Republicans are a near-lock to retake control of the Senate this fall, according to Election Lab, the statistical model built by three political scientists for The Washington Post.
The model says there is an 86 percent chance that Republicans will pick up the six seats they need to retake the majority, up from 82 percent earlier this year.
How does the Election Lab model get to those numbers, which are more bullish for Republicans than other models as well as predictions of nonpartisan political handicappers? John Sides, a political science professor at George Washington University and founder of The Washington Post’s Monkey Cage blog, explains:
“Our model suggests that the GOP has a very good chance of winning the Republican-leaning states: Alaska, Arkansas, Georgia, Kentucky, and Louisiana. That gives them five seats. They also have a better than 50-50 chance of winning Iowa, where Joni Ernst’s recent surge has made the race neck-and-neck – a trend that is consistent with what our model suggested about the Iowa race back in May. Meanwhile, Democrats have a good chance of winning Colorado, Michigan, and North Carolina.”
Here’s what the model looks like in the six seats mentioned above:• Alaska (Democratic-held): 65 percent chance of GOP win
• Arkansas (D): 85 percent chance of GOP win
• Georgia (Republican-held): 99 percent chance of GOP win
• Iowa (D): 78 percent chance of GOP win
• Kentucky (R): 99 percent chance of GOP win
• Louisiana (D): 93 percent chance of GOP win
Add those results to near-certain Republican pickups in Montana, South Dakota and West Virginia, and Republicans would start the 114th Congress with 52 seats.
Pretty simple. It all makes perfect sense.
But, it also might be wrong. Or, at least, more favorable to Republican chances than it should be. Here’s why.
Models are, by their nature, data driven. That’s why models tend to get better the closer the election gets. There’s just more raw material – poll numbers, fundraising numbers, etc. – to mine. Because of that reality, models tend to favor elements of races that can be easily quantified (presidential approval, GDP growth, fundraising) and diminish less easily quantifiable factors like candidate quality and the sort of campaigns being run on the ground.
Sides and his team use three data points aimed at ensuring the Election Lab model takes those candidate/campaign factors into account: 1) polling in the race 2) fundraising by the candidates 3) experience in elective office. Historically, all three have functioned as solid predictors of success or failure.
And yet, those three data points alone can miss other realities that help to decide elections. How candidates do on the stump, how they come across in TV ads, how smartly they are spending their money, how cohesive their campaign teams are, how on message they can be – all of these things matter to the final outcome of races. They don’t decide the final outcome but neither are they irrelevant in it.
Take the North Dakota Senate race in 2012. In a world in which only easily quantifiable data points were relied on, Heidi Heitkamp would have no chance of winning. She was a Democrat running in a Republican state against a GOP nominee who held an at-large House seat. But, if you dug deeper into less easily quantifiable factors, it was clear that Heitkamp was a far superior candidate to her Republican opponent and was running a strategically smarter race. She won – despite the fact that President Barack Obama got just 39 percent of the vote statewide.
Now, that’s not to say that the so-called “eye test” is a better way to predict winners and losers in campaigns than models like Election Lab’s. If I had to rely exclusively on either the eye test or a model, I’d take the model every time. (There’s a reason sabermetricians are preferred over scouts in sports these days.) But, it does mean that the model is only as good as the data that you put into it. And there are some things – candidate/campaign quality being at the top of the list – that are not easily plugged into a model.
And so, the Election Lab model is almost certainly right about the vast majority of Senate races on the ballot this November. But, what if, to pick two races at random, Mark Begich in Alaska and Michelle Nunn in Georgia wind up being better candidates who run better campaigns than their Republican opponents and, in so doing, squeeze out victories? Then Democrats would keep control of the Senate.
Models need to be understood for what they can tell us – and what they can’t. What the Election Lab model tells us is that the environment is ripe for a Republican takeover of the Senate. What it doesn’t tell us is how the specifics of each campaign and candidate can change that dynamic in small but potentially important ways.