Briefing: benchmarking by-elections

Sometimes there isn’t one single obvious “right” way to solve a problem – there can be lots of ways of analysing something. So this is how I’m thinking about the upcoming by-election at Copeland in Cumbria.

On any historical comparison, a governing party – even one heading for general election victory – would expect to lose ground in a Westminster by-election. Statistically, if we knew nothing else about what was going on, we’d say that the Tories starting in 6.5 points behind Labour would mean them coming out considerably further behind.

But the more detailed the model gets – the more you fit it to the recent data – the closer it starts. But equally, as you do that, the risk of underfitting the model (leaving too much noise in) goes down, while the risk of overfitting (mistaking noise for signal) goes up.

So really the outcome in Copeland will depend on whether benchmarking it to the parties’ performances in Sleaford, the most recent by-election and in a seat with a similar EU referendum result, is overfitting or not.

You can read the full analysis here.

Candidates tend not to get so much public attention in by-elections these days, but selections still matter – campaigning well and being a good prospective MP are important, and of course the candidate’s own positions matter too. LabourList has five declared names for Labour’s Copeland nomination, but as applications are open until Monday, there could yet be more.

And finally, thursday is polling day and most weeks we have local authority by-elections for every partisan on Twitter to cherry-pick and spin. This week there don’t appear to be any – the first confirmed contests in 2017 are next week.

Don’t wait for it to appear on the website – sign up for this briefing:


Please send me (select as required):

Email Format





About The Author

Related Posts