No One Has a Clue in the Primary, Except This One Hopkins Student

1
Share the News


bernie-sanders-hillary-clinton

Bernie Sanders’ upset win over Hillary Clinton in Michigan’s Democratic presidential primary surprised virtually everybody — everybody except one Hopkins graduate student.

FiveThirtyEight‘s Harry Enten declared that he and other political forecasters were “eat[ing] a stack of humble pie” after Sanders pulled off a primary win in Michigan. Why? Because “not a single poll taken over the last month had Clinton leading by less than 5 percentage points.” The statistics-oriented site predicted a Clinton win with greater than 99-percent confidence.

But blogger Tyler Pedigo, a student at Johns Hopkins University’s School of Advanced International Studies*, predicted a Sanders win. His statistical model doesn’t rely on polls, which has been a strength in this pundit-stumping election cycle.

That’s not to say his projections have been flawless. After each round of primaries and caucuses, Pedigo discusses what he got right, what he got wrong, and how he’s adjusting his algorithms to improve his forecasting.

If you’re into the statistical end of politics, Pedigo’s blog is a refreshing dose of unassuming rationality.

—–

*: located in a little town about 40 miles southwest of Baltimore



Share the News

1 COMMENT

  1. Tyler Pedigo has had far worse forecasting than other major pundits as far as I can see. Yes, he got Michigan right, where the polls where further off than they have been in decades. But one state is not a reasonable way to measure a model. He has only given a confidence interval in three states as far as I can see. On one of them, Arizona, he was over 99.5% confident that Bernie would win, and he was completely wrong. In Michigan, Fivethirtyeight projected Bernie 12.3% below his actual, by far their worst projection. In Arizona, Tyler Pedigo projected Bernie 16.3% ahead of his actual. Ohio, Florida, and Missouri are other recent states he was way off on (unfortunately, no confidence intervals are available for those).

    He isn’t necessarily updating his model to include correct information. He assumed before Arizona that Hispanics don’t tend to vote for Bernie or Hillary, while there was plenty of evidence to the contrary. He still doesn’t seem to have updated it, since he seems to think long voting lines in Arizona magically made Clinton win by an enormous amount. Based on Bernie’s higher performance in caucus states and the large number of individual donations he gets, the opposite could be the case, since Bernie supporters may be more willing to spend more time for their candidate. He makes some other bad assumptions too, like “a closed primary… doesn’t help Hillary Clinton as much as open primaries do”, which are the opposite of reality.

Comments are closed.