Obesity is a global health problem. Individuals with obesity suffer greater mortality compared with individuals without obesity. This is a socioeconomic problem and needs better management strategies both at the individual and governmental levels. Previous research has investigated several intervention strategies including exercise and fiscal interventions. Though these interventions have been shown some degree of success in mitigating obesity, it is important to note the financial implications, extensive governmental planning and time- consuming implementation strategies associated with these intervention processes.

Behavioral and self-imposed health management strategies to control obesity have increased over time. With the increase in the number of smartphone users in recent periods, smartphone apps have become a health management tool for many users and researchers [1].

Later, smartphone apps have shown limited effectiveness in promoting weight reduction strategies and health management. Though we observe some results related to the role of smartphone apps and weight management, all the studies so far used limited data from small-scale studies, with a short period of study. This gives us a large discrepancy in the literature [2], concluded a systematic relationship between smartphone app users and weight loss. On the other hand [3], found no difference in weight between app users and non-app users. The objective of this study is twofold. We investigate the effect of text-based coach messaging on app users’ health and weight change. We estimate the change in weight at the county level to analyze the effect of the Shelter-In-Place (SIP) orders and the behavior change between pre-and post-COVID periods. We further look at the welfare effect of weight management under a pandemic.

To empirically examine this causal relationship in the first place, we leverage the staggered adoption of the SIP orders across counties and states throughout the COVID-19 crisis. We then employ a generalized Difference- In-Differences (DID) design and use rich consumer-level data to measure personal health behaviors and outcomes. Overall, we find that introducing the SIP order leads to a significant decline in the difference between one’s weekly calorie budget and the actual calories consumed per week (76.729 units), an increase in the weekly number of calories burned by physical exercise by 39.845 units, and a one percentage point reduction in the probability of being obese. Meanwhile, we also provide evidence that individuals with active text-based coach messaging are more likely to achieve their health investment goals. Specifically, more digitally active people increase their inputs in managing calorie consumption by setting a more restrictive calorie budget than those receiving less text- based coach messaging.

To assess the causal interpretation and robustness of identification, we conduct several exercises including implementing an event-study specification, employing alternative robust difference-in-differences estimators, and addressing the concern of selection bias. These results suggest that the SIP orders triggered by the COVID-19 crisis may raise people’s expected costs of obesity and overweight and thereby incentivize individuals to take actions to reduce the obesity risk. To probe the validity of this hypothesis, we collect Google Trends data and leverage the search queries for topics related to weight management as a proxy for such expected costs. Along this line, we empirically provide some tentative evidence: Following the enactment of a SIP order, there is an increase in the daily search intensity for topics including anti- obesity medication, overweight, and weight management, reflecting rising public concerns about being obese and overweight.

Meanwhile, our estimated results in weight management and obesity could shed new light on social welfare of the SIP orders. To gauge the social benefits in terms of health, we benchmark the magnitude of reductions in obesity prevalence: the estimated state-level decline associated with a SIP order is equivalent to a decrease of one percentage point in the national obesity prevalence. In turn, such reductions suggest that increased weight management/reducing obesity during the COVID-19 pandemic saved the U.S. health care system $1.47 billion (in 2008 dollars) a year.

Our paper contributes to an emerging strand of literature investigating the impact of the COVID-19 pandemic (as well as relevant policies) on personal health behaviors and health outcomes. Most studies, to the best of our knowledge, have focused on mental health and the level of well-being [4-9]. Although some literature has provided suggestive evidence concerning the association between COVID-19 and lifestyle health behaviors, there is relatively sparse research establishing the casual relationship. The closest economic literature to ours is [10], in which the authors find that school closures, restaurant restrictions, and stay-at-home orders lead to a significant increase in searches for workout, physical activity, and exercise, but a decline in concerns for weight loss, diet, nutrition, etc. In contrast to this paper purely focusing on search behaviors regarding weight management, we take one step forward and directly explore the impacts on physical exercise, dietary intake adherence, and the probability of being obese. In other words, our paper provides the very first study investigating the effect of COVID-19 related policies on personal weight management and the prevalence of obesity.

This paper also adds value to the public health literature exploring the role of digital interventions in health behaviors and outcomes, especially personal weight management see for example [1,11-24]. Compared with prior research utilizing relatively small-scale Randomized Controlled Trails (RCTs), our paper leverages a large-scale, unexpected public health shock, the COVID-19 pandemic, as a quasi-experiment to provide plausibly causal evidence. Furthermore, we advance this strand of research by shedding new light on the welfare implications associated with both the SIP orders and digital health interventions.

