Impact of Weight Status Reporting on Childhood Body Mass Index

Abstract

Background:

Many states have adopted school-based BMI screening or surveillance programs in an effort to address high rates of childhood obesity, some of which involve provision of confidential BMI reports to parents. While there is evidence that parents are attuned to information in the reports, there is less evidence showing that the reports are effective in preventing excess childhood weight gain.

Methods:

Data from Arkansas, the state with the nation's first and longest running and BMI screening program, were used to measure the impact of BMI reports. This was done through a regression discontinuity design that compared future BMI z-scores among children falling within a narrow band around the obese and overweight thresholds. We derived the effects of BMI reports by comparing students who received different types of reports around the relevant threshold.

Results:

While we are unable to detect any differences in BMI z-scores between the children who received the overweight report and the children who received the healthy weight report, we detected some differences between children who received the obese report and children who received the overweight report. These findings hold across subsamples by age, minority status, and school meal status.

Conclusions:

Based on these data, overweight or obese reports to do not meaningfully impact future BMI z-scores. This may be due, in part, to the format of parental reports, which may dampen the surprise element of an overweight or obese report.

Introduction

Due to concerns about childhood obesity, there has been a push for BMI screening and surveillance programs for public schoolchildren. Over the past two decades, half of all states have implemented such programs.^1,2 Many of these programs include confidential reporting of BMI and/or weight status information to parents or guardians with the goal of enabling parents to help their children achieve or maintain a healthy weight status.²

Most parents are supportive of these programs and are attuned to BMI and weight status information provided.³ However, there is also little evidence that the reports change parents’ self-reported behaviors to improve children's weight status. For example, Jones et al.⁴ report that roughly 60% of parental respondents agreed with the BMI reporting programs and that 92% of respondents indicated having read all of the information on the BMI report. However, only 24% of parents attempted to make changes after receiving the BMI report.

The seeming contradiction between large majorities of parents agreeing that reports are valuable but at the same time indicating little if any change in behaviors can be resolved by recognizing that most reports contain no new actionable information. With this in mind, we use a regression discontinuity (RD) design to investigate whether BMI reports that indicate the child is overweight or obese in one period influence the child's BMI in the following period. Based on age and sex specific growth charts, overweight is defined as a BMI at or above the 85th percentile and below the 95th percentile and obesity is defined as a BMI above the 95th percentile.⁵ An RD design involves analyzing samples in a narrow band just above and below these thresholds to assess the impact of an unhealthy weight-status report, that is, an overweight or obese report.

Children who are just below the 85th (or 95th) percentile and consequently receive a healthy weight (or overweight) BMI report should have similar characteristics to the children who are just above the 85th (or 95th) percentile and consequently receive an overweight (or obese) BMI report. The only difference would be the weight status classification conveyed to parents via the BMI report. If the report has an effect, it could be revealed in a discontinuity or shift in a regression of future BMI on current BMI percentile. Because children around the threshold share similar characteristics except for the classification contained in the report, an RD design can be a strong quasi-experimental method that can elicit the causal effects of receiving different types of BMI reports.

Note that the estimated effects in our study show the causal effect of overweight report (obese report) on children's weight compared to the healthy weight report (overweight report), not the causal effect of the BMI report program itself.

Specifically, we focus on the Arkansas BMI screening program. In 2003, Act 1220 was passed by the Arkansas Assembly. This act mandated the measurement and provision of BMI information to the parents or guardians of public schoolchildren through confidential reports. The BMI reporting program was established to comply with this act.⁶ Arkansas has the nation's longest running and most comprehensive BMI screening and reporting program.

There are five types of reports for parents, one for children falling in each of the four BMI classifications: underweight, healthy weight, overweight, and obese and another for those children who were not assessed.⁷ In each report, background information covering the negative health consequences of childhood obesity, the state law mandating BMI measurement in school, and recommendations for improving nutrition and physical activities are specified. Figure 1 provides BMI report templates for children whose BMI fell into the healthy weight and obese ranges.

