Childhood Overweight and Parent- and Teacher-Reported Behavior Problems: Evidence From a Prospective Study of Kindergartners

Objectives  To determine if there is a relationship between overweight and behavior problems among children as young as 5 years old by studying the association between overweight and behavioral health at entry into kindergarten and to determine whether overweight status is a risk factor for the onset of new behavior problems during the first 2 years in school.

Design  We use data from a nationally representative sample of kindergartners in the United States—the Early Childhood Longitudinal Study–Kindergarten class. Data on height, weight, and parent- and teacher-reported behavior problems were collected 3 times during their first 2 years in school for 9949 children. We use a multivariate regression analysis that controls for sociodemographic characteristics, parent-child interaction, birth weight, and mother's mental health.

Results  Among girls, but not boys, there is a significant association between overweight and teacher-reported externalizing behavior problems (odds ratio [OR], 1.81; 95% confidence interval [CI], 1.23-2.68), teacher-reported internalizing behavior problems (OR, 1.54; 95% CI, 1.09-2.17), and parent-reported internalizing behavior problems (OR, 1.49; 95% CI, 1.08-2.06) at the beginning of kindergarten. However, overweight status was not a risk factor for the onset of new behavior problems over time for either girls (teacher-reported externalizing behavior problems: OR, 0.58 [95% CI, 0.25-1.33]; teacher-reported internalizing behavior problems: OR, 1.34 [95% CI, 0.88-2.03]; and parent-reported internalizing behavior problems: OR, 1.29 [95% CI, 0.82-2.01]) or boys (teacher-reported externalizing behavior problems: OR, 1.02 [95% CI, 0.67-1.57]; teacher-reported internalizing behavior problems: OR, 1.02 [95% CI, 0.68-1.52]; and parent-reported internalizing behavior problems: OR, 1.42 [95% CI, 0.94-2.15]), whereas low family income and maternal depression were strong predictors of such problems.

Conclusions  Childhood overweight is already associated with behavior problems when girls start school, but not boys. In contrast to common belief, overweight status does not predict the onset of new internalizing or externalizing behavior problems during the first 2 years of school.

Psychosocial problems have been hypothesized to be among the most serious consequences of childhood overweight.^1,2 Overweight children and adolescents may be teased and ridiculed or may experience social marginalization, leading to low self-esteem.^3-7 Most data on the relationship between overweight and mental health have focused on adolescents or adults, and, to our knowledge, no nationally representative data have been published for younger children. Erickson et al,⁸ using a sample from 13 schools in northern California, report a significant cross-sectional association between depressive symptoms and body mass index among third-grade girls, but not among third-grade boys.

In this study, we use data from a national probability sample of kindergartners in the United States to examine whether overweight children are at a higher risk for parent- and teacher-reported internalizing and externalizing behavior problems. This study helps us examine whether the association found by Erickson et al⁸ in northern California schools holds across the nation. Specifically, we examine the following: (1) whether overweight status is associated with greater behavior problems at the beginning of kindergarten and (2) whether overweight status at the beginning of kindergarten is a significant predictor of behavior problems after 2 years in school among children with no significant behavior problems when they entered kindergarten.

We analyze data from the Early Childhood Longitudinal Study–Kindergarten (ECLS-K) class. The ECLS-K surveyed a nationally representative cohort of kindergartners from about 1000 kindergarten programs in the United States in the fall and spring of the 1998-1999 school year. This sample will be followed up through grade 5, with data collection on the full sample in the spring of grades 1, 3, and 5. At this point, the first 2 years of data are available. The ECLS-K sample was collected using a dual-frame multistage sampling design with oversampling of Asians and children with a disability. The National Center for Educational Statistics⁹ has more information on the survey design and instruments.

We use data collected in the fall of kindergarten and the spring of first grade in our analysis. The behavior problem variables came from the parent and teacher questionnaires. Typically, the respondent for parent interviews was the child's mother. If the mother was not available, then the next selected respondent was another parent or guardian, followed by another household member. The response rates for the fall kindergarten teacher and parent questionnaires were 91.4% and 85.3%, respectively. There were no significant differences in the characteristics (race, sex, mother's educational level, and family income) of children with and without teacher data, although nonresponse on the parent questionnaire was more prevalent among nonwhites. However, nonresponse on the teacher and parent questionnaires was not associated with child's overweight status. Follow-up data in the spring of first grade on behavior problems were only available for 82.1% of these children. Children with no follow-up data between kindergarten and first grade were those who changed schools. The ECLS-K followed up all movers from a random 50% of schools in the sample. Therefore, most of the children with no follow-up data were those randomly selected for no follow-up and would be unlikely to bias the results. Our analysis sample consists of 9949 kindergartners with data on parent- and teacher-reported behavior problems in the fall of kindergarten and the spring of first grade. Because our study uses public use data, it has been exempted from institutional review board approval.

