The Covariance Structure of Earnings and Income, Compensatory Behavior and On-the-Job Investments
Observationally alike individuals who make different choices about on-the-job investments should have earnings profiles that differ in systematic ways. In particular, investments in non-specific human capital should result in lower initial earnings but higher earnings growth rates. Human capital models of this sort admit testing, then, by examining the covariance between the level of earnings and the growth rate of earnings. This paper reports estimates of this covariance using the sample covariance among income observations across time for the same individuals. The sample covariances are drawn from the Utah Panel Data, a panel of some 16,000 households with income and wealth observations at various intervals over the period from 1850 to 1900. The parameter of interest is negative. This estimate is robust to various specifications of the model. I also reexamine earlier work by Lillard and Weiss and Hause, who use data on earnings, and conclude that there is strong support for the on-the-job investment hypothesis using data from thre equite different sources covering different economies and different time periods.