We conducted a retrospective survey that took place between March 11 and 16, 2022. Of the 1205 respondents, 977 (81.1%) provided informed consent, completed the survey, and were adults between the ages of 18 and 74 years, living in Canada. The minimum completion time was 5 minutes, and the median completion time was 21 minutes. A third-party market research company Maru/Blue was used to recruit participants. Maru/Blue has an online polling database of 120,000 Canadians with a representative sample compared to the Canadian census [19,20]. Panel participants are recruited through a variety of online and offline methods and receive small cash incentives and prize opportunities from the company after completing surveys.

Prior to recruiting participants, ethics approval was obtained from the Institutional Ethics Board of University of Victoria (#21-0405). Participants were presented with an informed consent statement on the front page of the electronic survey. Only those who consented were allowed to proceed to the survey items. Informed consent was obtained from all participants. Participants were not compensated for survey participation or completion by our research team but may have been compensated by Maru/Blue for survey completion. These data were not shared with the research team. Survey responses were anonymous as no potentially identifying data were collected in this study.

Demographics, including age, sex, education, ethnicity, employment status, and income, were collected using self-report measures. Motivation for PA was assessed using the Behavioral Regulation in Exercise Questionnaire-3 (BREQ-3), which has been validated for use in this population [21,22]. The BREQ-3 assesses motivation for exercise using the 6 SDT regulatory subtypes: amotivation, external regulation, introjected regulation, identified regulation, integrated regulation, and intrinsic motivation. Each subtype is assessed on a 5-point Likert scale ranging from 0 to 4, where 0=“not true for me” and 4=“very true for me.” We calculated the mean scores for each regulatory subtype and used those values to identify distinct motivational profiles. In this study, the Cronbach α (95% CI) values for the internal consistency for each motivational subtype were as follows: amotivation: 0.91 (0.90-0.93); external regulation: 0.88 (0.87-0.90); introjected regulation: 0.88 (0.86-0.89); identified regulation: 0.83 (0.81-0.85); integrated regulation: 0.89 (0.87-0.91); and intrinsic motivation: 0.93 (0.92-0.94).

To assess PA at 3 time points (pre–COVID-19, prior to March 2020; during the most recent COVID-19 stay-at-home orders in Canada, January-early March 2022; and in the past 7 days, March 2022), participants responded to items from the International Physical Activity Questionnaire (IPAQ) [23]. Participants were asked to report the number of hours per week that they participated in vigorous, moderate, and mild PA. We summed the hours of vigorous and moderate PA and multiplied the total by 60 minutes to obtain a value for MVPA minutes at all 3 time points.

To account for extreme outliers, we chose to perform 95% Winsorization on the minutes of MVPA per week at all 3 time points. We also used the Mahalanobis distance with a P value of .001 as a cutoff to identify multivariate outliers for the motivational subtype variables. After identifying these outliers, we performed LPA, both with and without them in the sample. We chose to proceed with the multivariate outliers included in the sample as the outcomes were the same with and without them. We then calculated means and SDs for continuous demographic variables and frequencies and percentages for categorical variables.

To derive motivational profiles, we used LPA. To obtain the most appropriate model and number of classes, LPA uses a number of quantitative fit statistics, including the log-likelihood, the Akaike information criterion (AIC), the Bayesian information criterion (BIC), and entropy [14]. Higher log-likelihood values and lower AIC and BIC values indicate a better model fit [24]. Entropy represents the probability that members of a sample are able to fit into 1 of the resulting classes or profiles, and the magnitude of difference between classes. An entropy value of 1 would indicate that 100% of the sample fits into 1 of the classes; therefore, entropy values close to 1 are preferred [25]. After considering fit statistics, we examined each model to determine which one contained a set of substantive and distinct profiles with differing levels of each subtype.

We used participant mean scores for each motivation subscale to calculate z-scores, which were then used to conduct LPA. After choosing the model with the best quantitative and qualitative fit both, we conducted 1-way ANOVA to determine differences in each motivational subtype among the resultant classes. When the overall main effects of class were detected, we used the Tukey honestly significant difference (HSD) test for pairwise comparisons between classes. We also calculated the partial η² to determine the effect sizes of the profile on each motivational subtype.