The rest of the paper proceeds as follows. Section 2 describes the key data sources we utilized in our empirical analysis. Section 3 provides the features and characteristics of the sample. Section 4 shows the baseline empirical strategies we rely on, and Section 5 presents the main results on personal management of calorie consumption, calorie expenditure via physical exercise, and the probability of being obese. Section 6 assesses the role of text-based coach messaging in weight management, and Section 7 probes the validity of our estimates by conducting several robustness checks. Section 8 explores one potential channel through which personal health behaviors are altered: by raising the expected costs of obesity, the SIP order incentivizes people to self-invest in weight management and reduce the risk of obesity. Section 9 examines the welfare implications of these SIP orders, and eventually, Section 10 concludes.

To conduct the empirical analysis, we collect data on (1) the timing of adopting the shelter-in-place orders at the county level, (2) measures of personal health behaviors and outcomes related to weight management, (3) measures of digital social support using text-based coach messaging, and (4) measures of concerns for personal weight management. In this section, we describe the corresponding data sources and how we construct the key variables.

Adoption of Shelter-in-Place Orders

This study aims to capture the differential effect of the “shelter-in-place” (SIP) orders related to COVID-19 on the health management behavior of the Noom Weight users. Shelter-in-place orders require residents to stay at home, except for essential work, essential shopping, and other permitted duties. We obtained the shelter-in-place orders from the New York Times 2020 data page titled: See Which States and Cities Have Told Residents to stay at Home. The shelter-in-place orders consist of both state and local government-imposed COVID-19 related social distancing policies executed, between February 1, 2020, and April 5,2020.

In addition, we collect data on state and local emergency lockdown orders, the orders lifting original SIP, and the closure and reopening orders for gyms and restaurants. We also obtain county-level monthly unemployment rates from the Bureau of Labor Statistics’ (BLS) Local Area Unemployment Statistics (LAUS) program. To merge all data, we first use the crosswalk files from the U.S. Department of Housing and Urban Development (HUD) to map each zip code to the corresponding county in the Noom sample. We then merge all control variables to the Noom data using the county FIPS code.

Measuring Health Behaviors and Health Outcomes Related to Weight Management

Throughout the analysis, we rely on detailed consumer- level data provided by Noom, Inc. to measure individuals’ health behaviors and health outcomes. Noom Weight saves daily activity, exercise (frequency and calories burned), food intake, weight trends (initial weight and weight change), calorie intake by meals (breakfast, lunch, and dinner), and nutritional summaries of the app users. The app then records how users adhered to the recommended calorie budget and ratios of low to high calorie density foods (eg dietary intake adherence). Therefore, the data covers self-monitored observance data, including body weight, target body weight, daily food intake, steps, and an activity check built into the app.

The database provides information on instructional text-based coach messaging (coach messages received) and internal messages exchanged between app users. Our data is weekly for app users between January 2019 and April 2021, including the pre and post COVID-19 periods mentioned above. The data also provides information on the gender, age, and height (BMI) of each user. We have 287,447 unique Noom Weight users in the database. The benefit of the Noom Weight data over other similar databases is the nature of the detailed information available per individual app users. Compared to the initial and targeted weights, weekly weight change allows us to evaluate the most effective strategies.

Our first measure of personal health behaviors is dietary intake adherence, which is the difference between the weekly calorie budget and the actual calories consumed per week (i.e., calorie budget minus calories consumed through food intake). Specifically, we utilize the rich information on individuals’ daily consumption of calories via breakfasts, lunches, dinners, and snacks and derive the weekly calorie consumption by aggregating the amounts of calories for each consumer. The target calorie consumption, which is proxied by calorie budget, is set and updated by the Noom Weight based on personal information and recorded data. We also aggregated these amounts of calories at the individual-week level to mirror the construction of the actual caloric intake. Along this line, we argue this measure records how users adhered to the recommended calorie budget and ratios of low to high calorie density. Put it differently, the more the value of dietary intake adherence approaches zero, the better one performs in following the recommendation.

Besides calorie consumption, regular physical activity and exercise are especially important as well if one is trying to lose weight or to maintain a healthy weight. To examine the effects of the SIP orders on energy expenditure, we employ a second measure as the actual number of calories burned via physical exercise per week. All variables are assembled from the Noom sample.

Finally, we also utilize the probability of being obese, the probability of being overweight, and weekly BMI to measure the key health outcome: weight loss. Following the definition provided by the World Health Organization (WHO), we define the probability of being obese as one if one’s weekly BMI is at least 30 and zero otherwise. The probability of being overweight is constructed as one if the weekly BMI is between 25 and 30 and zero if the weekly BMI is below 25. Again, the variables are all constructed with the Noom sample on a weekly basis.

Measuring Text-based Coach Messaging

To investigate one channel associated with weight management, text-based coach messaging, we rely on consumer information provided by Noom. The data provides the number of coach messages received by users via the Noom Weight and we aggregate this information at the week level for each Noom user. We consider messages exchanged between users and the coaches. To avoid the potential that the message is sent automatically by the system, we define an individual as “social active” if the number of messages received by users per week is at least two and zero otherwise. In the empirical analysis, we first test whether one’s status of getting text-based coach messaging changes following the adoption of SIP orders, and then directly explore the differential effects of the SIP order by the status of text-based coach messaging.

