Trends in the associations between family income, height and body mass index in US children and adolescents: 1971–1980 and 1999–2008

Abstract

Aim: This paper evaluates trends in the childhood and adolescent associations between family income, height and body mass index (BMI) between the periods 1971–1980 and 1999–2008.

Subjects and methods: US-born children (aged 2–11) and adolescents (aged 12–19) in the NHANES I, NHANES II and 1999–2008 NHANES are evaluated for BMI and height using flexibly-estimated structured additive regression models.

Results: Three trends are identified: BMI gains have been greater in lower-income relative to higher-income children and adolescents; height has increased more in lower-income relative to higher-income children (but not in adolescents); and BMI has increased more in taller children than shorter children (but not in adolescents). Following from these trends is that contemporary children exhibit a negative height-income interaction effect on BMI such that the inverse association between income and BMI is larger in taller children. Similar results hold when categorical obesity is considered but with modest height-related changes and income-height interaction effect in adolescents. Race–sex differences are explored and descriptive evidence on the potential role of changes in caloric intake is presented.

Conclusion: The relationship between the economic environment and growth in US youth has changed in recent decades. Contemporary taller and lower-income children exhibit greater body mass and obesity.

Keywords::

• BMI
• children
• height
• income
• obesity

Declaration of interest: The author reports no conflicts of interest. The author alone is responsible for the content and writing of the paper.

Notes

1. A third survey, NHANES III, was conducted over the years 1988–1994. This survey is not used here so as to facilitate a longer time differential between the periods being studied.

2. The functional estimation over PIR is limited to 10 knots as the range of values for PIR is narrow. Increasing the number of knots for PIR or height did not result in better model fit as determined by the deviation information criterion (Spiegelhalter et al. 2002).

3. The LPM has an advantage in this framework to facilitate the use of sampling weights in the BayesX software package where logit/probit specifications are not supported for such weights. Unweighted results were compared between the LPM and logit specifications with no notable differences.

4. The figures are constrained to height z-scores between − 2.5 and 2.5 as few children fall outside of this range to compare over both periods of time.

5. A separate analysis was also performed on adolescents aged 14–19. The coefficient on height was not significant for this group, suggesting that the height–BMI association fades during the teenage years. However, the marginal effect from height on categorical obesity (see probit results) was significant for this older group, as was the income–height interaction effect on obesity.

6. Due to the non-linearity of the probit model, the interaction effect is not given by the interaction coefficient β₃, but by the cross-partial derivative of the estimated equation with respect to any interacted variables X₁ and X₂. Ai and Norton (2003) show that this will equal . The statistical significance of this effect is tested by the significance of the overall cross-partial derivative rather than only the t-statistic on β_?. The specifics of the estimation procedure are given in Ai and Norton (2003). Interaction effects in equation (2) and their standard errors are estimated in STATA/SE 8.2 using the ‘inteff’ command module written by Norton et al. (2004). Average marginal effects for the non-interacted variables are estimated with the ‘margeff’ command in Stata SE (Bartus 2005).

7. Mexican-American and Hispanic children and adolescents were also evaluated for income–height interactions in the 1999–2008 data. Neither Mexican-American age group of either sex exhibits significant interaction effects for either z-score or categorical obesity. Hispanic males show negative income–height interactions on categorical obesity in both age groups. However, this is a small group (247 children, 256 adolescents) and even when they are excluded from the overall samples the interaction effects remain negative and significant.