73

The Economist model never showed Trump as a clear favorite. In recent weeks, it said he has had less than 10% chance of winning the electoral college. It currently says his chances are particularly bad; 4% chance of winning the electoral college, and <1% chance of winning the popular vote.

I am interested in the fundamentals of this model and why it is sure now of what is happening. Is it the early voting, polls, a combination, or something else?

  • 1
    See this discussion. – Alan Oct 27 at 19:49
  • 25
    This would be the same model they used to predict Hillary's "inevitable" victory in 2016? – Valorum Oct 27 at 20:34
  • 7
    Related: "Election Update: Why Our Model Is More Bullish Than Others On Trump", FiveThirtyEight (2016-10-24). (The title's a bit misleading; FiveThirtyEight likewise predicted Clinton as more likely, but they claimed less certainty than others. That blog post explains why.) – Nat Oct 28 at 6:53
  • 14
    @Valorum They didn't predict an "inevitable" victory, they predicted odds of her winning. And even 99% odds don't mean the other 1% can't happen; that's how I understand it anyways. – fweigl Oct 28 at 12:17
  • 4
    This isn't an answer because the question is about the methodology of a particular model, but it's worth noting that given that an incumbent up for re-election has lost only once in the last 40 years (and the challenger was the almost inhumanly charismatic Bill Clinton), People are also somewhat famously more conservative than they'll admit in polls. I have a low prior on the accuracy of a model that gives another upset at 1 in 20 odds. If you live in the US make sure you go vote no matter how many polls tell you not to bother, prediction is hard. – Jared Smith Oct 28 at 20:04
168

In comparison to the 538 model, the economist model uses less "fat tails".

538 uses a t-distribution to account for "black swan" events: things that, although they are unlikely, would have a big impact on the polls. This means that 538 assigns a small probability to some very unlikely outcomes (Trump wins California, or Biden wins Utah) The Economist model doesn't. It uses a normal distribution which has thinner tails, and so treats these kind of results as essentially impossible. Moreover the 538 model uses data back to 1936 to try to estimate how much polls can change between now and the Election. The Economist uses only more recent data.

The effect of these modelling assumptions is to make extreme events more likely. For Trump to win now would require either a very significant failure in polling or a large swing in several states. The economist model essentially says "A swing like that has never happened, so I'll assume it won't" This results in about 95% chance for Biden. The 538 model says "Very large swings have occurred, if you look back to (for example) Truman v. Dewey. And sometimes something might happen that will completely change the course of the election (War was declared or a major scandal) and comes up with about 88% for Biden.

The change from a few weeks ago to now is that Trump is running out of time to make a big change. Both models were set to wait till after the debate, to see if it could make a change. It didn't (much). The assumption is that the gap will narrow as you get towards polling day. But this narrowing hasn't happened. This leads to greater confidence in a Biden win.

The difference is 538's "fat tails".

| improve this answer | |
  • Comments are not for extended discussion; this conversation has been moved to chat. – CDJB Oct 29 at 9:09
  • Probably also have to notate that none of these models are even remotely accurate this year, since any major event that would normally have influenced the election is reduced in effect, since many states have been collecting ballots for weeks, ie. Biden says something outrageous, making people who already cast their ballot regret their vote, even though it's technically before election day. – SnakeDoc Oct 29 at 22:54
5

The answer(s) could come in at almost any step in their forecasting process, probably several. By breaking down the inputs, the process, and the outputs, we can at least get a sense of the possible answers to the question.

Inputs

These models primarily use polls as input data. They may include other sources like prediction markets. The Economist model may use different sources or put more trust in some polls than others, compared to e.g. 538. There is also historical data, which may be relied on differently by different models.

Process

The Economist model simulates votes randomly to see what might happen. To do that, it needs to make a lot of assumptions about the process, including the distributions of votes for each state, how the states are correlated, and what the sources of error in their forecasts could be and how much impact they could have.

  • Maybe the polls don't represent the true population of "likely voters".
  • Maybe the likely voters don't represent the "actual voters", e.g. maybe people who work tough jobs systematically turn out less even though they intended to vote.
  • Maybe there are other impacts on vote totals, such barriers to vote stemming from too few drop boxes or polling places and long lines. Maybe a large number of mail-in votes will be rejected for signature mismatches, or a state court will decide to invalidate them for other reasons.
  • Maybe people lied to the polls in the first place (or were lying to themselves at the time).

You have to decide your sources of error around all these things. When I look at the Economist's forecast for Michigan, they predict vote totals that match the polls on average, and they predict a very high probability that the outcome will match the majority in the polls. That indicates they think their sources of uncertainty above are low (for some reason). This was a big concern after the 2016 election, and with 538's prominent place in the discussion, they've focused a lot on this issue.

But this is not enough. You also need to decide your sources of uncertainty around correlation between states. If Michigan is an upset, what is the chance that Wisconsin is also an upset? The Economist has a lot of details on their beliefs here, but it could be that they put less uncertainty on these factors than other models.

Outputs

The Economist model may be predicting something different than other models. Consider just the following two examples:

  • If nothing newsworthy happens before election day, who (if anyone) would get a majority of cast ballots in a set of states totalling 270+ electoral votes?
  • Who (if anyone) will record a majority of the officially-announced vote total in a set of states totalling 270+ electoral votes?
  • Who (if anyone) will be sworn in as President on January 20, 2021?

Between 1 and 2 are many sources of uncertainty: How does turnout among poll respondents compare to general population? How is that biased by party? Will there be any game-changing events such as a scandal unfolding in the next week? Etc. There are also sources of uncertainty between 2 and 3, such as whether the Electoral College votes actually match the state votes.


I want to add another point: incentives. It's not clear that the Economist optimizes their own goals by given the best possible forecast (whatever that means). People often evaluate forecasts in a winner-take-all manner: If you predicted Biden at 60% and he loses, you're heavily criticized for being "wrong". And if he wins in a landslide, you might still be considered "wrong". If you predict 90%, then at least if it does end up being a landslide for Biden then people will praise you. Another more nefarious possibility is that a forecaster biases their prediction in hopes of swaying the actual election outcome.

I don't actually think these are happening here, but it is something to be concerned about in election forecasting.

| improve this answer | |
  • 5
    I don't think "538 was burned by this in 2016" is a reasonable summary (particularly so since the design of their models have been publicly discussed in great detail both years, including reasons for higher uncertainty this year). – Michael Homer Oct 27 at 5:29
  • 1
    @MichaelHomer that's a fair point and I'll edit it; on the other hand, it is accurate that 538 and other forecasts (others more so) were widely criticized and perceived as not building in enough uncertainty around the polls, at least directly after the election. – usul Oct 27 at 5:47
  • 11
    Reminds me of weather forecasts. "It said 90% chance of rain and it didn't rain, this forecast sucks!" – user3067860 Oct 27 at 16:03
  • 11
    @usul 538 gave about 2:1 odds. Ultimately they got the winner wrong, but it seems pretty clear their model was showing the uncertainty. – Bryan Krause Oct 27 at 16:34
  • 12
    @usul: FiveThirtyEight explained it in October 2016. Amusingly, FiveThirtyEight also predicted that, if Trump won, many who don't understand statistics wouldn't understand what happened and still fault them for it. – Nat Oct 28 at 7:48

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .