This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
Walking is a simple but beneficial form of physical activity (PA). Self-monitoring and providing information about social norms are the 2 most widely used “mobile health (mHealth)” strategies to promote walking behavior. However, previous studies have failed to discriminate the effect of self-monitoring from the combination of the 2 strategies, and provide practical evidence within Chinese culture. Some essential moderators, such as gender and group identity, were also overlooked.
We aimed to investigate the effectiveness of social norm and self-monitoring interventions for walking behavior and assess the moderating effects of gender and group identity, which could guide optimal mHealth intervention projects in China.
In 2 longitudinal tracking studies (study 1, 22 days; study 2, 31 days), Chinese college students wore trackers for at least 8 hours per day (MASAI 3D Pedometer and Xiaomi Wristband 2) to record their daily step counts in baseline, intervention, and follow-up stages. In each study, participants (study 1: n=117, 54% female, mean age 25.60 years; study 2: n=180, 51% female, mean age 22.60 years) were randomly allocated to 1 of the following 3 groups: a self-monitoring group and 2 social norm intervention groups. In the 2 intervention groups and during the intervention stage, participants received different social norm information regarding group member step rankings corresponding to their grouping type of social norm information. In study 1, participants were grouped by within-group member PA levels (PA consistent vs PA inconsistent), and in study 2, participants were grouped by their received gender-specific social norm information (gender consistent vs gender inconsistent). Piece-wise linear mixed models were used to compare the difference in walking steps between groups.
In study 1, for males in the self-monitoring group, walking steps significantly decreased from the baseline stage to the intervention stage (change in slope=−1422.16;
Both gender and group identity moderated the effect of social norm information on walking. Among females, social norm information showed no benefit for walking behavior and may have exerted a backfire effect. Among males, while walking behavior decreased with self-monitoring only, the inclusion of social norm information held the level of walking behavior steady.
Walking is a simple but highly beneficial form of physical activity (PA) [
Providing social norm information and providing one's self-monitoring information (ie, a record of one’s behavior) are currently 2 of the most widely used strategies in mHealth. In self-monitoring interventions, pedometers or pedometer apps on smartphones (eg, Apple Health and Accupedo) provide users with daily step counts, which enable them to monitor and improve their own walking behavior [
In this study, we aimed to optimize mHealth walking interventions and provide more practical evidence about the effectiveness of interventions based on social norms. To address these aims, we implemented 2 longitudinal tracking studies in order to compare and isolate the mixed effects of social norm and self-monitoring interventions on promoting walking. In particular, we focused on 2 theoretically essential factors and ignored moderators regarding the influence of social norm interventions (group identity and gender).
Self-monitoring refers to a systematic self-observation or recording of target behaviors (eg, walking and food intake) [
Data from several studies suggest that a self-monitoring strategy, typically informing people how many steps they have walked, can be effective in boosting walking behavior [
However, there has been some controversy surrounding the effect of self-monitoring. For example, some researchers have argued that, despite its numerical effectiveness, the act of self-monitoring does not have a positive influence on walking behavior. Self-monitoring may also not guarantee high levels of exercise adherence, with rates of adherence shown to decline significantly over time [
More importantly, previous studies have paid little attention to exploring the mechanisms that underlie self-monitoring interventions in practice, thus limiting its application value and generalizability to other areas. For example, in order to optimize effectiveness, interventions have implemented a combination of strategies with self-monitoring included in a package of self-regulation processes [
Numerous theories of behavioral change, including the theory of planned behavior as an example, have listed social norms as an essential predictor of actual changes in behavior [
In particular, the perception of social norms about PA was positively correlated with the actual PA level of individuals [
However, the notion that social norm interventions are reliably effective has been complicated by some studies that have found its effect on behavioral change to be unstable or even negative [
With the development of social norm theories, researchers have increasingly declared that the relationship between social norms and target behavior cannot be explained by a simple one-way causal model. Recent studies have attempted to explain the weak effect of social norm interventions on PA by investigating previously ignored moderators such as group identity [
As one of the most widely used and effective theories in predicting and intervening in health behaviors, such as reducing alcohol intake [
As an innate label, information regarding individuals with the same gender may feel more relevant and relatable. Therefore, according to the TNSB, because of a higher group identity, social norms within males or females (gender-consistent or gender-specific social norms) may have a greater impact on the behaviors of individuals [
Moreover, the finding of systematic gender differences in the target behavior of this study (ie, PA) (see Pollard and Wagnild’s review [
Since 2018, China has been the home of over 500 million mHealth users [
Although there are increasing numbers of apps, smart devices, and high-tech companies trying to promote PA among individuals through systems that utilize self-monitoring or social norms, little is known about the comparison of the effects of social norm interventions with self-monitoring interventions. The absence of substantial research investigating the underlying mechanisms of social norm interventions has obstructed the development of more personalized and optimized mHealth interventions. This disconnection between up-to-date social norm theories (eg, the TNSB) and antiquated intervention techniques may have weakened the potential for practical benefits from the existing evidence base.
