Connect with us


Nate Silver’s forecasting model: partisan hack-job or garbage



Nate Silvers forecasting model partisan hack-job or garbage

Nate Silver is the founder of FiveThirtyEight, a widely known data analysis website. And while the website has a widely reputable reputation for quality infographics, it may be time to question the credibility of its founder and data analysis. America’s top political research (gambling) website PredictIt sent this message to its members.

You all know Nate Silver, the famous prognosticator and founder of Great guy. Except that he just called you “dumb”.

Okay, not exactly, he said prices on PredictIt are dumb. And it’s true there are some markets where your expectations and his are a bit out of whack.

For example, you’ve got the GOP at 58% to win Arizona’s Senate race. Nate is at only 38%. You’re pricing Claire McCaskill at only 39% to keep her Missouri Senate seat. Nate is much more confident at 57%. And, the Democrats to win the House? He thinks it’s nearly a foregone conclusion at 88%. You’re hedging your bets at 72%.

So who’s right? We’ll know that later today. In the meantime, are you comfortable being called “dumb”? If not, get back on the site and vote with your money. Or maybe you want to take the advice of an expert and hedge your bets.

This was great motivation to check out the current prices the day of the election. If you are unaware, PredictIt operates a stock market model. The price of a pick is what the last person paid for it. It’s putting your money where your mouth is. Digging deeper, FiveThirtyEight has a unique algorithm applied to polls and even non-polling factors for its more premium models. In contrast, a free market betting system is the aggregate of how likely people think this event will take place. So here’s the question: is the free market of election betting more reliable than the data analysis of Nate Silver and his infrastructure?

The answer is unequivocally yes. Nate Silver’s predictions were aggressive and extremely bearish on Republican candidates. The midterm results were anything but the historic Blue Wave we’ve been hearing about for two years. In the two Senate examples from the email, we saw Josh Hawley unseat Claire McCaskill with greater ease compared to the far tighter expectations of the Senate race in Missouri. We also saw Martha McSally ahead in a tight Arizona race.

Here Nate Silver explains that his model gave the Democrats an 18% chance at gaining control of the Senate. In my previous analysis Mapping the Senate in the midst of midterms elections I explain what the Democrats would have to do in order to achieve majority.

At a glance, the GOP already secures 49 seats, two fewer seats than what they currently hold. A Democratic majority would require the Democrats to win all 9 toss up races along with securing the competitive races that aren’t tossup. The Democrats max out at 51, maybe 52 if they drum up some rape allegations on Ted Cruz; it’s approaching that time in the race if they are going to do that. In contrast the Republicans max out at 59, all of the tossups and a W in both New Jersey and Michigan.

Giving these events an 1% chance is preposterous. If you had parlayed those odds, you would have become rich. A $100 parlay on the Democrats winning the Senate races in Arizona, Florida, Indiana, Missouri, and Nevada would yield $2036.35 according to odds found on Newsweek. Throw in Beto and the Blue Wave would have paid off over $10000. These odds come with an implied probability of 4.68%. This isn’t even all that the Democrats would need to win. Somehow, FiveThirtyEight had 18%, which was within 1 standard deviation of his model’s prediction. My Bookie had either -700 or -800 line on the GOP maintaining the Senate, in the 88-89% probability that the GOP retains the Senate, placing the Democratic control outside of one standard deviation away from their average.

And while boasting his mediocre system, Silver then proceeds to make woeful partisan hot takes.

(Uses popular vote)

(Neglects fact that Beto had $60 million dollars that could have gone to more viable campaigns, major celebrity endorsements, media attention. Such high cost for such a low reward.)

And then Silver reposts his shameless defense of polling. There a two specific fallacies. the first being that national polling has weight on Congressional or Presidential elections. In fact his defense uses national polling, and the results from such to average out the more volatile state results, when the state results in 2016 were the second most deviant results in his data set, the first being the disastrously polled 1980 presidential election. Secondly, Nate Silver defends polling by saying that polls predict the outcomes of races overwhelming majority of time. This is true, however, most political races aren’t competitive. Incumbents generally get reelected. So getting it right majority of the time is an easy accomplishment. It was fairly obvious Corey Stewart was not going to win in Virginia. Saying that polls got this right is unproductive. The thrill in predicting elections is in the competitive races.

In Silver’s Classic and Deluxe forecasts, he employs non-polling method, which of course include generic ballot polling… But his ratings here were flawed. In Florida, he gave a 1.6 fundraising advantage to Bill Nelson in one of few races where the GOP senate candidate had a bigger warchest, data if evaluated properly would have accounted for outside spending as well. The generic Ballot entitle Nelson to a +7.2 advantage. The overall advantage for these non-polling factors was 7.2 in Nelson’s favor. The Deluxe Model also employs experts who were more aggressive in their Nelson picks. This model gave Nelson a 73% chance of victory. My Bookie had this race even (50%). PredictIt was more bullish towards Scott. No fancy algorithm, just a collective intuition and a fiduciary desire.

The Classic model showed Mike Braun with a 28.2% chance of winning, Joe Donnelly’s performance outside of a reasonable margin of error. The same classic model gave the greatest edge Claire McCaskill (56.9%) in a race most had much tighter. For people financially invested, Josh Hawley was likely to win, a more aggressive prediction than could be derived from only polls. My Bookie had Josh Hawley’s moneyline at -160, an implied probability of nearly 62%. This race was easier to predict than Florida, Nevada, or Arizona, and Nate Silver still got in wrong, though his Deluxe Model which factored outside experts had Hawley winning. Out of all the tossup Senate races, Silver got Nevada correct, Arizona still pending. Nevada, as I pointed out was the second easiest seat for Democrats to flip, though this was before Kyrsten Sinema’s comments surfaced. Admittedly, I had Dean Heller winning. Jackie Rosen’s victory is hardly a bracket buster in an election where I foresaw GOP Senate gains. In the DeSantis race, Gillum was projected to win more aggressively than warranted, given his overtly corrupt nature as a candidate, but to mitigate the damage towards Silver’s credibility, My Bookie had DeSantis at +150 (40%). Upsets happen. Sometimes the 40% happens, but My Bookie foresaw the upset happening at nearly double the rate that Silver.

It’s time to admit, as a society, that Nate Silver’s FiveThirtyEight and his forecasting model is either partisan hackery or well animated garbage. Or perhaps, it is both because why assume the two options presented are mutually exclusive. If Silver really believed PredictIt’s prices were dumb, then the Arizona and Missouri races were bargains and he should have bought. Perhaps if Nate Silver put his wallet where his mouth was, he would have created a better model. But without such incentive, due shaky ethical grounds, he can continue operating his below average model, being celebrated as an expert to a partisan leftist media because he aggressively and incorrectly forecasts their favorite candidates to win.