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 http://www.jmir.org/, as well as this copyright and license information must be included.
Providing feedback is a technique to promote health behavior that is emphasized by behavior change theories. However, these theories make contradicting predictions regarding the effect of the feedback sign—that is, whether the feedback signals success or failure. Thus, it is unclear whether positive or negative feedback leads to more favorable behavior change in a health behavior intervention.
The aim of this study was to examine the effect of the feedback sign in a health behavior change intervention.
Data from participants (N=1623) of a 6-month physical activity intervention was used. Participants received a feedback email at the beginning of each month. Feedback was either positive or negative depending on the participants’ physical activity in the previous month. In an exploratory analysis, change in monthly step count averages was used to evaluate the feedback effect.
The feedback sign did not predict the change in monthly step count averages over the course of the intervention (
The feedback sign might not influence the effect of monthly feedback emails sent out to participants of a large-scale physical activity intervention. However, randomized studies are needed to further support this conclusion. Limitations as well as opportunities for future research are discussed.
In 2012, noncommunicable diseases (NCDs) such as diabetes, cardiovascular diseases, chronic respiratory diseases, or cancer were responsible for 68% of deaths worldwide [
Physical activity plays a crucial role in the prevention and management of NCDs, as it has been found to affect the incidence and course of NCDs such as diabetes [
Physical activity interventions often use feedback as a method to change behavior [
Behavioral theories provide a detailed specification of causal processes that lead to behavior change and can thus help to understand how feedback affects behavior [
CT provides a model of self-regulation for intentional (or goal-directed) behavior (eg, walking 10,000 steps a day). Self-regulation is vital for physical activity promotion as it constitutes the basis for self-directed change [
In contrast to CT, SCT assumes that the mere perception of behavior and standards is insufficient to regulate behavior. It rather posits that cognitions such as self-efficacy beliefs are central factors that impact goal pursuit and self-regulation [
Understandably, a major source of self-efficacy includes personal experiences of success and failure [
Effect of feedback according to control theory.
Effect of feedback according to social cognitive theory.
Both CT and SCT contradict each other in their implications for feedback design. Whereas CT predicts that negative evaluation of performance leads to favorable behavior change, SCT predicts the same for positive evaluation of performance. Whether the feedback contains a positive or a negative evaluation of performance is often referred to as the sign of the feedback message [
In order examine the effects of positive and negative feedback, we exploratory analyzed data from a cluster-randomized trial that primarily focused on the effects of different incentives on the acceptance of a digital physical activity intervention [
A total of 26,773 customers of a large Swiss health insurance company were invited through email, along with eligible family members, to participate in a physical activity intervention that was conducted from July 2015 to December 2015. In order to participate, customers had to be at least 18 years old, be registered in a complementary insurance program, accept participation conditions and privacy terms, and declare to be free of any medical condition that does not permit increased physical activity.
Before invitation, potential participants were clustered based on their state of residence and clusters were then randomly allocated to one out of three incentive conditions: In the financial incentive condition, participants received CHF10 (US $10) for each month they walked >10,000 steps a day on average. To prevent frustration, participants received CHF5 when their monthly step count average was below 10,000 but over 7500 steps, which matches the approximate minimum recommendation for daily physical activity [
In line with recommendations for health promotion, participants were advised to perform at least 150 min of moderate-intensity activity a week, which on average translates to a goal of 10,000 steps a day [
Exemplary feedback email (authors’ translation). A: Feedback message with positive or negative feedback depending on the performance of the participant. B: Season-based tip on how to increase physical activity (here: recommendation to participate in a geocaching activity).
Starting after the first month, every participant received a feedback message by email at the beginning of the month that contained information on goal achievement of the last month. Consequently, every participant received 5 feedbacks over the course of the intervention. If participants failed to reach an average step count of at least 7500 steps a day, a negative feedback was provided (eg, “Unfortunately you did not reach the goal of 7500 steps a day on average last month”). In all other cases the feedback was positive (eg, “Well done, you have achieved at least 7500 steps a day on average over the last month and did a lot for your health”). In the financial and charitable incentive condition, feedback emails also contained information about the amount of money earned in the past month. Moreover, and in line with theory [
Data from the baseline questionnaire was used to describe the sample of this study. We calculated means and standard deviations (SDs) for continuous variables and absolute and relative frequencies for categorical variables. Data on age and gender of the participants was provided by the insurance company. Statistics on monthly average step counts were obtained by calculating the mean of all participants’ mean step counts for each month.
