Overview

This US election campaign has simultaneously been very static and very fast moving. Biden’s lead has remained pretty steady and, if anything, actually expanded since Trump’s COVID-19 diagnosis - taking a 10% average lead in mid-October that has since declined to “only” 9%. To state the obvious this is a good lead to have a week out. Never say never but this hasn’t really been a close election so far, despite a lot of punditry implying it is.

However due to both noise from different pollsters chipping in in swing states and possibly some real movement the viable swing states are still in flux even this late on. At the time of the previous post on this Ohio looked like the most realistic win for the Democrats of the four “reach” states of Ohio, Iowa, Georgia and Texas. Now, at least purely on polling data, it’s likely the least. This may change over the next week but, although totally possible, it seems that it’ll be a bit of a reach whereas a few weeks ago I’d have thought Biden would carry it.

Similarly one poll in a rarely polled state may (at least temporarily) shift the narrative. As an example on Twitter Jaime Harrison for the South Carolina Senate seat was considered a real contender up to the 14th, when a poll from a credible firm put him 6% behind. Suddenly he was a loser and taking money from more viable contests. Then after two polls with his having a slight lead he once again became a real contender. In reality he’s a credible candidate who’s probably slightly odds off in his race. His winning wouldn’t be a surprise but it’s more likely he won’t.

Also of note is that the election has effectively been ongoing for a while now and the number of votes cast is at around 55% relative to total 2016 turnout, and at over 90% in Texas. It doesn’t feel like it watching from the outside but for many Americans their contribution to the election is done. This high turnout is good news for Democrats not only for the usual reason - their coalition includes most ethnic minorities and also younger people, both groups that turn out less often than older whites - but also because COVID-19 is somehow partisan in the USA. This means Republicans are more likely to vote in-person so these votes cast in advance are disproportionately Democratic and, importantly, already banked.

For an up to date version, or to look at more detail at individual states or party registration click here. I also haven’t discussed party registration or the whispers about what states are particularly good or relatively bad for the Democrats based off the figures so far. Namely because I’m not convinced how useful these indicators actually are at predicting results. Doubtless someone will work it out after the fact.

As a final note on voting in advance - many states do not start counting these votes until after polls close and allow votes received after election day to be counted. So it may be a while before marginal results are known, and, once again, expect a more red election night compared to the final result - particularly in the north.

However one state that does count these votes in advance is Florida, so unless it’s its usual knife-edge self this year we may at least have one key state called on the night. However given it’s Florida best not tempt fate, its results often take weeks to come in during a normal year. Pennsylvania, by, contrast, will be a long wait unless the Democrats win very decisively with on-the-day in-person votes.

Updated states of note

States have shifted a tad (or perhaps the data is noisy) so a quick update. As Biden is running ahead of where he was in my last post the long shot states are also a bit more realistic, though I don’t think the realistic Democrat gains go beyond Texas, Georgia, Ohio and Iowa at the presidential level. As interesting as a blue Missouri or Montana may be I don’t think the swing to Biden is quite robust enough for that!

The group of four states I mentioned in the last paragraph are also the ones I’m by far the least certain about at the presidential level. In 2012 I felt confident calling every state save Florida - ultimately I did call all 50 states correctly but Florida was around a 55-45 call. This is almost like four 2012 Floridas!

Somehow I knew there’d be a Florida variant on this meme.

Michigan (Senate)

For a while it looked like this race was tightening up, even if you ignore Trafalgar (as you should). That seems to have unwound by now and as no reputable poll has shown the Republican challenger in the lead for a very long time I don’t expect anything other than a Democrat hold here.

I am, however, glad I bothered to include this state in my last post. Democrat incumbent Peters will probably win by a smaller margin than Joe Biden in this state - showing the value of a good candidate even in a no-hope race.

Georgia (both)

For the Senate seats there’s been a fair amount of polling. For the special election the Democrat Warnock looks very likely to proceed into the next round. Naturally the climate may be different at the time of the run-off but at present I think Warnock is the favourite to win overall. At the time of my last post considering this special election more likely Democratic than the ordinary election was a much bolder prediction than it is now.

