The UK General Election was expected in 2024, but not this soon. Recently, they set the date for July 4.
That’s not the best timing for my book (coming in September) but that won’t dishearten me too much.
Again, UK pollsters are vastly overstating Labour’s likely/probable winning margin - because they don’t understand the difference between polls and forecasts.
They just had a Mayoral Election in which almost every single forecast masquerading as a poll failed miserably. No one seems to mind.
Here is my first forecast (late May) for vote share of the two major parties:
Liberal Democrats and Reform are likely to be in or near double-digits nationally, but for clarity I didn’t include them in the chart.
June 7 update: some recent developments suggest Conservative vote share will be lower than my initial forecast - namely some Sunak controversy and the efforts of Reform to win over swing/soft voters who are unlikely to support Labour. I'm skeptical of the staying power of this movement, but my next forecast will incorporate it - along with some wider error bars. I would not be surprised to see Con vote share under 20% or over 30% at this point. Lab I'm more confident will be in the low-mid 40s.
Here’s a visual of the updated forecast range of outcomes.
Clearly, a Labour win (and outright majority in Parliament) is extremely likely, but the chances the margin of victory is in the teens is still greater than that it’s over 25. In fact, I’d rate the chances of it being 15-20 as greater than 25+.
Labour Victory Margin Probability
Lose 0.00 (this is greater than 0%, rounds to 0)
0-4.99 0.23%
5-9.99 1.88%
10-14.99 11.17%
15-19.99 30.47%
20-24.99 33.48%
25+ 22.77%
Another way to look at the forecast is to look at vote share probabilities. According to my forecast, there’s a 50/50 chance Labour ends between 41-46.5, and the same probability that Conservatives end between 20-26.
This means that if I rated 100 elections (and my forecasts are decent) that about 50% of them should fall within this range - it also means 50% shouldn’t.
To expand a bit wider, an 80% chance Labour ends between 39-48%, and the same 80% chance Conservatives end between 17.5%-28.5%. Likewise, if my forecasts are decent, about 80/100 forecasts should fall within this range. 20/100 shouldn’t.
While 80% is a strong probability, it’s important to note that 20% is the same as one in five. Meaning, given just ten forecasts, if none of them fall outside my “80% probability” range, while people who want forecasts to be “calls” might rate me highly, there’s a good chance my range is too wide.
I stated in my last article, despite the poll averages at that time being around “Labour up 20” that does not mean that “Labour by 20” is the most likely outcome.
The reason why, as my readers understand, is that the eventual preferences of undecideds are not given by polls.
My poll average when this article was originally posted had “decided” Labour voters between 36-44%, and “decided” Conservative voters between 17-23%.
But it also has “undecided” between 10-17%!
Of course, since math is different depending on your country’s traditions, US analysts say those undecideds should split 50/50, while non-US ones say they should split proportionally to the decideds. It’s that fallacious “proportional” mindset that leads many statistical illiterates to assert a poll says Labour is “up by” as much as 30. Yes, because the same illiterates produced a forecast from one poll.
I probably sound mean here - good. These same people who are lying to the public about what their polls say literally, this month, forecasted Labour by 22 in the London Mayoral Election (which I said in advance would be horribly wrong) and Labour ended up winning by only 11.
If the public won’t hold “experts” who do terrible work accountable - and possibly worse, if other experts won’t hold them accountable - nothing will ever improve. Their “methodology” is unscientific garbage and it’s time they either admit it and fix it, take accountability for continued, predictable, terrible misses, or quit.
Unlike what they would have you believe with their “proportional” allocation, there is a big step between poll(s) and forecast. The biggest - especially when there are double-digit undecided - is forecasting if and how they will eventually vote.
When those undecideds are more likely to break for the party “behind” in the polls (which was true in the London Mayoral, and probably true in 2024 General), those proportional allocations will miss terribly.
Forecasting an unknown number of undecided voters, from an uncertain base of decided voters, is a glimpse into why the range of possible outcomes is necessarily wide. Yes, there is a nonzero chance Conservatives get more votes than Labour - though I wouldn’t bet on it.
But the large number of undecided voters makes the range of possible outcomes very wide.
This uncertainty is not conveyed to the consumers of this data, when poll companies lie by reporting a forecast instead of their poll, and don’t even report a margin of error (which is, by definition, much higher than their poll had!)
In this case, it’s very unlikely Conservatives win. It’s also unlikely that Labour doesn’t end up with a majority in Parliament, though that’s more in the realm of “possible.” But all of these outcomes would be derided as a poll failure, when in reality it’d be nothing more than another forecast & analyst failure.
As for my vote total forecast (updated June 16th):
50% chance Labour finishes with 41%-46.5%
50% chance Conservatives finish with 20%-26%
80% chance Labour finishes with 39%-48%
80% chance Conservatives finish with 17.5%-28.5%
Total Seats
My first run at this produced some weird results, so I’m borrowing some things from Electoral Calculus’s model for converting vote share to seats. While vote share for each party is (in my opinion) much easier to forecast, it’s the seat layout people are interested in.
Conservatives have dipped in the polls, but undecideds remain very high (mid teens). Accounting for the possibility of tactical voting - along with those who have defected from Conservative to Reform to come back to Conservative - there’s still a good chance Cons outperform the “pollcasts” put out by UK pollsters.
Labour winning an outright majority (326+ seats) is extremely likely - but (if you believe my forecast) not certain. There’s a nearly 2% chance they don’t hit that threshold, in which case the pollsters are going to probably all quit. But this is certainly a possibility greater than 0, regardless of what you think of the quality of my forecast - and why it’s essential to understand forecasting as a range of possible outcomes, not a “call” or “prediction.”
50% chance Labour ends between 401-467 seats
50% chance Conservatives end between 80-127 seats
80% chance Labour ends between 383-495 seats
80% chance Conservatives end between 60-150 seats
Like, update on this post? Or with a new post??