Below is from the US Census site. It appears that since the 2010 census, the type of data you are looking for is not available at the level of detail it was prior:
http://www.census.gov/hhes/www/poverty/about/datasources/description.html#cps
American Community Survey (ACS) ACS homepage
Starting in 2006, the ACS releases annual subnational estimates of income and poverty for all places, counties, and metropolitan areas with a population of at least 65,000 as well as the nation and the states. The sample size of this survey is about 2.9 million addresses per year, making the ACS exceptionally useful for subnational analyses. Three-year period estimates are available for areas and subpopulations as small as 20,000. Starting in December 2010, five-year period estimates are available for census tracts/block groups and for small subgroups of the population. ACS estimates are updated every year. Because of its large sample size, estimates from the fully implemented ACS provide the best survey-based state level income and poverty estimates. Time series trend data will be available for all geographic areas, and for small population subgroups, beginning with the 2006 ACS for geographic areas with population of 65,000 or more. (Background on ACS)
Survey of Income and Program Participation (SIPP) SIPP homepage
The Survey of Income and Program Participation (SIPP), a longitudinal survey, is most useful for understanding the dynamics of income and poverty (changes in income and poverty status for the same households over time, typically 3 to 4 years) and for examining the nature and frequency of poverty spells. Unlike the ACS and CPS ASEC, the SIPP also permits researchers to look at estimates of income and poverty for periods more than or less than one year. (Background on SIPP)
Census 2000 long form
The best measure of change between 1990 and 2000 for subnational areas and for subpopulations are the comparisons of Census 2000 long form estimates with those from the 1990 Census long form. Since the ACS eliminated the need for a long form, the 2010 census will not provide income and poverty estimates. For small areas and subpopulations, ACS multiyear estimates are available for comparison to Census 2000 and earlier. ACS 5-year data products provide data at the census tract level which can be compared “with caution” to earlier decennial census estimates. (We suggest that comparisons focus on the direction of change, e.g. whether poverty rates went up or down, rather than the magnitude of changes). (Background on Census 2000 long form)