The Midterms, An Unpopular President, and Politics Not as Local as You Think: Forecasting Montana's Legislative Races

Do campaigns really matter? Can politicians, much as Merida tried to do in the Disney movie Brave, change their "fate" and win elections despite unfavorable fundamentals? Or are they destined to blow a bunch of money fighting windmills a la Don Quixote?

It should be clear that I believe campaigns matter; at least, that's exactly the story I tell in Battle for the Big Sky. Jon Tester won despite fundamentals clearly favoring the Republican Party and Congressman Rehberg in the epic 2012 battle for Montana's Senate seat.

But I also make plain in that book that campaigns very often do not affect the outcomes of elections, and in fact, cite approvingly the work by John Sides and Lynn Vavreck who--in their fantastic book the Gamble--show that fundamental factors (the economy, presidential popularity, and the like) pretty much determine presidential election outcomes regardless of what the campaigns do. As the 2014 congressional elections demonstrated, forecasters knew for quite some time that Republican odds of taking control of the Senate were fairly high. If anything, as the election drew closer to November, the odds got decidedly worse for Democratic incumbents in the Senate despite running strong campaigns (see Mark Begich and Kay Hagan).

As I sat and watched the state legislative races here in Montana unfold, I kept getting media inquiries about various competitive races. Did the Democrats have a shot to win SD 14 in Havre with Greg Jergeson? Could Jebediah Hinkle, who unexpectedly won the Republican primary for SD 32, beat known and former congressional candidate Franke Wilmer (a colleague of mine here at MSU)?

Invariably, I had to plead ignorance to reporters. I hadn't a clue. I know what the parties told me, but I had very little information to give reporters other than that and, let's face it, the parties had their own spin to advance. Incumbents tend to get reelected. Midterm elections are bad for the president's party generally speaking. Fundraising gives some sense of the candidate's talents and support.

Then--nothing. That's all I had.

Fact of the matter is, I had very little information with which to make any informed judgment about specific races. And to be frank, most voters had even less information with which to make their own judgments. Truth is, down ballot races like state legislative races are low information environment elections. State legislative candidates can generally raise enough money to get some name id and help voters associate their name with a party id, but other than that, most voters simply don't know much about the candidates other than that.

So I was skeptical every time I heard so and so had knocked on more doors, or candidate Y had more signs than candidate X. I just didn't think, in a low information environment election, it would make much difference on the final outcome of most of these races. And, to prove my point, I set out to forecast the outcomes of each state legislative race in Montana.

And I would do it purely with publicly available information.

Here's what I did. I set out to predict the probability that the Democratic candidate in each race would win their seat. In statistical terms, I ran a logistic regression with the output the mean probability of a Democratic win in that seat. Associated with that mean statistic is a confidence interval surrounding that estimate. For example, my model predicted that Diane Sands in Missoula had a 54 percent probability of winning her Senate seat, but the probability of a win could be as low as 31 percent or as high as 76 percent. On average, her chances were a tad better than a flip of a coin.

How did I develop that estimation? I gathered information on past elections in Montana, from 2004 through 2012. For each race, I gathered the following variables:

Percentage of the Vote for the Democratic Presidential Candidate
The Type of Seat (State Senate or State House)
Whether there was a midterm election (-1 if the President was a Democrat, 1 if the President was a Republican)
Democratic candidate spending as percentage of the total spending in the race (all data gathered from Follow the Money's website)
Whether the Democratic candidate was an incumbent
Whether the GOP candidate was an incumbent
Presidential Approval rating averaged for the year which I then interacted with the midterm election variable

I ran this model using the data from 2004 through 2012. It turns out that this model correctly predicted the outcomes 94 percent of the time.  Next, I collected the data for the 2014 cycle (using presidential vote estimates from 2012 for the newly drawn districts courtesy of Daily Kos--thanks Mike Jopek for reminding me that's where I found it). Then, using the estimates from the model for 2004-2012, I estimated the probability of a Democratic win in each 2014 race using the 2014 variables.

Using this, I predicted that the Democrats would lose two seats in the Senate and gain one in the House. In reality, the Democrats outperformed the model by holding even in the Senate and gaining two seats in the House.

But, my model--using no polls, no data about the ground game, and nothing specific about candidate efforts other than fundraising--came pretty close.

 All the variables behaved as one might expect. Incumbents perform better. The more Democrats dominated the spending in a race, they better they did. But was striking the drag President Obama had on Democratic prospects in the state legislature.

Let's look at the State Senate. Democrats had an excellent recruit in Greg Jergeson in SD 14. But Obama only got 40 percent of the vote in that district in 2012; my model gave Jergeson only 45 percent chance of winning. Jergeson lost.

Senate District 32 Obama got almost 49 percent of the vote, and yet--despite having substantially outraised Republican Hinkle--the model only predicted a mean 29 percent chance of Democrat Franke Wilmer winning. Hinkle won.

Carlie Boland, who ran in a district that gave Obama 51 percent of the vote in 2012, had only a 12 percent chance of winning. She lost to Republican Brian Hoven.

Only Mary McNally (D) surprised in my model, as it only gave her 22 percent chance of winning a district that had gone for Mitt Romney by nearly 52 percent in 2012. She beat Tonya Shellnutt, 54-46.

Campaigns matter. Except when they don't. And in a year when the win was firmly at the back of one party and in an era where redistricting has become far more sophisticated and voters increasingly vote straight ticket, campaigns mattered in very few of the state legislative races. That said, my model predicted that the Democrats would do worse than they did, so it would seem in a select few places, individual efforts and candidates may have made the difference. Diane Sands' winning by only a few vote and Mary McNally's victory in Billings very likely had to do with their own and the Democratic Party's concerted efforts to stem the red tide.

Next time a reporter calls me about state legislative races, I'll have a bit more to say.

PS: I ran my model on election day, but decided not to share my predictions until after the polls closed. I didn't want to inadvertently affect any outcomes, particularly after the big row over the Stanford-Dartmouth experiments.