Measuring Public Concerns for Weight Management

In order to explore potential mechanisms driving the observed reduction in weight, we also collected Google Trends data to construct measures for public concerns related to weight management. Google Trends data provides an index for search intensity by topics over the sample period in each area. Such an index is the raw number of daily searches for a target topic divided by the maximum number of daily searches for this topic over the sample period. And the index is scaled from 0 (there is not sufficient information regarding the search for a specific term/topic) to 100 (this is the day with the maximum volume of searches over a specific period). In particular, we select several topics related to weight management: overweight, obesity, anti-obesity medication and weight loss. We choose to submit the topic queries, including all related search terms in any language, which are better proxies for public concerns.

One limitation of the Google Trends data is that daily data on search intensity is only offered for a query period shorter than 9 months [6]. Therefore, the scaling factors used to calculate the search intensity over different periods are not identical to each other. To obtain comparable daily data from January 1st, 2019, to June 30th, 2020, we followed the rescaling approach described in [6]: assembling the raw daily search intensity data via submitting two queries (1/1/2019-6/30/2019 and 1/1/2020-6/30/2020), collecting the raw weekly search intensity data and calculate the weekly search interest weights, and finally rescaling the daily data for each period with the weights.

In this section, we first provide the summary statistics of the data described in Section 2, and then compare the patterns of these variables between the treatment and control groups. To begin with, Table 1 presents the  summary statistics of the full sample,

Table 1: Summary Statistic (full sample).

Notes: All data are collected from the Noom sample (from 1/1/2019 to 4/30/2021). See text for details of how variables are constructed.

which includes 8,852,308 observations from 2878 counties from January 1st, 2019, to April 30th, 2021. Throughout the empirical analysis, we measure personal health choices and weight management from three dimensions: dietary intake adherence, calorie expenditure via physical exercise, and the probability of being obese and overweight. Specifically, the difference between the weekly calorie budget and the actual calories consumed per week is approximately 1631.056 units; on average, the weekly calories used through physical exercise is 1,164.539. The portion of users who have received at least two messages from coaches via Noom Weight per week is 34% and the average number of messages received from personal coaches is approximately 1.3 per week. In terms of the sample characteristics, the average age is 50.2 and the portion of male consumers is 19%. The BMI (weekly) among all consumers, on average, is 31.562, and both the prevalence of obesity and the prevalence of overweight are relatively high (54.2% and 70.9%, respectively).

To provide some motivating evidence that the SIP orders related to the COVID-19 pandemic indeed affect personal health behaviors in weight management, we compare the summary statistics of key variables for the treatment and control groups in Table 2.

Table 2: Summary Statistics (by the status of treatment).

Notes: All data are collected from the Noom sample (from 1/1/2019 to 4/30/2021). See text for details of how variables are constructed.

On average, the difference between calories budget and calories consumed, and the number of calories burned via physical exercise are both slightly higher after introducing an SIP order. Meanwhile, one can tell that compared with consumers in the control group (i.e., counties without the adoption of a SIP order), those in the treatment group (i.e., counties with a SIP order in place) are less active digitally: they tend to communicate less with coaches by receiving fewer messages and being less likely to interact with their coaches. Furthermore, when it comes to the features of personal weight, one can observe a decline in all three dimensions following a SIP order: the weekly BMI, the probability of being obese, and the probability of being overweight.

The main goal of this paper is to empirically test whether and how the SIP orders induced by COVID-19 would affect people’s health behaviors related to weight management and the corresponding health outcome (i.e., weight loss). To this end, we leverage the staggered adoption of SIP orders across counties and employ a generalized Difference-In-Differences (DID) specification as follows:

The main goal of this paper is to empirically test whether and how the SIP orders induced by COVID-19 would affect people’s health behaviors related to weight management and the corresponding health outcome (i.e., weight loss). To this end, we leverage the staggered adoption of SIP orders across counties and employ a generalized Difference-In-Differences (DID) specification as follows:

(1)

where i denotes individual, g represents the county and w and y refers to week (of a year) and year, respectively.

management due to the introduction of SIP orders.

In addition, we include individual- and county-level controls in some specifications. Xigwy refers to a vector of individual-level characteristics, including age, gender, and the number of weeks since one’s sign-up of the Noom Weight. And τgy is the unemployment rate for county g in January of each year. Standard errors are clustered at the county level in all regressions to allow for heteroskedasticity and correlation of the error terms within a county.

