Thursday 11 April 2019

Continued census whoas

Stats has started providing some detail on the missing response figures in the last Census.

Motu has followed Brian Easton's earlier call to bring forward the next Census to 2021, returning us to the timings we were on prior to the Christchurch earthquakes.

A correspondent raises a few further questions, which I hope are answered in the full report on the 2018 Census:

  • There are substantial differences between responses with a missing question or two, and responses where the respondent has only filled in their name. How big of a mess are we looking at?
  • The web-based Census may have had some problems, but it may also have provided some opportunities. Like, how many responses suggest the respondent isn't really taking things seriously - for example, by ticking the first box on every question. This could have been done in real-time, prompting a follow-up visit. Did Stats do any of this?
  • My correspondent finds it odd that there are more people who didn't do the Census at all than that left at least one question unanswered. I'm not so sure there - if you're filling in the web form, it can prompt that you've missed a question before it lets you move on to the last page. I'm not sure if the survey did that or not though. 
A have a few others:
  • If they're interpolating data from other administrative sources, are they filling blank cells from relevant admin data, or putting in an average based on matched other respondents' answers? How are these generated answers flagged in the data? This will matter for researchers doing things like checking the correspondence between census responses and administrative data to see how well the two line up. But it'll also matter for any research using those generated answers. I expect Stats will be well on top of this. 
  • We keep hearing Labour partisans blaming National for not sufficiently funding Census. Did the Government Statistician raise any specific risks in the shift to the online Census under that funding arrangement? If the Government Statistician had raised a lot of concerns about the viability of the Census under the funding provided, and National went ahead anyway, then pretty fair to sheet it back to National. I do not know what warnings of risks were provided.
  • I had thought that the funding allocation for the 2018 Census built in some of the expected costs of shifting to an online census model, with the prospect then of lower cost censuses to come. Given the problems in 2018, will that have to be revisited? 
  • We hear rumours around that non-responses were concentrated among particular communities. Stats' statement doesn't get into that yet, as they weren't asked. We should expect to see more detail on that in the full reporting yet to come.

No comments:

Post a Comment