I Sit but I Don’t Know Why: Integrating Controlled and Automatic Motivational Precursors Within a Socioecological Approach to Predict Sedentary Behaviors

Background. Precursors driving leisure-time sedentary behaviors remain poorly investigated, despite their detrimental consequences. This study aimed to investigate the predictive validity of controlled and automatic motivational precursors toward reducing sedentary behaviors and being physically active on leisure-time sedentary behaviors. The inuence of demographic, physical, socio-professional, interpersonal, and environmental variables on leisure-time sedentary behaviors was also examined and compared with the associations of motivational precursors. Methods. 125 adults completed questionnaires measuring controlled motivational precursors (i.e., attitudes, intentions, perceived competence), demographical (i.e., sex and age), physical (i.e., body mass index), and interpersonal (i.e., number of children) variables. Regarding automatic motivational precursors, habit strength and approach-avoidance tendencies were captured using the Self-Report Behavioral Automaticity Index and a manikin task. Leisure time, time and physical activity at work were computed as socio-professional variables, days of the week and weather conditions were recorded as environmental precursors. Participants wore an accelerometer for seven days and leisure time was identied thanks to notebooks. Associations between the different precursors and the leisure-time sedentary behaviors were examined in linear mixed effect models. the global effect of the weather conditions) precursors were more strongly associated with leisure-time sedentary behaviors. Conclusion. Our ndings show that, in comparison with demographical, socio-professional, interpersonal and environmental variables, the inuence of motivational precursors on leisure-time sedentary behaviors is limited. This study lends support for the adoption of a broad-spectrum of precursors when predicting sedentary behaviors.


Introduction
Sedentary behaviors, de ned as any waking behavior in a reclining, sitting, or lying position that requires an energy expenditure lower than 1.5 Metabolic Equivalent Task (Sedentary Behavior Research Network, 2012), are associated with a wide range of detrimental health consequences, including adverse metabolic conditions (Hamilton et al., 2007), depression (Teychenne et al., 2010), and cognitive decline (Olanrewaju et al., 2020). Adults spend about 77% of their waking time being sedentary (Diaz et al., 2016) and workplace settings account for a large amount of this daily time (Saidj et al., 2015). To mitigate the detrimental consequences associated with such patterns of activity at work, reducing sedentary behaviors during leisure time seems particularly important (Patel et al., 2010). However, the main precursors underlying leisure-time sedentary behaviors remain poorly investigated.

Explaining Sedentary Behaviors Through Sociocognitive Models
In the past decade, a growing number of studies investigated the motivational precursors of sedentary behaviors (Biddle, 2011). Most of these studies were anchored within sociocognitive models, which are based on the premise that imagined end states (expectancies, goals) are proximal variables of behaviors (Brand & Cheval, 2019). These models suggest that attitudes, intentions, and self-perceptions (e.g.,  (Rollo et al., 2016). In this line, additional theoretical perspectives, such as the dual-process models (Rhodes et al., 2019), have been mobilized to explain sedentary behaviors.

Explaining Sedentary Behaviors Through Dual-process Models
Dual-process models contend that behaviors are governed by both controlled and automatic motivational processes (Strack & Deutsch, 2004). Controlled processes are slow, initiated intentionally, require cognitive resources and effort, and operate within conscious awareness. The key aforementioned sociocognitive constructs are assumed to be "plugged" in this controlled dimension (Conroy & Berry, 2017). Conversely, automatic processes are fast, initiated unintentionally, require relatively less cognitive resources and effort, occur outside conscious awareness (e.g., habits, automatic affective reactions, approach-avoidance tendencies).
Despite the incidental enactment of sedentary behaviors (Spence et al., 2017), explained by the profusion of attention-grabbing cues in our modern environment (Levine, 2015), the in uence of automatic motivational processes remains overlooked. Indeed, only a few studies have mobilized dual-process models to explain sedentary behaviors (

