Child bodyweight, cognitive abilities, and well-being: evidence from West Bank schools

ABSTRACT

The present study investigates the effect of obesity and overweight on children’s cognitive abilities and well-being using survey data from West Bank schools. The results show the significant adverse impact of obesity on a child's well-being by raising externalizing (behavioral) problems and increasing the probability of classifying a child with abnormal mental health difficulties, according to the Strength and Difficulties Questionnaire (SDQ), while no such impact was founded on the cognitive test results. The analysis further exposes that teachers’ negative attitude toward obese students is an important factor that contributes to the association between obesity/overweight and higher SDQ scores.

KEYWORDS:

• Obesity
• cognitive abilities
• well-being
• the West Bank

JEL Classifications:

• I10
• I12
• I20
• I31

Acknowledgments

Many thanks to the Editor and three anonymous referees for their helpful comments and suggestions. I am also grateful to Geneva 2021 Workshop participants at the Institute for Global Law and Policy, Harvard Law School for their useful comments. The analysis presented in this paper is part of the project “Determinants of Cognitive Development in Deprived Environments: Evidence from the West Bank” funded by the German Research Foundation (DFG) under grant number JU 2769/2.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Notes

1 Others have used panel data and controlled individual fixed effect to account for unobserved individual-time-specific heterogeneity, see (Sabia 2007).

2 The data was collected as part of the research project entitled Determinants of Cognitive Development in Deprived Environments: Evidence from the West Bank, funded by the German Research Foundation (DFG) under grant number JU 2769/2.

3 Gender-separated schools make up most of primary governmental (public) schools (76%) and UNRWA schools (85%) (Ministry of Education and Higher Education; MoEHE 2012). Female students were over-represented in the survey because 59 percent of the 436 schools that meet the selection criteria were girls’ schools. Single-sex schools contain Grades 5 to 9.

4 The data were collected by cooperation between Al-Quds university Al-Quds Nutrition and Health Research Institute at Al-Quds university researchers and the General Administration of Health in Primary Schools in the Palestinian Ministry of Education.

5 BMI is calculated as the body weight of an individual in kilograms divided by the height in meters squared.

6 The main analysis includes the status of obese (8.5%) and overweight (11%) children in one category. However, in the robustness check, the analysis shows the effect of the obesity and overweight on a child’s outcomes separately.

7 See https://www.cdc.gov/growthcharts/data/set1clinical/cj41l023.pdf.

8 The subtests were selected and adapted from established tests of general ability: The Cognitive Ability Test (CAT; Thorndike, Hagen, and Lorge 1971; Standard Progressive Matrices (Raven 1983); and the Culture Fair Intelligence Test (Cattell and Cattell 1960)

9 The Strengths and Difficulties Questionnaire (SDQ) is a brief behavioral screening questionnaire for 3–16-year-olds. There are currently three self-reported versions of the SDQ in the following categories: children; parents; and teachers. Each version includes between one and three of the following components: 25 items on psychological attributes, an impact supplement, and follow-up questions (Goodman, Lamping, and Ploubidis 2010; Goodman 1997).

10 The total score for assessing mental health problems does not include the prosocial behavior score (Goodman 1997).

11 The standard of living index shows whether a household owns fixed assets, such as a TV, a mobile phone, a DVD player, air conditioning, a car, and so on, among other belongings.

12 The reason for not adding these controls in the main model is to reduce the number of deleted observations since not all students provided answers to these questions.

13 For more details about the measurement of the big five personality traits (John and Srivastava 1999) Table B.5 (in the on-line appendix) presents the survey statements.

14 Table (A.3.a). In the robustness checks, the regression for the overweight and obese child was run separately on the outcome variables.

15 The answers to the personality traits questionnaire were provided by the children’s mothers. Each trait was evaluated by three statements using a 5-point Likert scale. More points for openness, conscientiousness, and agreeableness indicate that the child has positive attributes; a higher score for neuroticism, which is associated with a lack of emotional stability and negative emotional responses, including anxiety, sadness, and irritability, indicates a negative attribute (Mitchell et al. 2021).

