Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Even odds?

August 31st, 2020, 11:38am by Sam Wang

Betting markets used to favor Biden over Trump. But in the days after the Democratic and Republican conventions, the markets now report 50-50 odds, a toss-up. Is that a realistic assessment?

Based on polling metrics alone, the data so far don’t support this shift. It seems that bettors are factoring in “pre-poll” possibilities such as difficulties in voter access in a way that helps Trump; or “post-poll” disruption such as post-election unrest arising from a slow count, eventually leading to an Electoral College outcome that doesn’t match true state-by-state voter intent. We must do everything we can to guard against these threats to democracy.

But today, let me put that aside. Ignoring the factors above, what would it take for conditions to suggest a true even-odds Presidential election, assuming the election is held without major incident? Based on polls only, here are some thresholds to watch for:

  • Biden Meta-Margin at 0.0%
  • Biden ahead in national popular vote by 2.5%
  • Democrats lead generic Congressional ballot by 3%
  • Trump job approval at 46-47%

Here is why.

The Meta-Margin is self-explanatory. As we’ve calculated here at PEC since 2008, the Meta-Margin is defined as what across-the-board polling shift would be needed to create a perfect toss-up between the Democratic and Republican candidates based on all available state polls. Currently the Meta-Margin is Biden +5.1%. (For comparison, Clinton had +2% on Election Eve, and this was off by about 3% based on real election results).

The popular vote requires a little explanation. It is well-known that the Electoral College doesn’t always give the Presidency to the popular-vote winner. In elections where the popular vote margin is less than 3% of votes cast, it fails to do so about 1 in 3 times (for a history of how this almost-popular-vote system came about, see our recent law article). In 2016 it favored the Republicans; in 2004 it favored Democrats (though they still lost). Who does it favor this year?

Comparing national polls and the state-poll Meta-Margin, it looks like the Electoral College currently favors Republicans by 2 or 3 percentage points: 

This is a plot of our Meta-Margin for the 1st of the month from March until now (the last data point, August 30) plotted against the FiveThirtyEight average for Biden-vs-Trump in national polls. The median difference is 2.4 percentage points; that offset is indicated by the red diagonal line. In other words, if Biden leads Trump by 2.4 points in national polls, the Electoral College is likely to be close to a toss-up.

As of September 1, with 8 post-RNC surveys, Biden is up by 7.1 points, about one point smaller than it’s been for most of June through August. His peak was a lead of 9.3 points right between the Democratic and Republican conventions. That 2-point range is amazingly narrow. I wonder if we’re in some kind of heat death of modern politics.

By the same token, the generic Congressional ballot would need to be at around the same threshold, since people have a strong tendency these days to vote the same way up and down ballot. Currently the generic Congressional ballot is at D+9.8%. Republicans have a structural advantage in the House of Representatives from population clustering and partisan gerrymandering that amounts, coincidentally, to 3 points. So Democrats are currently about 7 points above threshold to keep control of the House (that’s why the graph says D+6.8%).

Finally, we come to the Presidential approval rating. As Kyle Kondik pointed out, an incumbent President’s popular-vote share is typically 0 to 3 points above his approval rating at the time of the election. If we assume that 2 percent of the vote will go to minor candidates, Trump needs an approval rating of about 45% to enter the edge of the range of losing the popular vote, Trump 48% to Biden 50%, a margin that would correspond to a toss-up in the Electoral College. The more likely range for such an outcome is approval of 46-47%. Trump is currently at 43.3%.

In short, the indicators don’t seem to reflect what the markets are saying. Either the markets are being emotional in some way, or they are factoring in the possibility of extra-democratic events.

If you want to help, the banner above lists the “Moneyball” states where citizen effort has the most leverage in getting to 270 electoral votes – for either candidate. Leverage is measured in terms of per-voter power. For the Presidency the banner currently lists “AK AR GA.” Therefore the most effective places to get out the vote, phone-bank, register new voters – or secure an honest election – are Alaska, Arkansas, and Georgia.

Tags: 2020 Election · President

26 Comments so far ↓

  • Tim Vandehey

    I’m a first-time reader, and I’m impressed. I’m a journalist, and I much prefer sober analysis based on data and fundamentals than the partisan shrieking that seems to pass for analysis today. That said, there are 3 popular presidential election models floating around online right now—FiveThirtyEight, The Economist, and ElectoralVote—and I’d love to know what your thoughts are on the accuracy, methodology and impartiality of each. Btw, I don’t include RealClearPolitics because, unless I’m mistaken, they don’t weight polls and so take nakedly partisan polls like Trafalgar, Rasmussen and Civiqs at face value. Thanks, and KUTGW!