Figure 1.

Example of BMI reports for healthy weight or obesity. (a) Healthy weight report. (b) Obesity report.

This article contributes to a growing literature focusing on whether BMI reporting programs can improve the weight status of children, especially those who are classified as under the overweight or obese category. Only a few studies have used empirical strategies to examine the causal effect of BMI reporting programs on children's weight.^8–12 The duration of the Arkansas program permits us to explore the long-run effects of the program.

Literature Review

There are a number of earlier studies that have examined the effect of BMI reporting programs. However, the literature on the causal effects of BMI reports on children's weight is relatively small. Prina and Royer⁸ conducted a randomized field experiment in Puebla, Mexico. While their results showed that information provision through BMI reports significantly increased parental knowledge of their child's weight status, they did not find that this translated into changed behavior. Almond et al.⁹ investigated the effect of BMI reports in New York City through an RD design. Using a sample that comprised public schoolchildren from kindergarten to 12th grade during 2007–2012, they found that BMI reports had no effect on weight status.

Exploiting the change in implementation of the BMI program in Arkansas, Gee¹⁰ compared the students who were measured 2 years in a row with the students who were measured in their 1st year but not measured in the 2nd year. Using data from the Youth Risk Behavior Survey for students between 11th and 12th grades, he found no significant difference. More recently, Madsen et al.¹² conducted a randomized trial from the Fit Study to examine the causal effect of a BMI report program. In the Fit study conducted on 79 elementary and middle schools in California, students were randomly assigned to one of three groups: (1) those subjected to both BMI screening and reporting, (2) those subjected only to BMI screening, and (3) those subjected to neither BMI screening nor reporting. They found no significant difference in BMI z-score between groups 1 and 2 students.

Our study is similar to Almond et al.,⁹ in which we use a RD design. However, given that the data we use in this study cover a longer period of time than the data in Almond et al.,⁹ we provide additional evidence about the long-term effects and the effects over different periods, including time before major obesity intervention initiatives, such as the First Lady's “Let's Move” campaign during Obama administration. Arkansas also differs in many respects from New York City, California, and city of Puebla in Mexico. For example, relative to New York City and California, a much larger proportion of the population resides in rural areas, and the state has a higher percentage of low-income residents.

As of 2015, 62.3% of the students in Arkansas are eligible for free or reduced price lunch.¹³ Arkansas also has one of the highest childhood obesity rates in the United States, which makes it an interesting and important case to study. Compared with the study of Gee,¹⁰ our data include a broader and younger range of children that include those in kindergarten through 10th grade under the care of parents. Also, while the BMI data used by Gee¹⁰ are based on self-reported height and weight, we use measured BMI data. Finally, our study is different from Madsen et al.,¹² in which their results show the effect of a BMI report program itself, whereas the results in this study represent the effect of different types of BMI reports.

Methods

Data

Our dataset covers the 2003/2004 through 2014/2015 school years. During the in-school assessments, a school nurse conducted height and weight measurements while another individual served as the recorder. Measurements used a stadiometer, stabilized against a wall, and a digital scale. During the measurements, the recorder verbally verified entry of the height data. If there was a disagreement between the two height measures that exceeded one inch, the measurement protocol was repeated; otherwise, the height used in BMI calculation was the average of the two measurements. A complete description of protocols and equipment used to assess children's heights and weights is documented in the Arkansas Center for Health Improvement training document entitled BMI Screening Program: Height and Weight Measurement Training Manual.¹⁴

Along with information on BMI, these data include information on children's demographic characteristics, and whether the child qualifies for free or reduced-price school meals. During the first 4 years of the Arkansas BMI reporting program, children were measured annually. Afterward, children were measured biennially. To take into account this change in reporting frequency, we examine the earlier (2004–2007) and later (2008–2015) program periods separately, in addition to the full 12-year sample. We also separately examine the effect of reports indicating that the child was in the overweight category (overweight report) and reports indicating that the child was in the obese category (obese report). The use of these data for this research project was reviewed by the institutional review board at the University of Arkansas and was determined to meet exemption 4 for “research involving the collection or study of existing data or specimens if publicly available or information recorded such that subjects cannot be identified.” The University of Arkansas Institutional Review Board protocol number is 14-07-026.

