Aaron S. Veenstra Profile picture
Sep 16, 2018 26 tweets 6 min read Twitter logo Read on Twitter
Some interesting #twill things from @UpshotNTY's #IL12 data collected last week: github.com/TheUpshot/2018…
For reference, the overall results show a very close race in a swingy district that went Obama/Obama/Trump, and was originally drawn for a conservative Dem incumbent, but has been Republican since 2014. nytimes.com/interactive/20…
The candidates are two-term incumbent Mike Bost (R) and St. Clair County state's attorney Brendan Kelly (D). Bost won 54-40 in 2016, with the Green Party candidate taking 6%.
#IL12 is a compelling test district because it's split between suburbs (Illinois side of metro St. Louis) and rural counties. It's 79% white, and has a research university deep in the rural part of its area. But broadly speaking, the rural part is hard Trump country.
I've done some rough turnout exploration of the district previously, and non-white turnout and third party voting are each significant issues for Democrats here. St. Clair (East St. Louis and adjoining towns) and Jackson Counties (Carbondale) have a lot of "extra" votes lurking.
My analyses are multivariate regression models, which allow us to see the influence of multiple variables at once, and figure out which are simply proxies for others. For example, party identification may proxy race and gender, because voters are not evenly distributed on them.
Basic demographics first. Upshot didn't ask income, but among age, gender, race, and education, only race predicts vote intention (white people more likely to choose Bost controlling for age, gender, and education), and does it fairly strongly.
When we add party ID to the model, race is no longer statistically significant, but remains fairly close to the conventional threshold (avoiding the ongoing argument about significance here to report as fully as possible). Party ID explains 39.5% of variance in vote intention.
Fairly obvious on the baselines. Let's look at what might be complicating things in #IL12. First, the urban/rural split. Democrats are trying to poach from traditional GOP strongholds in suburbs, where educated, well-off whites have been turned off by Trumpism.
The data suggests this is a good move. Those in the Metro East counties are more likely to choose Kelly, controlling for demos and party ID, and the effect is about half the size of the bare effect that race had before adding party to the model.
Upshot asked several issue-related questions that have bearing on how the campaign has been run. Among them, "Have you, or someone you know well, struggled with addiction to opioids?" with 26% saying yes. Kelly has the opioid crisis as a centerpiece of his campaign.
The data don't show this as something that is determining vote intention. The relationship isn't close to significance (p = .590), and the effect size is minuscule anyway. There are many ways to interpret this – message issue, engagement issue. But it's not driving vote choice.
They also asked, "Is your economic situation much better today than your parents’ economic situation was when they were your age?" to which 63% in this Trump district said yes. This is the closest we get to an item testing the "economic anxiety" hypothesis.
Adding it to the demographic and party ID model, it is a marginally significant predictor of intention to vote for Bost, at about the same strength as being white. The economic anxiety is actually on the Dem-voting side of #IL12.
How about the racial anxiety hypothesis? Upshot asked, "Has discrimination against whites become as big a problem as discrimination toward blacks and other minorities?" 48% agreed that is has.
As background, I want to point out that there's strong evidence the current era of racial polarization began with the killing of Michael Brown in Ferguson, MO, which is about eight miles as the crow flies from #IL12. fivethirtyeight.com/features/charl…
In this context, it should be no surprise that the relationship between racial anxiety and vote intention is huge. This item explains 3.3% of vote intention variance on its own (demographics combined are only 1.9%). Its effect size is more than a third that of party ID.
Could either of these be proxies for the urban/rural split? It doesn't look that way. Including Metro East in the model, economic situation is a slightly stronger predictor in the same direction, and racial anxiety is unchanged. These appear to be unique effects.
The only issue that tops racial anxiety is an assault weapons/high capacity magazine ban. 52% of respondents support it, and support strongly predicts intention to vote for Kelly, above and beyond demographics and party ID.
This issue actually takes away some of the predictive power of urban/rural, but does little to change the effect of racial anxiety when added to that model.
OK, this has been fun, but before wrapping I must address the elephant in the room: Donald Trump, running at 48% approval (net +2). We know that "fundamental"-based models of midterm vote choice put a lot of weight on presidential approval.
Turns out they're right! Trump approval is an enormous predictor when added to the baseline model, explaining 12.6% of additional variance, and cutting down a ton of the explanatory power of party ID. It actually winds up nearly as strong as party ID was when added to demos.
When I add in racial anxiety, it still has a decent amount of unique explanatory power, but economic anxiety does not. This is consistent with basically all the data we have from both during and after the 2016 campaign.
Takeaways: House votes are filtered through a presidential framework. People doing well want to vote Republican. White people feeling racial identity threat want to vote Republican. Suburbanites want to consolidate their turn to the Democrats. Opioids have little traction.
Caveats: This was a live phone poll, and the response rate (like all such polls today) was terrible. Some statistical power was lost as a result of non-response bias, and the sample wasn't huge to begin with. Still, this is the best #IL12 data that will be made public in 2018.
Also, this survey can't explain causality, and relationships between the limited number of issues it measures and vote intention are likely influencing one another in complex and ongoing ways.

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Aaron S. Veenstra

Aaron S. Veenstra Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!


0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy


3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!