. The effects of past behavior on T1 and T2 automaticity were significantly greater in the flossing sample compared to the binge drinking sample. The effect of past behavior on 13 intention was significantly smaller in the binge drinking and flossing samples relative to the sun safety sample. Relative to the sun safety sample, the effect of past behavior on behavior was larger in the binge drinking sample For completion, we compared model effects without imputation of missing values, using listwise deletion of data with missing cases instead. The model estimated with listwisedeleted data did not result in substantive differences in the pattern of effects across the samples. Full results of these analyses are presented in Appendix D (supplemental materials). Discussion Social cognition theories tend to focus on a narrow range of determinants of health behaviors. As a consequence, they may not account for effects of other variables that could be potentially informative when it comes to predicting health behaviors. One potential variable whose effects within social cognition models may provide important information on the determinants of behavior is past behavior. There is already a substantive body of research examining past behavior effects within social cognition theories. Studies including frequency of past behavior within these theories have consistently shown that past behavior attenuates effects of intention and other social cognition constructs on subsequently measured (future) behavior. However, there is some debate over what such effects represent. Some researchers have argued that past behavior may model non-conscious and automatic processes, such as habit (Ouellette & Wood, 1998). Others have argued that past behavior effects should be mediated by constructs from social cognition theories if the theory is to be considered sufficient as an account for further behavior (Ajzen, 1991, 2002). Drawing on Ouellette and Wood’s (1998) propositions, we tested a set of key hypotheses related to reasoned action and automatic processes in a social cognition model including past behavior in three health behaviors, with three, independent samples: binge drinking in university students, dental flossing in community-dwelling adults, and sun safety behaviors by parents for their 2 to 5 14 year-old children. Furthermore, we adopted a comprehensive measure of past behavior, which encompassed frequency, recency, and routine patterns of previous behavior. Results revealed a consistent pattern of effects for the proposed model in the binge drinking and dental flossing behavior samples consistent with the TPB. These findings suggest that individuals are more likely to engage in these behaviors when they have positive attitudes toward performing the behavior in future and believe doing so is within their control. Subjective norms did not predict any of the behaviors, suggesting a lesser role for the perceived influence of significant others for these behaviors. Importantly, behavior was also predicted by automaticity in all three samples, suggesting, that the very least, all three behaviors were somewhat determined by habits or routine. These findings suggest that constructs representing both reasoned and automatic processes determine behavior simultaneously. A possible interpretation of this pattern of effects, provided by Hagger et al. (2016), is that these behaviors are determined by constructs representing one or the other of the processes for groups of participants within the sample, each with sufficient strength so as to be presented as statistically significant overall. The key challenge for future research is to identify the moderator variables that determine when each pattern of effect pervades. In contrast with previous studies (Hamilton, Kirkpatrick, Rebar, & Hagger, 2017), none of the TPB variables predicted parents’ intentions, and intentions did not predict behavior for parental sun safety behaviors. One possible reason for this pattern of effects may be that these behaviors are highly routinized and habitual, particularly in an Australian context where exposure to the sun is both likely and regular. A means to test this hypothesis would be to examine effects of the constructs representing the reasoned process predict behavior when effects of constructs representing the automatic process are removed. We therefore re-estimated the model removing effects of automaticity in the model for this behavior. As predicted, results revealed intentions predicted parental sun safety behavior (β = 15 .227, p < .01). This attenuation effect has been observed consistently in previous studies (see Hagger et al., 2016, 2018), and suggests automaticity is the pervading determinant of behavior in this context. Research has suggested that past behavior-future behavior relations effectively represent habits (Hagger, 2019; Hagger et al., 2016, 2018; Ouellette & Wood, 1998). We reasoned that if this was the case, residual effects of past behavior on future behavior should be mediated by automaticity, to the extent that individual’s reflections on automaticity sufficiently capture a key component of habit. Consistent with previous research (van Bree et al., 2015), our results revealed consistent indirect effects of past behavior on future behavior through automaticity at both time points. This result substantiates one of the propositions set by Ouellette and Wood (1998), that effects of past behavior model habitual or automatic actions. However, contrary to previous research (Hagger et al., 2016, 2018) and the proposals by Ouellette and Wood, effects of past behavior were not found to be mediated by the social cognition constructs in the current model. Ajzen (2002) suggested that the TPB constructs should account for the effects of past behavior if it is to provide a sufficient account of behavior. He suggested that indirect effects may reflect having made similar decisions in the past or the effect of past experience in