To examine the effects of motivational profiles on the weekly minutes of PA pre–COVID-19, during the COVID stay-at-home orders, and in the past 7 days, we used multilevel regression models using the participant ID as a random variable. We added biological sex, age, ethnicity, education, income, and current employment as covariates to the model. We examined the interaction between class and time (pre–COVID-19, during COVID-19, and in the past 7 days). When the overall main effects of class were detected, we used the Tukey HSD test for pairwise comparisons between classes. We also calculated partial η² to determine effect sizes of the class on MVPA. We used the benchmarks defined by Cohen for small (η²=0.01), medium (η²=0.06), and large (η²=0.14) effect sizes for interpretation. We used R version 4.1.3 (R Foundation for Statistical Computing), and statistical significance was set to P<.01 for all analyses.

The final sample consisted of 977 adults living in Canada in March 2022. The mean age of sample participants was 46.79 (SD 15.23) years. The majority of the sample identified as female (n=497, 50.9%), were of European (n=408, 41.8%) or other North American (n=374, 38.3%) origins, were employed full-time (n=537, 55.0%), had at least a bachelor’s degree (n=488, 49.9%), and had an income above CA $80,000 (US $58,358) per year (n=564, 57.7%).

Using quantitative fit statistics log-likelihood, AIC, BIC, and entropy, we determined that the model of best fit was either a 4-, 5-, or 6-class option. Although the 5- and 6-class solutions had lower AIC and BIC values, we determined that the 4-class solution contained qualitatively distinct profiles that better represent the full spectrum of motivational subtypes of SDT. Both 5- and 6-class solutions contained profiles that were similar to other resulting profiles in the same model (see Table 1).

Figure 1 shows the 4 resulting profiles as depicted by their motivational subtype z-scores. These scores indicate how much each profile differs from the individual subtype sample means. In this figure, 0 on the y axis represents the sample mean for each distinct motivational subtype. The “high controlled and high autonomous” (HCHA) regulation profile (n=56, 5.7%) had higher-than-expected scores on all 6 motivational subtypes but relatively higher controlled versus autonomous scores compared to the sample mean. The “low overall motivation” (LOM) profile (n=130, 13.3%) had lower-than-expected scores on all motivational subtypes. The “high autonomous and introjected” (HAI) regulation profile (n=566, 57.9%) had higher-than-expected scores on introjected regulation and the autonomous motivational subtypes. The HAE regulation profile (n=225, 23.0%) had higher-than-expected scores on both amotivation and external regulation and lower-than-expected scores on autonomous subtypes.

Table 2 shows the differences in motivational subtypes among each of the 4 profiles. We detected the main effect of profile on amotivation (F_3,973=1067.7, P<.01, =0.77), external regulation (F_3,973=171.40, P<.01, =0.35), introjected regulation (F_3,973=48.16, P<.01, =0.13), identified regulation (F_3,973=234.96, P<.01, =0.42), integrated regulation (F_3,973=120.57, P<.01, =0.27), and intrinsic regulation (F_3,973=437.07, P<.01, =0.57). Pairwise comparisons revealed that HCHA is significantly different than LOM and HAE or all subtypes and significantly different than HAI for all subtypes except identified regulation. LOM is significantly different than HAI for all subtypes and is significantly different than HAE for all subtypes except introjected regulation and identified regulation. HAI and HAE are significantly different for all regulatory subtypes.

Table 3 shows the means and SDs for weekly minutes of MVPA by profile pre–COVID-19, during COVID-19, and in the past 7 days. Figure 2 shows the results of the multilevel model with minutes of MVPA pre–COVID-19, during COVID-19, and in the past 7 days as the outcome variable. We detected a significant interaction between profile and time (F_6,1945.03=3.08, P=.005, =0.01). According to Agerström et al [26], the resulting η² represents a small, but meaningful, effect size. Pairwise comparisons revealed that LOM, HAI, and HAE had significantly fewer hours of MVPA at all 3 time points compared to HCHA. LOM and HAE had significantly fewer minutes of MVPA at all 3 time points compared to HAI. No significant differences in the minutes of MVPA were detected between LOM and HAE at any time point.

