The following items represent changes in the estimation procedures of state and county income and poverty estimates from the 2000 estimates to the estimates for 2001 and 2002. These changes reduce the comparability of the estimates between years and should be considered when making such comparisons. (See General Cautions about Comparisons of Estimates.)

The most important change to the state and county models was the acceleration of the production schedule by one year to produce estimates for 2002 as well as 2001 in 2004. The model-based estimates of income and poverty for state and counties are now available with only a one-year lag, measured from the release of national estimates from the Annual Social and Economic Supplement (ASEC) of the Current Population Survey (CPS) for the relevant reference year.

We were able to make this change due to a successful effort to acquire input data on an earlier schedule. Specifically, we were able to acquire the food stamp data from the Food and Nutrition Service that are used in the county poverty models a year earlier than in previous years. As a result of obtaining the food stamp data on an earlier schedule, we also changed to using an earlier version of the state and county demographic population estimates. In previous years, we used the "second vintage" release of population estimates that included revisions from their first release. Although this wasn't necessary, we did so because the revised data were available and were generally thought to be more precise. Unpublished internal research on the impact of changing to the unrevised or "first vintage" estimates indicated minimal impact on the SAIPE program estimates from this change. Regardless of the impact on the estimates, we believe it is appropriate to use the first vintage population estimates since they are the official population estimates and are used for other purposes by the Census Bureau and others.

Some changes were made to the CPS ASEC sampling weighting scheme that affected the direct CPS ASEC estimates for income year (IY) 2002. Though we did not apply the poverty and income models any differently because of these changes, we did modify our approach to estimating sampling error variances of the direct CPS ASEC estimates in two ways. First, in fitting sampling error models to direct estimated CPS ASEC sampling covariance matrices for 2001-2002, we estimated design effect parameters for 2002 that were distinct from those for 2000 - 2001 (within each age group). Previously design effects were assumed constant over time (but not age) when fitting the sampling error models. Out of concerns that the weighting changes in 2002 might affect the sampling variances, we allowed for distinct design effect in 2002. Second, the direct variance estimates for 2002 used in the sampling error modeling did not fully reflect the CPS ASEC weighting changes, as we did not have enough time to modify the variance estimation program to accomplish this. We were, however, able to obtain some variance estimates that did fully reflect the 2002 CPS ASEC weighting, though these variance estimates were approximate in other respects. Comparing these variances to others produced via the same methodology but without fully reflecting the weighting changes provided measures of the effects on variances of the weighting changes. These measures were used to multiplicatively adjust the 2002 poverty ratio variances obtained from the fitted sampling error models to reflect the estimated effects on variances of the weighting changes. The adjustments were constant over states within each age group. Such adjustments were not made to the sampling variances for the median income estimates because the approach used to obtain approximate variances (which used a linearization approach) was thought to be less appropriate for the median income estimates.

The effects of these variance adjustments were typically small. Design effects estimated for 2002 did not differ substantially from those estimated for 2000 - 2001. Also, the variance adjustments discussed had mostly small effects. They decreased the estimated sampling variances for age 18 - 64 poverty ratios by about 10 percent, but effects on the other age groups were only about a 3.5 percent reduction or less. In addition, sampling variances are updated as part of the iterative estimation scheme used for the poverty ratio models as described in the paper "Accounting for Uncertainty About Variance in Small Area Estimation," Bell (1999). This iterative updating almost certainly had more important effects in the sampling variances than the adjustments just discussed.

Due to the small number of years of sampling variances used in the sampling error modeling, and to the estimation of different design effects for 2002 versus 2000 - 2001, we decided not to attempt to estimate random state effects on the variances. The same decision was made for the production of the 2000 estimates last year, but this differed from what was done prior to 2000.

Sampling variances for IY 2001 CPS ASEC estimates were taken from sampling error modeling results for 2000 - 2001 produced last year, and were not affected by the issues just discussed. For further discussion of the sampling error models see "Sampling Error Modeling of Poverty and Income Statistics for States," (Otto and Bell, 1995).

There were no methodology changes to the county models.