The Logic and Practice of Growth Curve Analysis: Modeling Strategies for Life Course Dynamics

Abstract

The expansion of a life course perspective has occurred alongside the development of a number of important statistical innovations. One of particular significance is growth curve analysis for modeling heterogeneity in trajectories over time. In this chapter, we articulate the logic and practice of conventional growth curve analysis and tie it to a set of key theoretical principles in life course social science. Following a brief discussion of the core statistical elements, we use the example of trajectories of body mass as an example of how a life course perspective can be applied to the key social problem of excess weight and obesity. Using data from the National Longitudinal Survey of Youth – 1997, we examine the social dynamics associated with changes in body mass in the transition to adulthood as a vehicle for elaborating how growth curve analysis works and how one interprets the key statistical quantities. In doing so, we stress the important affinities between key components of life course social science and a growth curve approach as a basis for important empirical assessment of developmental processes over time.

Appendix: Data and Measures

The data that we use in this research come from the National Longitudinal Survey of Youth – 1997 (hereafter NLSY97). The NLSY97 consists of an initial sample of 8,984 youths who were between the ages of 12 and 16 in 1997. When possible the respondents were re-interviewed annually and data were collected on a range of topics on the transition to adulthood. As of 2014, there are 15 waves of data that cover an age range of 12 to 31. In addition to non-Hispanic whites, the NLSY97 oversampled blacks and Hispanics such that there are relatively large samples of six race-sex groups. Compared to other national surveys, panel retention is excellent with 83 % of the sample retained at wave 15.

For a study of the dynamics of BMI, we capitalize on the record structure of the NLSY97 data and its position in the history of population health in America. For the former, the multi-panel record structure provides annual, repeated measures of self-reported height and weight, coupled a rich set of time-stable and time varying measures. In the latter case, the obesity epidemic in the United States has had profound effects on the age structure of health liabilities. As Harris (2010) notes, numerous data, including studies such as the National Longitudinal Study of Adolescent Health show unequivocally that obesity is harbinger of both short-term and longer-term chronic health problems and that a range of serious health problems (e.g., type II diabetes, hypertension) are increasingly visible through the early adult years.

The key outcome of interest is annual measures of respondent’s body mass index score or BMI constructed from self-reported height and weight. There are clear anomalies and apparent coding errors in a small subset of person-periods and these introduce a number of extreme and unrealistic values for BMI. We deal with these by setting all values less than 12 and all values greater than 50 to be missing.

The central parameter of interest from which we extrapolate interpretation is aging. This is a measure that indexes the passage of time and when included in a linear model fit to panel data captures the nature of change over time in the independent variable. This measure is the key parameter in that we can elaborate its effects in a number of ways to study the ways in which various life course dynamics are implicated in stability and change in BMI over time.

We capture various aspects of social structure by measuring race -sex group that differentiates white males, white females, Black males, Black females, Hispanic males, and Hispanic females. We also include the highest level of father’s educational attainment based on the highest attainment of either the residential father in the household or the biological father if the former is missing. We capture an alternative measure of social structure and stratification through the respondent’s educational attainment. Although there are a number of conceptualizations, we treat attainment as a set of dummy variables indexing ‘high school/GED,’ ‘some college,’ a ‘two-year degree,’ or a ‘four-year college degree or greater’ with the reference category being ‘less than a high school degree.’ This allows us to capture a range of meaningful contrasts in education as they relate to health and allows us to assess linearity or consider nonlinearities if apparent.

Finally, we capture key life course transitions and their effects on BMI. These include school enrolment, independent residence, employment, marriage, marital disruption, and parenthood. Each of these is measured as a dichotomous time-varying variable.