Moreover, previous studies have suffered from some limitations, both methodologically and practically. For example, since most previous studies on social norms lacked a corresponding self-monitoring control group [
To address this, we aimed to answer the following 2 primary questions in this study: (1) In China, is an intervention based on social norms more effective than self-monitoring at promoting walking behavior? (2) Do gender and group identity moderate the effect of social norm interventions on walking? Specifically, we conducted 2 longitudinal tracking studies to compare the effects of self-monitoring and social norms on walking behavior. In study 1, we compared the effects of social norm and self-monitoring interventions on PA between genders by randomly assigning participants to a self-monitoring group, a PA-consistent intervention group (ie, similar levels of PA among group members), or a PA-inconsistent group (ie, varying levels of PA among group members). In study 2, after observing a gender-specific effect of social norm interventions on PA, we attempted to replicate the main finding of study 1 using a more precise measure of steps walked. We also assigned participants to a self-monitoring group, a gender-consistent intervention group (ie, providing social norm information that is gender-specific), or a gender-inconsistent intervention group (ie, providing social norm information that is not gender-specific). Our hypotheses were as follows: hypothesis 1 (H1), in China, providing social norm information, in the form of step ranking, can promote walking more effectively compared with self-monitoring; hypothesis 2 (H2), gender will moderate the effect of social norm interventions on the promotion of walking, and the walking behavior of males will be more impacted by social norms; and hypothesis 3 (H3), group identity will moderate the effect of social norm interventions on the promotion of walking, and a higher sense of group identity will strengthen the effect of social norm interventions. Furthermore, H3 had the following 2 parts: H3a, an intervention using PA-consistent norms will have a larger effect on walking behavior than that using PA-inconsistent norms; H3b, an intervention using gender-consistent norms will have a larger effect on walking behavior than that using gender-inconsistent norms.
In both studies, we recruited graduate students from the University of Chinese Academy of Sciences, Beijing, China as our participants by sending advertisements during university courses (744 students in study 1, and 986 students in study 2). To avoid the possible confounding effect of walking habits or previous mHealth app experience, all recruited students completed a 5-minute online screening and customized sports habit questionnaire. The inclusion criteria were as follows: (1) used WeRun or other walk-related mHealth apps less than twice a week; (2) not a member of any WeChat sports group; (3) self-reported being mentally and physically healthy; (4) consented to participate in the entirety of the experiment. Next, according to their self-reported sports habits, participants were labeled as either low PA (sports fewer than twice a week and run a total distance of less than 5 km per week) or high PA (sports more than three times per week and run a total distance of more than 10 km per week).
Given that the motivation of this study was to promote walking behavior among the low-PA sample, we only selected low-PA students as our formal participants (high-PA students were only recruited in study 1 to set up the PA-inconsistent group, but were not included for further analysis). We preset our sample size to at least 40 participants in each group. Thus, we included a total of 127 low-PA participants in study 1 (with another 42 high-PA students), while 182 low-PA participants were recruited for study 2.
This study was approved by the Institutional Review Board of the Institution of Psychology, Chinese Academy of Sciences, and all participants took part in the study voluntarily and completed written informed consent. To record the number of steps walked, all participants were asked to wear step trackers (MASAI 3D Pedometer, study 1) or smart bands (Xiaomi Wristband 2, study 2) during the entire experimental period. After the study, participants were allowed to keep their step trackers (and approximately US $18 in study 1) or smart bands as payments for participation.
To compare the effects of self-monitoring and social norms, both studies consisted of 3 stages (baseline, intervention, and follow-up;
Flow diagram for study 1. The diagram shows the complete experimental procedure including enrollment, randomization, and intervention. All participants (valid n=117) in 3 groups were tracked for 22 days. PA: physical activity.