Alternatively, we consider the potential heterogeneity across counties during the COVID-19 pandemic and propose another specification. In this model, we interact the week-fixed effect and year-effect with a full set of county dummies, respectively. These are the weekg and yearg variables included in equation 2 below, respectively. These account for any weekly or annually changing county-level heterogeneities, including changing patterns in weather, political attitudes, and other unobservable variances such as demographics and the county populations health-related differences, including health infrastructural differences varying within the counties over the time frame of our study. These variables also control the migration of individuals from one county to another Yigwy is the key outcome of interest (i.e., calorie consumption or the number of people traveling in and out of the county, carrying information and potentially spreading the virus. management, calories burned via physical exercise, and whether an individual is overweight or obese). The key explanatory variable, SIPgwy, is the indicator equaling to one if there is a “shelter-in-place” order in effect in county g at time wy. ρg represents the county fixed effect, which captures the within-period variation across counties; week and year are the week-of-year and year fixed effects, respectively. We include these two dimensions of time fixed effects to capture the within-county variations over time. Therefore, the main parameter of interest, β represents the change in our outcomes related to weight

Therefore, this specification controls for heterogeneities across counties’ economic, political, health demographics and health services infrastructure, social conditions, etc. in a more comprehensive way.

 (2)

where weekg and yearg are the corresponding fixed effects mentioned above. All others remain the same as those listed in equation (1). Standard errors are still clustered at the county level.  We also estimate the additional effects of ‘coach-interactions’ in Noom Weight. We call this effect “text-based coach messaging” as this involves exchanges of messages using the internet (like text messages) related to health behavior. We include those users who are present both in the pre and post pandemic periods.

The baseline results for the effects of SIP orders on personal health behaviors are presented in Section 5.1 (calories consumption management) and Section 5.2 (calories burned via physical exercises). In Section 5.3, we further assess its impact on personal health outcomes, which is proxied by the probability of being obese.

Effects of Shelter-in-Place Orders on Dietary Intake Adherence

To begin, we examine whether the SIP order affects the intake of calories through food consumption. Table 3 shows the estimates for dietary

Table 3: Effects of SIP Orders on Dietary Intake Adherence.

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Notes: This table presents the results from a regression on dietary intake adherence based on Equation (1) and Equation (2). To measure one’s dietary intake adherence, these models use the weekly calorie budget minus the actual calories consumed (via breakfasts, lunches, dinners, and snacks) as a proxy. The first two models rely on the specification listed in equation (1) and the next two models estimate the regression specification listed in equation (2). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

intake adherence based on our main specifications (equation (1) and (2)). We start with estimating a version of equation (1) that excludes all individual-level and county-level controls. As reported in column (1) of Table 3, we find that following the introduction of a SIP order, the difference between one’s weekly calorie budget and the actual calories consumed per week reduces by 110.005 units. After including both individual- and county-level controls described in Section 4, such an average effect slightly increases in magnitude (111.043 units, shown in column (2)) but remains statistically significant. Allowing the coefficients on week_gt and ?year?_gt to vary by county, one can observe that a SIP order still significantly reduces the difference between calorie budget and calorie consumption. That is, after the enactment of the SIP orders, individuals who are in the treatment group, exceeded their calorie consumption past their budgeted calories. Such estimates (-113.855 in column (3) and -107.976 in column (4)) are not substantially different from the baseline point estimate in column (2), thereby alleviating the concern that omitted variables are likely to play a meaningful role in the estimation. This set of our main results emphasized the direct effect of SIP orders on health behaviors that are directly related to personal weight management. In the following analysis, we will present some evidence that SIP orders also affect other key margins of health behaviors substantially and, therefore, have the potential to change the key outcome of interest --- the prevalence of obesity.

Effects of Shelter-in-Place Orders on Calories Burned via Physical Exercise

Turning to another key margin related to weight management, the calories used through physical activities, we present the corresponding results in Table 4. As shown in column (1) of Table 4,

Table 4: Effects of SIP Orders on Calories Burned via Physical Exercise.

 Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Notes: This table presents the results from regression on calories burned via physical exercise based on equation (1) and equation (2). The first two models rely on the specification listed in equation (1) and the next two models estimate the regression specification listed in equation (2). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

the adoption of SIP orders leads to an increase of 91.339 calories burned by taking physical exercise. Even after including a full set of individual and county controls, the result is still statistically significant and almost identical in terms of the magnitude (101.264, shown in column (2). In our preferred, more flexible specifications, which take into account the unobserved geographic heterogeneity across counties (column (3) and column (4), one can tell that introducing a SIP order, on average, is associated with an increase in weekly calories burned via exercise by 114.293 or 105.686 units.

