Nate Silver (AP/Nam Y. Huh)

Nate Silver unloads on Princeton rival in blistering critique

The war among wonks continues as Silver pans Princeton forecaster's "flawed" Senate model


Luke Brinker
October 3, 2014 8:03PM (UTC)

Nate Silver is bringing out the knives for fellow political prognosticator and bitter rival Sam Wang, penning a sharply-worded critique that assails Wang’s model as fundamentally “flawed.”

A primer on the war among wonks: Thirty-two days out from Election Day, most forecasters expect Republicans to net the six seats necessary to win control of the U.S. Senate. Silver’s FiveThirtyEight.com, for instance, currently gives Republicans a 58 percent chance of capturing the majority. And as Silver notes in a post for Taegan Goddard’s Political Wire, virtually every other leading forecaster pegs the GOP’s chances for Senate control at somewhere between 58 and 78 percent. Wang’s model, however, gives Democrats a 58 percent chance of maintaining Senate power (down from 70 percent last month).

Advertisement:

Why the discrepancy? Wang, a Princeton neuroscientist, says it boils down to polling vs. “fundamentals.” In forecasting electoral outcomes, FiveThirtyEight weighs both recent poll results and the dynamics of each contest – the partisan lean of the state, demographic makeup, and so on. Wang’s model relies only on polls.

But as Silver points out in his Political Wire post, other models – including the Huffington Post’s and the Daily Kos’s – are also exclusively polling-based, but they project a GOP Senate come January 2015.

There’s a simple reason Wang’s forecast diverges so sharply from other polling-based models, Silver writes: Wang’s model includes polls dating all the way back to June, when the outlook was better for many Democratic Senate candidates.

Wang’s model includes both a “snapshot” and a “forecast,” Silver notes; the former predicts what would happen if an election were held today, while the latter  is how Wang “assesses the Election Day odds.” But, Silver explains, “Wang's forecast is based on his snapshots.”

“In particular, his forecasts are based on an average of his past snapshots since June,” Silver writes. “Since Wang's is a ‘polls only’ model, this is equivalent to looking at polls back to June.”

Using this kind of model, Silver contends, would have suggested in October 1988 that Michael Dukakis was slated to defeat George H. W. Bush for president, because even though Bush was solidly ahead by this point, a Wang-type model would have weighed polls from the summer showing Dukakis with a healthy lead.

Advertisement:

Prior to Silver’s Political Wire post, Wang wrote a lengthy response to Silver’s concerns this week, suggesting that their dispute “could be solved over a beer or two.”


Luke Brinker

MORE FROM Luke BrinkerFOLLOW LukeBrinker



Fearless journalism
in your inbox every day

Sign up for our free newsletter

• • •