Flow diagram for study 2. The diagram shows the complete experimental procedure including enrollment, randomization, and intervention. All participants (valid n=180) in 3 groups were tracked for 31 days. PA: physical activity.
To compare the effects of self-monitoring and social norms, we randomized all participants in both studies into 3 groups according to the interventions they received, namely, a self-monitoring intervention group and 2 social norm intervention groups. Participants in the 2 social norm intervention groups received social norm information from experimenters during the intervention stage, whereas participants in the self-monitoring intervention group did not receive any extra message during the experimental stages.
To test the moderating effect of group identity on the effect of social norm interventions, we aimed to compare the effects of providing social norm information among groups that should theoretically have a stronger sense of group identity (ie, PA-consistent and gender-consistent groups) against those that should theoretically have a weaker sense of group identity (ie, PA-inconsistent and gender-inconsistent groups). In particular, in study 1, we randomly assigned participants to either the self-monitoring group or 2 social norm intervention groups (the PA-consistent intervention group and PA-inconsistent intervention group). In the PA-consistent intervention group, all participants had a low PA level, while in the PA-inconsistent intervention group, half of the participants had a low PA level and half had a high PA level. In study 2, we randomly assigned participants to either a self-monitoring group or 2 social norm intervention groups (the gender-consistent intervention group and gender-inconsistent intervention group). In the gender-inconsistent (but PA-consistent) intervention group, participants received general gender-mixed social norm information as in study 1 (ie, gender-mixed information simply showing the rank of each member in the WeChat group), while in the gender-consistent (and PA-consistent) intervention group, participants received 2 gender-separate social norm information sets (ie, 2 gender-specific information sets showing the rank of each female student among all the female members and each male student among all the male members). Through this design, participants in the PA-consistent group in study 1 and gender-consistent group in study 2 will theoretically have a stronger sense of group identity than participants in the other groups.
Except for high-PA participants in study 1, all low-PA participants in both studies were randomly allocated to 1 of the 3 groups according to randomly generated numbers. All participants were single-blind to the nature of our experimental design during the entire experimental period.
During the whole experimental period (including baseline, intervention, and follow-up), each participant was able to monitor his/her own steps by checking the trackers at any time. In addition, in order to send social norm information within the 2 social norm intervention groups of both studies, during the intervention stage, we set up separate WeChat groups for participants within each intervention group. Participants in the WeChat groups were asked to send a picture of their step trackers (study 1) or smart bands (study 2) before sleep every day. The next day at 10 AM, experimenters sent a personalized step ranking to each WeChat group member regarding their performance from the previous day.
To ensure that participants received the daily step ranking and paid sufficient attention to it, they were asked to finish an online daily tracking questionnaire containing 3 questions about step rankings (eg, How many steps did the person rank first in the group walk yesterday?) and their step tracker wearing habits (eg, When did you start to wear the tracker today?).
To measure social norms as our secondary outcomes, participants also completed a social norm questionnaire on the last day of each stage (total of 3 times) that was adapted from a previous study [
Consistent with previous studies [
In study 1, we excluded a total of 14 participants from all analyses for the following reasons: 10 were lost in the follow-up stage, 1 did not answer the tracking questionnaire during the intervention stage, 1 had no valid baseline step data, and 2 lacked sufficient intervention step data (more than five missing values). The final valid sample included 155 participants (75 males and 80 females; mean age 25.75 years, SD 1.27 years; age range 23-33 years), of which 117 participants were from low-PA groups. In study 2, 2 participants quit the experiment halfway through and were thus excluded from all analyses. This left a valid sample of 180 participants (92 males and 88 females; mean age 22.60 years, SD 1.16 years; age range 20-30 years).
In order to ensure the homogeneity of all the experimental groups, we conducted analysis of variance (ANOVA) to compare participant characteristics (age, subjective health level, and BMI) and baseline average step counts between the 3 groups. The homogeneity of variance was tested and satisfied. The significant factor (BMI in study 1) was included as a covariate in the analyses for the main outcomes.
In order to compare changes in social norms (injunctive norms, male descriptive norms, and female descriptive norms) among groups across different experiment stages, we used a linear mixed-effects (LM) model with the stage (baseline, intervention, and follow-up) as a within-group factor, group and gender (male and female) as between-group factors, BMI in study 1 as a covariate, and the participant as a random intercept term in the model.