Explaining Sedentary Behaviors Through Socioecological Models
Socioecological models are based on the premise that behaviors are jointly driven by multiple determinants (Glass & McAtee, 2006), which can be classi ed as intrapersonal, interpersonal, and environmental factors (O'Donoghue et al., 2016). Intrapersonal variables refer to demographic (e.g., gender, age) and physical factors (e.g., body mass index), as well as motivational and socio-professional factors (e.g., leisure time, working time, physical activity at work). Interpersonal variables include familial determinants, such as, the number of children. Alongside with built environmental determinants (e.g., accessibility to facilities), natural environmental factors can refer, to the days of the week or to weather conditions. Hence, far from competing with models focusing on motivational variables, the socioecological model integrates the aforementioned motivational precursors by considering individuals as actors amidst broader networks (Rhodes et  Although studies have jointly investigated the effects of motivational precursors with others variables (e.g., sex, body mass index, day of the week) (e.g., Conroy et al., 2013), no study has integrated socio-professional, interpersonal, and environmental factors alongside with controlled and automatic motivational variables to predict leisure-time sedentary behaviors. Importantly, recent ndings suggested that demographic, physical, socio-professional, interpersonal, and environmental factors could exert a greater in uence on sedentary behaviors than motivational variables (Buck et al., 2019). However, no study has directly compared the predictive validity of these variables within the same sample.

The Current Study: An Integrative Approach
The current study aimed to investigate the predictive validity of motivational (controlled and automatic), demographic, physical, socio-professional, interpersonal, and environmental precursors of leisure-time sedentary behaviors. Therefore, this study provides an integrative approach contributing to provide better understanding of the relative weight of motivational precursors in the regulation of sedentary behaviors ( Fig. 1). To this end, 135 healthy workers' leisure-time sedentary behaviors were monitored for one week using an accelerometer and associations with aforementioned precursors were examined.
We hypothesized that both controlled and automatic motivational determinants predict leisure-time sedentary behaviors (H1). Speci cally, higher controlled (H1a) (i.e., attitudes, intention, perceived competence) and automatic (H1b) (i.e., habit strength and approach-avoidance tendencies) motivation to reduce sedentary behaviors and to be physically active should negatively predict leisure-time sedentary behaviors. We also expected that demographic, physical, socio-professional, interpersonal, and environmental variables should predict leisure-time sedentary behaviors (H2). Finally, we compared the strength of the associations between these variables and leisure-time sedentary behaviors. We did not formulate an priori hypothesis on the relative weight of these precursors, although recent work suggests that the association between motivational precursors and leisure-time sedentary behaviors may be weaker than associations with the other variables (Buck et al., 2019).

Participants And Procedure
This manuscript contains data from two published studies (Cheval et al., 2015. A detailed description of the recruitment and experimental procedure is provided in supplementary material. First, 135 working adults completed a computerized reaction-time task and questionnaires assessing the different motivational precursors. Then, each participant received an accelerometer and received a notebook. They were invited to indicate on the notebook the time at which they woke up, put the accelerometer on their hip, arrived at their workplace, quit their workplace, removed the accelerometer, and went to bed. Participants were also asked to indicate whether they felt ill or injured during the monitoring period. Daily weather conditions, day of the week, and day in the monitoring sequence were collected over the course of the week by a research assistant. Eight days later, participants gave back their accelerometer and their notebook, and were debriefed.

Measures
Device-based measure of sedentary behaviors during leisure time A three-axis accelerometer (Actigraph GT3X+; Pensacola, USA) was used to quantify sedentary behaviors for seven days in free-living conditions. When data met the inclusion, criteria listed below, the eighth day of wear (i.e., when participants came back to the laboratory), was included in the analyses. One-minute epochs were used for data analyses and non-wear time was de ned as ≥ 59 consecutive minutes of zero counts. Daily data were included if they met two conditions: a wear time ≥ ten waking hours per day (Evenson & Terry, 2009) Fig. 2). Time spent in sedentary behaviors was determined through previously validated cut-points (i.e., 0 to 100 counts/min) (Freedson et al., 1998). To standardize for differences in leisure time between and within participants, the dependent variable was the daily percentage of leisure time spent in sedentary behaviors (Healy et al., 2011).

