I’m doing a nonresponse analysis using the ATUS-CPS dataset including both the respondent and non-respondents of ATUS. In this circumstance, I’m only applying a base weight.
However, I find that there are many 0 in BWT. Here is how I construct an indicator variable and cross-tabulate it with year. In 2004-2006 and 2008-2009, nearly half of the weights are equal to 0.
When I investigate on these years that contain weight ‘0’, I find they correspond with the non-respondents in those survey years.
(‘C’ is “complete interviews” which represent a respondent while the others are all non-respondents, not eligible or unknown eligibility)
Suspecting it is an error from the original dataset, I downloaded the ATUS-CPS data file from the website of Bureau of Labor Statistics and found no 0 value in the base weight variable (TUBWGT) from 2003-2018.
I wonder if it might come from any coding error or processing error when IPUMS harmonized the data?