香京julia种子在线播放

An a priori power analysis conducted using G. Power 3.1.9.4 (Faul et al., 2007; Erdfelder et al., 2009) indicated that 308 participants were needed, considering 47 predictors (21 single predictors and 26 interactions), an R² of 0.40 (based on similar research, Sniehotta et al., 2013), and 90% power. Participants aged 18 and over and residing in France were recruited to answer an online survey (about 20 min). Recruitment was done using social media (i.e., Facebook and Twitter) and by word of mouth. To encourage participation, our research laboratory committed to donating 0.50€ to bioclinical research on COVID-19 for each completely fulfilled questionnaire. The survey was available between March 30 (2 weeks after the French government announced the lockdown) and April 10, 2020.

Three-hundred-and-eighty-six people (65.54% women; M_age = 33.09, SD = 13.18) completed the survey, after reading and signing an online informed consent form.

Physical activity during lockdown was assessed based on the International Physical Activity Questionnaire (IPAQ, Craig et al., 2003), which was adapted to better reflect the extraordinary circumstances of COVID-19 lockdown. Participants reported the time in minutes on different physical activity categories. These categories were chosen based on a recent opinion article about how to maintain physical activity levels during COVID-19 (Chen et al., 2020). Participants were also asked to add the time spent doing any other physical activities and, in this case, to define these activities. We then classified each activity into moderate-to-vigorous physical activity when it was superior or equal to 3 METS (metabolic equivalent task, which is the amount of energy that is used during an activity) using the compendium of physical activities of Ainsworth et al. (2011).

Usual physical activity before the lockdown was assessed using the Saltin–Grimby Physical Activity Questionnaire (Grimby et al., 2015).

Intention to do physical activity was assessed using one item from Godin (2012), and self-efficacy related to physical activity was assessed using one item (Schwarzer et al., 2015). Autonomous and controlled motivation toward physical activity was assessed using a short version of the “motivation scale toward health-oriented physical activity” (Boiché et al., 2019). The eight items reflected four motivational regulations: intrinsic, identified, introjected, and external regulations. Intrinsic and identified regulations were averaged to obtain autonomous motivation, and introjected and external regulations were averaged to obtain controlled motivation (Brunet et al., 2015). Autonomous motivation showed good reliability (α = 0.89). However, because controlled motivation did not show good reliability (α = 0.55), we decided to remove one item. Reliability after removing this item was acceptable (α = 0.61). Subjective vitality was assessed using the Subjective Vitality Scale (Ryan and Frederick, 1997), showing good reliability (α = 0.90), and perceived stress was assessed using a French translation of the short form of Perceived Stress Scale (PSS-4, Warttig et al., 2013), showing good reliability (α = 0.81). Finally, the perceived risks of getting coronavirus were assessed using perceived susceptibility and perceived severity scales. Perceived susceptibility of getting coronavirus disease was adapted from a scale related to the susceptibility of getting influenza infection (Nexøe et al., 1999). This scale did not show good reliability (α = 0.48). Therefore, we decided not to include it in our analyses. Perceived severity of coronavirus disease was assessed and adapted from perceived severity scale of getting influenza infection (Nexøe et al., 1999). Reliability was good for this scale (α = 0.77).

Media exposure was assessed to gather information about the extent to which the search of information has or has not increased since the start of lockdown. Four items measured four different sources of information (e.g., television, Internet, social networks, and press). Reliability was acceptable for this scale (α = 0.64) (more details of the scales in Supplementary Material).

Sociodemographic information included age, gender, number of children, employment status (full-time work, partial-time work, partial unemployment, or no job), educational attainment, type of housing (housing with access to green areas or terrace and housing without access to green areas or terrace), habitat surface area, region's degree of contamination (regions most affected by coronavirus were classified as red, regions less affected as yellow, and the regions the least affected as green), and access to sports equipment at home (yes or no).

Moderate-to-vigorous physical activity (MVPA) did not have a normal distribution, and squared root transformation was applied to approximate a normal curve. Once MVPA was transformed, skewness and kurtosis were examined to check for normality.

All hypotheses were tested using multiple linear regressions in R version 3.6.0. The “Lm” function was used to test the first and second hypotheses, and “olsrr” package (Hebbali, 2020) was used to do stepwise regression analyses (Hypothesis 3). Dummy variables were created for the categorical variables (gender, employment status, type of housing, and access to sports equipment at home).

Hypothesis 1 was tested using hierarchical regression analyses. In the first step, all the sociodemographic and environmental variables were included as predictors. In the second step, individual variables (intention, self-efficacy, autonomous motivation, controlled motivation, subjective vitality, perceived stress, perceived severity of COVID-19, and usual physical activity before lockdown) were additionally included following the methodology used by Sniehotta et al. (2013). Finally, both models were compared using chi-square difference tests to decide which model better explained behavior.