Collectively, these results in Sections 5.1 and 5.2 imply that following the SIP orders related to COVID-19, individuals who utilize the Noom Weight consume more calories per week but simultaneously perform more physical exercises. These two findings thus shed new light on a key margin of weight management: balancing one’s calorie consumption and expenditure. One potential mechanism is that the enactment of COVID-19 lockdowns might increase calorie consumption, but it also increases calories burned. COVID-19 lockdowns increase one’s expected costs associated with obesity (e.g., reduced life expectancy and increased healthcare expenditure) by delaying the delivery of medical treatment. As a result, a priori, individuals with a high risk of being obese or overweight are likely to make more efforts to manage their weight and maintain a healthier status under the developing COVID-19 threats. This set of results thus emphasizes the importance of incentivizing/encouraging people’s self-investment in health, especially in light of the shortage of formal medical resources during the public health crisis.

Effects of Shelter-in-Place Orders on the Prevalence of Obesity

Thus far, we have examined the impacts of SIP orders on personal health behaviors related to weight management. While both dietary intake adherence and the expenditure of calories by exercise are good measures of weight management, more relevant to welfare purposes is whether SIP orders lead to changes in actual health outcomes. These results are presented in Table 5,

Table 5: Effects of SIP Orders on the Incidence of Obesity.

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Notes: This table presents the results from regression on the incidence of obesity based on equation (1) and equation (2). The first two models rely on the specification listed in equation (1) and the next two models estimate the regression specification listed in equation (2). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

which provides the baseline estimates for the impact of the SIP orders on the probability of being obese. After including all control variables, the introduction of the SIP orders results in a statistically significant decrease in the probability of getting obese by one percentage point (column (2)). According to the results shown in column (3) (-0.013) and column (4) (-0.012), other variations at the county level are not likely to explain such an effect.

Taking together, we have assessed the plausibly causal effect of these COVID-19-induced SIP orders on the key outcome of interest: the prevalence of obesity. The results derived above are not only crucial to weigh the costs and benefits of such public policies during a public health crisis, but also enable us to further explore the potential mechanisms driving such causal effects.

The above results described in Section 5 reveal that SIP orders have improved one’s performance in weight management. This section outlines and examines one channel that could create substantially differential effects: the app users who subscribe to digital guidance are more motivated and educated in managing weight and maintaining personal health scientifically. To empirically test it, we leverage the information regarding consumers’ status of digital social interaction with personal coaches and check if the imposition of SIP orders has substantially influenced the probability of exchanging texts with coaches.

Did the SIP Order Change the Status of Receiving Text Based Coach Messaging?

If the adoption of SIP orders raises individuals’ expected costs of obesity and encourages people to make more effort in weight management, then one might expect an increase in the frequency of exchanging messages and seeking guidance from coaches after the lockdowns. We test for such compositional changes in Table 6 and find some evidence to support

Table 6: Effects of SIP Orders on Text-based Coach Messaging.

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Notes: This table presents the results from a regression on the probability of exchanging texts with coaches based on equation (1) and equation (2). The dependent variable is the probability of receiving at least two messages per consumer. The first two models rely on the specification listed in equation (1) and the next two models estimate the regression specification listed in equation

(2). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

such a shift in the composition of people engaged in getting support from coaches via Noom. These point estimates are positive and significant across all specifications. It confirms our hypothesis holds in this case: exchanging texts via the app, as a kind of digital support, can be a key channel driving changes documented in personal weight management.

In this section, we implement several tests to assess the causal interpretation of our estimates and probe the robustness of these results.

Does the SIP Order Affect the Probability of Signing up?

Another potential threat to identification is that the COVID-19 pandemic and related policies could substantially change the composition of Noom users and thereby introduce selection bias on the margin of entry. Put it differently, people with higher risks of obesity and overweight are more likely to sign up for the Noom Weight. To alleviate this concern, we focus on the sample of newly signed users and implement the following regression analyses:

where log (#of newly signed usersst) if the logarithm of the number of consumers newly signed up the Noom Weight in state s at time t. SIPst is an indicator whether there is a SIP order in place in state s at time t. tst is the  state-level controls, including the enactment of a state emergency order and the introduction of lifting a SIP order. We also incorporate the state fixed effect to control time-invariant heterogeneities across states. And the key difference between equation (3) and (4) is that the latter one allows us to within-state variants with a full set of week x year fixed effects in a flexible way.

Table 7 provides the results:

Table 7: Effects of SIP Orders on Signing-up.

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Notes: The key independent variable Table 8 is log(# of newly signed consumers). Column (1) presents results based on the specification with year fixed effect and week fixed effect. Results based on a specification with a full set of week×year fixed effects are shown in column (2). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

without controlling for other state-level, COVID-19 relevant policies and more flexible time fixed effects, the number of newly signed consumers has increased by 11.44% following the SIP order at the 10% significance level. However, in column (2), we present the evidence that in our preferred, more conservative specification (i.e., eq (4)), the impact of an SIP order on the prevalence of signing up for the Noom Weight is statistically insignificant. Thus, we prefer to claim that there is little supportive evidence for a compositional effect on the entry margin.

Does the SIP Order Affect Recommended Calorie Budget?