The key variables in this study are child overweight status and teacher- and parent-reported behavior problems. The ECLS-K includes 3 behavior problem scales based on the teacher- or parent-reported Social Rating Scale (SRS), which we use to construct our outcome measures. These scales were adapted from the Social Skills Rating System instrument developed by Gresham and Elliot.¹⁰ The Social Skills Rating System is a well-established and useful tool for assessing and targeting social skills deficits and competing problem behaviors.

The teacher SRS is a self-administered questionnaire and provides 2 scales—externalizing and internalizing behavior problems. The externalizing scale includes information on the acting out behaviors of children and is based on 5 items that rate the frequency with which a child argues, fights, gets angry, acts impulsively, and disturbs ongoing activities. The internalizing scale consists of 4 items that ask about the apparent presence of anxiety, loneliness, low self-esteem, and sadness in the child. Teachers used a response scale to report how often the child demonstrated the behavior described: 1, never; 2, occasionally or sometimes; 3, regularly, but not all the time; and 4, most of the time. The scores on both scales are the mean rating on the items included in the scale. The split half reliabilities for the teacher SRS are high: 0.90 for externalizing problems and 0.80 for internalizing problems. We create dichotomous variables for externalizing and internalizing problems to simplify the presentation. Children who scored at the 95th percentile or higher on each scale were classified as having substantial behavior problems. The scales are not fully continuous and, therefore, slightly more than 5% of children will be identified as having either substantial internalizing or substantial externalizing problems. These scales do not provide clinical diagnoses, and the choice of a cut point on a scale is always somewhat arbitrary. By using the 95th percentile on each scale, we can be confident that children classified as having substantial problems are sufficiently symptomatic to warrant at least a psychological examination. To put these numbers into perspective, the treated prevalence of mental health disorders based on pediatric primary care diagnosis tends to be in the 5% range.¹¹ The exact cut points and their interpretation are not central to our analysis because we only want to compare the relative prevalence in different populations, not provide an absolute estimate of need. As a sensitivity analysis, we also used the 90th and 97.5th percentile points as alternate cutoff points.

This scale is derived from the parent SRS and consists of 4 items that ask parents about children's problems with being accepted and liked by others, sadness, loneliness, and low self-esteem. Parents used the same 4-point response scale used in the teacher SRS previously described. The reliability for the parent SRS is lower than those for the teacher SRS (0.60). We dichotomize the scale similarly. Because of the way responses are clustered on this scale, about 7% of the children display the same or more symptoms as the child at the 95th percentile. Parents were also asked about externalizing behavior problems in their children. The reliability of this scale was rather low (0.47), and we, therefore, did not include it in our analysis.

We use the body mass index to classify children. Height and weight were measured twice in the ECLS-K study at each period. Children who had a body mass index, calculated as weight in kilograms divided by the square of height in meters, at the 95th percentile or higher for their age and sex were classified as overweight.^12,13

Additional explanatory variables included in the multivariate regression models included race-ethnicity, annual family income, whether the child belongs to a single-parent household, maternal depression, parent-child interaction, and child's birth weight. These variables are discussed in detail in the "Data Analysis" subsection of the "Methods" section.

We first study the association between overweight status and behavior problems cross-sectionally at the beginning of kindergarten. We provide descriptive statistics and associations adjusted for potentially confounding factors that may be correlated with childhood overweight and children's mental health.

