HWTSUPP for Ann_Arbor metarea code 2-3x larger between 1996-2005. Why?

Hi. I use the ASEC data from 1986 through 2016 for Ann Arbor, MI (METAREA=440). Looking at only the PERNUM=1 entries and aggregating across years I find that the sum of HWTSUPP is roughly stable around 80k-110k expect for the years 1996 through 2005 where I see values of about 220k to 260k. This increase is roughly in line with the increase in rows I’m observing from roughly 50-60 to about 110-150. However, I thought that summing over HWTSUPP would give me an estimate for the number of households in this metarea. So I’m wondering what I’m missing to filter out or correct for for the years 1996-2005… I do, by the way, see a similar pattern in some other metarea codes, so not limited to Ann Arbor - just thought it might be helpful to limit investigation to just one for the time being. Thanks for any pointers / help / advice!

Note that the CPS sample is designed for national and (for more modern samples) state level analysis. The Census Bureau generally warns that estimates produced using lower-level geographic identification can be quite inaccurate due to a relatively large amount of sampling variability. This warning certainly applies to metropolitan areas, and in particular metropolitan areas with populations of less than 500,000. This is a limitation of the CPS. A potential alternative is the ACS samples, available via IPUMS USA, which are much better equipped (due to a larger sample size) for sub-state level geographic analysis.

Thanks, Jeff. That makes sense. As far as I understand, the ACS data started rather recently so unfortunately only 2000 and later years appear to be available. Is that correct? We were, however, interested in a longer time series so I started looking at CPS data. I did read the comment about METAREA which you are also highlighting in terms of unreliability for smaller MSAs. Will have to think about what to do given this limitation… Thanks for your help!