This cross-sectional study recruited university students (n=76) from the University of Victoria, British Columbia, Canada, using research posters and online advertisements. The rolling recruitment strategy took place between June 30, 2018, and July 1, 2019. Inclusion criteria required that participants be Instagram users, and between the ages of 18 and 29. Exclusion criteria included the inability to comprehend English. Interested participants were asked to contact the research team. The research team obtained consent from all eligible participants prior to data collection. During the consent process, participants were notified of the study objectives and that the sharing of their Instagram data was voluntary. All data obtained in connection with the study remained confidential. Confidentiality was maintained by means of encrypting on a secure server. All survey data and ID numbers were used on all databases. Researchers had access to participant Instagram data for up to 1 year from the start of the study.

Eligible participants who gave consent to participate in the study were asked to complete a questionnaire to record their demographic characteristics, daily Instagram use duration, and evaluate their current exercise identity and PA levels. Participants were also asked if they are willing to share their Instagram data. Participants that refused to share their Instagram data were given the option to write down reasons for not sharing. Those participants who agreed to share their Instagram data were asked to follow an Instagram research account in order for researchers to extract their Instagram data (photos, likes, followings). Researchers analyzed the previous 12 months of Instagram data from the completion date of the exercise identity and PA questionnaires. This study received research ethics approval from the research ethics board at the University of Victoria (protocol number: 17-294).

Demographic information (age, sex, and ethnicity) and daily Instagram usage time were collected using self-report instruments. PA levels were assessed using the Recent Physical Activity Questionnaire (RPAQ) [42]. Minutes of PA at low (metabolic equivalent of task [MET] < 6/week), moderate (MET 3-6/week), and high (MET > 6/week) intensities were calculated based on the RPAQ scoring instruction [42].

Exercise identity was assessed using the Exercise Identity Scale [16], which is a 9-item questionnaire on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree) (α=.92). Participants’ exercise identities were determined by calculating a total score from the 9-item questionnaire; thus, scores ranged from 9 to 45 whereby a higher score represented a stronger exercise identity.

Participants’ Instagram data were extracted using a customized program [43], which enabled the researchers to export users’ posts, likes received for each post, and followings into separate Microsoft Excel files. Instagram posts extracted during the past 12 months before the completion of the study questionnaire were manually coded by 2 trained research assistants using standardized criteria to determine whether posts were PA related or not (eg, binary coded). PA-related posts were defined as those posts suggesting participants performed physical activities defined by the guidelines for exercise testing published by the American College of Sports Medicine [44]. Specifically, Instagram picture and video posts were identified as PA related if they included the person in sporting attire or the person in a location to perform PA (eg, kayaking, skiing, swimsuit, hiking) individually or in a group setting. Any discrepancies between the coders were resolved through discussion. The percentage of the number of PA-related posts relative to the total number of Instagram posts was calculated for each participant. The average number of likes for PA-related posts for each participant was also calculated.

An analysis of participants’ Instagram “followings” was completed to identify the number of verified fitness or exercise-related Instagram accounts each participant followed. The verified fitness- or exercise-related Instagram accounts means Instagram has confirmed that an account is the authentic presence of the public figure, celebrity, or global brand it represents. A list of the most popular 300 verified fitness- or exercise-related Instagram accounts was compiled from publicly available Instagram rankings [45]. The percentage of exercise-related Instagram accounts following relative to the total number of Instagram followings was computed for each participant.

All analyses were performed using the standard SPSS version 26.0 for Mac (SPSS, Inc./IBM), with a significance level set at P<.05. Descriptive statistical analyses were used to compare the number of willing participants with those who were not willing to share their Instagram data. Participants that were not willing to share their Instagram data were asked to provide reasons for not sharing in an open-ended question. Responses from these questions were grouped to create common themes. Independent t tests were used to compare differences between those groups (Instagram data shared versus Instagram data not shared) on demographics, PA, and exercise identity. A chi-square test was used to compare differences between categorical variables.

We examined normality (eg, skewness > 2 and kurtosis > 3) of all variables to determine whether any transformations were required by conversion to z-scores [46]. Pearson correlation analyses were used to evaluate the relationship among exercise identity, PA, and Instagram use metrics (percentage of PA-related Instagram post, percentage of fitness-related Instagram followings, and average number of “likes” on PA-related posts). Mediation analyses were based on 5000 bootstrapped samples using Hayes PROCESS Macro version 3.5 [47]. Multiple mediation analyses was used to examine whether exercise identity mediated the relationship between Instagram use variables and PA level [47]. This process involved examining path a, the association between Instagram use (independent variable) and exercise identity; path b, the impact of exercise identity (mediator variable) on PA level; and path c, the effect of Instagram use on PA levels (outcome variable). All analyses were controlled for age and sex. The 95% CIs must not cross 0 to satisfy the criteria for mediation.