For the multivariate analysis, we estimate a logit model with behavior problem as the dependent variable and baseline overweight status as the right-hand side predictor variable. The first set of variables to control for confounding effects are standard sociodemographic variables: race-ethnicity (white, black, Hispanic, Asian, and others), mother's educational level (<high school diploma, high school diploma, some college, bachelor's degree, and >a bachelor's degree), annual family income, whether the child belongs to a single-parent family, and the number of siblings. The second set includes measures that may be directly related to psychological functioning: parent-child interaction, birth weight, and maternal depressive symptoms. The ECLS-K asked parents how often they participated in 8 different activities with their child, such as reading books to their children, telling stories, playing sports, and singing songs. Our measure of parent-child interaction is defined as the number of activities that the parent participates in with the child at least once a week. We include the child's birth weight because there is evidence that low-birth-weight children may be at a greater risk for psychological problems.^14-16 In addition, parental depression has been shown to be a strong predictor of children's psychological problems.^17-21 The ECLS-K asked the parent respondent 12 items from the Center for Epidemiological Studies Depression Scale.²² The full Center for Epidemiological Studies Depression Scale consists of 20 items, with scores ranging between 0 and 60, on which scoring at 16 or higher is considered to indicate elevated levels of depression. However, because the ECLS-K survey did not consist of all 20 items, we were unable to use the suggested cutoff. Instead, we use a continuous depression scale with scores from 0 to 36, with higher scores indicating more depressive symptoms. We convert the raw scores into standardized scores by dividing them by the standard deviation (5.31) of these scores.

Our second analysis approach exploits the longitudinal nature of the ECLS-K data to examine whether overweight status at baseline is an important predictor of the onset of behavior problems during the first 2 years in school. This prospective design allows us to examine the direction of the relationship between overweight and behavior problems. A logit regression model is estimated, in which the dependent variable is the presence of behavior problems at the end of first grade. We used the subset of children who were not described as having any substantial behavior problems by the mother or teacher at baseline. We estimate separate regression models for the 3 behavior problem measures. We also estimate separate models for boys and girls, using the same explanatory variables as in the cross-sectional analysis.

Standard errors are corrected using the Huber-White correction to account for the correlation between children from the same school.

The ECLS-K data show that about 1 in 10 kindergartners (11.1%) was overweight when he or she entered kindergarten in 1998; the percentage was slightly higher among boys (11.7%) compared with girls (10.6%). Descriptively, the prevalence of teacher-reported substantial externalizing behavior problems was more than twice as high among boys (8.3%) compared with girls (3.5%). The prevalence of substantial internalizing behavior problems did not differ significantly for boys vs girls (5.8% vs 5.5% for teacher report [P = .48] and 7.0% vs 6.3% for parent report [P = .17]).

Table 1 presents descriptive statistics for boys and girls separately, by overweight status. Overweight boys were no more likely to have parent- or teacher-reported behavior problems in the beginning of kindergarten compared with boys who were not overweight. However, overweight girls were significantly more likely to have any behavior problems at the beginning of kindergarten compared with girls who were not overweight. These descriptive statistics obviously include the effect of other confounding factors as well. For example, Table 1 indicates that overweight girls were less likely to be white and more likely to have a family annual income of less than $25 000, mothers with a high school diploma or less, fewer siblings, come from single-parent families, and a higher birth weight compared with their peers who were not overweight.

Table 2 shows the odds ratio of behavior problems between overweight and non-overweight children, adjusted for sociodemographic characteristics, parent-child interaction, birth weight, and maternal depression. Overweight girls have about 81% greater odds of having substantial teacher-reported externalizing behavior problems, 54% greater odds of having teacher-reported internalizing behavior problems, and 49% greater odds of having parent-reported internalizing behavior problems compared with girls who are not overweight. In contrast to girls, Table 3 shows no significant differences between overweight and non-overweight boys. In fact, after adjusting for sociodemographic characteristics, parent-child interaction, birth weight, and maternal depressive symptoms, the prevalence of behavior problems is slightly lower among overweight boys, but none of these differences are close to significance.

For the longitudinal analysis, we restricted our sample to only those children who did not exhibit substantial behavior problems in the fall of kindergarten (81.9% among boys and 86.3% among girls). The outcome variable is whether a child displays substantial behavior problems at the end of first grade. Results from adjusted models are reported in Table 4 for girls and in Table 5 for boys. There is no evidence that overweight is a risk factor for developing behavior problems for girls and boys. In contrast, low family income and maternal depression are much stronger predictors of the onset of behavior problems over time. For example, the odds of developing externalizing behavior problems during the first 2 years in school are more than 3 times greater for girls in the lowest income quartile compared with girls in the highest income quartile. Not surprisingly, maternal depression had the strongest association with parent-reported internalizing behavior problems, but it may not be a reliable indicator of child health. Maternal educational level was not a significant predictor of the onset of behavior problems, with one exception. Boys with mothers who have a bachelor's degree or more have about half the odds of developing teacher-reported externalizing behavior problems compared with boys with mothers who have a high school diploma or less. We also tested whether there are interactions of overweight status with race-ethnicity, but found no significant effects. In addition, the results from unadjusted models are similar to those from the adjusted regression models.