    • Sam Wang

      tl;dr – From a math standpoint, I think all three models are fine. To my knowledge (and I have communicated with all of them at some point) they are all honest brokers.

      I prefer because it is relatively unvarnished, there are no hidden parts, and also no conversion to probability. I think probability is harder to understand and makes people passive consumers. See my CJR and NYT pieces, linked elsewhere on this thread.

      Thank you!


  • Tim Vandehey

    By the way, in asking about 538, I am aware of your history with Nate Silver. But I know they’ve allegedly changed their model and I’m curious about your take. Thanks.

    • Sam Wang

      I wrote about what I think would be a good approach in the Columbia Journalism Review this year, and in the New York Times in 2016.

      FiveThirtyEight has to have increased the uncertainties in their model. Otherwise a Biden 5-point lead would give a larger win probability than they are showing. In retrospect their Clinton win probability was fine. It comes with an implied amount of implicit uncertainty. Applying that to this year would give a Biden win probability well above 80%. That would run into obvious marketing problems – they would get yelled at by readers of all stripes. So they have to be more cautious.

      In part, they are dealing with this by putting the win probability on their home page using English-language words rather than a percentage number. I think that is fine. I would rather they give something like a Meta-Margin…but maybe that’s too hard for a mass market to understand.

  • Emigre

    Hi Sam,
    thank you for evaluating the betting markets predictions. In addition can you also enlighten us what is going on with the USC/Dornsife poll again, please?
    As you remember in 2016 they were the outliers who had Trump winning before the election.
    Now they are the opposite having Biden significantly ahead.
    I am aware of your reluctance to spent time on single polls and pollsters but based on their track record it might be worth your effort.

    • Sam Wang

      If Dornsife’s methods are the same, then I think they might have a potential problem with microweighting, which can give them more stable results but could also lead to weird offsets in small demographics. I wrote about this in 2016.

  • Ebenezer Scrooge

    I’m not impressed by the betting markets. They are thin markets with many noise traders. (The noise probably is biased.) They are also subject to manipulation, because an investment of a few thousand $ can move the odds enough to generate some juicy headlines.

  • Matt McIrvin

    The betting markets generally had Trump favored to win back in the spring, even though his approval was already about as low as it is now and he polled badly against most possible Democratic nominees. I assume a lot of that was just a simple prior favoring the incumbent.

  • Richard Wiener

    There is a third possibility other than markets being “emotional in some way or factoring in the possibility of extra-democratic events.” Markets are forward looking. A substantial number of people playing the market think the race is going to narrow substantially in the next two months and mirror 2016 with Biden fading and Trump coming on strong. For example, some think Trump’s law and order, racist campaign message is going to work in the Midwest swing states. They might also think there is going to be another significant polling error in Trump’s favor. I don’t agree, but it is certainly possible and about half the market is betting on it.

    • Richard Wiener

      Just to be clear, what I’m saying is that markets don’t always reflect current information. People try to make money by anticipating future information. That’s not necessarily emotional nor counting on extra-democratic events. In most I most cases I suspect it is rational. My guess is that all three of these factors (emotion, the possibility of extra-democratic events, and using information other than polling data to determine which way the wind is blowing) are involved in the market movement to Trump.

  • J Jacques

    I don’t see mention in this thread of the effect a robust 3rd party vote had on the 2016 outcome, and how that’s not present this time around. If Trump gets 47% of the vote this time, he’s doomed. Realistically, he was lucky with some very close wins that even his people didn’t anticipate.

    Assuming a 51-47 national vote split this time, is it still your view that results in an electoral vote tossup? That was the margin in 2012 and the Electoral College was not close.

    The Stein and Johnson tally of just shy of 6% was the factor that swung the EC to Trump in 2016, and few prognosticators are talking much about it. Without that scenario playing out in 2020, I would think Trump would need to get to 49% or so before having a coin flip shot at winning the EC. And he’s a long way from getting there.

    • Sam Wang

      In this piece I assumed 2% minor-party support, in which case Trump needs about 48% of the vote (and Biden 50%) to make the Electoral College an even-odds proposition.

  • Siobhan McLaughlin

    dr wang: thank you for giving some of us ( even those like me who have NO CLUE) hope. I very much appreciate all your efforts and hope to sway my family in NC…

  • Amitabh Lath

    The Iowa Electronic Market is one of the ancient real-money markets (you can tell it’s old, who calls anything Electronic anymore).

    I guess I trust it more because run by academics. (yeah I hear you Sam, I should know better…)

    They have a thinly traded vote share market and a higher volume winner-take-all market. Trading at 73% Dem, pretty stable all of August.

