In 2009, two major surveys in the Governments Division of the Census Bureau were redesigned to reduce sample size, save resources, and improve the precision of the estimates. We developed a new decision-based estimation method in which the collapsing of strata either by state and government type or by small and large size was determined by a series of hypothesis tests of the equality of fitted coefficients in linear relationships between attributes and their values in the previous census year. In this research, we study design-based variance estimation by a bootstrap method, for the new decision-based stratified regression estimates applied to the Annual Survey of Public Employment and Payroll. The bootstrap method, which goes beyond available theory, is validated through a small simulation study. We use the data from the 2007 Census of Governments Employment to illustrate our methods.