The data around the main Senate election (Ossoff vs Purdue) is bouncing around a bit - truth be told I think it’s likely a dead heat with the leads for either candidate largely being noise.

On the presidential level it’s got very close indeed. Biden a sliver ahead.

Montana (Senate)

The polling seems to have settled into a pattern of very small leads for Republican incumbent Steve Daines.

Naturally I think this means Daines is the narrow favourite but as his leads are well within the margin of error we may not have seen a Bullock lead in a while just because there haven’t been many polls. Truth be told neither possible result would truly surprise me, there’s a strong candidate running against the partisan lean of the state on a good year for his party versus a weak candidate with partisanship on his side. A recipe for a close race.

Texas (both)

Based purely on the presidential polling a Democrat win here is thoroughly realistic but odds-off. However between a history of Democrats being underpolled in Texas as recently as 2018 (including in 2016), the sheer scale of early voting in the state and the huge last minute push planned for the state I think it’s currently a true toss-up. In fact I’m very close to saying I think a Democrat gain slightly more likely than a Republican hold and, had I wrote this a couple of days later, perhaps I would say that. There’s definitely something going on here.

However, as we’re already talking about Texas, in stark contrast to the presidential race I don’t think there’s much chance of a Democrat win at the Senate level - Hegar is running far behind Biden. It’s certainly not impossible that she somehow wins but the leads Republican incumbent Cornyn has built up are large enough that it would be quite the upset.

Mississippi (Senate)

I drafted this on Monday when there’d been new polls. The very next day we finally get some follow up and we can finally know if this race is genuinely competitive.

Long story short: barring polling error probably not. The latest poll shows Democrat challenger Mike Espy 8% behind, which is a great result for Mississippi. However if that’s where the race genuinely is then it’s pretty unlikely he’ll make it over the winning line. He is, however, slightly more likely than his colleague Hegar in Texas and is running several points ahead of Biden - once again first-past-the-post proves its cruelty to a talented candidate in the wrong place.

Alaska (Senate)

This is now looking more firm for the Republican. Upsets happen, and certainly it’d take a smaller one that Texas or Mississippi for this to flip, but I think Sullivan will likely win. In terms of likelihood to flip I’d say it’s slightly less likely than the Senate seats in South Carolina or Kansas.

Alabama (Senate)

One single poll shows the Democrat Jones ahead, but most show him double-digits underwater.

I’d be very surprised if the single poll was correct, it looks like an outlier, but I thought it worth mentioning given how scarcely this state is polled. Given how uncompetitive this state is I could easily see a systemic underestimation of Democrat vote share. I can’t really see a big enough one to carry Jones over the winning line.

Apparently Jones is being considered for the attorney general role in the new Biden government, should the Democrats win, so his career is looking healthy either way.

Iowa (both)

In hindsight my prediction that Iowa’s Senate seat would go for Theresa Greenfield (Democrat - challenger) on an average night was a bit bold, it really was a bit more marginal than that. However, though I can never be 100% certain, the data does look better now than it did then so I think, purely by accident, the prediction holds up, luck has corrected my misstep and I don’t need to update that map.

On the presidential side Iowa’s definitely got a bit rosier for Biden and he’s gained moved into the lead. This is confirmed by very reputable pollsters but it still seems odd to see Iowa to the left of Ohio. However this isn’t totally unprecedented as Iowa voted for Gore in 2000 while Ohio voted for Bush Jr.

North Carolina (Senate)

On the 4th of October Cal Cunningham (the Democrat challenger) had a sexting scandal and I wondered if it would have any impact. Turns out no.

Ohio (presidential)

At the time of my last post Biden looked like the narrow leader in this state - at the moment he looks like a marginally less narrow loser here. Notably some of the firms that gave him a firm lead then give him a narrow loss now so, although I can’t rule it out, I think this is probably a real shift rather than noise in the data.

