Gary Chamberlain, "Econometric Applications of Maxmin Expected Utility", Journal of Applied Econometrics, Vol. 15, No. 6, 2000, pp. 625-644. The file log_earn_resid.txt contains the data. It is an ASCII file, in DOS format, consisting of 1000 lines of data with one number in each line. This file is zipped in the file gcdata.zip. Our sample is drawn from the Panel Study of Income Dynamics (PSID). We work with a sample of males who were heads of household at ages 24 through 33. The earnings measure begins in the calendar year prior to the interview when the individual was 25 years old. We assume that ``serious'' labor force attachment begins in that year (t=1). The selection criteria are that the sample individuals be in the labor force, not in school, with positive earnings for the T = 10 consecutive years beginning with t=1. In addition, there should be no missing values for race or education. Note that t=1 may refer to a different calendar year (between 1967 and 1991) for different individuals. We work with a subset of this sample consisting of N = 100 white, high school graduates, who were not part of the Survey of Economic Opportunity (SEO) sample, which oversampled poor households. Y_{it} is the logarithm of annual labor earnings for individual i in his year t in the sample. The earnings are in 1991 dollars, based on the urban consumer price index. Log earnings residuals are formed by subtracting out the mean separately for each year t of experience: log_earn_resid_{it} = Y_{it} - \sum_{i=1}^N Y_{it}/N. The 100 by 10 matrix is converted to a vector by stacking the columns: log_earn_resid_{11}, ..., log_earn_resid_{N1}, ..., log_earn_resid_{1T}, ..., log_earn_resid_{NT}. The ascii file log_earn_resid.txt contains the 1000 elements of this vector, with one number per line.