    As for the the more well-known markets I wouldn’t put it past the deep pockets to throw $$ at it to generate column-inches. Similar thing happened with the Romney market late in the day.

  • Brent

    538 does have the race tightening, with the spread now at 7.1% from a high of 9.3% as recently as August 24th. That is a significant shift if it is the beginning of a trend.

    • Sam Wang

      Rather than speculate if it’s part of a trend, look at the input data. It supports your view, but allows you to have a little more confidence. There are 8 national surveys whose dates span August 28 or later, i.e. after the Republican convention. That’s a lot of data!

      However, your benchmark appears to be a moment between the Democratic and Republican conventions, so you might be capturing the full-scale range of fluctuation in both directions, rather than a trend over time. Data from earlier times is consistent with that. If so, I wonder if it’s a 2-point maximum range, which is amazingly narrow.

      There are only 2 surveys of the generic Congressional ballot that meet the same criterion.

      For now, national surveys are the fastest-responding indicator, generic Congressional next-fastest, and Senate…there’s only 1 post-convention survey, though it’s a juicy one – a tie in Alaska.

  • Michael D

    I think the discrepancy is due to markets considering things that polls don’t, like vote machine tampering, broken machines, power outages, ballots not showing up in the mail, ballots lost in transit, voter intimidation and voter suppression.

  • Investorian

    “I have seen you get it wrong badly twice now but yet you still stick to your beliefs that polling data and shifts are reflective of voting behavior (within well-behaved error bounds). Both voting choice and decision to vote operate more like a virus than a random walk and hence are highly non-linear. This non-linearity becomes increasingly pronounced as election day approaches. One day you will heed this message.”

    (I’m allowing this comment even though it seems to be antiscientific word salad. -Sam)

    • Ed Fernandez

      In your article you posit that betting markets are factoring in some anti democratic behavior or are emotional. Both are possible but I think it’s much more likely that bettors are looking at the trends over the past three weeks and projecting the race to tighten further. At this point Trump only needs to gain 1-2 points to put the majority of the battleground states into tossup territory. And the polling in battlegrounds has been consistently tightening as Trump edges back towards his pre pandemic approval ratting of 44 or 45. That doesn’t seem overly emotional of the markets.

    • Sam Wang

      By now it is pretty clear that polling data provides a view of voter behavior that changes in response to events, and can be predictive of eventual outcomes.

      To learn more about polls’ predictive value in the last few years, see this CJR article.

  • J Jacques

    Nate Silver just posted Biden’s odds of winning EC based upon national vote split:

    0-1 points: just 6%!
    1-2 points: 22%
    2-3 points: 46%
    3-4 points: 74%
    4-5 points: 89%
    5-6 points: 98%
    6-7 points: 99%

    The top numbers seem hard to believe this time around, what with the lack of a robust 3rd party candidate and the good fortune that Trump had last time in the rust belt. But maybe I’m delusional.

    If true, that means that really needs to be up 4+ to have robust shot at winning. I don’t buy it.

    • Sam Wang

      Yes, this is exactly what I said in this post. I tried to do it transparently so you could follow along, instead of taking someone else’s word for it.

  • Michael K

    Hi Sam,

    Can you clarify how you calculate “Presidential race vote power by state” this time around?

    Intuitively and based on past observation, I would expect states closer to the meta-margin to appear higher on list, all else being equal.

    For example, the current meta-margin is 5.5%. You have Arkansas with 6 electoral votes at R+2.0%, which is 7.5% from the meta-margin.
    Nevada also has 6 electoral votes and is at D+4.0%, or just 1.5% from the meta-margin. Both states have virtually identical populations (~3 million people).

    So shouldn’t Nevada, which is much closer to being a tipping point state than Arkansas, be ranked higher in voter power? Right now you have Arkansas as the #1 Presidential Moneyball state. What am I missing?

    Does this have something to do with the way you are modelling uncertainty and because Arkansas has just one qualifying poll, which was also not a recent one?

    • Mediaglyphic

      Dr. Wang is there a brief overview of the presidential power calculation somewhere? I see a link for the code, something in prose would be interesting. I believe you posted something a while Ba k, but I can’t seem to locate it.

  • Ray

    Interesting thought: how big are those betting markets, actually, on the scale of Presidential campaign money?

    I ask because I’m curious how much it would take to manipulate them as a subtle form of campaign manipulation (no, I’m not sure which direction that effect would be… I’m just curious how plausible it is).

    • ArcticStones

      On a related note, I suspect some “pollsters” are primarily in the business to influence public opinion rather than to objectively measure it.

Leave a Comment