The differences between the original IY 1993 estimates and the modified estimates released in January, 1998 were as follows:

• We changed one predictor variable in the county-level model for related children ages 5-17 in families in poverty. The change replaced the 1994 population under age 21 years with the 1994 population under age 18 years. In the model, this population variable, in conjunction with the number of dependent child exemptions reported on income tax returns, serves to represent children in families too poor to file tax returns.
• We replaced the 1994 series of resident and group quarters population estimates by age for states and counties in 1994, with the 1996 series of estimates of the population in 1994. These estimates are based on improved methodology and revised input data for 1994. (Go to Revised population estimates.)
• We changed the statistical technique of estimating the coefficients in the models. The income and poverty estimates issued in April 1997 used weighted least squares at the county level. The new estimates use maximum likelihood at both the state and county level.
• We corrected a misrepresentation of the uncertainty in the 1990 Census estimates of poverty in 1989.

While the practical effect of these changes on estimated numbers of people in poverty for income year 1993 was relatively small, the impact was substantial for particular places. Because the estimates are model-based, even a small change in a single input has some statistical impact on the estimates for every place. Our research suggested these changes would reduce statistical bias and improve the precision of the estimates. This is born out by the fact that the estimated confidence regions for numbers of poor are smaller in the revised estimates.