Additionally, we perform another exercise to test whether introducing the SIP order would affect the target calorie budget recommended by Noom Weight. Table 8 presents the result:

Table 8: Differential Effects of SIP Orders on Calorie Budget by Text-based Coach Messaging.

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

Overall, we find little evidence that the SIP order per se would meaningfully change this index (column (1), (2), and (3)). Yet, when separately exploring the effects by users’ status of receiving text-based coach messaging, it suggests that compared with text-based coach messaging users who experienced a 1.2% decline in this target budget following the SIP order (column (4) – (6)), the population communicating less frequently with coaches are more likely to have a higher level of calorie budget per week (an increase of approximately 1.3%, column (7) – (9)). According to Noom Weight, the calorie budget remains unchanged from week to week, but it can change when users request the coach to lower their calorie budget. Our results thus suggest that text-based coach messaging users are requesting lower budget following the SIP order and thereby imply compared with their less- active counterparts receiving digital social support, users frequently communicating with coaches are setting more ambitious goals in managing personal weight.

In terms of relevant policy implications, our results in this exercise highlight the potential role of digital platforms in altering individual health behaviors. Specifically, encouraging the adoption of the digital social support system during the pandemic might leave room for additional positive health externalities. In the meantime, expanding this digital social networking feature in relevant products can improve the quality and thereby add value to consumer wellbeing, especially to personal health outcomes.

Event-Study Specification

One key assumption for our difference-in-differences design is that treatment and control groups would follow common trends in the absence of a SIP change in the treated states. In the context of the COVID-19 pandemic, the validity of this assumption might be violated if changes in COVID-19 cases led to both the implementation of various policies and concerns of infection that lead to care avoidance that would have occurred even in the absence of any formal policy changes.

To formally address this concern, we employ an event study specification to test for pre- and post-treatment trends. Allowing for a 13-weeks pre-treatment window and a 10-weeks post-treatment window, we implement the analysis with the following regression model:

In equation (5), a series of coefficients βk would capture the impacts of the SIP orders in the weeks before and after the formal implementation. In Figure 1A – 1B,

Figure 1 A: Event Study (Dietary Intake Adherence).

Figure 1 B: Event Study (Calorie Expenditure via Physical Exercise).

we plot the coefficients on the leads and lags of adopting a SIP order. We normalize the coefficient on SIP(gt-1) to be zero. Overall, there is little evidence suggesting that prior to introducing the SIP orders, the trends in treatment and control groups are substantially different. Meanwhile, our results based on those lag coefficients indicate the impacts of such SIP orders can persist for at least 10 weeks.

Alternative Choices of Specification

In the appendix, we also probe the robustness of our results by checking if the baseline estimates are sensitive to changes along the following dimensions: the choice of the reference date, including other relevant state policies, and addressing the role of county urbanicity. We report these results in Table A1 – A5. Our main estimates hold up against alternative specifications and thereby suggest that the impacts on personal health behaviors and outcomes driven by COVID-19 SIP orders are plausibly causal and convincing.

The above results reveal that the SIP order related to the COVID-19 pandemic substantially affected individuals’ health behaviors and outcomes in weight management. In Section 5, we have discussed one mechanism that can drive such effects: by delaying the medical service, the COVID-19 pandemic can raise individuals’ expected costs of being obese and overweight. Consequently, people with high risks of being obese are more likely to self-invest in weight management in advance, leading to economically significant improvements in health outcomes.

In this section, we explicitly test whether this mechanism holds. Formally, we consider a measure of public concerns for weight management: search queries for weight-related terms and topics via Google. As described in Section 2.4, we collect Google Trends data and construct the number of Google daily searches for the following topics: obesity, overweight, anti-obesity medication, and weight loss.

where yst is the aggregate number of daily searches for topic y in state s at time t. SIPst is an indicator if a SIP order is in place in state s at time t. τst is a set of state-level controls, including the enactment of state emergency declaration and the introduction of lifting a SIP order. We include state fixed effect, day-of-week (i.e., Monday to Sunday) fixed effect, week-of-year fixed effect and year fixed effect.

In Table 9,

Table 9: Effects of SIP Orders on Search Queries.

Notes: This table presents the results from linear regressions on the search queries for anti-obesity medication, overweight, and weight loss based on equation (6). Standard errors, adjusted for clustering at the county level, are reported in parentheses.

our results indicate that enacting the SIP order is associated with a rise in search intensity for anti-obesity medication, overweight, and weight loss. Such estimates are both statistically and economically significant, reflecting a shift in public concerns/interests for these health topics during the pandemic. We thus assert that this exercise provides some suggestive evidence that the SIP orders triggered by the COVID-19 pandemic are altering people’s health behaviors through the psychological channel.

