Three months out, Biden looking good

With just three months now to go until the United States presidential election, it’s a good time to have a look at what the polls are saying and what we should be focusing on during the campaign.

My preview three months ago gives some statistical background, but here’s a quick recap. In 2016 the Democrats won 51.1% of the national two-party vote, but lost narrowly in the electoral college. To win this time, they need a uniform two-party swing of 0.4%, which would deliver them the three most marginal Republican states: Michigan, Pennsylvania and Wisconsin.

For about the last two months, national polling has been very stable; Democrat Joe Biden has a consistent lead of around eight or nine points over Republican incumbent Donald Trump. (I use FiveThirtyEight‘s polling averages throughout; RealClearPolitics will give you very similar numbers.) The most recent figures show a swing of 3.3%.

Swings, of course, are never uniform, so it’s necessary to look at state results as well – with the caveat that state-by-state polling is not as accurate as national polling. FiveThirtyEight is showing polling averages from 16 states; basically the most marginal on each side of the pendulum. That’s less than a third of the states, but because they tend to be larger ones (including seven of the ten most populous) they cover a bit over 40% of the population.

Even that is overstating the number of voters who will actually have an impact on the result; there’s really not much point, for example, in polling Colorado or Virginia (if Trump gets close there he will have already won). But it means that well over half of the American electorate can and will be simply ignored in the campaign. Not all votes are equal.

Anyway, here are the 16 battleground states, showing for each the number of electoral college votes in brackets, the winner last time with their margin, the current position of the polls and the two-party swing (to the Democrats, in each case) that that represents.

State 2016 winner Margin Leading Margin Swing
Virginia (13) DEM 2.8% DEM 5.9% 3.1%
Colorado (9) DEM 2.7% DEM 7.1% 4.4%
Nevada (6) DEM 1.3% DEM 5.4% 4.1%
Minnesota (10) DEM 0.8% DEM 4.1% 3.3%
New Hampshire (4) DEM 0.2% DEM 5.0% 4.8%
Michigan (16) REP 0.1% DEM 4.3% 4.4%
Wisconsin (10) REP 0.4% DEM 3.8% 4.2%
Pennsylvania (20) REP 0.4% DEM 3.6% 4.0%
Florida (29) REP 0.6% DEM 3.1% 3.7%
North Carolina (15) REP 1.9% DEM 1.2% 3.1%
Arizona (11) REP 1.9% DEM 2.0% 3.9%
Georgia (16) REP 2.7% REP 0.4% 2.3%
Ohio (18) REP 4.3% REP 0.1% 4.2%
Texas (38) REP 4.7% DEM 0.3% 5.0%
Iowa (6) REP 5.1% REP 0.6% 4.5%
South Carolina (9) REP 7.5% REP 4.3% 3.2%

The first thing to notice about these numbers is that they look good for Biden. He’s ahead in not just the three Republican ultra-marginals but also in Florida, North Carolina, Arizona and (very narrowly) Texas, and he’s very close to the lead in another three. If election day panned out like that, he would win a majority of probably 372 to 166 in the electoral college.*

For what it’s worth, that closely matches the current verdict of both the betting markets and expert opinion (not surprisingly, since both of those rely primarily on the polls). Maxim Lott and John Stossel’s aggregation of betting odds gives 334 to 204, the same for every state except Texas. The forecast map at RealClearPolitics says 352 to 186, also keeping Texas Republican but giving Ohio to the Democrats.

The second thing to notice is that the swings are remarkably consistent. Every state but one is shifting Democrat by between three and five points, and even Georgia, the outlier on 2.3%, is pretty close to that. By comparison, in the 2016 election swings in those states ranged from 8.0% to the Republicans in Iowa right through to 2.7% the other way in Arizona.

Consistency in polling should always ring at least a small alarm bell; if I were a Democrat strategist I would worry that something in the pollsters’ methodology (or perhaps in FiveThirtyEight‘s aggregation) was ironing out state-by-state differences too aggressively and perhaps therefore missing out something important. But it could just mean that this is an election where national issues are overshadowing regional differences more than usual.

The third thing, and the one that surprised me the most, is that there’s no sign of the pattern of swing that’s been apparent in recent elections. Over the last two decades the Democrats have lost ground significantly in marginal states in the north and midwest, balanced by gains in those in the south and west. From 1996 to 2016 they went backwards compared to the national average by 4.3% in Ohio, 4.9% in Minnesota and 7.2% in Iowa, but they picked up 4.2% in North Carolina, 7.0% in Colorado and 7.5% in Virginia.

But on the evidence of these polls, that trend has spent itself. The southern and western marginals show a weighted average swing of 3.8%, fractionally less than the 4.1% for the north and midwest. Nor is there any obvious correlation with the degree of urbanisation, another powerful driver of partisanship in recent years.