The first aim of this study was to identify the motivational profiles for PA in a sample of 977 Canadian adults. We hypothesized that we would detect profiles identified in other PA motivation research characterized by high amotivation and external regulation, high autonomous forms (ie, identified and integrated regulation, and intrinsic motivation), and low on all forms of regulation [14,17]. The study identified 4 distinct motivational profiles for PA in this sample, confirming our first hypothesis. The resulting profiles were high controlled and high autonomous (HCHA), low overall motivation (LOM), high autonomous and introjected (HAI), and high amotivation and external (HAE) regulation. Other studies assessing motivational profiles and PA have identified profiles characterized by high z-scores on autonomous regulatory forms and low z-scores on controlled forms, profiles characterized by high z-scores on controlled regulatory forms and low z-scores on autonomous forms, or combinations of all high or all low z-scores [16,17,27-33]. Although other studies have identified profiles with both HCHA regulatory forms [16,31], to the best of our knowledge, only Gourlan et al [16] also identified a profile with elevated amotivation, controlled regulatory subtypes, and autonomous regulatory subtypes.

The second aim of this study was to investigate the association between resulting motivational profiles and MVPA pre–COVID-19, during the most recent COVID-19–related stay-at-home orders in Canada (early 2022), and after those orders had been lifted (March 2022). Our second hypothesis was that specific profiles would predict MVPA participation at all 3 time points. We also detected a significant effect of profile on MVPA at all 3 time points. Specifically, HCHA, HAI, and HAE reported more weekly minutes of MVPA during the most recent stay-at-home orders in Canada than their pre–COVID-19 levels, whereas LOM decreased their minutes of MVPA. Further, LOM, HAI, and HAE reported fewer weekly minutes of MVPA in the past 7 days compared to their levels during the COVID-19 lockdown and pre–COVID-19. HCHA, however, reported a decrease in MVPA in the past 7 days in comparison to COVID-19 levels but an increase in comparison to pre–COVID-19 levels.

We were surprised to see that it was not HAI, with higher-than-average autonomous regulatory subtype scores, with the most minutes of MVPA but rather HCHA regulatory forms. Our findings are inconsistent with the literature in that, typically, amotivation and controlled regulatory forms are not associated with regular PA engagement [14,28,29,34,35]. Furthermore, controlled regulatory forms are heavily reliant on external factors (eg, family and friends who encourage the behavior) for behavioral engagement. These behaviors tend to extinguish once those factors are removed, resulting in lower levels of behavioral persistence [18]. Although controlled regulatory forms may be sufficient to initiate PA behavior, evidence does not suggest they are sufficient to support long-term engagement [36]. Similarly, a systematic review exploring SDT and PA found negative or null effects of controlled motivation on long-term PA maintenance [35].

Specific to MVPA during the COVID-19 pandemic, our results differed from those of Chirico et al [37], who conducted an assessment of motivation using a metric called the relative autonomy index and its association with the intention for MVPA during the lockdowns in Italy in the initial months of the pandemic. They determined that higher and more positive relative autonomy index scores, which represent higher scores on more autonomous regulatory forms of motivation, predicted the intention to perform PA. In our sample, the HAI profile would likely have the highest relative autonomy index score but only was associated with the second-highest levels of MVPA across time. Interestingly, the profile in our sample with the highest levels of MVPA at all 3 time points, HCHA, would likely result in a relative autonomy index score around 0. Chirico et al [37] assessed the intention for MVPA, whereas we assessed self-reported behavior, which we recognize are not synonymous. Although the association between intention and PA behavior is sizeable, it is not perfect, and several studies have identified a significant gap between intention and behavior in the realm of PA research [38].

