Combining The American Community Survey and the National Health and Nutrition Examination Survey?

[Roman Johnson]

Hi, All:

I am wondering if it is possible to combine two national surveys, the American Community Survey, and the National Health and Nutrition Examination Survey. I want to combine the two surveys to weight according to county level information.

[Clyde Schechter]

I'm not sure what you mean when you speak of combining them. They can't be merged at the level of individual responses because there is no common identifier to link them. If you want to separately aggregate them up to the county level in some way, and if there is a variable identifying counties in each data set that is coded the same way in both of them (or if you can create a crosswalk of the county identifiers), then you could -merge- the county-level responses. Is that what you mean? If so, the major obstacle I can think of is that both of these are surveys and they may not sample the same counties. Do you know if they do?

[Roman Johnson]

Correction, that should be the National Health Interview Survey and the American Community Survey. I will be able to find out that information this weekend and will post here.

[Weiwen Ng]

Clyde's question still applies. Some of the National Health Interview Survey's (NHIS) individual respondents later are asked to participate in the Medical Expenditures Panel Survey (MEPS). So, it's year 1, everyone takes NHIS. Then years 2 and 3, a subset of those people take MEPS. To my knowledge, the ACS takes a sample that's independent of NHIS. It's mathematically possible to get sampled for both. However, I will bet you money that individual survey respondents can't be linked. By anyone, anywhere in the US government. Except maybe the National Security Agency. (THIS IS A JOKE.)

You could use the American Community Survey (ACS) to derive county- or state-level characteristics. In principle, you could link the NHIS survey respondents to those characteristics, but you would not be able to do so using public data at the county level. I'm not sure if you can even link identify NHIS respondents at the state level in public data. This would be a restricted dataset, meaning you would need to submit an application to the US National Center for Health Statistics. I don't know what fees are involved, if any. It's not as burdensome as obtaining Medicare claims data, I assure you.

For other readers, all datasets discussed here are national datasets collected by the United States. Their sample frames are designed to represent the entire US.

[Roman Johnson]

Thank for your help. It seems that it is not possible.