December 8, 2019 📊 Britain's election isn't over til it's over

The Tories are favored to win a majority, but a 2015-style polling error could still surprise us

Welcome! I’m G. Elliott Morris, a data journalist for The Economist and blogger of polls, elections, and political science. Happy Sunday! This is my weekly email where I write about politics using data and share links to what I’ve been reading and writing. Thoughts? Drop me a line (or just respond to this email). Like what you’re reading? Tap the ❤️ below the title and share with your friends!

Share


Dear Reader,

This week’s main read: The United Kingdom will vote to elect all 650 members of parliament on Thursday. I’ve been running a poll aggregator at work for a few months now, and have some related wisdom to bestow to inquisitive minds. The TL;DR? Pretty much every forecasting method is predicting a Tory majority, but a Conservative minority is still in the cards—and a large polling error could surprise us even more.

Plus, I came across some very interesting work on voters’ issue priorities, the culture war and impeachment last week. I’ll take you through it.

Thanks all for reading my weekly email. Please consider sharing online and/or forwarding to a friend. The more readers, the merrier! If you’re shy, the best way you can support my newsletter is to press the heart button below the title (this makes it rank higher in Substack’s curation). If you’d like to read more of my blogging I publish subscriber-only content 1-3x a week on this platform. Click the button below to learn more!

—Elliott


Britain's election isn't over til it's over

The Tories are favored to win a majority, but a 2015-style polling error could still surprise us

The United Kingdom’s politics have been completely absorbed by Brexit. Over the past three and a half years, the question of the country’s exit from the European Union has poisoned every political question of the day. Those hoping for a break from the topic might be excited—disregarding the downstream consequences—that the Conservative party is poised to expand their position in Parliament and finally “deliver Brexit” next year.

The statistical models certainly point to this outcome—a Tory majority—as the most likely. One of such models programmed by Jack Bailey of University of Manchester and Patrick English at the University of Exeter currently gives a 93% chance of a Conservative majority of seats in the House of Commons. Their average forecast is for the party to gain 30 seats relative to their 2017 showing:

Models from Election Maps UK and the social scientists over at Elections Etc show similar results:

These “traditional” forecasts rely on polls, betting markets, crowd wisdom and an average of them all to produce expectations of seats based on the national environment (IE: polls). These methods are largely “swing-based”; if the Tories are expected to do 10 points better in the whole of Great Britain than they did in 2017, then every constituency would be swung 10 points toward the Tories and the winner decided by this new predicted highest vote-getter.

These traditional methods met their match in 2017 when a “new” game in town in (mainstream) UK election forecasting showed up: Mr P. Sophisticated modelers, most notably at YouGov, turned to a method called “multilevel regression and post-stratification” (MRP) that uses polling micro-data to determine the relationship between voters’ demographic traits and geographic location and project results at the constituency-level after knowing each seat’s demographic breakdowns on the same variables. If YouGov’s models learned that older voters were more likely to vote Tory, for example, they got a boost in whiter constituencies.

In 2017, YouGov’s MRP model differed from usual models and correctly called a hung parliament (others were predicting 95 and 99% chances of a Tory majority). But this year, both types of models agree: the Tories are headed for a big victory.

However, we’re forgetting something really really important! Polls can be wrong! Even very wrong! In the 2015 UK elections, just for one cherry-picked example, pollsters under-estimated the Conservative Party’s vote margin by about 6 percentage points; the final poll from YouGov measured a tied GB-wide vote when they won by six. In 2017, errors were reversed as pollsters underestimated Labour by 6-7 percentage points. Nate Silver finds that UK election polls have about twice as much error as US ones do.

I am thus compelled to remind you, dear reader, that the predictive margin of error for the Tory-Labour vote margin UK polling averages is 9 percentage points. (I arrive at this conclusion by calculating the root-mean-squared error of UK polling averages since 1945 and multiplying it by 1.96, the z-score for a 95% confidence interval). And the question then becomes what happens if Labour beats their polls by 9 points? What if the Tories beat theirs by 9? The former would surely end in a hung parliament, and the latter perhaps the biggest Tory landslide since Margaret Thatcher’s in 1983.

Of course, a 9-point polling error is unlikely—but not so unlikely that we should rule it out. And even a 6, or 5, or 4-point error in favor of Labour could drastically alter the electoral landscape. So I would advise you not to discount any outcome, except maybe those of a Labour win or a 20+ percentage point margin for the Tories. Everything else, I reckon, is fair game.


And here are some selected links to the work I read and wrote last week:


Posts for subscribers:


Links and Other Stuff

While we’re on the subject of UK elections…

  • I liked this post from Stephen Fisher on the reliability of “the old swingometer”. That is, the “traditional” or swing-based forecasting methods. As it turns out, voters aren’t really that unique and forecasts that use uniform national (or regional!) swings perform admirably.

  • Fisher includes this image as evidence of his point.

  • Nate Silver had a Twitter thread on this a few weeks ago about how the strength behind YouGov’s MRP model was mostly in more closely approximating the national vote margin than other pollsters. For what it’s worth, I don’t necessarily agree with the premise behind his tweet—I think that MRP is worthwhile and we have proof in the US of how it can match, or even beat, the traditional methods—though it is plainly true that YouGov did indeed “beat” many other pollsters.

On impeachment…

  • Amelia Thomson-DeVeaux and Laura Bronner over at FiveThirtyEight had a good piece titled “Plenty Of People Are Persuadable On Impeachment” out last week. They show that the people who aren’t tuning in to the impeachment debate mostly haven’t made up their minds about the president’s guilt.

And then there was this super interesting paper on historical public opinion, polarization and the cultural war from Delia Baldassarri and Barum Park, both at NYU.

  • Here’s the lead image from their study:

  • Matt Grossman of Michigan State had an interesting take on Twitter, writing “Liberals slowly win culture wars, with public opinion moving leftward over time. It only looks like polarization at the beginning of the process because Republicans liberalize their opinions at a slower pace than Democrats”.

Then finally there was a piece on issue prioritization in The Upshot from Lynn Vavreck, John Sides and Chris Tausanovitch.

  • They used conjoint experiments to determine which issues voters ranked as more important when they were exposed to a tradeoff between two of them. This project is also massive; they interviewed more than 110,000 people!

    Here’s which issues are “actually” important to Republicans, scattered against what they say is most important:

  • And here’s the same plot for Democrats:


What I'm Reading and Working On

Last week my colleagues and I worked on a big analysis of what would happen in this week’s UK election if there was a late surge to the Liberal Democrats. Here’s the big plot that we featured:

I’m working on MRP, impeachment data and UK election night shenanigans this week. You’ll probably want to keep an eye on my Twitter feed.


Thanks for reading!

Thanks for reading. I’ll be back in your inbox next Sunday. In the meantime, follow me online or reach out via email. I’d love to hear from you!

If you want more content, I publish subscribers-only posts on Substack 1-3 times each week. Sign up today for $5/month (or $50/year) by clicking on the following button. Even if you don't want the extra posts, the funds go toward supporting the time spent writing this free, weekly letter. Your support makes this all possible!