Hypothesis 2 was investigated using mediation analysis following recommendations of Yzerbyt et al. (2018), which showed that the joint significance test has a better balance of type I error and statistical power, compared to other approaches such as the bias-corrected bootstrap method. Intention and self-efficacy were tested as mediators between all other variables (environmental, sociodemographic, and individual) and physical activity. In the first step, we tested whether sociodemographic, environmental, and individual variables (except intention and self-efficacy that were tested as hypothesized mediators) predicted physical activity. In the second step, we tested whether sociodemographic, environmental, and individual (except the hypothesized mediators) variables predicted each of the hypothesized mediators (intention and self-efficacy). In the third step, we tested whether each mediator predicted physical activity when controlling for sociodemographic, environmental, and individual variables. According to the joint significance method, an indirect effect is claimed when regression coefficients in the second and third steps are significant.

Hypothesis 3 was tested using stepwise forward regression analyses. Stepwise forward regression is a method that selects and retains predictors based on mathematical criteria (e.g., Akaike information criterion), the final model containing the best predictors of the outcome and the best fitting indices (Field et al., 2012). In the first step, we centered all predictors using subtract mean to avoid multicollinearity problems (e.g., Shieh, 2011; Iacobucci et al., 2016). In the second step, physical activity was regressed on all sociodemographic, environmental, and individual variables. In the third step, interactions between sociodemographic and environmental variables, on the one hand, and intention and self-efficacy, on the other hand, were included. Finally, significant interactions were decomposed into simple slope analyses and Johnson-Newman plots using the package “interactions” (Long, 2019). To simplify these analyses, all the variables were scaled using the scale function in R (this function subtracts the mean and divides each value by the standard deviation).

After testing each hypothesis, we followed recommendations to assess the independence of residuals (using Durbin–Watson test), normal distribution of residuals (using bar plot and q–q plot), and non-multicollinearity (using VIF function in “car” package, Fox and Weisberg, 2019).

The sample population reported performing an average of 368 min of moderate-to-vigorous physical activity (MVPA) per week (SD = 251.12). The means, standard deviations, and the description of our variables are presented in Table 1. Correlations between variables are displayed in Supplementary Material.

Hierarchical multiple linear regression analyses were performed to test Hypothesis 1 (see Table 2). The first model including all sociodemographic and environmental variables was significant [F_{(12, 354)} = 4.27, p < 0.001], with an R² of 0.13. Gender (β = 0.11^*, p = 0.040), habitat surface area (β = 0.13^*, p = 0.035), and not having access to sports equipment at home (β = −0.24^***, p < 0.001) were significantly associated with physical activity.

The second model, which included individual variables in addition to sociodemographic/environmental ones, was significant [F_{(20, 330)} = 10.95, p < 0.001] with an R² of 0.40. Usual physical activity before the lockdown (β = 0.32^***, p < 0.001), intention (β = 0.24^***, p < 0.001), habitat surface area (β = 0.11^*, p = 0.037), and controlled motivation (β = −0.09^*, p = 0.048) were significantly associated with physical activity during COVID-19 lockdown. Durbin–Watson test (Durbin and Watson, 1971) (Durbin–Watson_{Model 1} = 1.84, Durbin–Watson_{Model 2} = 1.97), quantile–quantile plot (available in Supplementary Material), as well as VIF tests (Mansfield and Helms, 1982) (average VIF_{Model 1} = 1.18, average VIF_{Model 2} = 1.24) suggested that residuals were normally distributed and not autocorrelated (i.e., Durbin–Watson values should be between 1.5 and 2.5; Field et al., 2012 and VIF values should not be bigger than 10; Field et al., 2012). Finally, the chi-squared tests showed that the second model (the extended one) better explained physical activity than the first model.

The first multiple regression of the mediation analysis (see Table 3, model 3) tested whether sociodemographic/environmental and individual variables (excluding intention and self-efficacy) predicted physical activity. This regression was significant [F_{(18, 332)} = 9.19, p < 0.001] with an R² of 0.33. Usual physical activity before the lockdown (β = 0.38^***, p < 0.001), subjective vitality (β = 0.15^**, p = 0.003), autonomous motivation (β = 0.13^*, p = 0.015), and controlled motivation (β = −0.10^*, p = 0.042) were significant predictors. Durbin–Watson_{Model 3} = 1.95 and average VIF_{Model 3} = 1.18.

Second, in model 3.1 (Table 3), intention was regressed on the same predictors used in model 3. The regression was significant [F_{(18, 334)} = 5.97, p < 0.001] with an R² of 0.24. Autonomous motivation (β = 0.33^***, p < 0.001), usual physical activity before lockdown (β = 0.17^**, p = 0.004), subjective vitality (β = 0.14^*, p = 0.011), and perceived stress (β = −0.10^*, p = 0.041) were significantly associated with intention to do physical activity. Durbin–Watson_{Model 3.1} = 2.01 and average VIF_{Model 3.1} = 1.18.