Data Analysis

We use an RD design to examine the causal effect of an overweight or obese BMI report on the child's BMI. Specifically, we estimate the following nonparametric local linear regression:

$$Y_{it+1}={\beta }_0+{\beta }_1{T}_{it}+{\beta }_2\left(S_{it}-c\right)+{\beta }_3\left[T_{it}\times \left(S_{it}-c\right)\right]+\rho X_{it}+E_{it},$$ (1)

where i indicates a child, and t is the year that the child's BMI is measured and a report is delivered to his/her parents. The dependent variable, Y, is the child's BMI z-score. T is the indicator for a child in the overweight category (or a child in the obese category), which is based on reference CDC age and sex-specific growth charts. S is the BMI percentile, which is the running variable in parlance of RD estimation, and c is the relevant threshold for an overweight or obese classification.

The coefficient of interest is β₁, which is an estimate of the difference in the BMI z-score for children who receive overweight (or obese) BMI reports around the cutoff point. Finally, X is a vector of covariates used to control for individual characteristics. These include sex, race, grade, and school meal status.

For the baseline specification, we estimate Equation (1) using the triangular kernel with mean square error optimal bandwidth developed by Calonico et al.¹⁵ Robust standard errors are clustered by school. To account for the change in reporting periodicity of the program that occurred between 2007 and 2008, we include a period dummy and an interaction term between the period dummy and T when estimating the model using the full 12-year sample. Because the probability of receiving a BMI report for overweight or obesity ranges from zero to one across each threshold, we implement a sharp RD design.

Validity of Analysis Strategy

The identification assumption for a valid RD design is that children have imprecise control over the running variable, which is their BMI percentile. If this could be strategically manipulated, children who receive overweight or obesity reports may have different characteristics from those who do not, and this would result in biased estimates. However, as Almond et al.⁹ point out, such manipulation is unlikely because BMI percentile thresholds for children in the overweight or obese category are age and gender specific, and the actual BMI at the threshold is not widely known and seldom falls on round numbers. Moreover, in addition to the day-to-day and diurnal variations in weight, children are unlikely to have advance notice of the BMI measurement date.

Nevertheless, we test the identification assumption in two ways. First, following McCrary,¹⁶ we assess whether there is discontinuity in the density of the running variable around the overweight or obesity threshold. Second, we examine whether there is a difference in observed characteristics of children around the overweight and obesity thresholds by estimating a series of balancing regressions, wherein Equation (1) is estimated, but the dependent variable, instead of being the BMI z-score, is replaced with one of the control covariates in X. These tests confirm that strategical manipulation of weight is unlikely. The results of these tests can be found in the Supplementary Data (see Supplementary Fig. S2; Supplementary Tables S1 and S2).

Results

Summary Statistics

Data used in the RD models are summarized in Table 1 by weight status and sample subperiod. These data show some noteworthy patterns in the incidence of overweight and obesity from the 12-year panel. One is a racial/ethnic disparity in overweight and obesity. When we imputed the number of minority children from the descriptive statistics reported in the table, we find that 228,799 of 518,487 or 44% of minority children (not White or Hispanic) were overweight or obese during the full study period compared to 37% of nonminority children (White and not Hispanic). Most of this difference can be attributed to the difference in obesity prevalence, which was 25% and 20% among minority and nonminority children, respectively.

Table 1.

