Weights for mortality analysis (IPUMS_NHIS)

First of all, congratulations on your wonderful work on this platform.
Although I have read several posts about weights for analyzing IPUMS_HNIS data, I am not sure I have found an answer to my specific question.
I plan to perform a analysis with mortality as an outcome (Cox). In this case, I know I have to use MORTWT when analyzing mortality in conjunction with variables originally included in the NHIS person files, or MORTWTSA when analyzing mortality in conjunction with variables from NHIS adult files. However, my main exposure variable is in the adult files, but most of co-variables are from the person files. I think that, in the case that variables come from two different types of files (person and adult files), the most restrictive weight should be used. If I am correct, should I use MORTWTSA as the weight for all analyzes? Finally, to calculate the data for the descriptive table (eg, percentage by sex, mean age, etc.), should I also use the MORTWTSA weight, or the PERWEIGHT or SAMPWEIGHT weights depending if each variable is from the person files or from the adult files, respectively?

Thank you very much in advance. Best regards,


Hello again. I have another question about the mortality data from IPUMS_NHIS. I have located the year and quarter of death, but is the exact date of death (mm/ dd/yyyy) available in the public files?

Thank you. Sincerely,


In regards to your questions about weights, you are correct that in cases when variables are coming from different types of files (adult and person), the most restrictive weight should be used. So, in your case, if one of your variables requires MORTWTSA, you should use MORTWTSA in your analysis. It is also a good idea to compare the universe statements for each variable. For your descriptive table, if you are summarizing a variable that is not in conjunction with a mortality variable, you should use the appropriate weight depending on if it is from a person level (PERWEIGHT) or adult level (SAMPWEIGHT) file. Read more on the specifics of NHIS sample weights here.

As per your question about mortality data, the exact date of death (mm/dd/yyyy) is not available. Quarter of death (MORTDODQ) and year of death (MORTDODY) are determined by NCHS based on probabilistic matches of survey participants’ NHIS records to the National Death Index (NDI) records. For more information on the creation of mortality variables, see the NHIS Linked Mortality Files.