We are examining the characteristics of seniors who qualify for both Medicaid and Medicare. But we are showing much higher incomes than expected. Is it possible that lower income individuals are underrepresented in the sample?
If we weight the data by PERWT, will that take household income into account?
The Accuracy Statement for any ACS sample has a very detailed description of the weighting and sampling procedures.
Also, this may not be the case, but high income aggregates may be due to NIU values being present in the aggregation. Individuals who are NIU (Not In Universe) or NA (Not Applicable) are given values of 99999 or 999999, etc. which if included in means will produce very high values.
I hope this helps.
Yes, of course. We are taking account of missing values (999999). But we are seeing a bit of an underrepresentation for low income individuals when compared to the Census breakdowns. It is difficult to get a good representation of low income Medicaid enrollees, and especially the disabled. We are attempting to weight the data by income and age band to match the Census.
You may also consider using the 5-year files to increase the number of respondents in the sample, if you are not already. More information on the multi-year files can be found here.