These findings, then, suggest that the combination of high levels of all regulatory subtypes is somehow a more influential combination that either controlled or autonomous regulatory subtypes alone. Phillips and Johnson [17] were among the first to assess motivation for PA as a multidimensional construct rather than as a single index score or by individual regulatory subtype score. In their work, they determined that the relationship between controlled and autonomous regulatory subtypes and PA engagement is more complicated than originally thought. Traditionally, researchers have recognized high-quality motivation for PA as having lower levels of controlled regulatory forms and higher levels of autonomous regulatory forms. Phillips and Johnson [17], however, discovered that moderate levels of controlled regulatory forms, combined with any amount of autonomous regulatory forms, predict the highest levels of PA engagement. As such, the traditional stance that controlled regulatory forms of motivation are detrimental to PA may be incorrect, as the data presented here, along with the evidence provided by Phillips and Johnson [17], indicate that controlled regulatory forms in combination with autonomous forms may support PA engagement in some individuals. It is unclear as to why a combination of controlled and autonomous regulatory forms seems to be more supportive of PA among some people. We speculate, however, that the potential negative impacts of controlled motivation for PA that have been identified in the literature, such as decreased enjoyment, persistence, and PA intention [39], were buffered by the elevated autonomous forms. Autonomous regulatory forms, incidentally, are associated with increased enjoyment, persistence, and PA intention [39]. Future research should focus on the interplay between controlled and autonomous forms in PA motivation.

Although it remains unclear as to why individuals within the HCHA and HAE profiles increased their minutes of MVPA from pre–COVID-19 to during COVID-19, we speculate that the overall emphasis on PA during the COVID-19 stay-at-home orders may have been a factor. For example, news outlets, social media, and other media sources emphasized the importance of maintaining PA to reduce the impact of a COVID-19 infection [5] Further, despite the closure of in-person fitness facilities, online and virtual offerings from personal trainers and group fitness instructors increased over the course of the stay-at-home order period [40,41]. Potentially, individuals with elevated doses of controlled regulatory forms were subject to the social pressure (via media and social media outlets) to continue being physically active and took advantage of the novel and widely available fitness and exercise resources offered online. Further investigation is necessary to confirm this interpretation of these study results, however.

We identified that our results are applicable to other contexts beyond the life interruptions of the COVID-19 pandemic. In general, clinicians and practitioners may not want to reduce controlled regulatory forms of PA motivation when desiring to influence PA participation [17]. Perhaps interventions intending to increase both more controlled and more autonomous regulatory forms may be more beneficial for some groups. Further, future research can focus on how motivational profiles influence PA engagement during other stressful life episodes, such as job transitions, relationship changes (eg, change in marital status), or the death of a loved one. Research indicates that these life changes can negatively impact PA behavior [42]. Perhaps motivational profiles that include both controlled and autonomous regulatory forms of motivation provide resiliency against reductions in PA.

We recognize that this study is not without limitations. First, we implemented a cross-sectional design, limiting our interpretation of the results to association rather than cause and effect. Second, we asked participations to reflect on their MVPA at 3 different time points, including pre–COVID-19, which was more than 2 years prior to survey administration. Despite the use of the IPAQ, a validated self-report instrument for PA assessment [23], self-report instruments are still vulnerable to overreporting of both the time and the intensity of PA behavior [43]. In our case, the mean reported weekly minutes of MVPA were well above the 150-minute-per-week recommendation for health in Canada [4] at all 3 time points. Finally, we assessed motivation regulatory subtypes at the third time point (March 2022) but then asked participants to reflect on MVPA from prior to March 2020 and then in early 2022. A recent study investigated the stability of motivational profiles and MVPA across a 5-year period and discovered that movement between profiles is not only possible but likely [44]. Therefore, to more clearly investigate motivational profiles and their associations with MVPA at different time points, researchers should undertake a robust longitudinal design like that of Emm-Collison et al [44].

We derived 4 distinct motivational profiles from our sample. These profiles were characterized by HCHA, LOM, HAI, and HAE regulatory forms. Our findings suggest that motivational profiles predicted PA activity engagement before, during, and after the COVID-19 pandemic and associated stay-at-home orders. Participants characterized by high levels of both controlled and autonomous regulatory forms reported the highest levels of MVPA at all 3 time points, whereas all other groups saw slight increases or decreases from pre–COVID-19 to during COVID-19 but then decreased in the past 7 days to below pre–COVID-19 levels. These results suggest that contrary to previously held beliefs about motivation regulatory subtypes, controlled motivation may not be detrimental to MVPA engagement.