Discrepancy between IPUMS and census.gov data for same survey

Hello IPUMS World,

I’m looking back at data I previously analyzed from the 2017 ACS via IPUMS and I’m finding odd discrepancies between that data and the 2017 ACS data accessed though data.census.gov.

For instance, the 2017 ACS from data.census.gov shows 4,926 people ages 65+ below the poverty line in Rochester, NY.

Via IPUMS (using the poverty variable, ages 65+, and city(5930), it shows 7,798. That’s a pretty large discrepancy (FWIW, outside the margin of error displayed on the Census online table).

Can anyone help me understand the discrepancy?



I suspect that you inadvertently retained cases where POVERTY is equal to 0 in your analysis. As noted on the codes tab for POVERTY, the 0 values does not represent 0% of the poverty threshold, but rather is an N/A category. Using the 2017 1-year ACS microdata, when I recode the POVERTY values into N/A, at or below the threshold, and above the threshold, I estimate 5,758 persons who are aged 65 and older and living in Rochester, NY and are at or below the threshold. I am not certain I found the exact table you were referencing from census.data.gov, but suspect this is probably within the margin of error for such a specific subpopulation in a single-year ACS dataset.

Incredibly helpful, thanks Kari! I think you got this exactly right. Really appreciate the guidance.



1 Like