• More on the erosion of employer-based coverage – ctd. (Wonky)

    Last week, I posted on the difference between the CPS data on employer-based coverage and an analysis from the NIHCR of a RWJF funded survey . The NIHCR had shown a more conservative estimate (higher) of employer-based coverage in the past, but looked more dire in 2010. I have been exchanging some emails with Chapin White, who today suggested it might be because the survey does not include under over 65 year old’s in their data, while the CPS data I used did. So I’ve remade my chart using CPS data for the non elderly:

    This certainly brings the CPS data more into line with the NIHCR  survey in the past. But I still wonder what happened in 2010. Why all of a sudden the discrepancy?

    UPDATE: Fixed a stupid mistake and the title of the chart.

    UPDATE #2: A very polite reader who works at the NIHCR  pointed out the gazillion errors I made in referencing the organization and the work. My apologies. Fixed. I should know better than to try and fix/post from a school function for my daughter.


    • Since the report with NIHRC data does not include standard errors I do not think it is possible to rule out sampling error. That is, the CPS interviews about 200k people for its insurance estimates and the NIHRC interviewed about 14k in 2010. These will have very different standard errors. The difference in estimates may still be significant but it is difficult to speculate about potential survey effects without more information.

      Common sources of difference in surveys of insurance include: point in time coverage (NIHRC) versus any-time in year coverage (CPS) question structures, non-response rates (9% for CPS, ~51% for NIHRC), control totals for weights, question wording, and countless other effects.

      But first, I would rule out sampling error. You might also compare with the ACS estimates for that period which have 62.9 with ESI in 2008 and 59.3 in 2010 for nonelderly. http://www.shadac.org/datacenter/tables. The ACS has almost 3 million observations for these estimates.

      Typically, a survey like NIHRC will have richer questions that can answer other things besides just the employer coverage rate. That said, the trends are the same: employer coverage is not working…

      • According to the paper from which Aaron took the data, the NIHRC changed methodology in 2010 to better accommodate cell phones. I don’t know how much that mattered, but it certainly could be a lot.

    • Could at least part of the answer behind 2010’s decline be directly related to the unemployment situation?