Walking step data in our studies had 2 key features. First, there were 3 different stages (baseline, intervention, and follow-up) in the entire experiment. Second, each stage consisted of a segment of repeatedly measured daily step counts from each participant (in study 1, a total of 22 daily step counts, while in study 2, a total of 31 daily step counts). Given these features, to better capture the variability of participants’ daily step counts, both at the individual level and the trend during each stage, we applied a piece-wise linear mixed-effects (PLM) model. Compared with a traditional simple linear model with a constant slope, a piece-wise linear model implements several segmented regression lines with different slopes, and thus is suitable to perform staged and especially time-segmented prediction [
The PLM model for walking step data was conducted for each gender separately. The predictors of the model were group, time, and their interactions, with BMI as a covariate in study 1 and the participant as a random intercept term in the model. The predictor “time” was defined by combining the continuous variable “day” during the entire observation period (22 days for study 1; 31 days for study 2) and the dummy categorical variables to reflect “stage” (defined as baseline days 1-3, intervention days 4-17, and follow-up days 18-22 for study 1; and baseline days 1-10, intervention days 11-25, and follow-up days 26-31 for study 2) to allow different linear trends over different stages. The resulting 3 independent slopes represented the different changes of each group from baseline to the intervention stage, as well as from the intervention stage to the follow-up stage. The model also allowed different intercepts for each participant to control for possible variability in individuals’ overall levels of steps. In this study, as participants in all 3 groups could self-monitor their steps during the entire experimental period (baseline, intervention, and follow-up), the changes in the slopes from baseline to the intervention stage in the 2 social norm intervention groups will represent the extra effects of social norm information on walking (isolated from the effect of self-monitoring information in the entire experimental period).
Next, to more closely examine the intervention effect, we also created an LM model by only including the data from the intervention stage (14 days in study 1 and 15 days in study 2). The predictors of the model were group, day, and their interactions, with BMI as a covariate in study 1 and the participant as a random intercept term in the model. Then, by comparing the slopes of “day,” we were able to compare the effects on walking performance between providing self-monitoring alone and providing social norm information and self-monitoring.
Considering that the ranking position may impact the effect of social norms, we also added ranking position as a new covariate in the PLM. The results remained the same with the ranking position as a covariate (see detailed information in Table S1 in
We performed all data analyses using R 4.0.3 (R Project for Statistical Computing). We set the level of significance for all analyses to .05. A
Detailed information about participant characteristics is provided in
Participant demographic characteristics and baseline steps in study 1.
Variable | Male (N=54) | Female (N=63) | ||||||
|
Self-monitoring (n=15) | PAa-consistent intervention (n=21) | PA-inconsistent intervention (n=18) | Self-monitoring (n=22) | PA-consistent intervention (n=19) | PA-inconsistent intervention (n=22) | ||
Age (years), mean (SD) | 25.87 (1.25) | 25.57 (1.03) | 25.72 (1.02) | .72 | 25.32 (0.89) | 25.42 (0.96) | 25.77 (1.02) | .27 |
Subjective health, mean (SD) | 3.47 (0.92) | 3.38 (0.86) | 3.67 (0.69) | .55 | 3.32 (0.72) | 3.32 (0.58) | 3.05 (0.72) | .33 |
BMI, mean (SD) | 22.54 (2.13) | 22.86 (3.58) | 20.51 (1.95) | .03 | 20.23 (2.29) | 20.76 (1.23) | 19.18 (1.83) | .03 |
Baseline steps, mean (SD) | 8687 (3759) | 7670 (3130) | 8634 (3105) | .57 | 8356 (3717) | 7527 (3407) | 7623 (2376) | .66 |
aPA: physical activity.
The LM model results (Table S1 in
As shown in
Step counts (piece-wise linear mixed-effects model predicted) of the self-monitoring and 2 social norm intervention groups in study 1 (A) and study 2 (B). Each polyline represents a participant. One highlighted polyline is used for each group and gender (male, blue line; female, red line). All (marginally) significant changes of slope and slope of the intervention are labeled. *
These results partially supported H1 and H2, suggesting that the effect of social norm interventions is gender-specific. For males in the self-monitoring group, the promoting effect of self-monitoring on walking was short-lived. Although the social norm intervention did not robustly increase walking in males, the extra social norm information, regardless of its PA consistency, could at least prevent their walking steps from decreasing. For females, self-monitoring showed a stable but nonsignificant effect on walking behavior, and we found no evidence that social norms worked better than self-monitoring.
Results of the piece-wise linear mixed-effects model in study 1 across all experimental stages.
