IG correlations in pension participation tutorial

1 Intergenerational (IG) Correlations in Earnings Ngina Chiteji, Elena Gouskova and Frank Stafford I. Overview Intergenerational analysis represents an important area of research for social scientists who study health or labor market outcomes. Researchers in this field have discovered that, upon reaching adulthood, many children have outcomes that are similar to their parents'. For example, existing research on earnings reveals that the elasticity of sons' earnings with respect to fathers'--often called the "intergenerational income elasticity"--is about 0.4 in the United States (Solon 1992, 1999; Lee and Solon 2006). This result implies that, on average, the earnings of a son whose father's income was $50,000 a year would be expected to be about 40 percent higher than the earnings of someone else whose father earned only $25,000 a year. Hence, a son's earnings are positively correlated with his parent's, suggesting that high- income parents will have high-income children, even though one might think it more natural to expect earnings to be determined solely by individual characteristics (not by family or class background). Researchers have found similar correlations across generations in other labor market outcomes, such as occupation, union membership, hours worked (Tabb, 2004), and participation in pension plans (Treiman and Robinson 1981; Blanden and Machin 2003; Gouskova, Stafford and Chiteji, 2006). In the areas of health we know that there is a substantial correlation in rates of obesity between children and their parents and grandparents (Kim McGonagle and Stafford, 2001). This tutorial will teach you how to conduct these types of analyses. The tutorial uses cross-generational connections in earnings of Baby Boom males and their dads as a case study in order to highlight key features of the Panel Study of Income Dynamics (PSID) that make it one of the premier datasets used for intergenerational analyses. This tutorial shows how to construct a file by merging data from two generations when the objective is to see the measures at the same age or life cycle point. In the case of earnings it is well-known that earnings peak later in the life cycle. “Since earnings reach a plateau at later ages in the most highly educated groups, both dollar and relative annual earnings differentials among schooling groups grow with age until 45-50, and later still for weekly earnings.” (Mincer, 1974, p. 70). In addition to differences by gender, race and education, data on obesity rates as measured by Body Mass Index (BMI) also have a life cycle signature. The pattern is better represented in a semi-parametric model than the quadratic often used in earnings. Cross-sectional BMI rates are rather flat over the 20’s then rise quite steadily to age 40 and have more of a plateau but some rise from age 40 onward as shown in Figure 1 for white males by education level (Kim, McGonagle and Stafford, 2001). In this case too, the study of intergenerational BMI could benefit from the observation of BMI at specified ages rather than the ages of the different generations at a given time point. In the case of the PSID, BMI was measured in 1986 and as recently as 2005. This allows the researcher to align the ages of the different generations to some extent. Labor earnings have been measured throughout the PSID, 1968 to present. This allows better life cycle alignment across the generations and, as we will show, this better alignment is of great importance for studying the intergenerational correlation in earnings.  2 In this tutorial we demonstrate how one can use the earnings data contained in the PSID--which Figure 1. Age B M I Profile for White M en: By Education 30 292929 High school 28 27 26 College 25 24 252525 272727 222999 313131 333333 333555 373737 333999 414141 434343 444555 474747 494949 555111 535353 555555 575757 595959 666111 636363 Age represent an example of data that the PSID has collected every survey year--to estimate the extent to which earnings are correlated across generations. More generally, this case also serves to illustrate how researchers conduct intergenerational analyses, and it enables the tutorial to highlight some of the challenges that researchers encounter when attempting to study labor market, wealth or health outcomes in a multi-generational setting. The tutorial is lengthy. So, here is a guide to help you understand how it is structured (in case you feel there are sections you want to skip or de-emphasize). And you can always skip past or back to a section or subsection by using the selection bar on the left of your screen. Or you can download the document and companion files in PDF or Microsoft Word format if that is more convenient. Section II provides an overview of the PSID. If you already have done some of the other tutorials, you probably know a lot about the PSID already and you can breeze through Section II. Section III discusses the PSID's labor income data, and explains how one can navigate the data files to learn what data the survey has collected over the years and archived in the Data Center. It concludes (in Part III-D) with a general discussion of the different research strategies and methodological issues that a researcher might contemplate when setting up an intergenerational study, and a discussion of the explicit strategy that the tutorial will adopt. Section IV --by far the longest section of the tutorial--presents two different examples of ways intergenerational analysis can be conducted. For each example the tutorial shows the steps you need to take in order to get the data that you need to test the hypothesis identified above. Upon completion of each example you will have an actual numerical estimate of the degree to which fathers' and sons' earnings are associated. The first example (Example 1--presented in IV-B) illustrates an approach to conducting intergenerational analysis that relies on a measure of earnings taken from a fixed calendar year for each generation. The second example (Example 2--presented in IV-C) shows how to construct measures of earnings that incorporate information from many years of a worker's life, instead Before getting to these examples however, Section IV starts with a discussion of a new PSID tool that allows one to match fathers and sons to each other (Part BMI 3 A of Section IV). Note that you cannot do either example without going through the steps needed to match fathers and sons, so make sure you definitely read Section IV-A. II. Background about the PSID Why use the PSID for intergenerational analyses? The PSID is a nationally representative sample of U.S. individuals (men, women, and children) and the family units in which they reside. (See the overview for more about