If not allowed to cop-out by claiming a toss-up I’d expect Ohio to stay red, particularly as its early voting turnout has been relatively low compared to many states (41% of 2016 total turnout vs 77% in Georgia or 91% in Texas). But Trump’s margin is firmly within the margin of error. It’s not certain.

The same applies for similarly close states, no matter whose nose is ahead at the moment.

Polling in the USA

I realise that few of my readers are dedicated poll-watchers in the UK but I think it’s worth mentioning that the landscape for pollsters between the UK and USA is quite different and poll-watching American elections feels very different to doing so with most other countries, UK included.

For a start there are far more pollsters in the USA than in the UK, with American polls you never seem to stop seeing new names. American pollsters are also far less regulated - in the UK a polling firm cannot withhold polls on request of the publisher under BPC rules. In the USA, so far as I’m aware, it’s rare but not unheard of for this to occur - with Gravis’ polls for One America News Network reportedly released selectively earlier this year. This is not the only mini-scandal in the world of US polls.

Two other things that are legal in the USA but not here are robo-caller polls and explicitly partisan polls. In general I’ve found Democratic-leaning partisan polls are mostly to be taken with a slight pinch of salt and Republican-leaning ones are to be ignored altogether - often having strange methodologies such as apportioning arbitrary, variable percentages to the Republicans to account for “the silent majority” or dropping all data from newly registered voters. As with many topics you can correctly claim both sides aren’t ideal but that one is far worse than the other. Polls with a lower sample size than the industry standard of 1000 and therefore a higher margin of error are also more frequently published, with 400 or even 200 participant polls being quite common.

So it’s good to have some idea of what pollsters are good before trusting a poll. Unfortunately a good firm in one market may not always translate to others - YouGov operates in both the UK and USA and has a firmly good reputation in the UK, aside from with a few vocal Corbyn fans. In particular its MRP models make other election models in the UK completely obsolete with some real surprises in 2017 and 2019 being accurately predicted. However in the USA YouGov’s nothing special. Other assumptions you might make aren’t always true - perhaps the most surprising one being Fox News polls, completely unlike the news station, are actually extremely reputable and Fox News is one of the most trustworthy names in the business in my view.

Unfortunately there’s no quick and easy way to know if a poll is good, short of following them for a while and testing them against real results, but a quick and dirty shortcut could be to go to 538 and check out their polling ratings. These are based off prior results and assessments of methodology but are not perfect - Emerson in particular is well-rated on 538 but seems to have a poor recent reputation. Nevertheless it’s probably the simplest shortcut to knowing what to trust despite it being a rather imperfect site.

An example of 538 ratings. I haven’t discussed likely voters (LV) versus registered voters (RV) versus voters (V), though I think the idea is reasonably self-explanatory.

An alternative is to use Twitter but for some reason US polling accounts are often poor quality. Some make dramatic statements about polling shifts that are within the margin of error, a big no-no, for instance. Generally poll watchers in the states (“Election Twitter”) tend to be more engaged than in the UK, with far more amateur cartographers out there, but also in my view more reckless, with a tendency to break into the crosstabs of polls to try and glean more data or ascertain its reliability. This can be legitimate but as crosstabs have large margins of error even with large samples it’s worth taking much of their discussion with a pinch of salt. UK polling Twitter would tend to point that out but I’ve yet to see it on American polling Twitter.

Having said most US polling accounts are a little sketchy I would recommend PollingUSA and Stella2020. Neither are perfect, PollingUSA are affiliated with LeanTossup, an election model I’m not a fan of (partially for the model itself, partially for their rather snotty attitude to their competitors), and they also don’t incorporate 538 poll rankings or any alternatives. Stella2020 is partially automated and tends to be a chore to read whenever SurveyMonkey vomits out a dozen poor quality polls at the same time. But on the plus side PollingUSA tends to provide links to the PDFs of polls and Stella2020 is a good girl, yes she is!

Stella.