In model 3.2 (Table 3), self-efficacy was regressed on the same predictors. This model was significant [F_{(18, 334)} = 9.52, p < 0.001] with an R² of 0.34. Subjective vitality (β = 0.30^***, p < 0.001), autonomous motivation (β = 0.29^***, p < 0.001), and usual physical activity before lockdown (β = 0.19^**, p = 0.001) were significantly related to self-efficacy to do physical activity. Durbin–Watson_{Model 3.2} = 2.06 and average VIF_{Model 3.2} = 1.19.

We decided to stop the mediation analyses at this stage because there was no sociodemographic or environmental factor that was significantly associated to both physical activity and one of the potential mediators (intention or self-efficacy).

Given the high number of predictors when adding interactive terms, a stepwise forward multiple regression analysis was performed to test Hypothesis 3. The final model is detailed in Table 4. This model was significant [F_{(29, 321)} = 8.64, p < 0.001] with an R² of 0.44. Usual physical activity before the lockdown (β = 0.28^***, p < 0.001), intention (β = 0.20^*, p = 0.022), media exposure (β = 0.10^*, p = 0.031), and controlled motivation (β = −0.11^*, p = 0.020) were significantly related to physical activity.

Concerning the moderating role of self-efficacy and intention, the interaction between gender and intention (β = 0.12^*, p = 0.041) and the interaction between people having a part-time job and intention (β = −0.10^*, p = 0.042) were significantly related to physical activity.

Durbin–Watson test [Durbin and Watson, 1971; Durbin–Watson_{Model 4} = 2.00, quantile–quantile plot (displayed in Supplementary Material)] as well as VIF tests (average VIF_{Model 4} = 1.35) suggested that residuals were normally distributed and not autocorrelated.

To simplify simple slopes analyses interpretations, all independent variables were scaled before analyses. All the Johnson Neyman plots are displayed in Supplementary Material. We then decomposed gender × intention and partial-time job × intention interactions using “Interactions” package (Long, 2019) (details of the interactions are displayed in Supplementary Table 2). Intention significantly moderated the association between gender and physical activity (Figure 1). This association was significant when intention was lower or equal to SD = −1.37. In other words, women were more physically active than men when intention was low. Moreover, intention significantly moderated the association between partial-time job and physical activity (Figure 2). This association was significant when intention was inferior to SD = −1.22. In other words, participants with partial-time jobs were less physically active than participants with full-time jobs, but again, only when intention was low.

Results provide partial support to the hypothesis that individual, sociodemographic, and environmental factors independently predict physical activity (H1). More particularly, we observed a significant role of only one environmental variable (habitat surface area). In contrast, three individual-level variables (usual physical activity, intention, and controlled motivation) significantly predicted physical activity. In other words, people were less physically active when they were little physically active before the COVID-19 lockdown, when they had low intention to be physically active, when they had a high controlled motivation, and when they lived in a small housing.

In contrast, our findings do not provide support to the hypothesis that intention and self-efficacy mediate the association between sociodemographic/environmental factors and physical activity, which contradicts previous studies (Sniehotta et al., 2013; Hagger and Hamilton, 2020). This may be due to the lack of statistical power to carry out mediation analyses.

Finally, intention moderated the association between some sociodemographic variables (i.e., gender and partial-time job) and physical activity, providing some support to H3. More particularly, when intention was low, women and participants with full-time jobs were more physically active than men and participants with partial-time jobs.

The main contribution of this study is to show that individual factors predicted physical activity more than environmental and sociodemographic ones during lockdown, corroborating the results of Rhodes et al. (2020). Although the lockdown has caused sudden changes in people's work, family, and living environment, usual physical activity before the lockdown remained a major predictor of physical activity during this period. This suggests the importance of habits in order to maintain regular physical activity in a suddenly changing environment. Whereas one could have expected external factors to be particularly important in this situation, only one environmental factor (i.e., habitat surface area) significantly predicted physical activity.

At first glance, these results may seem contradictory with several studies showing that the diminution of physical activity during lockdown mostly affected people who were usually physically active (Barkley et al., 2020; Bourdas and Zacharakis, 2020; Castañeda-Babarro et al., 2020; Maltagliati et al., 2020; Meyer et al., 2020; Martínez-de-Quel et al., 2021). Instead, we believe our results nicely complements this line of research. While the lockdown may have negatively affected the evolution of physical activity mostly in usually active individuals, the present study indicates that these individuals were still more active than usually inactive individuals (see also Maltagliati et al., 2020). This suggests that although past physical activity did not completely prevent the damaging impact of lockdown on physical activity, it still had a protective role during lockdown.