informing beliefs regarding future performance of the behavior. However, it seems that for the current set of behaviors, beliefs regarding future participation jn behavior are not based on past experience. One possible interpretation is that a minority of individuals in these samples perhaps have low levels of previous experience and, thus, their beliefs toward performing the behaviors in future are not based on their past experience. However, for the majority, these behaviors are likely determined by habits. For example, it is well documented that university students frequently engage in hazardous binge drinking (Davoren et al., 2015), even compared to their non-student peers (Kypri et al., 16 2005). Similarly, sun safety practices is a relevant behavior for most parents in an Australian context (Hamilton et al., 2016; Hamilton, Kirkpatrick, Rebar, White, et al., 2017). It is therefore likely that the behaviors are likely those that are largely habitually determined. Unlike other proposed models that include deliberative and automatic pathways (Caudwell et al., 2019; Hamilton, Kirkpatrick, Rebar, & Hagger, 2017; van Bree et al., 2015), a novel aspect of the current research is the measurement of automaticity over two time points, this may have contributed to the full mediation of past behavior effects by automaticity for the dental flossing and parental sun safety behavior samples. A residual effect of past behavior was still observed in the binge-drinking sample, which suggest that automaticity may not fully account for effects of automatic, non-conscious behaviors. For example, it could suggest that individuals self-report of automaticity for binge drinking may not be entirely precise because they do not take into account of ‘in-the-moment’ decision making. Other constructs that reflect automatic evaluations of binge drinking behavior such as implicit attitudes toward alcohol, impulsivity, and implicit alcohol identity may be further important mediators of past behavior effects (Caudwell & Hagger, 2014; Houben & Wiers, 2009). An innovative contribution of the current study is the use of a second-order latent variable of past behavior that included long-term frequency of performance, recent performance (up to 1 month), and routine performance of the target behavior. Typically, past behavior is measured using an arbitrary time frame, such as 1 week. Such measures may miss previous patterns of behavior that may be important in the determination of future behavior. Bagozzi and Warsaw (1990) in their theory of trying, and later Perugini and Bagozzi (2001) in their model of goal-directed behaviors, argued that past behavior should be separated into long-term and recent components. They argued that while the two components may be related, they are conceptually different and therefore add important independent information 17 in the prediction of behavior. For example, an individual may have only recently taken up an activity (e.g., flossing after advice from their dentist), or could have regularly engaged in an activity over a long period, but has not been able to recently (e.g., an individual who usually binge drinks each weekend but has recently reduced spending to save for their university textbooks). Perugini and Bagozzi (2001) proposed that recency of behavioral engagement may influence future behavior by anchoring biases that may carry implicit information about intentions to a degree higher than by what is consciously available. Furthermore, consistent with Ouellette and Wood’s (1998) premise, they suggest that long-term frequency of behavioral engagement likely maps to habitual occurrences of the behavior. Despite these premises, we found that these conceptually distinct components of past behavior were highly correlated and served as indicators of a second-order past behavior latent variable. These findings suggest that all three components are captured by an overall past behavior construct. Measures of the different past behavior components tend to converge in large samples, and separation of the different components do not offer additional information in terms of the prediction of future behavior. In fact, separation of the different components would likely confound analyses due to multi-collinearity due to the high intercorrelations. Nevertheless, justification for separating past behavior into separate components may be justified when studying past behavior effects of infrequently performed behaviors, like blood donation, or new behaviors that have only just been initiated. The pervasive effects of habit on the health behaviors in the current study have important implications for practitioners and clinicians. Findings suggest that once an individual has adopted a health-behavior, such as applying sun-safe practices to their children or flossing, they should focus their efforts on building behavioral automaticity to maintain the health-behavior. Strategies that promote habits such as increasing behavioral performance in the presence of a cue could therefore be adopted. Furthermore, previous research has 18 demonstrated that habit formation follows an asymptotic curve; automaticity grows fastest in the initial weeks of habit formation, and then plateaus (Lally et al., 2010). This pattern may mean that while it can take several months for a habit to fully form (Lally et al., 2010), the more intensive work to build the habit (e.g., focusing on building a response to a stable and applicable cue) can be performed in the initial few weeks of an intervention before sufficient self-regulatory actions (e.g., development of self-monitoring charts) could enable persistent repetition of the habitual action. In addition, interventions may need to focus on building the awareness of when an unhealthy habit is being cued. The focus of interventions may be to break or swap a habitual response, for example, if an individual regularly begins bingedrinking after work on a Friday, they may change their response