Which party will win control of the Senate in the upcoming midterm elections is a close call and hinges on senatorial races in five key states: Alaska, Colorado, Iowa, Arkansas and Louisiana, molecular biology professor and co-founder of the online blog the Princeton Election Consortium Samuel Wang said.
Wang co-founded the blog in 2004 to use statistical analysis of polling data to predict the outcome of the then presidential race. His blog has gained prominence lately due to a widely-publicizedstatistics dispute with fellow forecaster Nate Silver, a statistician who gained mainstream fame when his blog was picked up by The New York Times.
In all five key states, the Senate races show a three to four percentage point difference between candidates, and their outcomes may determine which party will have control of the Senate.There are 36 Senate seats up for election in November; 15 of those seats currently belong to Republicans and 21 to Democrats. To claim a majority in the Senate, Republicans would have to win six of the current Democratic seats.
Wang forecasts that the elections will leave the Democrats in control of 50 Senate seats, while Silver, who now runs the blog FiveThirtyEight,predicts that the Democrats will likely have only 49 seats.The distinction between the two outcomes may seem slight, but will determine if Democrats keep their hold on the Senate, Wang said in an interview.
Wang remains one of the few statisticians who predicts Democrats will maintain control of the Senate. He recently predicted a 70 percent chance that Democrats and Independents will control 50 seats, while Nate Silver recently predicted a 57.1 percent chance that Republicans will win control.
The source of variation in the forecasts of different models is likely caused by the uncertainty in a number of key races, statistician and former assistant professor of political science at Emory University Drew Linzer said.
“The fact that there is uncertainty about a handful of races that lead to this close outcome statisticians are pointing at means that small differences in the models are going to add up to what seem like big differences in the predictions,” he said.
Different methodology in constructing forecasting models for the midterm elections is a subject of debate. Attention was cast on Wang’s predictions when Silver published a column on FiveThirtyEight criticizing Wang’s methodology, calling his model “wrong” because it underestimates empirical uncertainty in polls.
The main difference in forecasting methodology between the two models is that Silver’s accounts for fundamentals, or correcting polling data by accounting for public opinion through examining other statistics, while Wang relies solely on aggregating data from polls as a direct measure of voter opinion.
The use of fundamentals is risky and could bias results when predicting a close election, Wang said, noting that in 2012 his polls-only model correctly predicted ten out of ten close Senate races.
“In some sense using fundamentals is building a political science model of where the race should be. Polls tell us where the race is,” Wang said.
Nate Silver did not respond to a request for comment.
Wang and Silver differ on their opinions of fundamentals versus poll-only data, but other statisticians are not necessarily in either camp.
Linzer, for example, argued that the use of fundamentals in a close race can be tricky. Making additional assumptions, he noted, introduces potential for error, but there is nothing inherently wrong with fundamentals if they are done well.
Wang said that the focus on probabilities and variations in predictions amongst forecasters has led to the appearance of disagreement between Silver and himself, Wang said, while the more important aspect is that the models are all predicting a close race.
“Nearly every model that I’m aware of shows that it’s a very close call who is favored to win, and no model is certain,” Wang said.
Another wildcard in the upcoming race that polls cannot account for is independent candidate in the Kansas Senate election Greg Orman ’91.
Polls indicate Orman has 39 percent of the vote compared to 44 percent for his competitor Republican incumbent Pat Roberts, yet if Orman wins, which party he chooses to caucus with can also influence who holds a majority in Congress.
While polls may predict who will win the race, polling data is powerless to predict which way Orman will choose if he wins, Wang noted in an article on the midterm elections in The New Yorker.
Nonetheless, the point of forecasting remains to inform voters, he said.
Voters in the four key states in this upcoming Senate election should be aware of the power their votes hold in determining which party will have control of the Senate and focus their energies appropriately, Wang said.
“The reason I do this is to help voters decide where to put their time, energy and attention. I think it’s important to understand what is at stake and what they can do about it,” he said.