PUMS poverty rate differs more than expected from ACS published rate

As part of an analysis to determine the poverty rate exclusive of college and graduate students who do not live in group quarters, I created a 2010-14 ACS variable called PovStatusGQ, which splits the household and group quarters population into categories by housing type (household, group quarters) and poverty status (not evaluated for poverty, below threshold, above threshold). I followed the advice in the answer to the question from this past April about “Poverty and Group Quarters”, which states that persons with a GQ status of 4 or 5 and POVERTY status of greater than 0 should be included in the poverty calculation. This also appears to match the defintion provided in documentation available through the ACS web site.

When I run a simple table for the population of Cambridge, MA (statefip(25), puma(3200,0506)) using just the PovStatusGQ variable I can see that the analysis covers the entire population. Doing so yields an overall individual poverty rate of 14.6%. The 2010-14 ACS itself reports a poverty rate of 15.0%. The difference of 0.4%, while not large, was a bit more than I expected. Have I erred in some way in estimating the overall individual poverty rate?

Hello, I’ve run the PovStatusGQ variable you created and seem to be getting different numbers than you’ve recorded in this question. Can you provide some more specific information on how you calculated poverty?

PovStatusGQ divides cases into 6 categories, along with a small number of “no data” cases. Categories 1 and 2 are the household population. Categories 3 are institutionalized group quarters residents. Categories 4, 5 and 6 are nononstitutionalized group residents whose poverty rate is 0, 1-100% of poverty or greater than 100% of poverty respectively.

If my reading of the IPUMS defintion of the POVERTY variable and the ACS discussion of “Poverty Status in the Past 12 Months” are correct then group quarters residents who are not institutionalized or living in dorms or milatary barrackes are included in the ppopulation for whom poverty is determined. The defintion of the POVERTY variable indicates that such cases have a poverty rate greater than 0, which is equivalent to categories 5 and 6 in PovStatusGQ.

Thus, the population for whom poverty is determined in equal to the total number of cases in caetgories 1,2,5 and 6. For Cambridge, MA these total 91,951, compared to the universe in ACS 2010-14 Table S1701, which is 90,989. The persons in poverty comprise catgeories 1 and 5, totaling 13,333 compared to 13656 in Table S1701. The table generated through IPUMS yields a poverty rate of 14.6 vs 15.0% reported by Table S1701.

Thank you for providing additional information on the calculations behind the PovStatusGQ variable. Although the IPUMS-USA tables do not necessarily exactly replicate published figures, the estimates produced by your variable are within the margin of error reported by the published ACS figures. Therefore, I do not suspect you have made any substantial mistakes in generating this variable.