THIRTY-TWO MILLION votes were cast in last year’s general election. But how high was turnout? That is a harder question than it sounds. A new paper* suggests that most analyses have been getting the answer badly wrong.

Perhaps the truest measure of turnout would divide the number of votes by the number of people eligible to vote. But eligible voters are a hard group to quantify. They include Britons over 18 (among them expats who have been away for less than 15 years) as well as resident citizens of Ireland and the Commonwealth; prisoners, peers and electoral fraudsters are barred. No one keeps track of how many people meet these conditions, so two imperfect proxies are normally used as denominators when calculating turnout. One is the voting-age population, which includes a lot of people who are ineligible to vote, such as most immigrants. The other is the electoral roll, which suffers from multiple inaccuracies.

Jonathan Mellon of Manchester University and colleagues argue that both methods seriously underestimate turnout. Using population surveys, they estimate the number of people who were eligible to vote in general elections going back to 1979. Until recently the voting-age population was a reasonably close substitute for this figure. But in the past decade, as immigration has risen to record levels, the voting-age population has grown much larger than the number of eligible voters. The result is that, whereas in 2005 measures of turnout using the voting-age population underestimated the true turnout by only about 1.7 percentage points, by 2017 the gap was 5.6 points.

Turnout as measured against the electoral roll has other problems. Accidental double-counting and fraud mean that almost a tenth of entries on the roll are bogus, according to the Electoral Commission. The roll is further inflated by people legitimately registering twice: students, for instance, can sign up both at home and on campus, though they are not allowed to vote more than once in general elections. The authors find that, using a more realistic, lower estimate of the electoral roll, turnout as a share of registered voters has been an average of 9.4 percentage points higher than the official tally since 1979. The official turnout in last year’s election was 69%. Mr Mellon and his colleagues suggest that it was actually in the range of 76-80%.

The findings render previous theories of turnout shaky. For instance, past analyses based on the voting-age population have suggested that turnout has not increased much since a big drop in 2001. In fact, when one strips out foreign residents, it appears that it has risen quite dramatically. The problem is not confined to Britain. The authors believe that such errors are even larger in America, and suggest that other countries with opt-in electoral rolls probably have the same issues. “Substantive conclusions based on this flawed data may easily be entirely meaningless,” they conclude.

One takeaway for policymakers is that efforts to increase turnout may have been misdirected. Much work goes into persuading registered voters to go to the polls. If, as it now seems, registered people vote in higher numbers than once thought, it is far more important to get eligible voters to register in the first place.

* “Opening the can of worms: Most existing studies of aggregate-level turnout are meaningless”, by Jonathan Mellon, Geoffrey Evans, Edward Fieldhouse, Jane Green and Christopher Prosser, available at