In the meantime, the implications of our findings can shed some new light on policymaking as well. For example, the burden of a public health crisis might be far-reaching and go beyond one single dimension. In the case of the COVID-19 pandemic, due to the crowd-out of medical resources and healthcare capacity, people with other chronic diseases (e.g., obesity) would suffer from unexpected negative externalities both physically and psychologically. As a result, it emphasizes the necessity to take into account these extra costs when designing relevant policies aiming to mitigate the adverse effects during these pandemics.

Obesity, as a notorious public health epidemic, has long been receiving intensive concerns. Public debates regarding this issue have centered on how to provide efficient interventions in medical treatment, legal framework, and social safety net programs. In this paper, we have provided empirical evidence that one target policy related to the COVID-19 crisis can create unexpected, substantial spillover effects on the prevention of obesity. By reducing the obesity risk among the population, it thus leaves the potential to raise social welfare.

In the empirical portion of this paper, we have documented that adopting a SIP order leads to substantial improvements in preventing obesity: on average, one adoption would result in a one percentage point decline in the probability of getting obese. Though the COVID-19 SIP orders were a curse for all societies and economies alike but analyzing this measurable positive impact of these SIP orders on welfare-improvement might suggest further policy for controlling obesity. To put this magnitude in perspective, one helpful benchmark is to rescale the result at the national level. According to the Behavioral Risk Factor Surveillance System (BRFSS), the U.S. obesity prevalence was 27.8% in 2011 and 30.9% in 2018, with the average as 29.54%. Among the states with a SIP order during the COVID-19 pandemic, the average prevalence of obesity was 29.4% from 2011 through 2018. Our estimate (a decrease of 1 percentage point shown in Table 5) is thus equivalent to a decline of 3.40% (= 1/29.423*100%) throughout the sample period. After applying this percentage response to the national level, adopting the SIP order also implies a decrease of 1 percentage point in the national obesity prevalence. Therefore, such a decline induced by the SIP orders is comparable to preventing the aggregate growth of obesity prevalence by 2.26 years. Meanwhile, as pointed out by the Center of Disease Control and Prevention (CDC), the medical costs of obesity are $147 billion per year (in 2008 dollars). The estimated reduction in the U.S. aggregate obesity prevalence thus implies that increased weight management/ reducing obesity during the pandemic saved the U.S. health care system $1.47 billion (= $147 billion*1%, in 2008 dollars) a year.

After the pandemic, users active via ‘digital social support’ achieved more beneficial effects of Noom Weight vis-à-vis their health choices and outcomes. Social networking on the app helped users better achieve their calorie consumption goals, comparing to their pre pandemic goals. ‘Text-based coach messaging’ in the weight loss App also helped users exercise more comparing to their pre pandemic goals. ‘Text-based coach messaging’. ‘Coach message system’ users enjoyed significantly more calories burn and weight loss. Non-coach system users do not enjoy significant calorie burn or clear weight loss. Overall Noom has had a positive effect on users’ health management during the pandemic but not without the digital messaging feature of it.

Ethics approval and consent to participate: IRB approval was not necessary as the data received was completely identified. IRB approval waiver was verified from Auburn University IRB.

Consent for publication: Not applicable.

Availability of Data and Materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Authors’ Contributions

All authors contributed equally for the preparation of the manuscript. Dr. Banerjee and Dr. Nayak collected and prepared the data.

Acknowledgements

We thank Noom Inc. for providing the data.

