Fivethirtyeight on electability

Three days after I posted my thoughts about the idea of "electability," Nate Silver's FiveThirtyEight (link) made a similar observation. Their article went much deeper into a pile of academic research concluding that you don't know someone's electability until after the election. That's another way of saying what I wrote: predicting someone's electability is the same as predicting the winner of the election.

If you missed the post, it was here I explained why electability is a meaningless concept, a tautology, a circular logic. A candidate claiming "I am more electable" is the same as saying "I will win the election". A voter selecting a "more electable" candidate is the same as believing his/her own prediction that said candidate will win the election.

Even specialists like Nate Silver sometimes gets it wrong. But he and his imitators update their predictions all the way to the night before the general election. Their skills are then evaluated on their last forecasts. A voter in the primaries who selects a candidate based on electability is taking on a challenge that is 10 times harder. That voter is making a prediction up to 12 months out, and with access to very limited and unreliable data.

The key quote of the FiveThirtyEight article is this:

Political scientists study electability, but electability ain't no science. Instead, researchers say, it's basically a layer of ex post facto rationalization that we slather over a stack of psychological biases, media influence and self-fulfilling poll prophecies.

The article then survey the main conclusions of the political science literature on this subject. The following points resonate with me:

"Voters' conception of who can get elected appears to be based on who has been elected in the past." This is entirely unsurprising, and applies to predictive models too. A model can't predict Obama's first nomination a year out because the historical record had never had a non-white President.

What are the underlying factors that determine someone's voting preference? One study said voters elect people with deeper voices. Participants in this research were asked who they would vote for after hearing pairs of voices, one lower-pitched and one higher.

This sounds like a well-designed study that isolates the effect of voice as a factor. That's also why it's problematic. The conclusion relies on the idea of "all else being equal". But in real life, all else isn't equal. There aren't two candidates with the same experience, the same platform, the same financial backing, the same everything else -- except for the pitch of their voice. The familiar setup in the academic study forced participants to select someone purely based on voice. The study measured what people would do if they made selections solely based on voice, but it did not account for how voters actually choose a candidate because most likely, voice is only one factor, if at all.

Another study concluded that women who get elected tended to be more qualified for their jobs than men who get elected. This can be true but the statement is meaningful only if voters based their preferences on who's more qualified. The 2016 election pretty much slayed that theory.

The cited studies fall into two types. One type is the simulated experiment. Conditions are artificial but the environment is carefully controlled, which produces more valid and reliable results. The relevance of experimental results depends on the gap between reality and the experiment.

The other type of study relies on counterfactuals - estimating what might have happened that which did not happen. For example, a study showed the existence of racial bias even though Obama won in 2008. That finding is based on comparing Obama's winning margin relative to what a hypothetical similar white candidate's margin would have been.

"Interaction of polls and media becomes its own self-fulfilling prophecy." This is definitely on point.

Electability is shifting. Ideologues and outliers are winning more often than in the past. A different way of saying this is that electability is specific to which election you're predicting. Candidates who won multiple Senate elections are necessarily competitive in a presidential race.

***

As I end this post, I let one of the voters quoted in FiveThirtyEight have the last words: "Was Trump electable?" he asked.

Comments

[nceladean]

In the context of US presidential elections, I always thought “electability” was a subjective, hypothetical probability judgement of how a given candidate would do in the general election, conditional on them winning their party’s nomination. At least that’s how I’ve been using it.

For example, you might get a candidate who strongly appeals to the party’s ideological base (e.g. Barry Goldwater in ‘64 or George McGovern in ‘74) or an “establishment” favorite (Mondale in ‘84) who ends up winning the nomination only to belly-flop in the general election. Or you might have a centrist or moderate candidate who fares poorly in the primary, but who might do well in the general election against an unpopular rival (e.g. Yang in the current election, who has virtually no chance of winning the primary, but, by virtue of him being a centrist taking on Trump, might fare better in the general election).

[Kaiser]

nceladean: yep, that is how I'd define electability; it's a prediction of who would win in the general election assuming the candidate wins in the primary. I agree with the 538 assessment that it is used to create self-fulfilling prophecy. If voters do pick someone based on "electability", that candidate becomes more electable. It's circular.