Because feedback on participants’ physical activity referred to monthly average step counts, we used the change in monthly step averages as the outcome variable to compare the effect of positive and negative feedback. Specifically, we calculated the difference between monthly step count averages before and after dispatch of the feedback email. Since each participant received 5 feedback emails, we consequently obtained 5 difference measures per participant. This difference indicates whether a participant increased or decreased his or her average monthly step count in the month after receiving feedback. Differences with an absolute value of more than 10,000 steps are likely to be the result of irregular recorded step counts (eg, very few and very low recorded step counts in 1 month) and were regarded as outliers and excluded from analysis. Exclusion of outliers resulted in the removal of 26/8115 (0.03%) observations and did not affect the results of the analyses.
To determine what analyses should best be used to examine the effect of the feedback sign, a 2-level hierarchical linear model with measurements as level-1 unit of analysis and participants as level-2 unit of analysis was fit to the data. However, the comparison of an intercept-only and a random-intercept model revealed that modeling the nested data structure did not significantly improve the model fit (χ21=0.0,
In total, 1410 directly invited customers and 213 family members participated in the program resulting in a sample of N=1623 participants. On average, participants were 42.66 (SD=13.06) years old and slightly more men (848/1623, 52.25%) than women (770/1623, 47.44%) participated in the prevention program. Five participants (0.31%, 5/1623) did not disclose their gender. Compared with the Swiss population [
Over the course of the intervention, participants walked on average 10,410 steps a day (
Descriptively, a clear pattern emerged from the data as it is apparent from
Participant characteristics.
Characteristics | T1 questionnaire, n (%) |
|
University | 548 (44.92) | |
Professional school | 208 (17.05) | |
High school | 389 (31.89) | |
Secondary school | 23 (1.89) | |
Primary school | 4 (0.33) | |
Not declared | 48 (3.93) | |
Town | 142 (11.64) | |
Outskirts of town | 300 (24.59) | |
Village | 598 (49.02) | |
Countryside | 180 (14.75) | |
<2500 | 62 (5.08) | |
2501-5000 | 184 (15.08) | |
5001-7500 | 383 (31.39) | |
7501-10,000 | 210 (17.21) | |
>10,000 | 127 (10.41) | |
Not declared | 254 (20.82) | |
Swiss | 1098 (90.00) | |
German | 55 (4.51) | |
Other | 53 (4.34) | |
Not declared | 14 (1.15) | |
Fitbit | 782 (64.10) | |
Fitbit app | 249 (20.41) | |
Garmin | 130 (10.66) | |
Jawbone | 59 (4.84) | |
Yes | 673 (55.16) | |
No | 511 (41.89) | |
Not declared | 36 (2.95) |
Descriptive statistics of monthly average step counts by feedback sign.
Monthly step count average (SD) | Dropout, n (%) | ||||
Positive feedbacka | Negative feedbacka | Totala | |||
Month 1 | - | - | 10967.02 (3744.64) | 53 (03.27) | |
Month 2 | 11581.79 (3273.48) | 6293.99 (2668.99) | 10710.19 (3732.68) | 68 (04.19) | |
Month 3 | 11639.40 (3145.38) | 6470.87 (2298.96) | 10714.99 (3597.49) | 65 (04.00) | |
Month 4 | 11533.57 (3326.82) | 6450.72 (2269.40) | 10657.20 (3717.10) | 53 (03.27) | |
Month 5 | 11216.21 (3395.22) | 6046.97 (1945.79) | 10366.47 (3742.52) | 91 (05.61) | |
Month 6 | 11308.43 (4283.81) | 5968.97 (2144.36) | 10299.51 (4479.57) | 49 (03.02) | |
Total | 11462.60 (3489.37) | 6252.79 (2291.78) | 10409.96 (3427.29) | 379 (23.35) |
aValues represent monthly average step counts depending on the feedback at the beginning of the month.
Descriptive statistics of change in monthly step count averages by feedback sign.
Mean difference in step counts (SD) | |||
Positive feedbacka | Negative feedbacka | Totala | |
Mail 1 | −482.71 (2035.95) | 363.30 (2114.51) | −349.37 (2070.88) |
Mail 2 | −328.77 (1858.73) | 717.76 (2153.88) | −148.82 (1952.29) |
Mail 3 | −399.11 (1869.87) | 587.71 (2012.82) | −234.52 (1929.12) |
Mail 4 | −497.77 (1854.33) | 215.26 (1654.28) | −382.55 (1842.09) |
Mail 5 | −422.96 (1607.90) | 329.19 (2112.57) | −285.12 (1735.20) |
Total | −425.34 (1858.38) | 450.29 (2032.06) | −279.51 (1916.24) |
aValues represent the mean change in monthly average step counts after dispatch of the feedback mail.
