First time user of NHIS. I’m looking at access & uninsured for various populations. I see that a lot of variables, such as “hinotcove” have various iterations of “unknown” as coded answers. Is the proper protocol to leave these answers as-is for reporting? I would be tempted to recode them as missing to remove them from the denominator but I don’t want to do this unless it is accepted practice. Thank you!
In general, there is no guideline about this and it is up to you, the researcher, to make a decision about how to handle “unknown” cases. In the case of HINOTCOVE, there are relatively few cases that are unknown so whatever you decide it shouldn’t matter much. This could change if you are analyzing a very narrow sub-population of the sample. In particular, if reporting “unknown” is not random in some way, then excluding these observations will bias your estimates. My only concrete advice is to perform your analysis both with the “unknown” cases included and excluded and make a note in your write-up about how much this changed (or didn’t change) your results.