Originally, SIPP did not oversample any groups within the population. Over the years, however, budget constraints dictated a reduction in the SIPP panel size. As a result, analysts found it difficult to conduct meaningful analyses of government programs for the low-income population because the sample sizes for the subpopulations were too small. In response to those concerns about the diminished usefulness of SIPP data, the Census Bureau pursued budget initiatives to increase the sample to its original size and to oversample the low-income population. Oversampling occurs when certain groups or units are sampled with higher probabilities than others. Analysts then have enough cases to complete analysis of subpopulations or subgroups of the population. The share of an oversampled group in the resulting sample is greater than its share in the population from which it was drawn. Although this imbalance addresses the need for increased sample sizes for certain subpopulations, analysts looking at the entire sample will need to use weights in their analyses to redress the imbalance (Chapter 8 of the SIPP Users' Guide).7
As detailed in the SIPP Quality Profile and discussed in Allen et al. (1993), oversampling was used with the 1990 Panel, which included about 3,900 predominantly low-income households from the truncated 1989 Panel (see Tables 2-1 and 2-4). In the 1990 Panel, the Census Bureau included all housing units from Wave 1 of the 1989 Panel in which the head of household was black, Hispanic, or female with no spouse present living with relatives (FHNSP). Such households tend to have higher poverty rates than the general population. The 1990 Panel also included a small sample of other housing units for the 1989 Panel. Table 2-4 shows the components of the 1990 Panel.
|Households in addresses originally to be interviewed first in the 1990 Panel||19,700|
|Households associated with sample addresses first interviewed in February through May 1989 (in the 1989 Panel ) and at the time headed by a black, Hispanic, or FHNSPa||2,700|
|Households in one-ninth of all other 1989 Panel sample addresses||1,200|
The Census Bureau also oversampled the low-income population for the 1996 Panel,8 using 1990 decennial census information. Housing units within each PSU were split into high- and low-poverty strata. If the housing unit received the Census long form that included income questions, the unit.s poverty status was determined directly; for other housing units, poverty status was assumed on the basis of responses to Census short-form items predictive of poverty rates. The Census Bureau then sampled the low-income stratum at 1.66 times the rate of the high-income stratum in each PSU. Compared with the number of cases produced without oversampling, this oversampling produced an 18 percent increase in the number of cases in and near poverty at Wave 1.9 Even greater gains occurred in some subgroups, such as blacks and Hispanics in poverty, with a gain in the number of sample cases as high as 24 percent. However, the increases in effective sample sizes were somewhat smaller after allowance was made for the increased variance associated with differential weighting. Also, the sample sizes for the higher income and higher age groups were reduced.
Introduction to SIPP |
SIPP Survey Content |
Technical Information |
Using & Linking Files |
SIPP Publications |
| Access SIPP Data | SIPP Users' Guide | SIPP Tutorial | User Notes/ListServe/News | SIPP Help |
Page Last Modified: May 9, 2006