Appendix

The Survey of Income and Program Participation

The estimates of wealth reported in this study are derived from the Survey of Income and Program Participation (SIPP). Conducted by the Census Bureau, SIPP is a longitudinal survey, i.e., it is designed to follow the same panel of households over several years. The data for this report are from the 1996 and 2001 panels of SIPP. The 1996 SIPP panel, initially consisting of over 90,000 individuals, was interviewed from April 1996 to March 2000. Interviews for the 2001 panel, initially featuring over 70,000 persons, began in February 2001 and ended in January 2004. Each SIPP panel is asked two sets of questions over time. A “core” set of questions is repeated with each round of interviews, called a “wave,” and a rotating set of questions, called “topical modules,” is asked at intermittent waves. The 1996 SIPP panel had 12 waves of interviews and the 2001 panel had nine waves of interviews. The topical modules on wealth for the 1996 SIPP panel were conducted during Waves 3, 6, 9 and 12, starting in December 1996 and conducted a year apart. Wealth data from the 2001 panel are currently available from Waves 3 and 6 conducted at the end of 2001 and 2002 respectively.35

The entire SIPP panel is not interviewed simultaneously but over a four-month period. One fourmonth period constitutes a wave. For example, Wave 3 of the 1996 SIPP refers to a set of interviews conducted from December 1996 to March 1997. In each round of interviews, respondents are asked to provide answers pertaining to the preceding four months. Thus, an interview conducted in December 1996 would collect data for August 1996 to November 1996, the so-called reference months. In the case of wealth, data are collected only for the fourth reference month. In other words, persons interviewed in December 1996 would report wealth data for November 1996, those interviewed in January 1997 would report wealth data for December 1996, and so on.

The time period references in this report pertain to four-month time periods and the specific months covered by each year mentioned in the report are as follows: 1996 – November 1996 to February 1997; 1997 – November 1997 to February 1998; 1998 – November 1998 to February 1999; 1999 – November 1999 to February 2000; 2001 – September 2001 to December 2001; 2002 – September 2002 to December 2002. This nomenclature differs slightly from that used by the Census Bureau in its wealth reports. More specifically, the Census Bureau uses 1998 to refer to November 1997 to February 1998 and 2000 to refer to November 1999 to February 2000. In other words, the Census Bureau prefers to place an emphasis on the beginning of a year whereas this report emphasizes the end of a year. Drawing attention to the end of the year is preferred in this report because of the use of both the 1996 and 2001 SIPP panels. The reference months for the wealth data from the 2001 SIPP panel are indisputably from the end of a year.

Most of the estimates in this report are from SIPP “cross-sections”, i.e., they are from the samples of households present in any particular wave. After data editing, the household sample sizes for the 1996 panel were as follows: Wave 3 – 33,899, Wave 6 – 30,805, Wave 9 – 29,058, Wave 12 – 28,269. For the 2001 panel, there were 27,401 and 26,601 households present in Waves 3 and 6 respectively. Wealth estimates for immigrants were prepared from subsets of these samples. The nativity status of individuals was ascertained only during Wave 2 interviews in both the 1996 and 2001 SIPP panels. Therefore, the wealth of immigrants, as determined in Waves 3, 6, 9 and 12, is only known for those who were also interviewed in Wave 2. This additional restriction led to the following household sample sizes for the 1996 panel: Wave 3 – 33,205, Wave 6 – 30,283, Wave 9 – 28,613, Wave 12 – 27,857. For the 2001 panel, the numbers of households present in Waves 3 and 6, as well as in Wave 2, were 26,439 and 25,452 respectively. The smaller subsets of households, i.e., those for whom immigration data are known, are also used in section 3 of the report.

Sample size diminishes over time because some households inevitably exit the sample for a variety of reasons. This can cause a bias, known as “attrition bias,” in the estimates over time if the households leaving the sample are systematically different from those remaining in the sample and those differences are related to household wealth. An examination of the characteristics of the SIPP panel at different points in time showed little cause for concern. Sample characteristics that might be related to wealth, such as, age, income, education, and gender, did not show any unusual trends.

A related area of concern was the “sample effect.” In this study, the wealth estimates for 1996, 1997, 1998 and 1999 are derived from the 1996 SIPP panel, and the wealth estimates for 2001 and 2002 are based on the 2001 SIPP panel. Therefore, there exists the possibility that differences in the compositions of the 1996 and 2001 SIPP panels, and not just the 2001 recession, are partly responsible for shifts in estimated wealth between 1999 and 2001. A key difference between the two panels is that the 1996 panel, by definition, does not incorporate immigrants who entered the U.S. in 1996 or later years. As discussed in the main body of the report, that led to an adjusted set of estimates derived from a subsample of the 2001 SIPP panel that excludes immigrants who arrived in the U.S. after 1996. When those immigrants are excluded, the resulting household sample sizes for the 2001 SIPP panels are as follows: Wave 3 – 26,009, and Wave 6 – 25,064.

The unit of observation in SIPP is a person but the wealth estimates are derived for households. Therefore, one issue with the estimation of household wealth is the identification of household characteristics. For example, how does one decide whether a household is Hispanic or immigrant? Does everyone in the household have to be Hispanic or foreign-born? The method used in the report is to equate the characteristics of the head of the household with the characteristics of the household. Thus, households are classified by the race, ethnicity, nativity, education, age, etc. of the head of the household. The same method is also used by the Census Bureau in preparing its estimates of wealth.

A principal alternative to using SIPP is to use data from the Survey of Consumer Finances (SCF). The SCF is a triennial survey conducted by the Federal Reserve Board. The three most recent surveys were conducted in 1995, 1998 and 2001. The SCF is more comprehensive in terms of the varieties of assets covered. For example, SCF collects data on equity in pension plans, and the value of household items, such as, artwork and jewelry. However, the SCF has a relatively small sample size, it over samples relatively wealthy households, and it does not record the immigration status of a person. SIPP offers a much larger sample size, it over samples the relatively poor, and it inquires about immigration status. These factors make SIPP much more useful for a study of the wealth of minority groups and foreign-born households. Because of the sample differences, the level of wealth estimated from SCF tends to be higher than the level of wealth estimated from SIPP. But there are also many similarities with respect to changes in wealth over time and the relative wealth of minority groups. Those are pointed out in this report at the relevant points.