Okay this is fascinating! The citizenship test that Census should have done a decade or two ago for the ACS finally got done for the Census question.... It did reduce Hispanic response rates *and Hispanic ethnic identification rates*.https://twitter.com/hansilowang/status/1211772725238386689 …
-
-
Throughout this whole process (go read my tweets and see my published articleS: you can can check me on this!) I've argued that: 1. The Census experts were wildly overstating their case 2. There would be some decline in response 3. It would be smaller than people were saying
Show this thread -
I WAS RIGHT PEOPLE I WAS SO RIGHT YOU WERE ALL WRONG AND I WAS RIGHT I'm gonna gloat about this for a while.
Show this thread -
Because Census ***had never conducted an actual test*** of the citizenship question, they overestated the effect on responses by ***400%***.
Show this thread -
Now, for SCOTUS' purposes, this doesn't matter, because SCOTUS shot down the question on APA and technical grounds, not the apportionment effects. And to be clear, as I've said for a long time, I think killing it on APA grounds was good and proper.
Show this thread -
And I'll remind you that I wrote and argued many times before SCOTUS that what SCOTUS WOULD DO was kill it on APA grounds and not apportionment effects. AND THAT'S WHAT THEY DID.
Show this thread -
On bothw hat SCOTUS would do and what a test would reveal I have made out-of-the-money calls and been right both times, because most of the commentary on this issue has been wildly partisan nonsense, even from experts.
Show this thread -
Anyways back to the test results: They ONLy found significant results in Los Angeles and New York, which is interested. Suggests different levels of political activation or concern.
Show this thread -
I'm just.... folks.... I'm shaking with how simultaneously angry I am at all the experts who propagated misinformation WHICH WAS OBVIOUSLY MISINFORMATION and also how VINDICATED I feel.
Show this thread -
The citizenship question WAS CONSTITUTIONAL IN PRINCIPLE, but the Trump Administration's unique mixture of malice and incompetence made it fall egregiously afoul of APA. The question BARELY lowered response rates, about 80% lower than was suggested would happen.
Show this thread -
Now, barely lowering response rates might be super bad still! We might really value the data we get from other questions and not regard this small response reduction as an acceptable cost! But drop the histrionics, folks.
Show this thread -
Oh man sweet justice. Vindication, thy name is randomized controlled trials.
Show this thread -
Here's a
@voxdotcom article from me on why the Census question addition probably would not reduce response rates very much: https://www.vox.com/the-big-idea/2018/4/5/17202092/census-question-citizenship-immigration-trump-deportation-undercount-latino-hispanic … And a@FDRLST article on constitutionality:https://thefederalist.com/2018/04/02/blue-states-sue-200-year-old-census-citizenship-question/ …Show this thread -
The Federalist article was early in the process so we didn't know how screwed up the administration's reasoning was; to see my view on that you can just search old tweets.
Show this thread -
I vividly remember being called a right-wing shill and a charlatan, a defender of racists and racist policy, for arguing that asking about citizenship was not prima facie unconstitutional, and that response rates would not fall by 2%.
Show this thread -
@Noahpinion I urge you to communicate to your legions that response rates to the citizenship question are only about 1/4 to 1/5 as sensitive as elite public opinion suggestedShow this thread -
(sorry not implying mocked by Noah here; just he's someone I feel like might be interested/might have followers who are interested)
Show this thread -
I'm gonna be basking in this one for a while. I still can't believe that a team of 5 Census experts published a paper arguing that adding a citizenshop question would reduce response by *2.2 percentage points*.
Show this thread -
I flagged it as, uh, preeeetty weird to use comparisons of response rates to the ACS and the Census short-form in 2010 as a benchmarkl given that the ACS contains a whoooole lot more differences vs. the short-form Census than just a citizenship question
Show this thread -
but nooooooo everybody was like, "these are experts, and this is a difference in difference model, therefore"
Show this thread -
And this is why you need to pay attention to the wider literature on an issue before just adopting one study! The general finding of most survey response data for Census products is that individual questions do *not* have large effects on response rates!
Show this thread -
Individual attitudes and characteristics and large variations in survey-length, as well as differences in survey instrument and response delivery, are the biggest drivers, not specific questions, which are *usually* found to have quite modest effects.
Show this thread -
Particularly because questions that make people worry about their identity are easy to JUST LIE ABOUT. Now, that may be an issue for other reasons: we may get unreliable data! Fair concern and fair critique of a citizenship question! Not the critique that was made!
Show this thread -
okay I could rage-tweet about this all day but I have real work to do so baaaack to the grind
Show this thread -
Given that the *sign* of these significant subsamples matches the *sign* of the insignificant aggregate result, and that the estimated effect sizes are in a similar range, I think it is reasonable to suppose the subsamples do have some "real" meaning.https://twitter.com/karlbykarlsmith/status/1211891611032072193 …
Show this thread -
HOWEVER.... It could be that the NYC and LA field offices administered the AB test differently somehow. Or any number of other factors. And the overall effect was quite small, which is the most important conclusion.
Show this thread -
Furthermore, the subsamples Census reports are fairly standardized; i.e. similar subsamples that they track and report for all similar tests. In other words, it's kind of like a de facto pre-registration.
Show this thread -
So given that I don't think they chose these subsamples selectively, but had a kind of implicit prereg, and given that sign/size match pretty well, I think it's reasonmable to say, "extremely small overall decline in responses, but modest decline in a few subgroups"
Show this thread -
Show this thread
-
I warned you, but did you listen to me? Oh, no, you knew it all, didn't you? Oh, it's just a harmless little overestimated effect size, isn't it? Well, it's always the same. I always tell them--pic.twitter.com/QZvAbSsgDt
Show this thread -
also this is a true-to-life picture of me; that's my Tuesday hat
Show this thread
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.