to the cue (i.e., ‘finishing work on a Friday) with something more healthy (e.g., eating at a restaurant that does not serve alcohol or by engaging in physical exercise). This work requires effortful, conscious reflection and decisions, and therefore future interventions may also need to use techniques that focus on the social–cognitive aspect of the current model (i.e., focus on building perceived behavioral control and positive attitudes to change). Strengths, Limitations, and Research Directions The current study has a number of strengths, including a comprehensive test of a model that included constructs representing reasoned and automatic processes in three distinct behaviors: dental flossing, binge drinking, and parents’ sun safety behaviors for their children. Importantly, this is the first study, to the authors’ knowledge, to use a measure of three distinct components of past-behavior: frequency, recency, and routine. The research is also innovative as it explores the extent to which effects of the past behavior factor on future behavior were accounted for by measures of behavioral automaticity, a key component of habit. Also unique was the measurement of automaticity at two points in time, which takes the temporal stability of this construct into account. 19 However, this study is not without limitations, which we outline next along with directions for future research. The current study used a prospective correlational design, so the direction of relations can only be inferred from the proposed relationships outlined in the relevant theories. Cross-lagged panel and experimental designs are needed to confirm the direction of causality (Chan et al., 2020; Liska, 1984). The current study primarily recruited relatively homogenous samples of participants that was low on diversity, which places limits on the generalizability of the results (Henrich et al., 2010). Another limitation is the reliance on self-report measures of behavior. Given one of the behaviors could be socially undesirable (i.e., binge drinking), and another could be seen as an evaluation of positive parenting practices (i.e., the use of sun safety behaviors on your child), social desirability effects have unduly inflated reports of these behaviors. Future studies should focus on collecting behavioral data that does not rely on self-reports which may obviate these biases (Buller & Borland, 1999). In addition, the parental sun safety behavior was defined as a collection of behaviors, which may have had unintended consequences in the way some participants reflected on the automaticity of the behavior. For example, it is plausible that different sun safety behaviors may vary in their degree of automaticity, which may have made it problematic for participants to respond to the measure. However, given each of these behaviors are likely cued in the same way (e.g., going outside during the day) and that the collection of behaviors have been advocated to occur together in a long-standing public health campaign in Australia (Montague et al., 2001), it is likely that each has a similar level of automaticity. Last, automatic processes in the current model were represented by selfreported automaticity alone. Future research should consider including other measures that tap into these processes, such as counter-intentional habits (Gardner et al., 2015) and implicit beliefs (Hagger & Chatzisarantis, 2014; Strack & Deutsch, 2004). These additional constructs may play an important role in accounting for effects of past behavior (Hagger et al., 2017). 20 Conclusions The current study tested a social cognition model that encompassed constructs representing reasoned action and automatic processes to predict three health behaviors in three separate samples: binge drinking in university students, dental flossing in communitydwelling adults, and parental sun safety behaviors for their 2 to 5 year-old children. Results indicated that constructs representing the reasoned action and automatic processes significantly predicted flossing, whereas binge drinking and sun safety behaviors were generally predicted by constructs representing automatic behaviors. The current investigation also found support for the mediation of the past behavior-future behavior relationship by automaticity. The current study fills a knowledge gap in the current literature on the multiple processes that guide behavior and provide further evidence that constructs that represent automatic processes play a key role. Future research should focus on exploring the role of other constructs that represent automatic processes such as counter-intentional habits and implicit beliefs. 27 Verplanken, B. (2006). Beyond frequency: Habit as mental construct. British Journal of Social Psychology, 45(3), 639–656. https://doi.org/10.1348/014466605X49122 Verplanken, B., & Orbell, S. (2003). Reflections on past behavior: A self‐report index of habit strength. Journal of Applied Social Psychology, 33(6), 1313–1330. https://doi.org/10.1111/j.1559-1816.2003.tb01951.x Xiang, F., Harrison, S., Nowak, M., Kimlin, M., Van der Mei, I., Neale, R. E., Sinclair, C., & Lucas, R. M. (2015). Weekend personal ultraviolet radiation exposure in four cities in Australia: Influence of temperature, humidity and ambient ultraviolet radiation. Journal of Photochemistry and Photobiology B: Biology, 143, 74–81. https://doi.org/10.1016/j.jphotobiol.2014.12.029 Study Highlights • Constructs representing reasoned action and automatic processes predicted health behaviors • Constructs representing automatic processes mediated the past-future behavior relationship • The study used a novel multi-component measure of past behavior Credit author statement Daniel J. Brown: conceptualization, methodology, data colletion, formal analysis, writing – original draft and edits. Martin S. Hagger: conceptualization, methodology, writing – edits, supervision. Kyra Hamilton: conceptualization, methodology, writing – edits, supervision.