16 Using parental BMI as an instrument is based on Mendelian randomization research. This method is widely used to measure variation in genes to interrogate the causal effect of an exposure on an outcome (such as fat mass, lipids, energy intake, number of cigarettes, etc.). If the assignment (instrument) is a continuous intermediate variable (IV), in that case, it is uncertain how to differentiate between the four IV categories (Always-takers, those who will take the treatment regardless of whether they are assigned to the treatment or not; never takers: those who never take treatment, whatever their assignment; and defiers, who always do the opposite of the assignment (see Angrist and Pischke 2009; Bartolucci and Grilli 2011; Jin and Rubin 2008; Marbach and Hangartner 2020; Sjölander et al. 2009). For this reason, it might be more beneficial to consider a situation where the instrument and treatment are binary (von Hinke et al. 2016).

17 von Hinke et al. (2016) use the genetic variants FTP and MC4R as instrumental variables to investigate the impact of children’s mass on their school performance.

18 This regression was implemented for the father only since the majority of the mothers in the sample were not working.

19 Kaestner and Grossman (2009) point out that obesity might be a consequence of government policies (e.g. household consumption behavior can be adversely affected if healthy food is not subsidized or if transportation costs change).

20 To test this assumption, I investigated the association between child parental BMI and child school performance assessed by school Grade Point Average (GPA) in the academic year 2011/2012. The regression results do not support any significant correlation between parental BMI and children’s academic performance (Table B.1 on – line appendix).

21 To save space, all analyses that show the correlation between parental obesity and their behavioral outcomes are reported in on-line appendix B of the paper.

22 The data set includes a list of 13 statements that indicate whether the child is maltreated by their family/parent. Each statement has three answers: never (0); sometimes (1), and mostly (2). Higher scores indicate a worse situation (more maltreatment). These statements include questions about whether the child lacks family support; is raised in an abusive and neglectful family. For more details about this instrument, see (Abdeen et al. 2018).

23 The third and the fourth assumption of having a valid instrument is monotonicity. All observations that are affected are affected in the same direction. The analysis shows a positive correlation between the parent’s/parents’ and their offspring’s obesity. However, one of the violations of this assumption could be that the role of this specific gene responsible for obesity depends on another environmental factor, such as education. For example, whether the potential BMI for an educated individual who is aware of the importance of nutritional habits with the specific genetic marker is smaller than the likely body mass for the same individual without this genetic marker. However, even with the previously mentioned case, the monotonicity would not be violated since the expected obesity for an individual with the genetic variant remains at least as high compared with the anticipated obesity for other individuals without these genetic markers (von von Hinke et al. 2016). The final assumption for having a valid instrument should have a non-zero treatment. As indicated earlier, biomedical literature concludes the association between genetic variants and BMI, which justifies using parental obesity as a valid instrument. Furthermore, the first stage regression shows a significant correlation between parental and offspring obesity.

24 The coefficients for the control variables are not reported in order to save space, but are available on request from the author.

25 Using father BMI as separate instrument does not provide any significant results. The coefficient results are not reported to save space, but are available on request from the author.

26 In separate regressions, I investigated the effect of the interaction indicator between a child’s BMI and being obese/overweight. Overall, the results are consistent with the results obtained from the main analysis. The only exception is that the 2SLS for the effect of the interaction variable between obsess/overweight status and BMI on child's cognitive abilities becomes significant with a positive sign. This result is also in line with articles that find those children who spend more time studying than their peers do fewer physical activities, improving their educational attainment (Nghiem et al. 2018). Results are not reported in order to save space, and are available upon request from the author.

27 When the effect of overweight status on child’s outcome was investigated, the observation with obesity status was excluded from the analysis.

28 The IV estimate may be larger than the OLS estimate because IV estimates the local average treatment effect (LATE) for the subgroup affected by the instrument. In other words, the instrument (parental obesity) shifts the behavior of a subset of individuals for whom treatment effects (overweight/obese separately) are larger than average. In contrast, OLS estimates the average treatment effect (ATE) over the entire population. Then IV estimates will be larger than OLS estimates due to the heterogeneity in the affected population.

29 The 2SLS coefficient in Tables A.5 and A.6 was obtained by using single instrument (mother is obese/overweight). The IV estimates using two instruments do not provide any significant results. Results are not reported in order to save space, and are available upon request from the author.