Descriptive Statistics by Subperiod and Type of Report for the Full Sample

	BMI screened annually (2004–2007)			BMI screened biennially (2008–2015)			Full study period (2004–2015)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Normal	Overweight	Obese	Normal	Overweight	Obese	Normal	Overweight	Obese
BMI z-score	0.09	1.33	2.12	0.07	1.33	2.12	0.08	1.33	2.12
	(0.63)	(0.18)	(0.32)	(0.64)	(0.17)	(0.32)	(0.63)	(0.17)	(0.32)
BMI percentile	53.77	90.49	97.89	53.17	90.49	97.87	53.50	90.49	97.88
	(21.87)	(2.89)	(1.34)	(22.18)	(2.89)	(1.34)	(22.01)	(2.89)	(1.34)
Male (%)	50.53	50.38	54.88	50.08	49.50	54.29	50.33	49.99	54.62
	(50.00)	(50.00)	(49.76)	(50.00)	(50.00)	(49.82)	(50.00)	(50.00)	(49.79)
Female (%)	49.47	49.62	45.11	49.92	50.50	45.71	49.67	50.01	45.38
	(50.00)	(50.00)	(49.76)	(50.00)	(50.50)	(49.82)	(50.00)	(50.00)	(49.79)
Younger cohorts (%)	58.37	55.82	55.20	51.71	47.72	46.20	55.41	52.54	51.23
	(49.29)	(49.66)	(49.73)	(49.97)	(49.95)	(49.86)	(49.71)	(49.95)	(49.99)
Older cohorts (%)	41.63	44.18	44.80	48.29	52.28	53.80	44.59	47.76	48.77
	(49.29)	(49.66)	(49.73)	(49.97)	(49.95)	(49.86)	(49.71)	(49.95)	(49.99)
Nonminority (%)	70.71	66.81	63.31	68.51	63.65	58.66	69.73	65.42	61.26
	(45.51)	(47.09)	(48.20)	(46.45)	(48.10)	(49.24)	(45.94)	(47.56)	(48.72)
Minority (%)	29.29	33.19	36.69	31.49	36.35	41.34	30.27	34.58	38.74
	(45.51)	(47.09)	(48.20)	(46.45)	(48.10)	(49.24)	(45.94)	(47.56)	(48.72)
Free meals (%)	41.25	43.97	47.84	43.36	47.55	52.92	42.19	45.55	50.08
	(49.23)	(49.64)	(49.95)	(49.56)	(49.94)	(49.91)	(49.39)	(49.80)	(50.00)
Nonfree meals (%)	58.75	56.03	52.16	56.64	52.45	47.08	57.81	54.45	49.92
	(49.23)	(49.64)	(49.95)	(49.56)	(49.94)	(49.91)	(49.39)	(49.80)	(50.00)
N	531,732	156,301	190,149	425,282	123,970	150,277	957,014	280,271	340,426

Note: Standard deviations are in parentheses.

Second, prevalence of overweight and obesity was higher among less affluent children, those who qualify for free school meals at 42% compared to 37% for those who did not qualify for free school meals. Again, most of this can be attributed differences in obesity prevalence. Third, a slightly larger percentage of boys (40%) and older children (41%) had an unhealthy weight status compared to girls (38%) and younger children (38%). In sum, children in the overweight or obese category are more likely to be minority, male, older (i.e., in 6th to 10th grades) and less affluent relative to children with a healthy weight. These patterns and overall prevalence of unhealthy weight status were similar during the 2004–2007 and 2008–2015 subperiods, during which BMI was reported annually and biennially, respectively.

RD Design Results

Figure 2 presents the conditional means of BMI z-score as a function of BMI percentile in the previous measurement period around the overweight or obesity thresholds. In the figure, hollow diamonds indicate mean of BMI z-scores for bin, solid diagonal lines represent fitted values, and dashed diagonal lines represent 95% confidence intervals. If the BMI reports had a significant effect on the children's weight status, there should be a discernible discontinuity in the fitted line at the threshold, which is indicated by the dashed vertical line in each panel.

Figure 2.

BMI report effects. (a) Overweight reports 2004–2007 (b) Obese reports 2004–2007. (c) Overweight reports 2008–2015. (d) Obese reports 2008–2015. (e) Overweight reports 2004–2015. (f) Obese reports 2004–2015. Note: Bin width for overweight is 0.1 percentile, and bin width for obesity is 0.05 percentile. The dashed vertical line indicates the threshold for an overweight report (a, c, e) or obese report (b, d, f).

