Thread by @dwschanz

Here's what we've learned so far about the spike in nonresponse in the Census Household Pulse Survey.

Of course, I'm focused on the Q on food sufficiency, but it would be straightforward to extend this to other sections.

https://twitter.com/dwschanz/status/1303706187406376963

New questionnaire is much longer. Like 20 min vs. old 12 min. People drop off as the survey goes on. Here's what nonresponse %'s are, by section, in week 12 vs. week 13.

Questions on food used to come 4th, after demographics, employment + spending. 4% nonresponse rate then.

Now food Q's come 6th, behind new sections on benefits and trips. 13% nonresponse rate now.

Aside: was it worth losing reliable info on food hardship to get answers to Q's like: in the last 7 days have you taken fewer trips to stores than you normally would have, due to Coronavirus?

Don't we have good data on this from cell phone locations, sales?

(facepalm!)

Nonresponse is up across the board, generally 6X what it was for the food questions I rely on.

Generally higher for more disadvantaged groups.

Groups that had larger increases in nonresponse rates have larger declines in food hardship responses, which suggests to me that sample selection is driving a lot of the change.

UPSHOT: I don't think we can compare week 13 to weeks 1-12 directly.

Food hardship is still high, but I don't think I can say anything much about direction.

We may be able to do better when the mircodata are released in 2 weeks (which is a million years in covid time!)

I'm not a survey expert, and surely the hard core methodologists will have more to say on these issues. Please do!

(Hey, I could send you a scented candle as a thank you! I have a whole stash.)

I suggest that Census make questionnaire changes ASAP.

Reorder the survey so food hardship is 4th module again!

OR, randomly rotate modules (after demo + emp sections) so some respondents see food, housing questions early.

Something! We need reliable data.

(end).