New Data Gives GOP 75 Percent Chance of Holding Senate

Yesterday Axios published a SurveyMonkey poll that measured head-to-head Senate matchups in Missouri, Tennessee, Indiana, Montana, North Dakota, West Virginia, and other key states. Elections Twitter and the political media immediately latched onto the poll and attempted to figure out exactly how good or bad it was for Democrats.

This is a situation where SwingSeat, THE WEEKLY STANDARD’s new data-driven Senate forecast, is very helpful. I fed the new data into the model and updated the main page with the results.

Here are the details:


In Tuesday’s update, the GOP had a 69 percent chance of taking the Senate. But now that we’ve factored the SurveyMonkey poll in, Republicans have a 75 percent win probability. (Important note: Polls are often published at least a few days after they leave the field, but the model looks at the date polls were in the field. The SurveyMonkey poll was in the field for most of June, so we counted the poll as happening in mid-June and had to re-estimate a few weeks of outputs.)

Right now, the median SwingSeat simulation gives Republicans 52 seats, which represents a one-seat gain from where they are now and no change from where they were coming out of the 2016 election (Roy Moore lost a special election to Democrat Doug Jones in Alabama). The change in the median simulation from 51 to 52 shouldn’t be overinterpreted. A small shift in the polling data could easily push the median back to 51. (Explanation for the mathematically inclined: If you sort the simulation from least to greatest number of GOP seats, the shift from 51 to 52 happens close to the median.) But the model also thinks that a wide spread of results are possible: anything from the Democrats netting a couple seats and taking the chamber to the GOP unseating some red state Democrats and netting a couple seats is on the table.

In probabilistic terms, the GOP has a roughly 75 percent chance of holding the upper chamber. That’s roughly 3-to-1 odds, which is s also the probability of flipping a coin twice and getting heads both times.

In some settings, there’s a real, practical difference between a 70 percent and 75 percent probability (e.g., if you’re playing a million hands of poker, it’s better to have a strategy that works 55 percent of the time than one that works 50 percent of the time). But if you’re looking for a basic, 30,000-foot view of the election, this difference isn’t that big. Republicans still have a real edge. Democrats are the underdogs, but they could very plausibly win. Events that have a 25 percent probability happen all the time.


The SwingSeat model doesn’t just output topline control probability; it also produces race-by-race estimates based on the data we have in each state. In some cases, these probabilities make sense intuitively and match either my gut feelings or the conventional wisdom. But other times it doesn’t. No model is capital T Truth, and in some states it’s working off sparse or old data.

So here’s a rundown of the major changes, what they mean and whether they make sense in context of everything else we know about the election:

The biggest change is in Tennessee. SurveyMonkey gave Republican representative Marsha Blackburn a 14-point lead against Phil Bredesen, the state’s Democratic former governor. That seemed like a lot to me: This is an open seat, and Bredesen seems like a high-quality candidate. I think the deep redness of the state gives Blackburn an advantage, but 14 points is very different than the Bredesen leads we’ve seen from other pollsters.

But SwingSeat looked at all the data and ended up closer to my assessment of the race. SwingSeat believes that SurveyMonkey is less accurate than other pollsters, so it reduced the poll’s weight. It also looked at this poll in the context of the previous polls (many of which showed Bredesen ahead), decided that the wide range between them was a cause for greater uncertainty. It concluded that Blackburn had a win probability of about 70 percent. Both Sabato’s Crystal Ball and Inside Elections say the race leans toward the GOP, and this probability seems like it’s in that range.

SwingSeat also moved North Dakota and Indiana slightly in the GOP’s direction. The model currently thinks that Republicans have a 60 percent chance of unseating Joe Donnelly in Indiana and a 70 percent chance of unseating Heidi Heitkamp in North Dakota. Those are both a bit higher on the GOP than I am personally. But in both states, the model is doing what it can with stale, scarce data. And it’s reacting to the SurveyMonkey data in a rational way: It’s modestly increasing the GOP win probability based on new data showing small to middling Republican leads.

SurveyMonkey also published numbers in Florida, Missouri, Nevada, and Arizona, but SwingSeat didn’t change the outlook much based on those numbers. Claire McCaskill and Jacky Rosen led their Republican opponents by small margins in Missouri and Nevada.* In each case, the Democratic candidate has a win probability somewhere between 50 and 65 percent. That translates to a small edge where the GOP candidate could easily make a comeback. In Arizona, Sinema led McSally by four points. But the model has more recent and more reliable polling in that state, so SurveyMonkey was discounted significantly and Sinema maintained a solid win probability.

In all these cases (and in some of the sleepier races that I don’t feel a great need to discuss), the model did a good job of integrating the SurveyMonkey data and producing realistic responses.

There are arguably two places where it didn’t.

SurveyMonkey also gave West Virginia Sen. Joe Manchin a 13-point lead. To me, that seemed like too big of a margin. I’d expect that sort of a margin from a Republican or Democratic candidate in one of the safer, sleepier races (e.g,. Tammy Baldwin leads Kevin Nicholson, the GOP primary poll leader, by 13 points in Wisconsin in this poll). And the model currently gives Manchin an 89 percent chance of winning.

Manchin has polled very well in the early phases of this campaign, but 89 percent seems like too high of a probability for that red of a state. It’s possible that Manchin will waltz to re-election. But that probability doesn’t align with my gut assessment of the race, so I’d take it with a handful of salt. I’ve been tinkering with versions of the model that integrate fundamentals more, and a good version of that model would likely pull Manchin’s win probability down significantly.

I’m also skeptical of the Montana estimate. SurveyMonkey said that Tester had a 12 point lead over Matt Rosendale, his GOP opponent. There hasn’t been much polling in Montana, so the model reacted strongly to that margin and gave him a win probability north of 90 percent. This isn’t where I think the race is—I would guess that Tester’s “true” win probability is north of 50 but well south of 90 based on the polling and non-polling data I’ve seen.

When I debuted the model, I warned that some state by state estimates may be off in the early, data-deprived phases of the campaign. I think that’s what’s happening here. There’s been very little polling in Montana, so SwingSeat may be overreacting to the new data and overestimating Tester’s win probability. So, like West Virginia, I’d take the Montana estimates with a fistful of salt.

Source link

Articles You May Like

Sanders’ Campaign Workers First Ever to Unionize
New Zealand prime minister announces semi-automatic weapons ban after Christchurch attack
Beto O’Rourke breaks one-day fundraising record, beating all Democratic rivals
Former California GOP Chairman Lists 9 Ways America Is Moving Towards Socialism
Canada ramps up border security to stop the flow of asylum-seekers coming from the US

Leave a Reply

Your email address will not be published. Required fields are marked *