Accounting for Sample Weights in Analyses (Just Add the Weighted Variables in the Regression)?

[Roman Johnson]

Greetings,

I would like to analyze my measures using sample weights within the 2016 National Interview Health Survey data. I am using the person and adult sample components from this data which I have merged into one data. The user guide says that WTFA is the weighted variable for person sample component and WFTA_SA is the weighted variable for the adult sample component (found on pg. 74-75 of the attached user guide for the above data.) ftp://ftp.cdc.gov/pub/Health_Statist...6/srvydesc.pdf

Is doing a weighted analysis of my data just as simple as adding the weighted variables (e.g. WTFA and WFTA_SA) to my analyses? Thank you for your help.

Roman

[Richard Williams]

I imagine you would want to use the svy commands. For a brief overview, see https://www3.nd.edu/~rwilliam/stats3/SvyCautionsX.pdf

[Steve Samuels]

I agree with Richard. There is a sample svyset command on page 89 of the NHIS User Guide. However, contrary to what is shown, all commands must be lowercase.

[Roman Johnson]

Thanks for the heads up on the svy and svyset commands. I know that I enter svyset whenever I want to find the weighted descriptives. But what is unclear to me of which weight variable is appropriate to use when you have variables that belong to different weight variables/ come from different sample files. (I have merged three files: the sample adult, person, and family within the 2016 National Health Interview Survey.)

I see that each of the sample files (person, adult, and family) samples that I need have different weight variables. However when I am running a logistic regression with variables that come from different files and therefore have different sample design weight variable names (e.g. wfta, wtfa_sa, wtfa_fam), how do I know which survey weight I should use with the logistic regression?

[Steve Samuels]

You analyze only one unit at a time (any person, adult, family) and you svyset with the weight for that unit. Suppose it is an adult. Then you start adult file and you'll use the adult weight for your analysis. However if the other two files contain information about the adult respondents, you will merge that information from those files into the adult file, creating a new analysis file. So, for example, you would merge variables about family income from the family file into the adult file. The weights in the other two files are irrelevant to analysis of adults. The bottom line is that the weight goes with the analysis unit not with the variables.

[Roman Johnson]

You're a rock star, Steve! Thank you. That makes perfect sense.