Changes in sampling over time skewing results?

I’m working with the November Supplement and noticed a huge uptick between 2016 and 2020 in both the registration and the voting rates of respondents who select Asian or Hawaiian/Pacific Islander.

While rates have been increasing, an increase of 11 points from one presidential year to another is unprecedented. Given the clustering of these populations, could a change in how samples are taken explain this?

The data below, for voter registration, are not weighted, but the issue is the same if they are.

**Race        2012  2016  2020**
Asian only  0.708 0.703 0.817
H/PI only    0.639 0.686 0.744

There was no change in CPS sampling during this time period. If you look at turnout statistics for the 2020 election, you’ll see large increases across the board. This is probably due to many factors, but mainly the huge increase in mail-in voting.

True, but the increase for API and Natives was far greater than for other groups. “White only” had an increase by about 4 points from 2016 to 2020. But the change for A/PI was 11 and for Native Americans was 8 points. Since these two groups are small and concentrated, I wonder if something changed in where samples were drawn from within states.

There was no change in the CPS sampling during this time. New PSU’s are drawn for the CPS sample only once every ten years, generally a few years after each census. Of course there is always sampling error, which can cause estimates to vary month to month due to chance, especially for small subpopulations. I did find this paper which might have some insight into this issue.

1 Like