difference in saving rates is attributable to the fact that average saving rates in the data rise with income and African Americans have lower incomes than whites, rather
than to whites having a higher saving rate conditional on income level.
II. Data
A. Measurement of Wealth
The main source of data used in this study is the PSID, which has followed about 5,000 U.S. families since 1968, interviewing them annually. As the PSID was origi-
nally based on two subsamples, one of which oversampled the low-income popula- tion, the use of weights enables analysts to make estimates that are representative
of the nonimmigrant U. S. population and its subpopulations. We use family weights for all calculations, even for multivariate analysis, as did Hurst, Luoh, and Stafford
1998, in their examination of the same data. While the use of weights for regression analysis is not without controversy, their application seems imperative here. As Hill
1992 notes in her monograph on the PSID, one should not use the low-income sample without weights if the dependent variable is income or another variable di-
rectly related to the PSID’s oversampling criteria.
Data on wealth were collected via special supplements carried out in 1984, 1989, and 1994;
3
a sequence of questions falling under the PSID rubric “active savings,” used to collect information on ows of money into and out of different assets, was
included in 1989 and 1994 Hurst, Luoh, and Stafford 1998; Juster, Smith, and Stafford 1998. For the purposes of this study, the PSID has several key advantages
over other data sets available to study race differences in wealth. First and foremost, given that families are followed over time and that questions are asked about move-
ments into and out of assets, one can, subject to certain caveats that will be discussed subsequently, attribute changes in net worth over time to components due to intergen-
erational transfers, saving, and capital gains. Second, in part because the PSID con- tains an oversample of the low-income population, the number of African American
families is larger than in the Survey of Consumer Finances SCF. Third, presumably owing to the rapport that PSID interviewers have developed with respondent families
over time, the rate of item nonresponse in the wealth questions is relatively low, no small consideration given the reluctance of many families to divulge information on
their net wealth Hurst, Luoh, and Stafford 1998.
We would be remiss, however, if we did not note some important limitations of the PSID data. Given that it was not designed as a wealth survey, the PSID does
not take steps to oversample the richest of the rich, necessary to obtain precise esti- mates of wealth for those in the upper tail in the distribution. Thus, with respect to
those on the right tail, estimates from the PSID are unavoidably less accurate and less precise than those from the SCF, which does oversample the very rich. Alleviating to
some extent concerns about the upper part of the distribution, Juster, Smith, and
3. At the time of the analysis, an early release of the 1999 PSID wealth data was available, but it was not used here because much of the data needed to study wealth accumulation between 1994 and 1999 had
not yet been released and because of concerns about comparability owing to changes in questionnaire and sample. See Lupton and Stafford 2000 for additional details.
Stafford 1998 nd that the PSID wealth data for 1989 stack up well next to those from the same year’s Survey of Consumer Finances through the 98
th
percentile. A second limitation of the PSID is that wealth components are grouped into seven
broad categories or eight if one includes net equity in the home, information on which is collected annually, just a small fraction of the number of categories in the
SCF; more detailed questions and the probing that accompanies them are thought to lead to a more accurate reporting of wealth.
Net worth is measured in the PSID by adding the net values of the home, real estate other than the main residence, the farm or business, and vehicles together
with holdings in stocks, checking and saving accounts and “other savings,” and then subtracting nonmortgage debt. Details on the assets and liabilities included in each
category can be found in the appendix. The appendix may be accessed on the JHR website:
www.ssc.wisc.edujhr .
The fact that social security and pension wealth are excluded is an important caveat to keep in mind when interpreting the results, a
point to which we return in the conclusion. In order to understand in some depth how wealth accumulation and its components
differ by race, it is essential to have information not only on family wealth at different points in time, but also enough additional details to determine the path a family
followed in order to arrive at its net worth.
4
Questions about the market value of the main home and the remaining principal of the mortgage are asked each year. In
1989 and 1994, in its active saving questions, the PSID also asked respondents about a number of different types of nancial transactions over the previous ve years
including: the amount invested in a business, in stocks or in real estate other than the main home; the value of additions to the main home or other real estate; and
the value of gifts or inheritances. Details on the active saving questions are contained in the appendix on the JHR website.
Information on asset levels and ows enables a division of changes in net worth into saving, capital gains, and transfers. For each asset, the change in value over a
period can be divided into two parts—the capital gain and the amount of additional funds invested in that asset, which we term gross saving. Details of the algorithm
used are contained in the appendix on the JHR website, but the basic approach is as follows: For those assets for which the amount of the net inow—that is, the
gross saving in that asset—is known, it is straightforward to calculate the capital gain, as it is simply the difference between the asset’s change in value over the
period and the net inow. For assets for which nothing is known about the net inow, an appropriate market-based rate of return is assigned in order to calculate the amount
of the capital gain, and, for these assets, the amount of gross saving is calculated as the difference between the change in the value of the asset over the period and
capital gains on the asset. Summing over the group of assets, one arrives at a total for capital gains and one for gross saving.
4. Kennickell and Starr-McCluer 1997a use the 1983– 89 panel of the Survey of Consumer Finances to study household wealth accumulation, but they use change in net worth as their denition of household
saving.
As the description indicates, gross saving is not the same as saving traditionally dened as the difference between income and expenditures, because gross saving
can be funded by any source of funds, not just income. As a result, to arrive at an estimate of the amount of saving that comes directly out of income, it is necessary
to subtract from gross saving the other ows into the household, the largest of which is inheritances and other gifts. We will refer to gross saving minus inows from
inheritances and other sources simply as saving. It is worth stressing that, though closely related, our measure of saving differs from the concepts of active saving
referred to in PSID literature and used in the analyses of Hurst, Luoh, and Stafford 1998 and Juster, Smith, and Stafford 1998. There are two main differences be-
tween our approach and that detailed in Survey Research Center 1992. First, we divide the change in net equity of the home less the value of any home improve-
ments into a portion that is saving the reduction in mortgage principal and one that is capital gains the change in the value of the house, rather than attributing
the entire amount to capital gains. Second, for the assets for which there are no specic questions about inows—those other than the main home, other real estate,
farmbusiness, and stocks—the PSID approach implicitly assumes that changes in net value are entirely attributable to saving, rather than allocating some portion into
capital gains by applying a market-based rate of return, as we do. In any case, with either method, it is apparent that saving can potentially ow into any assets, and not
just those for which there are specic questions in the PSID’s active saving module.
B. Sample Selection