Gender and group | Intercept | Baseline | Intervention | Follow-up | Conditional |
||||
|
|
Slope | Change in slope | Change in slope |
|
||||
|
|
|
|
|
|
|
|
0.27 | |
|
Self-monitoring (n=15) | 8169.98 | 1254.20 | .03 | −1422.16 | .02 | 48.66 | .84 |
|
|
PAa-consistent intervention (n=21) | 7930.21 | −142.53 | .77 | 181.10 | .71 | −148.91 | .46 |
|
|
PA-inconsistent intervention (n=18) | 9267.26 | 166.57 | .75 | −128.39 | .82 | −213.44 | .32 |
|
|
|
|
|
|
|
|
|
0.15 | |
|
Self-monitoring (n=22) | 8082.79 | −766.83 | .14 | 854.63 | .12 | −84.81 | .68 |
|
|
PA-consistent intervention (n=19) | 8181.32 | −37.13 | .95 | 152.45 | .79 | −54.04 | .81 |
|
|
PA-inconsistent intervention (n=22) | 8118.16 | −525.53 | .30 | 496.86 | .35 | 42.54 | .84 |
|
aPA: physical activity.
Results from the LM model at the intervention stage indicated (details in
Results of the intervention-focused linear mixed-effects model in study 1.
Gender and group | Intercept | Slope | SE | Conditional |
|||
|
|
|
|
|
0.31 | ||
|
Self-monitoring (n=15) | 8448.02 | −144.83 | 71.43 | .04 |
|
|
|
PAa-consistent intervention (n=21) | 8068.35 | 37.04 | 60.92 | .54 |
|
|
|
PA-inconsistent intervention (n=18) | 9520.74 | 48.70 | 65.76 | .46 |
|
|
|
|
|
|
|
0.21 | ||
|
Self-monitoring (n=22) | 7800.66 | 81.14 | 57.49 | .16 |
|
|
|
PA-consistent intervention (n=19) | 7973.10 | 32.73 | 61.70 | .60 |
|
|
|
PA-inconsistent intervention (n=22) | 8141.32 | −122.18 | 57.44 | .03 |
|
aPA: physical activity.
Taken together, these results partially supported H1 and H2, with several restrictions. For males in the self-monitoring group, self-monitoring only was not able to exert any long-term effect since steps decreased significantly, while for males in the 2 social norm intervention groups, steps were kept unchanged by the provision of social norm information. In contrast, for females, social norms did not work more effectively than self-monitoring. The hypothesized moderating effect of group identity (H3a) was supported, but it also interacted with gender. In particular, for females, PA-inconsistent social norms had a negative effect on walking behavior.
In summary, these results preliminarily confirmed H2 and H3, revealing a gender-specific moderating effect of group identity on social norms. Given the finding that gender did play an essential role in moderating the effect of the social norm intervention and considering the reliability of our results, we conducted study 2 to replicate this study. Moreover, we modulated gender consistency to further examine the effect of group identity.
Detailed information of participant characteristics can be found in
Participant demographic characteristics and baseline steps in study 2.
Variable | Male (N=88) | Female (N=92) | |||||||
|
Self-monitoring (n=23) | Gender-consistent intervention (n=33) | Gender-inconsistent intervention (n=32) | Self-monitoring (n=26) | Gender-consistent intervention (n=34) | Gender-inconsistent intervention (n=32) | |||
Age (years), mean (SD) | 22.57 (1.20) | 23.06 (1.50) | 22.88 (1.29) | .41 | 22.58 (0.95) | 22.24 (1.02) | 22.28 (0.63) | .29 | |
Subjective health, mean (SD) | 3.30 (0.82) | 3.42 (0.61) | 3.41 (0.50) | .77 | 3.35 (0.56) | 3.29 (0.72) | 3.44 (0.56) | .65 | |
BMI, mean (SD) | 22.40 (3.56) | 21.71 (2.52) | 21.41 (3.72) | .54 | 19.89 (2.59) | 19.42 (1.16) | 20.54 (2.52) | .14 | |
Baseline steps, mean (SD) | 8737 (2279) | 8750 (2211) | 8012 (2647) | .39 | 7358 (1518) | 8099 (1732) | 8307 (2788) | .22 |
Similar to study 1 (Table S1 in
In contrast to study 1, we found that females perceived higher descriptive gender norms than males. When asked “What is your estimation of the percentage of male/female students in the same university walking over 6000 steps?” (descriptive male/female gender norms), female participants estimated that 52.53% (SD 20.68%) of male students and 40.13% (SD 19.76%) of female students would walk over 6000 steps, while male participants estimated that 42.00% (SD 19.98%) of male students and 33.99% (SD 19.06%) of female students would walk over 6000 steps. In addition, there was a significant interaction between group and gender (
Again, these results indicated that descriptive gender norms kept increasing for both genders as the experiment progressed from baseline to intervention to follow-up. We found that the gender-consistent manipulation was effective in influencing the gender norm perceptions of males but not females.