Controlled and automatic motivational precursors
Affective and instrumental attitudes toward sedentary behaviors and physical activity, intention to reduce sedentary behaviors and to be physically active, perceived competence to adopt an active lifestyle, habit strength toward sedentary behaviors and physical activity were measured using questionnaires. A detailed description of the scales is provided in supplementary material.
The automatic approach-avoidance tendencies toward sedentary behaviors and toward physical activity were assessed using a manikin task (Krieglmeyer & Deutsch, 2010) (see Cheval et al., 2015 for a detailled description). The automatic tendency to approach sedentary behaviors was calculated by subtracting the median reaction time when approaching sedentary stimuli from the median reaction time when avoiding sedentary stimuli -a higher score indicating a higher tendency to approach (vs. avoid) sedentary stimuli. The same logic was applied to calculate the automatic approach tendency toward physical activity. Two participants were excluded from the study because they demonstrated extreme scores (i.e., more than four standard deviations away from the sample mean, which consists into a commonly used threshold to detect outliers, Cousineau & Chartier, 2010).
Demographic, physical, socio-professional, interpersonal and environmental precursors Sex and age were included as demographic variables, body mass index as a physical variable, and the number of children as an interpersonal precursor.

Leisure time
Daily leisure time, expressed in minutes, was computed on the basis of notebook reports. This variable had two levels: an average between-person level, centered on the average value of the sample, and a daily within-person level, centered on the average value of each participant. This bidimensional approach accounted for both inter-individual differences in leisure-time and intra-individual changes across monitoring days for a given participant.
Time spent at work and time spent in physical activity at work These two variables were computed by matching notebook reports with accelerometer measures. Time spent in moderate-to-vigorous physical activity was identi ed by using previously validated cut-points (i.e., > 1952 counts/min) (Freedson et al., 1998). As mentioned for leisure time, these variables were separated into two dimensions: an average between-person level and a daily within-person level ( Table  S1).

Day of the week
Days of the week were dummy coded. Saturdays served as the reference category as the lowest levels of leisure time sedentary were observed on this day (Table S2).

Daily weather conditions
Daily weather conditions were obtained over the study period from the website MeteoFrance.com and coded by a research assistant. Three levels were created: sunny, cloudy, or rainy. Sunny days served as the reference category because the lowest levels of leisure-time sedentary behaviors were observed on these days (Table S2).

Confounding variables
Illness or injury were added as a confounding variable because they were a potential source of higher time spent in sedentary behaviors (Maher & Conroy, 2016). Based upon participants' notebook, a dichotomous variable was coded (1 for people who reported an illness or an injury; 2 for people who did not report any illness or injury). Finally, day in the monitoring sequence was included as a confounding variable as self-monitoring procedures have been shown to in uence sedentary behaviors across days of wear (Motl et al., 2012).