Summary of multiple regression results predicting change in average step counts.
Model parameter | Standard error ( |
Beta | ||
Intercept | −139.97 | 98.39 | - | .16 |
Time | −24.98 | 15.84 | −0.018 | .12 |
Baseline physical activitya | −0.13 | 0.01 | −0.239 | <.001 |
Group: financial incentivesb | 3.71 | 80.30 | .002 | .96 |
Group: charitable incentivesb | 5.42 | 81.85 | .001 | .95 |
Feedback signc | −84.28 | 78.74 | −0.062 | .28 |
aBaseline physical activity was centered before entering the model.
bGroup membership was represented as 2 dummy variables with the control group serving as the reference group.
cFeedback sign was represented as 1 dummy variable with negative feedback serving as the reference group.
Results of the multiple regression analysis of change in monthly step counts on time, baseline physical activity, group, and feedback sign are presented in
Left: Scatterplot of changes in monthly step counts against baseline physical activity for the first feedback email; the dark solid line represents perfect agreement (no change) and the colored lines are regression lines for positive and negative feedback. Right: Overall difference between negative and positive feedback emails adjusted for baseline physical activity and other covariates.
In this paper we analyzed the effect of positive and negative feedback emails on physical activity of participants of a large-scale physical activity intervention. Using a quasi-experimental approach, we found no difference between the effect of positive and negative feedback emails. Substantial differences found on a descriptive level could be explained by regression to the mean. Contrary to the theory outlined in this paper, our results might suggest that the feedback sign does not influence the effect of feedback. Similar results have been found by previous research in other behavioral domains [
Both frequency and relevance of the feedback could have limited the general effect of feedback on participants’ physical activity levels in our study, thereby, explaining the missing effect of the feedback sign. Feedback in our study was only provided once at the beginning of each month, which might have not been frequent enough to affect monthly physical activity levels. Indeed, meta-analyses of feedback interventions in the area of health care demonstrate that feedback is more effective if it is delivered more frequently, for example, weekly [
Some methodological issues arise as part of the practical setting of our study. Participants were not randomly allocated to a negative feedback and a positive feedback condition. Rather, positive and negative feedback was dependent on participants’ physical activity. As a consequence, we must not infer causality as inherent group differences beyond the included control variables may have affected our results. Furthermore, the internal validity of our results is limited as we were not able to check whether participants actually read the feedback emails. If a substantial proportion of participants ignored the feedback, we might be able to observe an effect of the feedback sign only within a subgroup of our sample that read the feedback emails. Finally, we were not able to include a true control group that did not receive any feedback emails. However, providing evidence for the general effectiveness of feedback was not the primary focus of this paper as this has been investigated and confirmed in other studies [
Research regarding the effect of feedback in health behavior change interventions is in its infancy. Thus, we urgently call for the conduction of randomized controlled trials examining the effects of feedback on health behavior as well as related mediators and moderators. Results of those studies can help researchers and practitioners to decide how to best incorporate feedback in their health behavior interventions and thereby ensure a positive effect of feedback. In this context, digital technology might be a promising resource to maximize the effect of feedback [
There is no difference between the effect of positive and negative feedback emails that were sent out on a monthly basis in a large-scale physical activity intervention. Framing of the feedback in terms of success and failure may not be crucial when the feedback is given infrequently and in situations when individuals are likely to be aware of their levels of behavior. However, feedback characteristics, including the feedback sign, should be carefully considered when designing feedback to change health behaviors.
control theory
Just-in-Time adaptive Intervention
Non-communicable Disease
social-cognitive theory
standard deviation
We would like to thank the institutional review board of the University of St. Gallen for their valuable feedback and support. The study protocol was approved by the Ethics Committee of the University of St. Gallen, Switzerland (reference number: HSG-EC-2015-04-22-A; date of approval February, 17th, 2016). Informed consent to participate was obtained from all participants of the study.
The study was funded in part by the CSS insurance, Switzerland. The funder had no role in reviewing or approving the manuscript for publication.
None declared.