Results of the PLM model in
However, in contrast to study 1, we found a backfire effect of gender-consistent social norms among females. Females in the self-monitoring group walked more after baseline, and the slope changed in a positive direction over time from baseline to the intervention stage (
Results of the piece-wise linear mixed-effects model in study 2 across all experimental stages.
Gender and group | Intercept | Baseline | Intervention | Follow-up | Conditional |
||||
|
|
Slope | Change in slope | Change in slope |
|
||||
|
|
|
|
|
|
|
|
0.25 | |
|
Self-monitoring (n=23) | 8159.99 | 64.89 | .30 | −151.33 | .08 | 160.51 | .20 |
|
|
Gender-consistent intervention (n=33) | 8518.26 | −39.97 | .45 | 28.29 | .69 | −81.14 | .44 |
|
|
Gender-inconsistent intervention (n=32) | 7839.48 | −37.16 | .49 | 50.25 | .49 | −100.37 | .35 |
|
|
|
|
|
|
|
|
|
0.22 | |
|
Self-monitoring (n=26) | 7303.92 | −117.33 | .04 | 164.49 | .03 | −148.91 | .19 |
|
|
Gender-consistent intervention (n=34) | 8014.41 | 79.70 | .11 | −143.68 | .03 | 177.38 | .07 |
|
|
Gender-inconsistent intervention (n=32) | 7970.47 | 10.94 | .83 | −42.36 | .54 | −19.19 | .85 |
|
We did not find any significant slopes in the intervention-focused model. Detailed results from the LM model over the intervention period are shown in Table S1 and S2 in
Through 2 longitudinal tracking studies in a sample of Chinese college students, we made several findings. First, gender moderated the effect of the social norm intervention on walking. For males, the effect of the self-monitoring–only intervention was short-lived, while the addition of social norm information, regardless of its PA consistency or gender consistency, was able to keep the walking behavior unchanged. For females, we found no evidence to show that social norms performed better than self-monitoring. Second, an additional higher level of group identity in females did not consistently guarantee an improved effect of the social norm intervention on PA. With PA-inconsistent social norms (study 1) and gender-consistent social norms (study 2), the results suggested even a backfire effect on walking steps among females.
Most previous intervention studies considered self-monitoring and social norms as 2 independent and effective strategies for promoting PA [
As a meta-analysis [
An alternative explanation may be related to a counterproductive cultural stereotype, prevalent in Chinese culture, that females should be less active and petite than males [
It should be pointed out that, based on the current evidence, providing social norm information might have no more than a protective effect. Since neither study revealed a significantly positive change in slope in any social norm intervention group during the intervention stage, the provision of social norm information did not produce any additional enhancement in PA. Among males, however, both studies suggested that self-monitoring on its own may lead to a reduction in steps walked following the baseline stage. The addition of social norm information was, at least, shown to hold the level of PA steady in males, which can be considered as a protective effect.
In both studies, we found that group identity moderated the effect of social norm information on walking behavior. According to predictions of the TNSB, social norm information shared between members with a high group identity should make its effect more powerful [
In particular, in study 1, we found that providing social norms with lower group identity (PA-inconsistent social norms) led to a reduced effect on walking in females, a finding which is consistent with the TNSB. This finding among females who were part of the PA-inconsistent group engaging in a low level of PA may be driven by 2 possible factors. First, we have to note the large disparity in performance between high-PA participants and low-PA participants in our study. For high-PA participants, the mean count of steps walked during the intervention stage was 11,997 (SD 3165), but for low-PA participants, the mean count of steps walked was 7868 (SD 2341). The later participants may have been likely to perceive themselves as outgroup members, thus weakening the effect of any social norm intervention. Second, these members may also have perceived a greater sense of upward social comparison, that is, comparison with peers who are relatively better off [
In study 2, however, social norm information from a gender-consistent source, which should theoretically lead to a higher group identity, surprisingly led to a decrease in PA among females compared with PA in the self-monitoring control group. Although there are few interventions using gender-consistent social norms, our results are consistent with the finding in another health intervention study (providing gender-consistent social norm information is ineffective in reducing alcohol intake) [
Based on WeChat groups and step ranking information as an ecological intervention, our study can provide new empirical evidence with high ecological validity regarding the effect of social norm information in promoting PA within the Chinese context. Our study contributes to the field of mHealth and PA interventions in 3 ways. First, by using a self-monitoring control group, our study avoided the high statistical error rate typically brought about by a mixed-strategy intervention. Second, our findings highlighted salient gender differences associated with the effects of social norm information and self-monitoring on PA. These findings suggest the necessity to take gender into account when designing mHealth interventions for Chinese users. Third, we advise caution in using group identity to promote PA. This is especially true for gender-specific manipulations, since the increase in identification with females may, in the context of promoting walking or general PA, conflict with the target behavior.
