Nate Silver (FiveThirtyEight.com), Dave Wasserman (Cook Political Report), Nate Cohn (NY Times Upshot/The Needle) and Michael McDonald (ElectProject) have come to be relied upon by hundreds of thousands, perhaps millions, of people in trying to make sense of the data related to election forecasts and results.
They all have their strengths (and weaknesses) and areas of emphasis, and all provide content that is devoured by data junkies like me. But like the pollsters they rate for quality, it’s only fair that we take a look at how our favorite “fixes” performed in the last election cycle.
So I’m taking a shot, providing my grade on three phases:
- pre November 3rd
- the election night/week actual results
- the all-important Georgia Senate runoff.
I’ve also added a poll to allow readers to voice their selections as to the best data analyst of the cycle.
Dave Wasserman (Cook Political Report):
- Pre-November 3rd — F
- Election Night/Week — D
- Georgia Runoff — A+
Wasserman was without doubt the biggest disappointment of the cycle. The maestro of “I’ve seen enough” severely harmed his reputation.
- Leading up to the vote, Wasserman was THE person (supposedly armed with private internal polling data) who predicted large Democratic gains in the Congress; he was the one who referred to a “bloodbath”. Wasserman completely missed the dynamic on this.
- On Election Night, Wasserman’s contract with NBC News deprived him of his most famous platform (the ability to make calls on Twitter before others dared); he became just another apologist for why the networks were taking so long to call what was obvious to most.
- Wasserman did partially redeem himself during the Georgia Runoff vote count by being the first (by far) to “have seen enough” for both the Warnock and Ossoff races. His confidence was surely merited given that the final margins of victory appear to be >90,000 and >50,000 respectively. Maybe that will encourage him to focus once more on data and less on speculation and punditry.
Nate Cohn (NY Times Upshot and “The Needle”):
- Pre-November 3rd — B-
- Election Night/Week — B+
- Georgia Runoff — A+
With the resources of the NY Times at his disposal. Cohn and his work at the Upshot may have become the gold standard for comprehensive political analysis.
- Prior to November 3rd, The Upshot’s daily polling summaries and analysis were very well conceived and presented, with easy to understand — including showing the deviation from the 2016 actual results. But “garbage in/garbage out” still hampered the analysis; there were big misses on a number of states.
- The night of the election, “The Needle” actually performed pretty well, providing the depressing news about Florida very early, and the nearly definitive projection for North Carolina in a timely way. But “The Needle” did shine in calling Georgia as a tiny Biden tilt (.4%) while the popular vote was still lopsided in Trump’s favor. In fact, it was this small ray of hope that sustained me during the VERY dark hours late on November 3rd into the wee hours of November 4th. The letdown was that The Needle was stopped far too early and applied only to three states. I guess even the NY Times doesn’t have infinite resources.
- “The Needle’s” analysis and accuracy as the Georgia Senate runoff vote was tallied was simply extraordinary. The presentation which showed EACH vote batch vs. the “expected” results was genius — a print version of Steve Kornacki’s famous white board chicken scratch showing mail in ballot results in Pennsylvania vs. the same counties’ “day of” voting results. And the final (early am Jan. 6th) Needle predictions of +2.0 for Warnock and +1.1 for Ossoff were spot on, and were made when most others were speculating about whether the final margins would be outside the .5% threshold for a recount.
Nate Silver (Fivethirtyeight.com):
- Pre-November 3rd — C+
- Election Night/Week — B+
- Georgia Runoff — A+
Silver, the dean of the data gurus, is always a flashpoint for controversy. In his defense, like a baseball manager, he gets too much credit for his successes and too much blame for misses. After all, he is reliant on his players (source polls); if they don’t perform well, he’s doomed.
- I thought FiveThirtyEight.com’s graphics leading up to the election were just too cutesy (the the “snake”, etc.). On the other hand, Silver’s +/- 90% confidence level of a Biden win seemed prescient — it was very likely going to happen, but there was always a chance it wouldn’t. That’s truly how things played out — Biden led the popular vote comfortably, but the imbalance of the Electoral College nearly resulted in disaster.
- On election night, and the following days, the site evolves to a running series of commentaries, posts and some data analysis. This format is also useful as an aggregator of what others are saying — from the betting markets to sources like Wasserman and Cohn). I thought by and large the information was useful, particularly as we moved into analyzing final House results; I really liked the “races not decided” updates.
- For the Senate runoffs, Silver managed to take data from (supposedly) very few reputable pollsters and come up with final averages that were very, very close to the final result:
- +2.3% for Warnock vs. +2.0% actual
- +1.4% for Ossoff vs. +1.2% actual
Michael McDonald (US Election Project):
- Pre-November 3rd — A+
- Election Night/Week — n/a
- Georgia Runoff — B+
In my opinion, no one did more to burnish his “brand” this cycle than McDonald. For the past decade plus, he’s been building his reputation as THE source for advance voting data. But in 2020, he broke through with spectacular presentation that matched the raw numbers.
- The 2020 maps and tables were easy to understand and insightful. McDonald clearly predicted not only huge advance voting totals (both early vote/mail ballots) but ALSO that the total turnout would be enormous. What’s sometimes missed is that while McDonald is not a “pundit” in the traditional sense of predicting outcomes, yet he is sometimes chastened by the likes of Silver and company for trying to draw conclusions from the data. Take a look at McDonad’s election eve forecast. He was spot on about states like North Carolina and Nevada, which were very close, but for which the final result was telegraphed due to the huge proportion of advance votes.
- McDonald doesn’t analyze actual vote/turnout totals as they come in. There’s only so much one guy can do!
- During the runoff, it was McDonald’s data which first pointed to the potential overperformance of the Democratic base and the higher black vote% vs. November. He was a bit tepid in total turnout projections, saying only that it was clear that turnout would be lower than the general election. Duh. I don’t recall him speculating that turnout would exceed 90% of the general, however.