Panels (a) and (b) are for the 2004–2007 period, during which BMI reports were issued annually. For children near the overweight threshold in panel (a), we do not find meaningful discontinuity at the threshold. For those near the obesity threshold in panel (b), there is a more discernible drop in BMI z-score at the threshold, but the magnitude is small. The period of biennial measurement (2008–2015) is presented in panels (c) and (d).

These show a small increase in BMI at the overweight threshold but a discernible drop at the obese threshold similar to that observed for the 2004–2007 period. Results for the combined 12-year period are reported in panels (e) and (f). Consistent with findings from the two subperiods, there is discernible discontinuity only at the obesity threshold [panel (f)] and this is small. In sum, the mean of BMI z-score at each bin shows the expected positive relationship with the running variable but without a meaningful drop at the threshold. Thus, the graphical results in Figure 2 provide evidence that weight status-specific information contained in the BMI reports does not have considerable impact on children's BMI.

Corresponding estimates from the RD regressions are presented in Table 2 and reinforce the graphical evidence in Figure 2. Panel A contains results for overweight reports while panel B contains the results for obese reports. Across all subperiods, effects of an obese report are negative and are similar in magnitude, but are not statistically different from zero at the 5% level. Given the similarity in effect sizes over the two subperiods, this can be explained by the increased power that accompanies the larger full-period sample. While not significantly different from zero, the effects of obese reports are noticeably larger in absolute value in comparison to the effects of an overweight report. Even so, as a practical matter, the effect of an obese report is small, being <0.02 standard deviations of a BMI z-score.

Table 2.

Regression Discontinuity Estimates

	(1)	(2)	(3)
	BMI screened annually (2004–2007)	BMI screened biennially (2008–2015)	Full study period (2004–2015)
Panel A: Overweight reports
BMI z-score	−0.007	0.002	−0.005
	(0.008)	(0.011)	(0.007)
Bandwidth	2.702	3.046	2.704
N	42,863	37,932	76,578
Panel B: Obese reports
BMI z-score	−0.013	−0.016	−0.016*
	(0.009)	(0.012)	(0.008)
Bandwidth	0.976	1.114	0.918
N	26,756	24,943	46,749

Note: Robust standard errors are clustered by school. Choice of optimal bandwidth follows Calonico et al.¹⁵ Long-run effect indicates the effect of BMI report at kindergarten on BMI z-score at fourth grade. All estimations contain controls for age, gender, race, and school meal status. ^*p < 0.1, ^**p < 0.05, ^***p < 0.01.

We also examined a sample composed of children who received first-time overweight or obese reports. The rationale here is that first-time reports may carry more informational value. In addition to this, we check the possibility of a long-run effect of BMI reports by investigating the effect of overweight or obesity reports in kindergarten on weight status by fourth grade. We chose the K to fourth grade interval because childhood obesity starts to emerge at an early age,^17,18 and kindergarten is the first time BMI reports are issued to parents. The results are exhibited in Table 3.

Table 3.

Regression Discontinuity Estimates: First-Time Reports and Long-Run Effects

	First-time reports			Long-run effect
	(1)	(2)	(3)	(4)
	BMI screened annually (2004–2007)	BMI screened biennially (2008–2015)	Full study period (2004–2015)	Full study period (2004–2015)
Panel A: Overweight reports
BMI z-score	−0.015	0.007	−0.012	−0.026
	(0.011)	(0.015)	(0.010)	(0.030)
Bandwidth	2.975	3.281	2.859	4.576
N	23,984	23,835	43,852	11,775
Panel B: Obese reports
BMI z-score	−0.001	−0.018	−0.006	−0.022
	(0.011)	(0.016)	(0.011)	(0.032)
Bandwidth	1.209	1.339	1.046	2.100
N	16,850	16,683	27,437	6719

Note: Robust standard errors are clustered by school. Choice of optimal bandwidth follows Calonico et al.¹⁵ Long-run effect indicates the effect of BMI report at kindergarten on BMI z-score at fourth grade. All estimations contain controls for age, gender, race, and school meal status. ^*p < 0.1, ^**p < 0.05, ^***p < 0.01.