Given the critical role of personalization in mHealth interventions, our results provide some useful insights into PA-targeted mHealth intervention projects in China. First, a gender-specialized intervention is essential to the goal of nudging walking behavior in China. For males, compared with a more traditional passive presentation of walking data, proactively pushing PA-related social norm information to users may be a more effective method for pedometer apps and other self-monitoring interventions. For example, WeRun or other mHealth apps could push a daily reminder containing the walking data of the friends of male users in order to maintain the initial beneficial effect of self-monitoring. However, among females, since social norm information did not have a beneficial effect on walking behavior, rather than adding extra social norm manipulations, maximizing the effect of self-monitoring may be a better approach. Female-targeted interventions should make full use of self-monitoring strategies. For example, they could push a daily personal step count to female users and track the overall walking step count record for each week. Second, nonessential gender-specific social norm information should be used with caution in PA interventions, since heightened gender identity may conflict with the target behavior, thus leading to an undesired outcome. Therefore, further mHealth interventions or smart devices should tailor their plans according to the gender of users; the provision of social norm information may be “the icing on the cake” for males but would be superfluous for females.
Several limitations of this research should be noted. All of our participants were Chinese college students, and this may limit the external validity and generalizability of our results. Although the main results were replicated across 2 studies to ensure their reliability, more empirical evidence is needed before generalizing our findings to other samples (eg, older adults) or other cultural backgrounds (eg, those with relatively weak gender stereotypes or gender gaps). Given that college students have a relatively high PA level (the average step count recorded by WeRun was 6932 steps/day in 2019 [
In addition, previous studies have suggested that the effect of self-monitoring on PA can be attributed to the Hawthorne effect [
Finally, we did not provide sufficient empirical evidence to attribute our finding that females reduced their PA in the gender-consistent group to the presence of a backfire effect of high group identity on social norms. Other potential variables, such as trait competitiveness, may also impact the effect of social norm information. Further studies may benefit from controlling these variables to directly address this unexplained result.
Gender and group identity are 2 moderators of the effect of social norm interventions on walking. In female Chinese college students, a higher sense of group identity does not guarantee a better effect of social norm information on PA, that is, PA-inconsistent or gender-consistent social norm information reduces walking in females. However, in male Chinese college students, social norm information and group identity do not show such an effect. Specifically, while walking may decrease with self-monitoring only, the inclusion of social norm information may help to keep the level of walking steady in males.
Additional analysis of the effect of self-monitoring and social norm information on walking behavior (study 1).
Additional analysis of the effect of self-monitoring and social norm information on walking behavior (study 2).
Descriptive information of social norms (study 1).
Descriptive information of social norms (study 2).
analysis of variance
linear mixed-effects model
mobile health
physical activity
piece-wise linear mixed-effects model
theory of normative social behavior
We would like to thank J Murray and S Lippke for their constructive comments during peer review. We also acknowledge David O’Connor for his language editing. We also thank Wang Xiao-Ming, Chen Jun-Fang, Yang Shu-Wen, and Sui Xiao-Yang for their help during data collection. This research was supported by Beijing Natural Science Foundation (BNSF, 9172019), the National Natural Science Foundation of China (NSFC, 71471171), CAS Engineering Laboratory for Psychological Service (KFJ-PTXM-29), the Scientific Foundation of the Institute of Psychology, Chinese Academy of Sciences (Y9CX303008), and Major Projects of the National Social Science Foundation of China (19ZDA358).
None declared.