The predictive role of intention was in line with past research (e.g., Hagger et al., 2002). In contrast, the lack of significant association between autonomous motivation and physical activity (Teixeira et al., 2012) as well as the association between self-efficacy and physical activity (Hagger et al., 2002) were less expected.

Furthermore, the role of habitat surface area is less studied in the physical activity literature. Some research in leisure-time sitting (Saidj et al., 2015) showed that people living in smaller surfaces tended to spend more hours in a leisure-time sitting. Moreover, habitat surface and characteristics of housing might be an indirect measure of socioeconomic status (Juhn et al., 2011). If we link smaller surfaces with lower socioeconomic status and bigger surfaces with higher socioeconomic status, this could explain our results, as socioeconomic status is related with physical activity (e.g., Ford et al., 1991; Gidlow et al., 2006; Cerin and Leslie, 2008; Beenackers et al., 2012).

Contrary to past research (Cerin and Leslie, 2008; Sniehotta et al., 2013; Hagger and Hamilton, 2020), sociodemographic and environmental effects were not mediated by intention and self-efficacy. COVID-19 has provoked negative impacts on health, employment, and economy in most countries. Nevertheless, recent studies reveal negative impacts are greater for those with lower socioeconomic status (Chung et al., 2020), suggesting that social, health, and economic inequalities are exacerbated due to the epidemic (van Dorn et al., 2020). It seems plausible that the extraordinary challenges of the COVID-19 have revealed a direct association between sociodemographic/environmental and physical activity rather than an association mediated by intention and self-efficacy.

Finally, previous studies have shown that intention and self-efficacy moderate physical activity behaviors (Sniehotta et al., 2013; Schüz et al., 2019; Hagger and Hamilton, 2020); therefore, intention toward physical activity might moderate the effects of sociodemographic and environmental variables on physical activity. For instance, gender and intention toward physical activity have been shown to affect physical activity behaviors in previous work (for a review, see Rhodes and Dickau, 2013).

Measuring physical activity using self-reports was the main limitation of this study, as past research has shown an overestimation of the amount of physical activity when using self-reported physical activity (Dyrstad et al., 2014). In addition, while some methods of power analysis suggest that our study is sufficiently powered to detect mediation (e.g., Schoemann et al., 2017), our methods suggest instead that our study might be insufficiently powered to detected mediation (e.g., Fritz and MacKinnon, 2007). Accordingly, results of our mediation analyses should be interpreted with precaution and need to be replicated in future studies before concluding on the mediating role of intention and self-efficacy in the relationships between sociodemographic/environmental variables and physical activity. Furthermore, because the participants were recruited through social media, our sample was overeducated (i.e., individuals holding a diploma of more than 2 years in France, representing between 14 and 36% of the population; INSEE, 2019) and had fewer children than the average French person (i.e., 0.5 in our sample against the birth rate of 1.87; INSEE, 2019). As such, our results should be interpreted with caution as they are limited to this particular population, which may limit the generalization of our results. In addition, while we observed that individual and environmental/sociodemographic variables independently predicted physical activity, fully disentangling their role is difficult. Indeed, it is possible that usual exercise is determined by sociodemographic (e.g., gender and social status) and environmental factors. As such, some of the environmental/sociodemographic effects may have been partialled out by the inclusion of usual physical activity in the model. Finally, the cross-sectional nature of our study does not allow us to establish causal links. Further longitudinal research during and after the lockdown might allow having more insights about the barriers and levers to physical activity, as well as the mediation and moderation effects of psychological variables.

In terms of practical implications, identification of the sociodemographic, environmental, and individual factors of physical activity patterns and levels could benefit physical activity promotion programs. Most countries have implemented two or more lockdowns since the beginning of the pandemic, and the health situation seems to be far from over. Consequently, the promotion of healthy behaviors during lockdowns are critical to preserve mental and physical health, especially for people who have been impacted by unemployment and the economic crisis provoked by the COVID-19 pandemic. Future research should focus on understanding how the health behaviors of individuals from different socioeconomic backgrounds are affected by containment measures in order to better adapt intervention programs.

Broadly speaking, understanding how different levels of factors (i.e., individual, environmental, and sociodemographic) affect physical activity and other health behaviors might give us clues to address social inequalities in physical activity and health (e.g., Hunter et al., 2015). This could be done by targeting either individual-level factors or environmental-level ones. For example, developing intention to be physically active or autonomous forms of motivation for physical activity seems crucial during lockdowns. This may be done by fostering positive attitudes toward physical activity at the individual level or by implementing policies that enable secure, accessible, and child-friendly outdoor places (e.g., public parks) for people living in small and crowded housings. In summary, these findings provide some evidence for the importance of considering multi-level barriers and levers to healthy behavior.