Statistical Analyses
Associations of predictors with the percentage of leisure-time spent in sedentary behaviors were analyzed using mixed effects models. This approach allows to account for the nested structure of the data (here, multiple observations within a single participant). Moreover, mixed effects models do not require an equal number of observations from all participants and increases power compared with traditional approaches, such as linear regressions (Boisgontier & Cheval, 2016;Judd et al., 2017). All models had random intercepts for participants and random slopes were added for all the time-varying variables (Frossard & Renaud, 2019).
First, we estimated a base model (M0), which tested the associations between the confounding variables (i.e., day in the monitoring sequence, and illness or injury) and leisure-time sedentary behaviors. All the subsequent models were adjusted for these confounders. In a rst set of models (Ms1), we separately added each controlled and automatic motivational variables to M0. This "one-by-one" strategy was used to account for the substantial shared variance between some of these constructs (Table S3). In a second set of models (Ms2), we separately added each demographic and physical variable to M0. In a third set of models (Ms3), we separately added each socio-professional variable to M0. This set of models Ms3 was adjusted for the day of the week as socio-professional variables and days of the week were closely related (Table S2). In a fourth model (M4), we added interpersonal variables (i.e., number of children). In a fth set of models (Ms5), we separately added environmental variables. Variables which were signi cantly associated with leisure-time sedentary behaviors in Ms1, Ms2, Ms3, M4 and Ms5 were identi ed on the basis of the p-value (p < .05) and were gathered in a last parsimonious nal model (M6) 1 .
To examine whether the different variables included contributed to improve the t of the nal model M6, this nal model was tested against models in which the variables of interest were removed. Variables which increased the t of the models were identi ed on the basis of the Bayesian Information Criterion (BIC), -2-log-likehood (-2LL) and p-values (Bollen et al., 2014). All models were estimated using the lme4 and lmerTest packages in the R software (Bates et al., 2014;Kuznetsova et al., 2015). An estimate of the effect size for xed effects was reported using the marginal pseudo-R 2 , computed using the MuMin package (Barton, 2009). Statistical assumptions associated with mixed models were checked (including normality of the residuals, homogeneity of variance, linearity, multicollinearity, and undue in uence) and met for all the models.

Results
Descriptive results are presented in Table 1. The nal sample included 125 adults (age = 40 ± 9 years; body mass index = 24 ± 4; 75 women; 74% with at least one child). The average percentage of sedentary behaviors during leisure time was 58 ± 12%, corresponding to 5 h 38 min ± 2 h 21 min per day.

Strengths And Limitations
The present study includes the following strengths. First, we assessed and compared a broad range of potential precursors of leisure-time sedentary behaviors. In this perspective, the present work paves the way to future work interested in mapping potential interactions between these levels of in uence, as showed in the physical activity domain (Rhodes et al., 2006). Second, although Actigraph GT3-X + tends to overestimate sedentary behaviors, in comparison with the ActivPal (Migueles et al., 2017), it provides an acceptable evaluation of sedentary behaviors in comparison with self-reported measurement (Gardner et al., 2019). Finally, we used a re ned statistical analysis suited to examine daily-basis associations.
Several limitations should be considered. First, our ndings should be interpreted with caution as, unlike previous studies (e.g., Maher & Dunton, 2019), our study did not assess the daily (or hourly) uctuations in motivational variables. This feature may explain the weak associations observed in the current study (Rebar et al., 2020). Second, leisure time and working time were identi ed thanks to notebooks reports, which may have led to approximative segmentation of these periods. Combining accelerometric measurement with global-positioning systems (GPS) could enable future research to re ne the identi cation of the context in which sedentary behaviors occur (Jankowska et al., 2015).

Conclusion
This study shows that demographic, physical, interpersonal and environmental variables are associated with leisure-time sedentary behaviors, while the associations with controlled and automatic motivational precursors are weaker. As suggested by the socioecological framework, adopting a theoretical integrative perspective encompassing a large range of precursors can help to better characterizing the key variables involved in sedentary behaviors. These ndings also suggest that interventions aiming to reduce leisuretime sedentary behaviors should pay attention to the speci cities of the targeted people (i.e., men, individuals with higher body mass index, childless). Interventions could also bene t from strategies enabling individuals to cope with environmental conditions to reduce sedentary behaviors over the course of the week and on days with adverse weather conditions. By contrast, focusing on motivational variables could be rather ineffective.

Declarations
Ethics approval and consent to participate: Ethical approval for the study was granted by the University of Predictive variables and outcome included in the mixed models Note. Outcome (red), motivational (purple), demographic (black), physical (black), socio-professional (blue), interpersonal (orange), and environmental (green) used in this study. Controlled motivational variables included attitudes, intentions and competence. Automatic motivational variables included habit strength and approach-avoidance tendencies. BMI: body mass index.  Associations of variables entered in the nal model M6 with leisure-time sedentary behaviors Note.
Coe cients b, 95% con dence interval are reported. Female individuals, Saturdays and absence of injury or illness served as references for sex, days of the week and illness/injury, respectively.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.