It oftentimes happens in research into matters of social equity that there is a desire to estimate timeseries of mean access to social goods (such as higher education or health insurance) by quantiles of the income or asset distribution. Are the poor catching up the rich, are the rich getting richer, or are the gaps stable over time?
However, it is rare for questionnaires to ask for exact values of income or assets. There are exceptions such as the SIPP or the March supplement to the CPS, but most surveys will not have that level of income/asset detail. So for sample persons in the bins containing the quantiles, it will be unknown whether the person belongs the lower or upper quantile involving the bin. One particular example is educational attainment of young people by household income quartiles. The CPS education data are part of the October supplement to the CPS and it has only rough categories for income. If there were a good way of imputing continuous income, then quantile membership would follow, and producing the data points of the timeseries would be easy. One could also think about imputing quantile membership directly or some maximum likelihood solution to the problem, but imputation of continuous income seems easiest. Of course, one would want to repeat the imputation process (probably a large number of times) to reduce the variance. Anyone ever see such a thing or know of any other way of producing income-quantile dependent means of variables of interest? (It was hard for me to even come up with the right language for the datapoints in these timeseries. If others have heard of other names for them, please let me know.) Tom Mortenson of the Pell Institute for the Study of Opportunity in Higher Education has been producing time series of this nature for a number of years with a rather ingenious type of linear interpolation to the graph of mean educational outcomes by income bin, but I'd like something with less of an ad hoc feeling to it. David Judkins Senior Statistician Westat 1650 Research Boulevard Rockville, MD 20850 (301) 315-5970 [email protected]