If you’re not interested in working out potential results yourself from polling data there’s no shortage of independent projections for you to peruse. Some examples I won’t elaborate on are The Economist, JHK Forecasts and Our Progress. For familiar British names that don’t seem to have as much fame in the US there’s also The New Statesman and YouGov. Huffington Post seems to have stopped doing projections after they bombed last time.

538 has some flaws, notably Nate Silver has been quite bad for pushing a two-horse race narrative which doesn’t really match his own data, but the site has also a good reputation for being the reputable modeler showing the highest chance of a Trump victory in 2016. This year it gives him a measly 11% chance of victory. Some might say that’s still quite kind to him. It also has quite a fun way of showing their predictions without a map:

Iowa’s exactly even on this chart and was blue yesterday - most of these sites have similar rankings.

Broadly I think these are useful resources but due to the surprise of 2016 many have been coded to be a bit conservative in their outlook and, until recently, underestimated Republican strength assuming Biden’s leads would decline over time. That Arizona was a likely Democrat pickup has been clear for months but most projections only moved it out of the toss-up column pretty recently. The exception is LeanTossup which is, by contrast, overconfident about the Democrats and very vocal about it. They’ve projected a blue Texas for months and although I could easily see that being correct their model was very premature in predicting it, although it was close the evidence for it flipping simply wasn’t there at the time.

Technically RealClearPolitics has a prediction too but, to be honest, it’s an absolute joke and I’m including it because mostly because it’s funny how indecisive and unbalanced it is:

On this model Minnesota has a comparable lead for the Democrats as the Republicans have in Alaska. Somehow one’s a toss-up and the other likely Republican.

Finally I’ll talk briefly about polls at the national, state and district level. The dynamics between these and UK-wide, country-level and constituency-level aren’t really the same.

National polls tend to be more accurate than state polls as states without regular polling can often be especially difficult to work in. Misses are frankly more common in state polls (no matter what the office) than in national ones. The obvious downside is that the popular vote doesn’t win the election, so in itself it doesn’t tell you what you want to know.

State polls are more scarce than national polls and the quality varies wildly from state to state, with some quirks unique to each - for instance it’s often stated that Nevada polls tend to underestimate Democrats whereas Alaska is more known for simply being difficult to poll. The obvious upside is that this is directly what’s of interest when it comes to presidential elections (also senatorial or gubernatorial).

District level polling is a world I’ve not got into, note my lack of a House prediction (it’s incredibly unlikely to flip, that’s all I’ll say!), but by measuring the difference between many districts’ polling numbers and their last result an estimate of swing can be obtained. Reportedly doing this in 2016 showed a collapse in Clinton’s vote ahead of the election, although there’s no guarantee they’ll be accurate this time, they may be of interest. A good summary can be found here.

I personally tend to look at state polls as, frankly, I’m rarely interested in more than the win or the loss and, 2016 aside, they’re usually an adequate guide for that. However it is worth saying that this time the three levels of polling are arguably telling slightly different stories. District level polling implies a larger Biden lead than the national polls (though it’s worth mentioning one of its biggest advocates warns this may be illusory), whereas state polls vary between implying similar or worse results for Biden than the national polls. We’ll know soon!

New maps

I’m going to take this opportunity to update one of my predictions. Despite Biden’s figures being a smidge more rosy than during my last post neither candidate has really broken new ground in the last few weeks so the best and worst case scenarios for the Democrats are identical in my view.

However my “as expected” scenario does need an update as Ohio has slipped a bit for the Democrats while Georgia and Iowa have improved to the extent that the blue are slight favourites. They’re also doing better in Texas but in my view not quite enough to be the front-runners. At least yet. I’m half-expecting to change my mind on that one.

357-181 - still not quite beating Obama 2008!

Although how certain I am in my choices has shifted a bit since my last post on this topic I also don’t think my Senate predictions have shifted. I’m tempted to add Mississippi to the “very good day for the Democrats” case but, truth be told, I think it replaces the Texas Senate race as the closest race that’s just out of reach in any realistic scenario.

In any case, hopefully an enjoyable political event ahead of us in the next week. For anyone else watching the results I hope you’ve a great evening!