Hello,
We are doing a study of older adults and want to look at attrition from each MEPS Panel starting in 1996 through 2023.
We have examined the pstats variable which has reasonable loss of individuals due to mortality or institutionalization but very low non-response for other reasons. Is there another variable we should examine to understand people and households that attrite from the MEPS?
Thanks!
I’m replying to let you know that we’re looking into your question and will follow up soon when we have enough information to share. Thanks for your patience.
Thanks for your patience in awaiting a reply. We have consulted with our IPUMS MEPS PI, who provided the following information. Please let us know if you have additional questions.
MEPS respondents are organized into reporting units (RUs), which are groups of related people within a sampled dwelling. Response to the MEPS core interview happens at the RU level—with the exception of targeted supplemental questionnaires that are completed at a different time – individuals are not interviewed separately. An RU either completes the MEPS in each round or does not.
Technical variables capture whether someone moves into/out of scope and/or into/out of eligibility and therefore has valid data for some interview rounds but not others. More information about the concepts of scope and eligibility (and the related concept of “keyness”) – as well as the IPUMS MEPS variables that capture these concepts – are available in the IPUMS MEPS glossary of key concepts and terms. If people (or RUs) simply do not respond to one or more interview rounds in a given year, they are excluded from that year of data altogether and the person weight values of remaining people are inflated to make up for the deletion of nonrespondents. No attempt is made to re-enroll these nonrespondents in the MEPS panel. We do not know whether people have been deleted from the first year of the panel because they would never appear in the MEPS-HC data, but I imagine these people would be taken into consideration in the original MEPS technical documentation on response rate. We can infer if people (or entire RUs) have been deleted (i.e., attrit) due to nonresponse in the second year of the panel because they would have an observation in the data file corresponding to the first panel year, but not in the data file corresponding to the second panel year. This would be more appropriately considered nonresponse conditional on having responded to the first half of the MEPS panel. For more information, the user can refer to this report describing the treatment of part-year respondents, specifically, this passage:
“Sample persons who are classified as non-respondents to one or more rounds of data collection (i.e., initial non-respondents and drop outs over time) are not included in MEPS annual files, and survey weights for full-year respondents are inflated through statistical adjustment procedures to compensate for both full and part-year nonresponse (for more details, see Cohen et al., 1999).”