What a By-Election Result Can and Cannot Tell You About 2027

A by-election measures one seat, on one day, with far more resources per voter than that seat will ever see again.

That makes it strong evidence for a narrow set of questions and no evidence at all for the question everyone actually asks it, which is who will win the general election.

A by-election is held within 90 days of a vacancy

The rule is worth knowing because it sets the character of the event. When a seat falls vacant, a by-election is held within 90 days.

Ninety days is a compressed campaign in a single constituency or ward. Every party that wants to make a point can put its entire national apparatus into an area that would normally receive a fraction of that attention. Nothing about that resembles a general election, where the same apparatus is spread across 47 counties, 290 constituencies and 1,450 wards on the same morning.

What it genuinely tests

Machinery, and machinery is not a small thing.

A by-election shows whether a party can identify its supporters in an area, get them registered and transported, put an agent in every polling stream, and hold its vote together under concentrated pressure. Those are the exact capabilities that decide a general election, and they are extremely hard to fake.

It also tests whether a new party can convert attention into an actual ballot line and an actual result, which is a different thing from drawing a crowd. And it tests something easy to overlook: whether a side can stop its own people from standing against each other once the nomination is settled. Related: what recent by-elections have shown.

What it cannot measure

National vote share. Presidential preference. Anything about a county the by-election was not held in.

Turnout in a by-election is usually well below a general election in the same area, because five of the six offices are not on the ballot and the day is not a national event. A lower-turnout electorate is not a small random sample of the higher-turnout one. It is disproportionately the people most attached to a party or a candidate, which is precisely the group whose behaviour tells you least about persuadable voters.

There is a second limit that gets even less attention. A by-election is contested by whoever chose to contest that particular vacancy, which is often a smaller or differently shaped field than the seat attracts at a general election. Comparing a two-candidate by-election with a crowded general election measures the field at least as much as it measures the electorate.

So a result that looks overwhelming may be overwhelming among the committed, and silent about everyone else. Related: how to compare two elections in the same place.

How to compare one honestly

Compare like with like, which usually means comparing a by-election to another by-election.

If you must compare it to the general election in the same seat, compare shares rather than raw votes and say out loud that turnout differed. Note which office is being contested. Note whether the parties fielded one candidate each or several. Note whether a locally powerful name was on the ballot in one event and not the other.

And treat the margin the way you would treat any margin: as a proportion of valid votes, sized against the rejected-ballot count and against a single station's register. Related: what actually counts as close.

The one thing a by-election reliably predicts

Whether the winning side can do it again.

A party that wins a by-election with a full agent roster, clean forms and no disputes has demonstrated an operational capability it can repeat. A party that wins one on the strength of a visiting delegation and a fleet of hired vehicles has demonstrated something it cannot repeat 290 times on 10 August 2027.

That distinction is visible on the night if you look at the right thing, and the right thing is not the margin. It is whether every polling stream had a named agent who stayed to the end and came away with a signed form.

Ask that question about every by-election you read about between now and the general election, and about your own side first. If the answer for your ward is that you are not sure, then you have found the real finding, and you have time to fix it. Related: prove you can name an agent for every stream.