G Elliott Morris Profile picture
Sep 23, 2020 5 tweets 1 min read Read on X
10% chance of a trifecta? color me skeptical, very skeptical
The one huge thing I learned from modeling this year is that it is a mistake to model uncertainty as uniform over time. Polarization makes elections much easier to predict (on balance) today versus, say, 1970. I think that's making Nate's models too uncertain across the board.
If people in the media are to take one thing away from the theory behind our models, it's that how we think about certainty in a time when voter behavior is constrained by very predictable forces can lead to better predictions, especially earlier in the year, in modern elections.
(The other thing is that fundamentals models should definitely have presidential approval ratings in them.)
The thing about incorporating polarization into your model is that we have data other than just election returns to know it's the right decision. We have tons of polling microdata that says the coefficient on party -> vote has increased over time.

• • •

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

Keep Current with G Elliott Morris

G Elliott Morris 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!

PDF

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!

More from @gelliottmorris

Jan 25
📊Today 538 is releasing an updated set of our popular pollster ratings for the 2024 general election! Our new interactive presents grades for 540 polling organizations based on their (1) empirical record of accuracy + (2) methodological transparency. 1/n abcnews.go.com/538/best-polls…
There’s tons to say but I’ll hit a few main points. First, a methodological note. For these new ratings, we updated the way 538 measures both *empirical accuracy* and *methodological transparency.* Let me touch on each. (Methodology here: ) abcnews.go.com/538/538s-polls…
Image
(1) *Accuracy.* We now punish pollsters who show routine bias toward one party, regardless of whether they perform better in terms of absolute error. We find that bias predicts future error even if it’s helpful over a short time scale.
Read 19 tweets
Nov 21, 2023
if you want to understand polling today, you have to consider *both* the results and the data-generating process behind them. this is not a controversial statement (or shouldn't be). factors like nonresponse and measurement error are very real concerns stat.columbia.edu/~gelman/resear…
given the research on all the various ways error/bias can enter the DGP, if your defense against "polls show disproportionate shifts among X group. meh" are "well X group voted this way 20 years ago," i am going to weight that pretty low vs concerns about non-sampling error Image
at the same time, if a critical mass of surveys is showing you something ,you should give it a chance to be true. interrogate the data and see if there's something there. i see tendencies both to over-interpret crosstabs and to throw all polls out when they misfire. both are bad
Read 4 tweets
Oct 14, 2023
There is good stuff in this thread, and I’ve been making the first point too for some time. But remember a lot can change in a year, and some of the factors that look big now may not actually matter. Uncertainty is impossibly high this far out.
I took a look yesterday at how much Dem state-lvl POTUS margins tends to change from year to year. It’s about 7pp in our current high-polarization era. That’s a lot! With 2020 as our starting point simulating correlated changes across states, you get p(Biden >= 270) around 60%.
that is obviously not a good place to start if you are team Biden. But the range of outcomes is laughably large—a landslide for either party is more than plausible. So there is a pick your own adventure element to analyses like these: Dobbs, Jan 6 help Ds; Economy, Biden age hurt
Read 6 tweets
May 19, 2023
so, as they say... some personal news
Lots to share, but for now I'll just say FiveThirtyEight was one of the outlets that inspired me to be a data journalist. Nate Silver did great work & the team he led changed political journalism for the better. We will be iterating on that, but we start with a strong foundation.
2/3 ABC and I have been in talks for 6 months to ensure there will be as little disruption as possible in transitioning from the aggregation + forecasting models Silver is taking with him when his contract expires to our new in-house methods, developed w input across ABC & 538.
Read 4 tweets
May 18, 2023
pretty bleak picture for the GOP 10-20 years from now, unless the party changes its policy endorsements and messaging to shrink the gap in Gen Z/Millennial voting behavior catalist.us/whathappened20… Image
yes, however, rolling back convenience voting reforms for students is not going to be an effective voter suppression strategy when the average Gen Z voter is out of school (my back-of-envelope math says this should happen around 2028)
bad tweet!
the point is that crossing your fingers and pretending that young people just get more right-leaning as they age is not an effective electoral strategy,
not that there is a 100% probability of democratic electoral success for the next 30 years
Read 4 tweets
Apr 17, 2023
Just one poll… But IMO pundits are currently running the risk of spinning the proverbial wheel too far against Ron DeSantis in terms of his electability value above Trump (which is declining or completely negated by recent news, according to some accounts)
I do think it is a little early to be writing definitive takes about which GOP candidate would have the best chance against Biden (esp if you don’t do any demographic-lvl vote projections). + there are a lot of factors that are hard to mentally model. With history as our guide… Image
Read 4 tweets

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!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(