A total of 76 participants were recruited and completed the study. Participant characteristics are presented in Table 1. The mean age was 19.7 (SD 0.31) years and the sample consisted of 72% female (55/76). Overall, 46% (35/76) of participants did not agree to share their Instagram data. Of these participants, 14 participants provided reasons for not sharing. Privacy was a major concern; specifically, there were concerns with (1) data oversight to prevent misusage of data and technology (n=9), (2) limitations to remain anonymous using Instagram photos (n=4), and (3) data ownership (n=1). Participants that were not willing to share their Instagram data reported significantly fewer weekly PA minutes at moderate to high intensity (MET ≥3/week) relative to participants that agreed to share their Instagram data (P=.04); however, there were no significant differences between groups for demographic characteristics, exercise identity, weekly PA minutes at low intensity (MET <3/week), and daily Instagram usage (Table 1).

A total of 41 participants agreed to share their Instagram data. The mean number of posts during the 12-month study period was 29.6 (SD 4.7). Examination of shared Instagram data revealed that, on average 7.3 (SD 1.1) or 25% of Instagram posts shared during the previous 12 months were related to PA. The mean number of Instagram following per person was 439.3 (SD 258) accounts. The mean number of verified PA- or fitness-related Instagram accounts following per user was 3 (SD 3.14). The mean number of “likes” received on all posts was 102.2 (SD 72.4) and the mean number of “likes” received on PA-related Instagram posts was 99 (SD 73).

Normality analysis revealed that weekly PA minutes at moderate to high intensity (MET ≥3/week) and fitness-related Instagram followings percentage were kurtotic (ie, values ≥3). These variables were transformed by conversion to z-scores. The Pearson correlation analyses are shown in Table 2. The results indicated that the percentage of PA-related Instagram posts (r=0.38; P=.01) and fitness-related Instagram (r=0.39; P=.01) followings were significantly associated with exercise identity. The strengths of the associates were moderate. We did not find that the average number of “likes” received (r=0.05, P=.75) to be significantly associated with exercise identity.

Exercise identity significantly influenced the relationship between the percentage of PA-related Instagram posts and PA levels (indirect effect: 3.28, 95% CI 0.54 to 8.11; direct effect: –3.65, 95% CI –9.70 to 2.39). As indicated in Table 3, a unit increase in percentage of PA-related Instagram posts was associated with a 0.10-unit increase in exercise identity; a unit increase in exercise identity was associated with 34 minutes of PA (MET ≥3/week) increase. Approximately 29% of the variance was accounted for by the predictors (R²=0.29).

Exercise identity significantly influenced the relationship between the percentage of verified fitness-related Instagram following and PA level (MET ≥3/week; indirect effect: 91.9, 95% CI 36.93 to 174.82; direct effect: 60.53, 95% CI –115.15 to 236.21). A unit increase in percentage of fitness-related Instagram following leads to a 3.18-unit increase in exercise identity; a unit increase in exercise identity was associated with an increase of 29.34 minutes of PA (MET ≥3/week; Table 3). Approximately 26% of the variance was accounted for by the predictors (R²=0.26).

Exercise identity did not significantly influence the relationship between the average number of “likes” received for the PA-related Instagram posts and PA level (MET ≥3/week; indirect effect: 0.69, 95% CI –0.10 to 2.28). We did not observe a significant direct effect between the number of “likes” received on PA-related Instagram posts and PA minutes (direct effect: 0.28, 95% CI –2.06 to 2.61; Table 3).

The objectives of this study were to explore whether participants were willing to share their Instagram data with researchers to predict their lifestyle behaviors and examine the relationship among PA-related Instagram metrics, exercise identity, and PA. We found that the majority of participants were willing to share their Instagram data for monitoring lifestyle behaviors. We also found that only PA-related Instagram posts and fitness-related followings were significantly associated with an increase in exercise identity. Furthermore, we found that exercise identity significantly influenced the relationship between certain Instagram use metrics (eg, PA-related Instagram posts and fitness-related followings) and PA level. To our knowledge, this is one of the first studies to examine the relationship among Instagram use, exercise identity, and PA levels. The methods and the findings from this study can be used to inform future infodemiology PA-related studies.

Our results suggest that over 50% (41/76, 54%) of the users were willing to share their Instagram data for monitoring their PA and health-related behaviors. This observation was supported by our hypothesis. It was worth noting that participants that agreed to share their data reported significantly higher weekly PA minutes at moderate to high intensity (MET ≥3/week) compared with participants who were not willing to share their data. A potential explanation is that those who were less active felt guilty for not showing consistent behavior with their social media posts. Individuals may not always portray their real selves on social media sites due to factors such as lower self-esteem, self-reflection, and self-concept clarity [48]. Alternatively, those participants may feel being monitored or judged for what they were posting, given the purpose of this study [27,49]. Concerns over privacy were the main reasons for not sharing their Instagram data. Unlike Twitter data, Instagram data are not publicly available by default, and the users mainly share pictures instead of text. Thus, this poses a limitation to remain anonymous when analyzing Instagram photos. Our findings reinforced the ethical challenges faced in using social data for public health monitoring. There is a need to establish “best-practice” standards for analyzing social media data while simultaneously acknowledging peoples’ individual rights and respecting their privacy [50,51]. Previous studies have shown that users are more likely to accept the use of social media data for research if there is an oversight body that protects user data to ensure that their data are not being used beyond the intended purpose and that user data can remain anonymous to protect the identity of the user [52,53]. Overall, these privacy considerations need to be addressed for the future development of social data monitoring technology.

