About Last Night
The US election, polling, and how to not make assumptions when reading polls in volatile situations
The Smokeless Room is a newsletter by Rushaa Louise Hamid designed to help you clear the smoke from the air and better understand the tools of decision makers, with a special focus on all you need to know to sift the bunk from the gold of opinion surveys!
Thoughts on the US Election
Like a lot of people I spent last night through to the early hours of this morning following along with the US election. It's an election I've always thought was going to be close regardless of polling since platforming as “the sensible party” rarely is a vote winner in and of itself.
US polling is a nightmare in a lot of ways. I can attest to this once having to work on a methodology check to see how accurate predictions could be. State polls will often have radically different polling methods - robocalls when an electronic voice directs you to click a number for your preferred candidate vs online for instance - making aggregation incredibly difficult, and often requiring a lot of data modeling to get around since you're not comparing like-to-like.
There are national polls which can often inflate the opinion of single members of minority groups as they often use what are “interlocking” quotas even if they have very small sample sizes for a group. Basically these are quotas as we've discussed but instead of just checking you have the right amount of red vs blue and young vs old overall, you are looking for the amount young red vs young blue vs old red vs old blue. This can - and has - lead to dramatic effects if someone in a group has an outlier opinion.
And that's before you get onto the fact that the electoral college and the popular vote are two very, very different things.
Because of all of this I have never thought it appropriate to comment too much on American electoral polling, but last night does offer a good case study in issues less commonly brought up that create problems for polls in all countries.
The US is huge - phenomenally so, in both population and sheer breadth. And it is huge in a way that creates disruptions to turnout even for those who intend to vote. And this is before you factor in a pandemic.
Many states laws make it exceptionally hard to vote, requiring photoID, exceptionally early voter registration, and disenfranchising those with convictions. Over election day you will always get reports of people being denied a vote even when eligible and with the distances required to travel that can effectively mean that even if the solution is to simply collect a drivers licence in practical terms they have been denied their right.
Polling stations or mail ballot drop off boxes might be many many miles out from where you are, lengths which to some might be impossible to cross whilst balancing work, especially when factoring in wait times that can be multiple hours even with early voting. Weather, already a big factor in UK politics, can be an even bigger impediment in these situations.
It can mean that estimating likely turnout of an individual is very hard, even if you have how strong their intention to vote is.
The Ethnicity “Puzzle”
One of the discussions that regularly pops online - last night being no exception - is the surprise when minority ethnic groups vote in what is seen as surprising way. In this current election the strengthening turnout for Trump of the Latino vote was the focus.
A large part of this issue is a result of grouping voters in ethnic categories that don’t mean much in terms of highlighting alligned interests. The States is extraordinarily diverse in all their ethnic minoritiy communities, and the Latino categorisation encompasses a range of races, country backgrounds, wealth classes, and family histories. These alliances tend to mean much more than the overall ethnic group label.
For instance you may not know this but there is an entire Welsh town in Argentina where there are Welsh language schools. It makes sense that the white Argentinan from Y Wladfa now living in Pennsylvania working as an accountant with the surname Jones likely has different concerns to the indigenous Mexican-American who lives in California, has a low income job, and is visibily seen to be a minority.
You obviously get this diveristy in all groups, but there is a strong tendency to lump ethnic minorities in the same democratic box and then have pundits surprised when so areas behave different even if breaking it down it makes sense. Someone might prioritise their Evangelical faith in an election or be a high earner who is looking to pay less tax. Using quotas only makes sense if that quota has a strong correlation with what opinion someone might hold.
The States Themselves
It is often said that the individual states are like countries in and of themselves and in polling terms I think there's definitely a lot of truth in that. So often comparing turnout swings in one state can be as absurd as comparing Australian election results to the UK. What motivates voters in some areas can be so radically different it's not worth it.
A familiar example that is easy to grasp if we look at UK national trends is that we would never say Brighton Pavilion would go Green over Labour based on what we know of the national picture. Yet it does thanks to local factors, and the ground game of the Green Party. National polling is poor at predicting areas that will go against a norm for valid reasons - some of which are intricately linked to how many people are on the streets canvassing.
Something I’ve never understood about the States is that the news can discuss projections and votes coming out of the east coast before polls even shut in the west coast. This is especially pertinent in situations like the election now when you expected to - and did - see swings back and forth between the two main candidates.
It’s the age old problem writ large. How much do polls and other similar projections influence whether people bother to vote at all or who they put their faith in? If you think someone is a no-hoper do you even turn up to stand for hours? Do you switch from your preferred candidate to one who you think might have a fighting chance? The narrative is so important to motivation, so the fact that it can be played out this way during polling itself is concerning.
As we wait for the winner to be declared in this knife fight I think it would be unwise to completely discredit the American polling industry. I know from previously having to analyse it that it is a tricky area, and one in which numerous advances - in particular a model called multi-level regression and post-stratification (MRP) which we will discuss later on - have been created to try and remedy. In a lot of ways it simply highlights that polling can actually only tell part of the picture and institutions have a huge impact on how accurate they subseqently appear to be.
Next week we’ll be back to our regular programming and looking outwards to aspects of the polling industry!
As always, if you'd like to drop me a note, you can contact me by replying to this email or over on Twitter at @thesecondrussia.