In addition, boys and girls whose mothers scored 1 SD higher on the maternal depression scale were 15% and 14% more likely to exhibit teacher-reported internalizing behavior problems, respectively.

While there is much concern about a possible relationship between overweight status and mental health problems, especially if the latter affects readiness to learn, there are few data for younger children. The ECLS-K is the first large and nationally representative data set to study children in kindergarten and first grade. Beginning elementary schoolchildren are an important group because adverse events could permanently lower educational achievement. At the same time, elementary school is the first time all children are exposed to possible interventions or to screening for health risks that could precipitate referrals to health care.

Our cross-sectional results confirm the findings of one cross-sectional study⁸ of third-grade students in northern California that associated overweight status with increased depressive symptoms among girls, but not among boys, after controlling for socioeconomic differences. These findings suggest that overweight girls start school with significant parent- and teacher-reported behavior problems. In addition, the longitudinal aspect of the ECLS-K data allowed us to identify risk factors for the onset of behavior problems during the first 2 years of elementary school among children who did not exhibit any behavior problems at baseline. The longitudinal analysis is important for understanding cause-effect relationships between these 2 factors, and we are not aware of any similar published data. In light of the literature emphasizing the adverse psychosocial effects of obesity, we have focused on the temporal effects from overweight to the onset of new (previously not reported) behavior problems. While family income and maternal depression were strong predictors of the onset of behavior problems for girls and boys, we found no evidence that overweight status is a risk factor. These findings are similar to those in one recent study²³ among adolescents that found baseline obesity was not a predictor of follow-up depression.

We also conducted sensitivity analyses by using different cut points (90th and 97.5th percentiles) on the behavior problem scales or by determining overweight status relative to a child's classmates rather than using a national (age- and sex-specific) cut point. However, none of the conclusions about the associations between overweight status and behavior problems are affected.

Our study findings should be interpreted in light of several caveats. First, we focused on the onset of new significant behavior problems, but there may also be deterioration of emotional health among overweight children who already had significant problems at baseline. Second, including measures of parent-child interaction, birth weight, and maternal depression in our multivariate regression analyses may overcontrol for differences between overweight and not overweight children if the purpose is to identify children at risk of developing behavior problems. Finally, overweight status measured at one point may not be a significant risk factor for subsequent behavior problems among children, but persistent or long-term overweight may result in later mental health problems, as suggested by one study.²⁴ Future waves of the ECLS-K data will provide opportunities to test this hypothesis.

We also do not show results for the opposite causal pathway, namely, that behavior problems predict later weight gain. Recent data^23,25,26 for adolescents suggest an association between depression and later obesity, especially among women. The same pattern seems to hold in these data. Among overweight girls, the presence of internalizing behavior problems predicted large additional body mass index gains, but no similar effects existed for boys. Further exploration of these preliminary findings is recommended for future research.

There has been considerable attention on overweight status recently, often under the assumption that this can cause psychosocial problems for children and prevent learning. However, our results show that overweight girls already have more behavior problems before kindergarten; therefore, focusing on mental health and overweight is important in the early years.

To our knowledge, there are no nationally representative data on the association between overweight and mental health outcomes among young children despite growing concerns regarding increasing obesity and its likely psychosocial effects.

To our knowledge, our study is the first to examine this relationship prospectively among a large sample of kindergartners in the United States. We confirm that there exists a significant association between overweight and behavioral health problems as early as starting kindergarten. However, we find no evidence that overweight status at the beginning of kindergarten is a risk factor for behavior problems after 2 years in school among children with no behavior problems at baseline.

Correspondence: Ashlesha Datar, PhD, RAND, 1700 Main St, Santa Monica, CA 90407 (datar@rand.org).

Accepted for publication December 18, 2003.

This study was supported by the National Institute for Health Care Management, Washington, DC.