Major sample attrition during 1994 and 1995 basic monthly CPS files


I’m using basic monthly CPS data for women aged 38 - 52 and I am finding what appears to be extraordinary attrition during survey years 1994 and 1995. I am assuming that something peculiar happened during this period, but nothing turned up after a quick search.

Specifically, if I look at individuals (CPSIDP) who entered the survey (MISH=1) during 1994, 54% of them exited in the fourth month (max value of MISH within CPSIDP is 4). Among individuals who entered the survey in 1995, 10%, 6%, 8%, and 24% left after 1, 2, 3, and 4 months in the sample, respectively. These “attrition” rates are far higher than in any other year.

Meanwhile, there are large numbers of “new” individuals popping up in later months (min value of MISH within CPSIDP is 2, 3, 4, etc.) exactly where these missing individuals should have been.

My main concern is that some of these attrited and new individuals are actually the same people, with inconsistent personal identifiers. I would be interested in any information you can spare on the manner.

You are correct that these cases do not represent attrition, but rather that between 1994 and 1995 there are a subset of respondents that cannot be linked across the full interview period. This is due to the major survey redesign that was implemented during this period and is discussed in the CPS User Note for 1994, which can be downloaded here (note: the usernote.asc is a downloadable, plain-text file that can be opened in any plain-text editor).

I hope this helps.