The effects of old and new media on children’s weight

Abstract

Childhood obesity rates have recently been rising in many countries. It has been suggested in the literature that changes in children’s media exposure may contribute to explaining this trend. I investigate whether or not this hypothesis is supported by data. I contribute to the literature by focusing not only on television but also on new media – computers and video games. The Child Development Supplement to the Panel Study of Income Dynamics is used for the analysis. To address the endogeneity of children’s media exposure, I use dynamic and panel data models. This is another improvement upon the existing literature. Additionally, an extensive list of control variables is included in the regressions. I find that video game playing or computer use has no effect on children’s body weight. On the other hand, television viewing may increase children’s body weight slightly.

Keywords:

• obesity
• body weight
• media
• children

JEL Classification:

• I12
• J13

Acknowledgements

This study uses the data from the Panel Study of Income Dynamics. I am grateful to Denise Doiron, Hong Il Yoo, Charlene Kalenkoski, Susumu Imai, David Johnston, Deborah Cobb-Clark, Michael Shields and the participants of the 33rd Australian Conference of Health Economists, the 2011 PhD Conference in Economics and Business, and Empirical microeconomics group meeting at UNSW for their useful comments and suggestions. All opinions and any mistakes are my own.

Notes

¹ Basal metabolic rate is the amount of energy consumed at rest.

² I also attempted to estimate a dynamic correlated RE probit to account for persistence in overweight status, but unfortunately it was not possible to achieve convergence (likely due to small sample size).

³ All observations with missing information on weight, height or media time are omitted from the sample. I deal with missing values of control variables in two ways. If a variable has less than 1% of values missing, the observations with missing information on this variable are omitted from the sample. If a variable has more than 1% of values missing, a dummy variable for missing values is created and included in the regressions (to preserve sample size).

⁴ A child’s standardized BMI is calculated according to the following formulas: if and if , where M is the median of the BMI distribution corresponding to a child’s age and gender, S is the generalized coefficient of variation, and L is the power in the Box-Cox transformation, which accounts for the skewness of the BMI distribution (Centers for Disease Control and Prevention, 2002).

⁵ Not all of these variables can be included in all models. For example, the coefficients on the time-invariant variables cannot be identified in the child FE model.

⁶ The overall R-squared in the FE models is the R-squared of a ‘dummy variable’ regression that includes the FE as regressors.

⁷ The differences in the estimates across models are not explained by the differences in the samples. Estimating the child FE model on the same samples as the other models provides results that are very similar to the child FE estimates obtained using the full sample.

⁸ The coefficients in panel C are interpreted as the effects of an 1-hour per day increase in media time on standardized BMI. In Table 2 and panels A and B of Table 4, the coefficients are interpreted as the effects of an 1-hour per week increase in media time. That explains why the coefficients in panel C are larger in magnitude.