1. Allen Jerilyn K, Janna Stephens, Cheryl R Dennison Himmelfarb, Kerry J Stewart, Sara Hauck.. Randomized Controlled Pilot Study Testing Use of Smartphone Technology for Obesity Treatment. J Obes. 2013; 2013: 151597.
2. Flores Mateo, Gemma, Esther GranadoFont, Carme FerréGrau, Xavier MontañaCarreras. Mobile Phone Apps to Promote Weight Loss and Increase Physical Activity: A Systematic Review and Meta-Analysis. J Med Int Res. 2015; 17.
3. Semper H M, R Povey, D ClarkCarter. A systematic review of the effectiveness
4. of smartphone applications that encourage dietary self-regulatory strategies for weight loss in overweight and obese adults. Obesity Reviews. 2016; 17: 895-906.
5. AdamsPrassl Abi, Teodora Boneva, Marta Golin, Christopher Rauh. Inequality in the Impact of the Coronavirus Shock: Evidence from Real Time Surveys. J Pub Eco. 2020; 189: 1042-45.
6. Armbruster, Stephanie, Valentin Klotzbücher. Lost in Lockdown? COVID-19, Social Distancing, and Mental Health in Germany. Discussion Paper No. 2020–04. University of Freiburg, Wilfried Guth Endowed Chair for Constitutional Political Economy and Competition Policy, Germany. 2020.
7. Brodeur Abel, Andrew E Clark, Sarah Fleche, Nattavudh Powdthavee.. COVID-19, Lockdowns and Well-being: Evidence from Google Trends. J of Pub Eco. 2021; 193: 1043-46.
8. Davillas Apostolos, Andrew M Jones. The First Wave of the COVID-19 Pandemic and Its Impact on Socioeconomic Inequality in Psychological Distress in the UK. Health Economics. 2021; 30: 1668- 1683.
9. Hamermesh DS. Life satisfaction, loneliness and togetherness, with an application to Covid-19 lock-downs. Review of Economics of the Household. 2022; 18: 983–1000.
10. Tubadji A, Boy F, Webber D. Narrative Economics, Public policy and mental health. Center for Economic Policy Research. 2023; 18: 43-70.
11. Hasanzadeh Samira, Modjgan Alishahi. Public Health Shock, Intervention Policies,
12. and Health Behaviors: Evidence from COVID-19. Review of Economic Analysis. 2022; 14: 71–88.
13. Banerjee Tannista, Veena Chattaraman, Hao Zou, Gopikrishna Deshpande. A Neurobehavioral Study on the Efficacy of Price Interventions in Promoting Healthy Food Choices among Low Socioeconomic Families. Scientific Reports. 2020; 10.
14. Beratarrechea Andrea, Allison G Lee, Jonathan M Willner, Eiman Jahangir, Agustín Ciapponi, Adolfo Rubinstein. The Impact of Mobile Health Interventions on Chronic Disease Outcomes in Developing Countries: A Systematic Review. Telemedicine and e-Health. 2014; 20: 75–82.
15. Michelle Clare Carter, Victoria Jane Burley, Camilla Nykjaer, Janet Elizabeth Cade. Adherence to A Smartphone Application for Weight Loss Compared to Website and Paper Diary: Pilot Randomized Controlled Trial. J of Med Int Res. 2013; 15.
16. Czernichow Sébastien, Adeline Renuy, Claire RivesLange, Claire Carette, Guillaume Airagnes, Emmanuel Wiernik. et al. Evolution of the Prevalence of Obesity in the Adult Population in France, 2013- 2016: the Constances Study. Scientific Reports. 2021; 11; 141-52
17. de Jongh Thyra, Ipek Gurol-Urganci, Vlasta Vodopivec-Jamsek,Josip Car, Rifat Atun. Mobile Phone Messaging for Facilitating Self-management of Long-term Illnesses. Cochrane Database of Systematic Reviews. 2012; 12.
18. Ferraro KF, Jessica AKM. Cumulative Disadvantage and Health: Long- Term Consequences of Obesity? American Sociological Review. 2003; 68: 707-729.
19. Grave Riccardo Dalle, Simona Calugi, Elena Centis, Marwan El Ghoch, Marchesini. Cognitive-Behavioral Strategies to Increase the Adherence to Exercise in the Management of Obesity. J of Obes. 2011; 2011: 1-11
20. Kairey L, Cook A, H P van der Ploeg, King L, Bauman A, AllmanFarinelli M. A Mobile Health Intervention for Weight Management among Young Adults: A Pilot Randomized Controlled Trial. J Hum Nutt Diet. 2013; 27: 322-332.
21. Kim Yu Jin, Sang Youl Rhee, Jong Kyu Byun, So Young Park, Soo Min Hong, Sang Ouk Chin et al. A Smartphone Application Significantly Improved Diabetes Self-Care Activities with High User Satisfaction. Diabetes Metabolism Journal. 2015; 39: 207–217.
22. Klasnja Predrag, Wanda Pratt. Healthcare in the Pocket: Mapping the Space of
23. Mobile-Phone Health Interventions. J of Bio Inf. 2011; 45: 184–98.
24. Laing Brian Yoshio, Carol M Mangione, ChiHong Tseng, Mei Leng, Ekaterina Vaisberg, Megha Mahida. et al. Effectiveness of a Smartphone Application for Weight Loss Compared with Usual Care in Overweight Primary Care Patients. Ann of Int Med. 2014; 161: S5– S12.
25. Spring Bonnie, Jennifer M Duncan, E Amy Janke, Andrea T Kozak, H Gene McFadden, Andrew DeMott. et al. Integrating Technology into Standard Weight Loss Treatment: A Randomized Controlled Trial. JAMA Int Med. 2013; 173: 105-111.
26. Xu Hanfei, L Adrienne Cupples, Andrew Stokes, Ching-Ti Liu. Association of Obesity with Mortality Over 24 Years of Weight History: Findings from the Framingham Heart Study. JAMA Network Open. 2018; 1.
27. Mitchell Ellen S, Qiuchen Yang, Heather Behr, Laura Deluca, Paul Schaffer. Adherence to Healthy Food Choices during the COVID-19 Pandemic in A U.S. Population Attempting to Lose Weight. Nutrition, Metabolism and Cardiovascular Diseases. 2021; 31: 2165 – 2172