By Andrew Hyman, Class of 2019
“Is Nate Silver superhuman?” One audience member put that question to Mr. Silver during a conversation with Austan Goolsbee at the Harris School of Public Policy’s new Keller Center on February 7th. Silver did, after all, predict 49 of 50 states in the 2008 election and 50 of 50 in 2012.
But not everyone is so bullish on his work. Early on in the discussion, Goolsbee asked Silver about reactions to the 2016 election. Betting markets and most pollsters put the odds of Trump’s electoral college victory at 15%, and conventional wisdom said it was even lower. FiveThirtyEight, Nate Silver’s website, predicted a 30% chance of Trump being elected.
Yet Trump won, and the high expectations set by previous success led to a specific criticism of him that he’s “overrated”, and “an astrologer using silly algorithms”. “Is [that criticism] fair? Is it right that 2016 wasn’t able to be predicted?” Goolsbee asked. Silver responded, with characteristic understatement, “…life isn’t fair. We got too much credit for the 2012 election. But is it right? No, it’s dumb as fuck.” He went on, “We were on the right side of the market… and for the right reasons. He, not Clinton, had the electoral college advantage. Lots of undecided voters. Clinton actually won more votes than the polls predicted, but Trump won tons of undecideds across lots of different states.”
Silver was able to find a niche for FiveThirtyEight’s data-driven journalism because of the systemic problems that underlie qualitative political analysis. Political reporters often use street-level intelligence to build mental models of how the electorate behaves, but this can lead to the creation of just-so stories [explanations for phenomena that are divorced from their actual causes]. Often, people forget to connect individual anecdotes and datapoints to the bigger picture. “Outcomes in different states are correlated. Losing Pennsylvania, and Michigan, and Ohio, and Wisconsin, it’s not unlucky. You’re not rolling the dice 4 different times.”
Because we perceive the world as a series of discrete events, we forget to consider the trends and forces that are often driving, or at least influencing, individual behaviors and actions. Silver and team have shown a repeated ability to divorce their analysis of events from prior beliefs, avoiding confirmation bias to a great extent, allowing more accurate predictions about the future. This process of de-biasing perceptions and supporting evaluations with quantitative analysis can be useful not only when considering political questions but also in career and even personal decision making.
Given Silver’s track record of successful prognostication, it was no surprise that the conversation quickly turned to 2020. The odds for the current president to win re-election are “about even money.” However, presidents historically win re-election about 70% of the time, so “a coin flip is a pretty big shift from the prior that incumbents usually win.” One heuristic people often employ is to look at the economy, which means things look pretty good for Trump’s chances - but “if you look at how the economy is performing two years prior to the election, it’s actually negatively correlated with election.” If the current run of economic growth continues through November 2020, it will be good for the incumbent; if things turn south, so do his re-election prospects.
Silver’s comments about the potential Democratic candidates are included off to the side, and while his commentary is interesting (and I highly recommend the FiveThirtyEight Politics podcast to anyone who enjoys these sorts of topics), the most important takeaway I gleaned from the talk is this: we are all vulnerable to bias and inaccuracy. System 1 thinking, as described by Daniel Kahneman in his classic book Thinking Fast and Slow, predominates almost all of our interactions. Unless we consciously slow ourselves down - gaining perspective that frees us from our prior beliefs, assumptions, and hopes about what will happen - we won’t be able to separate the signal from the noise. We will find the signal we are looking to find, rather than the truth of the situation. And the worst thing is that we won’t even know that’s what happened.