Abstract

In the first two chapters of this thesis I review the literature concerning health and obesity (indexed by the Body Mass Index, henceforth BMI) drawing on readings from sociology, epidemiology, economics and behavioral genetics. In particular, I focus on two issues: the socio-economic determinants of health and BMI and the socio-economic gradient in health and BMI, trying to integrate the standard sociological approach with findings from behavior genetics. The third is a methodological chapter in which I describe the data and the model I use in the following chapters. As far as the data, I use both cross sectional (the Multiscopo “Aspetti della vita Quotidiana”, ISTAT and twins data coming from the “Italian Twin Registry”) and panel data (the European Community Household Panel). In the fourth chapter I use behavioral genetics model and multilevel model in order to estimate the heritability of health status/BMI. More precisely, I try to understand which proportion of the observed inter-individual variation in health/BMI can be attributed to unique environmental factors (E), common environmental factors (C) or genetic factors (A). Then I run a Cholesky decomposition model in order to understand whether the observed covariance between education and BMI is due to common genetic/environmental factors. I find out that bivariate heritability is 0.30, meaning that around 30% of the observed covariance between these two trait is due to common genetic factors. Since genetic factors are, by definition, prior to both education and BMI this result prompts us to rethink the usual interpretation of the relationship between education and BMI as a causal one. In the final chapter I used ECHP data in order to study the socio-economic gradient in health/BMI and to understand whether the effect of socio-economic factors on health/BMI varies over the life course or across cohorts. I first apply growth curves models, a special case of multilevel model for change, that enable us to model individual health trajectories in health and BMI. More precisely, I model the differences in these trajectories (i.e. in both their intercepts and slopes) as a function of a set of explanatory variables (income and education) and other covariates (wave, age, cohort, gender, area of residence). I then compare the results obtained from growth curves (that are random effect models) with the ones from fixed effects models. The main problem of sociological studies that aim to estimate causal effects is the problem of unobserved heterogeneity. Standard sociological models assumed the predictor to be uncorrelated with the error term. However when we study health this assumption may not hold: we know that there are different unobserved or unobservable factors (genetic or psychosocial characteristic like different time preferences or risk aversion) that may affect both an individual’s socio-economic status and his/her health, creating a correlation between the predictor and the error term. Through a process of “time demeaning the data”, the fixed effect model is able to control for time constant individual heterogeneity and to correctly estimate the effect of individual income that varies over time. With this kind of model the effect of annual income is no more significantly related to health/BMI. However, with fixed effects models we can no longer estimate the effect of time constant variables like gender and education –very interesting from a sociological point of view- that do not vary over time (and, hence, are cancelling out from the equation in the process of time demeaning the data). For my analysis I used the software Stata 11 and SPSS (AMOS).