Barbara Wolfe asked Hurd: If you compare the distribution of prime age individuals, say 25-45 years old, do you know what the distribution of the HRS looks like? What proportion, for example, in the lowest quintile, are actually in the HRS? Has anyone ever tried to do that kind of comparison? How useful would some of the numbers presented be, with respect to the entire age distribution? The HRS is a really rich data set, but if it misses the lowest tail, then it is less promising.
Hurd responded that the poverty rate in the HRS is very close to the CPS poverty rate, within half a percentage point. He explained that they reweight to CPS totals along a number of dimensions; and that has been studied a lot, and there is no known substantial bias in HRS recruitment. The baseline response rate was 80 percent. Very large differential nonresponse by some variable is therefore needed in order to get a lot of bias when the response rate is that high.
Wolfe wanted to emphasize that, in the work that she and colleagues have done at the University of Wisconsin, they looked at the risk of having income go below the poverty line or move into near poverty if someone has a health or cognitive effect. They found that the risk is strong for those who are very close to the poverty line, but it is not relevant for people who are at 400 times or even 300 times the poverty line. So the risk is important for a small group.
Although she is sure there is a lot of overlap between people with less than a high school education and those with low income, if he did that by initial level, maybe in the first year they are in the survey, or some average, he will probably find higher risk for people with low income than he calculated using education.
Hurd explained that they thought about doing the analysis by income or wealth but decided not to because of the classification error on income and wealth. They do as much as they can, but they have to admit that it is inaccurate. As a number of the presentations today have shown, people with low incomes have a high ratio of spending to income.
Michael O’Grady (NORC at the University of Chicago) asked Jessica Banthin when she moved to adjusted income and used 5 percent of assets, was there anything special about 5 percent or she just needed to pick something to move forward? Also, because part of what has been discussed or at least implied is that different assets are more or less fungible or available, or even originally planned to be used for things like health care spending, what are the options in terms of thinking about different kinds of assets using MEPS?
Banthin explained that, regarding the adjusted income and 5 percent of assets, that was just a rough approximation. She wanted to move forward and give a simple approach to provoke this conversation. She repeated that a more careful analysis would have separated by retirement status, not age.