Again, there is no evidence that a first time overweight or obese report leads to a reduction in BMI by the following measurement point. Moreover, there is less consistency in the point estimates of report effects over the subperiods. The long-term effects reported in the rightmost column of Table 3 are larger than those reported to this point in the article but are not statistically different from zero.

To assess heterogeneity in report effects across different subsamples, we estimate Equation (1) from subsamples by age (children in kindergarten to 5th grade vs. children in 6th to 10th grade), minority status, and school meal status. Overall, the effects are small in absolute value and are statistically insignificant. The results by subsample are available in the Supplementary Data accompanying to this article.

Discussion

Overall, our results do not provide strong evidence that receipt of an overweight or obese report reduced children's weight by the time of the next BMI measurement. These results are broadly consistent with the results of previous studies mentioned earlier.

The small impacts we find may be due to the layout of the reports (as shown earlier in Fig. 1). The reports indicate the child's BMI along a continuum making it easy to see how close a child is to the overweight or obese threshold. Because parents know the proximity of their child's BMI to the overweight or obese thresholds, those with children just below a given threshold may be acting on the report information in a manner that is similar to those with children just above the threshold. If this is in fact how parents respond to the reports, then our analysis strategy would not be able to detect a significant difference even if the reports were meaningful in motivating behaviors conducive to a healthy weight status. For this reason, our findings should not be taken as evidence that BMI reporting is ineffective.

Another possible explanation for our findings lies in the identical recommendations for maintaining healthy weight status in the BMI reports across the healthy weight, overweight, and obese reports. If parents lack specific knowledge on how to help children in the overweight or obese categories achieve a healthy weight as they grow, then the lack of recommendations contingent on weight status could be limiting the impact of overweight or obese reports. Finally, the effects of the reports on behavioral change of parents are more likely to be attenuated if they do not share the information with their children. Unfortunately, our data do not permit us to evaluate whether reports would have had greater impacts had information content differed or had more parents shared the information with their children. We think it important to acknowledge that this may be one factor behind our findings.

While we detected only a small effect of the BMI reporting program, it is important to mention that the overall benefits of the program extend beyond the information provided to parents. For example, the BMI reporting program has revealed marked differences in incidence of obesity rates across communities and school districts.¹⁹ This information can be used to improve and target programming designed to improve the health of schoolchildren. Benefits resulting from better intervention design and implementation will not be captured in our estimates of the overall program's effect. This is an important reason why our findings should not be taken as overall evidence that the BMI reporting program is ineffective. In fact, information from the BMI screening program could be used to prioritize schools with children at highest risk for unhealthy weight.

Given the findings above, recommendations for maintaining healthy weight status may need to be more specific and differentiated according to the children's weight status. Earlier work has shown that provision of information can be more effective in changing individual's health behavior if the message is personalized.²⁰ In addition, considering ways to convey information on adverse obesity-related outcomes in a manner that does not stigmatize students may increase the salience of the reports.^21–24 Finally, the BMI reports could also provide an opportunity to showcase how schools and communities are working to promote healthy weight status among children, thereby alerting parents to important resources in their child's school and in the broader community. At present, the BMI reports only provide parental recommendations but earlier studies have shown that improvements in school meals and in the built environment of communities can play a role in childhood obesity prevention.^25,26

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Funding Information

Financial support for this study was provided, in part, by a grant from the National Institute of General Medical Sciences of the National Institutes of Health under Award Number P20GM109096. The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report.

Author Disclosure Statement

All authors have no conflicts of interest that are directly relevant to the contents of this article.

Supplementary Material

Supplementary Figure S1

Supplementary Table S1

Supplementary Table S2