Based on the self-perception theory [24] and past research [25,26], we hypothesized that PA-related Instagram use would be positively associated with exercise identity and that exercise identity would mediate the relationship between PA-related Instagram use and PA minutes. Interestingly, we observed that only the percentage of PA-related Instagram posts and fitness-related followings were positively associated with exercise identity. Exercise identity also significantly influenced the relationship between these Instagram usage metrics and PA level (MET ≥3/week). Exercise identity did not influence the relationship between the number of “likes” and PA level. Previous research has demonstrated that social media “likes” can contribute to PA engagement [54,55]. The Instagram “likes” metric is unlike the number of followings and posts metrics because users do not have direct control over the number of “likes” received. The lack of consistency of our findings compared with previous literature may be attributed to the way individuals are using their Instagram in our sample. “Surveillance users” (eg, users focused on monitoring other users) may post lesser often and have a lesser number of followers, leading to a lower number of “likes” compared with “cool-ness users” who are motivated to become popular [56,57]. Thus, the types of users need to be taken into consideration for future studies.

Findings from this study have several implications for PA promotion research. First, Instagram may have the potential to target identity antecedents through the reception of both inspirational and informative posts about exercise from respected verified fitness accounts (ie, enhancing self-efficacy and congruency with the self) [58-60]. Based on the self-perception theory [24] and self-definition model [21], future PA promotion interventions using Instagram may target personal investment, perceived commitment (ie, engaging with relevant fitness groups/users), social activation (ie, contributing and receiving support from fitness groups/users), and self-expression (ie, share photos of their PA or fitness-related experiences or write about these activities using social media), as these may play a role in strengthening an individual’s exercise identity and subsequent PA behaviors [22,24]. However, it is worth noting that Instagram as a medium for expressing one’s PA behavior with the end goal being to strengthen individuals’ exercise identities may only be appealing to certain segments of the population. Second, results from this study extend previous Infodemiology research that Instagram data may be used to provide insights into health-related outcomes [39]. Future studies can build upon the methods used in this study and may also explore other Instagram metrics (eg, frequency of post) in PA infodemiology studies. Finally, PA researchers can leverage these social media analysis techniques to build prediction tools to monitor PA on a population level in real time. These tools may in the future aid public health agencies in identifying particular PA-related trends in various geographical areas on which to focus health and wellness initiatives.

There are potential negative aspects associated with using Instagram for the purpose of promoting PA and forming or strengthening one’s exercise identity that must be considered. For example, recent movements known as “fitspiration” and “fitspro” have become popular on Instagram to inspire and motivate others to eat a healthy diet and exercise regularly. Fitspiration provides many positive attributes, such as accountability for exercising and education about fitness [61]. However, some women sharing fitspiration posts on Instagram may indicate disordered eating and exercise behaviors [62]. Despite the potential of using Instagram to motivate adherence to a healthy lifestyle, it is important for future interventions to consider the potential negative effect of social media on mental and physical health. Future studies need to examine ways to create a positive environment where participants feel comfortable posting pictures of their exercise experience and journey to fitness.

In terms of the limitations of the study, our sample only included university students which may limit the ability to generalize our findings beyond our sample. Future studies should examine the relationship between social media posts and PA-related behavior in middle-aged and older adults. Another limitation is that we used a list of verified fitness-related Instagram account in our analysis. It is possible that users may follow fitness-related Instagram accounts that are not verified and did not make it onto our list. Thus, this means that the observed relationship between exercise identity and the percentage of the fitness-related Instagram following needs to be interpreted with caution. Self-report questionnaires may result in confirmation bias. Moreover, it may be possible that some users’ may have deleted photos, and thus affecting the number of PA-related photos in our analysis. Finally, the lack of longitudinal data for exercise identity and PA levels was another limitation when conducting mediation analysis. Thus, the mediation analysis result needs to be interpreted with caution. Future study with longitudinal data is needed to build more complex multivariate models to better examine mediation effects.

This study examined the relationships among Instagram use, exercise identity, and PA levels. Findings from this study demonstrated that over 50% (41/76, 54%) of participants were willing to share their Instagram data with researchers for the purpose of monitoring and predicting PA behaviors. Exercise identity significantly influenced the relationship between Instagram use (eg, PA-related Instagram posts and fitness-related followings) and PA level. Results from this study suggest that there is an association between Instagram posts and PA-related outcomes (ie, exercise identity and PA levels). Future studies should examine whether Instagram may be used for a future intervention to help form or strengthen an individual’s exercise identity, particularly if the individual can be supported by their “social world.”