- Comments and Discussion
Alan S. Blinder:
With one exception, which I will come to later, this paper is very readable, fact-based, and enlightening. It reminds me of one of those famous Yogi Berra quotations: "You can observe a lot just by watching." William Gale and Karen Pence watch the data on household wealth accumulation from 1989 to 2001 and observe two interesting phenomena that others have missed:
— The change in household wealth over this period was concentrated in the upper age groups, such as those 55 and older. The authors' figures 1 and 2 show this clearly and, by the way, indicate that the mean and the median exhibit the same basic pattern.
— These older groups did better because they "improved" their demographics, not because, for example, they held most of the stock as the S&P 500 climbed about 270 percent over those twelve years. Each of these observations is stated clearly and backed up by impressive data crunching. And, as the authors point out, the two findings are not just corollaries of the well-known fact that wealth has become increasingly concentrated. Something else was going on.
As I read the paper for the first time, I wondered why the editors asked me to be a discussant. Then I came to the first mention of the Blinder-Oaxaca decomposition, and it was "déjà vu all over again." Since macroeconomics-oriented readers may be unfamiliar with the Blinder-Oaxaca technique, let me just say that it is a simple, regression-based decomposition of the mean difference in the attainment of some left-hand-side variable (wages in my original application, wealth in this case) between two populations (blacks and whites in my original application, wealth holders in 1989 and 2001 in this case) into a portion attributable to differences in [End Page 214] the right-hand-side variables and a portion attributable to differences in the coefficients. So its use here, although probably unprecedented in the Brookings Papers, is appropriate.
Specifically, assume linear regressions explaining the wealth of individual i in year t, where t = 1989 or 2001:
The notation indicates that both the attributes X of individuals (such as health or marital status) and the regression coefficients ß on those attributes may change over time. The question is: How much of the change in mean wealth of one group versus another can be attributed to changes in X and how much to changes in ß? In this case the changes in X indicate how much individual i "improved" herself (changing her attributes so as to generate more wealth), and the changes in ß indicate how much the economy changed its valuations of those attributes (how much more or less wealth those attributes typically generated).
Conceptually, think of the "time derivative" of equation 1 as being ΔW = ßΔX + XΔß. There are two discrete-time decompositions:
In each decomposition the first term measures the portion attributable to changes in average characteristics between 1989 and 2001 (the major ones for Gale and Pence being education, health, and marital status), evaluated at one of the two "price vectors." The second term measures the portion attributable to changes in coefficients between 1989 and 2001, evaluated at one of the two "quantity vectors." Note that neither decomposition has any inherent claim to superiority. So, unless they lead to approximately the same conclusions, the evidence must be scored as inconclusive-analogous to when a Laspeyres index and a Paasche index give sharply different measures of inflation.
With this in mind, turn now to the upper panel of Gale and Pence's table 5, where the results from both versions of their Blinder-Oaxaca decomposition are displayed. The two versions of equation 2 agree quite well for four of the six age groups. In three of those cases (ages 25-34, 65-74, and 75-84), the story is that virtually all the action stems from changes in demographics-the second of Gale and Pence's findings above. [End Page 215] The two decompositions also agree for the 55- to 64-year-olds, but here the split is closer to 50-50 between changes in ß and changes in X. It is...