That’s probably bad news for a candidate, like Trump, who might have been hoping to snatch a narrow victory in the electoral college despite being (again) behind in the popular vote. If regional differences are not becoming more accentuated, that reduces the scope for such a mismatch between votes and seats.

That said, the two fundamental truths about the electoral college remain valid. First, that in a very close election it operates as a randomising factor; who it will end up favoring is essentially unpredictable. And second, that if it’s not close, it doesn’t matter: if a candidate is leading by eight points (or anything like that), they might not know exactly which states they’ll win, but they know it’ll be enough.

So the big problem Trump has at the moment is not specific to any one group of states: it’s the fact that voters nationwide seem to have soured on him and seem fairly set in their opinions.

A lot can happen in three months; there’s certainly still time in which that could change. But something will have to happen to change it. If things continue as they are, Biden is on track to be the next president.


* Technical note: I say “probably” because there are five seats in the electoral college (two in Maine and three in Nebraska) that are allocated by congressional district and so can’t be captured by state polling. Three of them are safe (two Republican, one Democrat) but two currently Republican-held are possible Democrat gains. I’m assuming that Nebraska’s 2nd district will go Democrat and Maine’s 2nd district will stay Republican.

12 thoughts on “Three months out, Biden looking good

  1. Interesting finding on the swings. Maybe what we’re seeing is that the 2016 Trump induced realignment is now baked in, and we’re just working off that baseline.

    My own take is that Trump hit fast forward on a realignment that was going to happen eventually anyway. It will probably still continue long term, but perhaps now we’re just hitting the pause button.


    1. I think that’s probably right, but the other point is that the “2016 realignment” wasn’t nearly as great as people think – it was more a modest continuation of a trend that was already in progress.


    1. The linked article says that Bari Weiss’s resignation letter is worth reading, so there’s obviously something wrong there.

      But on the larger point, is it possible that the polls are wrong about the election result? Yes, it is. Is it possible that somebody posting on his blog is wrong about the election result? Yes, it is? Of the two, which is more likely to be wrong?

      Opinion polls are an imperfect way of forecasting election results; but every other way of forecasting election results is even more imperfect than opinion polls.

      Liked by 1 person

    2. Things can change in three months; something might well happen to retrieve Trump’s chances. If it does, that won’t really amount to me being wrong, since I specifically made that point. All I’m saying is that if things continue more or less the way they are, Biden will win it.


  2. My comment was perhaps too brief. I intended that the “wrong” be taken to refer to the situation where Trump wins, notwithstanding that things do continue much as they are. Reasoning as per link.


    1. Opinion polls are an imperfect guide to the probabilities, but they have more value as a guide than the reasoning at your link.

      If the polls stay as they are and Trump still wins re-election, my reaction will be that the polls were inaccurate, not for the first time, but that people posting on their blogs are still more often inaccurate than the polls are.

      Liked by 1 person

      1. The article I linked to suggests a reason why the polls may be inaccurate. I find the reason persuasive. There is an element (although far less extreme) of similar intolerance of ‘right wing’ views here in Melbourne, Aus. Could even be why the polls, and our blogger host, wrongly predicted the outcome of our federal election last year.

        In your response to my previous post you say, of the article’s comment on the subject of Bari Weiss, “something wrong there”. A bit cryptic/enigmatic. Care to expand on that?


    2. Thanks Colin. Yes, I can see the argument for why the polls might be understating Trump’s vote, but I find it very hard to believe that the magnitude of that effect could be enough to counteract Biden’s current advantage. But I suppose if you add that to the slow drift towards Trump that looks as if it might already be in progress (Biden’s lead has dropped by about a point in the last month or so), then perhaps it could put him within striking distance.


  3. ‘The article I linked to suggests a reason why the polls may be inaccurate. I find the reason persuasive.’

    It’s not new. It’s even been given a technical name, ‘social desirability bias’. It is, or may be, one of the reasons for polling error; it’s one of several known factors that can contribute to polling error. However, knowing that the possibility exists of social desirability bias contributing to polling error is not in itself evidence that recent polls for the 2020 presidential election are worthless because of social desirability bias.

    It is worth knowing about factors that can contribute to polling error, but the fact that such errors occur is known independently of knowledge of the factors that contribute to them. None of this changes the fact that any other basis for estimating the election outcome is more unreliable than the polls.

    ‘In your response to my previous post you say, of the article’s comment on the subject of Bari Weiss, “something wrong there”. A bit cryptic/enigmatic. Care to expand on that?’

    The article states that it’s worth reading her resignation letter. That can’t be true. Of course, in itself that doesn’t demonstrate that other statements in the article are also inaccurate, but it’s grounds for suspicion.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.