Tomorrow we're releasing an update to our TargetEarly site, impacting the modeled partisanship classifications in MI, OH, and WI, improving their accuracy. I'll explain the process and rationale in a brief thread here, for anyone who is interested. targetearly.targetsmart.com/index.html
I've suspected for a while now that the generic partisan models weren't doing as good of a job as we'd like accurately capturing the generic partisanship of voters in those states. To be fair, those are tough states to model, with no party reg and relatively homogeneous.
To assess the models, first I looked at the party models in the context of the 2016 electorate. In MI the model was 8 points more GOP than the election outcome, 7 pts in OH, and 10 pts in WI. That's compared against the Clinton-Trump margin in each state.
Compared to the exit poll party ID numbers (an imperfect measure, but still helpful), the generic partisan model showed themselves to have a slightly larger GOP bias.
Thankfully, our polling team has conducted several thousand interviews we could leverage for this analysis. Matching those survey results to the party model provide its GOP bias. This process also demonstrated that our Biden support model did a much better job.
The Biden model improved upon our accuracy in predicting partisanship by about 7 pts, on average. So we will be putting that model in place on TargetEarly, replacing the generic party score. You should see these changes take effect tomorrow.
I don't know exactly what they will show (I'll know when you know, basically), but I do expect them to show a slightly more Democratic outlook in those states, though the margins will almost certainly vary.
I'm happy to answer any questions. Thanks for indulging me in this too long thread, but it's important to me that we are entirely transparent with the data we share on TargetEarly!
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Our update went live last night. We saw a slight turn towards Dems in OH and WI in the reported AVEV data, and MI appears to have generally stayed the same. targetearly.targetsmart.com/index.html
There's still evidence that the partisan model rollup in these states is overly Republican. If you look at the model distribution, you'll see that there is a cluster of voters at the low end of the score (GOP) but no such cluster at the Dem end.
That said, the model does a good job of rank ordering voters in terms of likely partisanship, an improvement upon the prior partisanship score.
We've surpassed 10M ballots cast nationally, with 21 days remaining. What's more, early voting will only accelerate as key states begin early in person voting.
We can clearly state that Dems are building a lead in the early vote, not just cannibalizing E-day votes.
Details...
We've added a filter on our TargetEarly site allowing users to filter just to aggregate data for Presidential or Senate battleground states. targetearly.targetsmart.com/index.html?vie…
Limiting the scope just to presidential battleground states, Dems have a lead of 15% among early votes cast thus far, almost doubling the 8% lead they had in those same states at this same point in '16.
142,510 Americans who have never voted before have already cast a ballot in the 2020 general election. At this point in the 2016 cycle only 38,394 first time voters had cast a ballot.
Among these first time voters, Dems have a modeled partisan ID advantage of 6.4%. At this same point in the '16 cycle Republicans had an advantage of 1.3% among these new voters.
This surge in first time voter turnout among the early votes cast thus far is being driven by big increases from black, asian, and latino voters.
Over 3.6M votes have already been cast in the 2020 general election.
While Democrats have a large lead overall among those early votes cast, it's reasonable to assume that Republicans will skew somewhat towards Election Day voting, nationally.
For me, the most important datapoint we can gather from the early vote is the partisan distribution among low likelihood turnout targets. At this point, 136k people have voted who never voted before. Dems have a 6.1% advantage.
In Pennsylvania, registered Democrats have built up an advantage of 835,578 mail ballot requests over Republicans. Donald Trump carried the state by 44,292 votes.
Not every voter requesting a ballot will return it. And yes, more GOPs will vote on Election Day. But the Biden campaign will bank these votes very early, allowing them to focus their resources on a much smaller target universe through November 3rd. That's a huge advantage.
223,612 of these Democrats didn't vote in the 2016 election.
A quick thread on where the vote by mail (VBM) requests stand at this point. Sorry for the megachart, I'll explain as I go!
First, all of the partisan data in here is modeled party, for the sake of consistency.
Looking at this column, we see that Dems lead in VBM requests in every state but 4 (AR, ID, IN, MI).
Keep in mind, this includes automatic/permanent VBM requests, which may explain MI.
Compare this to the partisan distribution of the 2016 electorate in these 15 states, where Dems trailed in all but 5 of the 15 states. Again, this is generic modeled partisanship of the voters who cast a ballot in '16.