Lies, the Evening Standard, and opaque election statistics

A recent Evening Standard (London) headline asserted, "500,000 votes not counted" in the London elections for mayor and London assembly. A bold statement, as there are only 5,197,647 people eligible to voted in London, and only 1,920,533 did so, implying that a high percentage of votes cast were invalid. So I decided to investigate.

Statistics for London elections indicate that marks on cast ballots are classified in a rather curious way. The statistical breakdown provided by London Elects for the 2000 elections, for example, breaks down votes rejected in the mayoral election into

- multiple votes
- uncertain or blank
- no valid first choice
- voter ID discernible

each then in turn broken down by whether the vote was cast for a first choice or second choice candidate.

That year, almost 25,000 electors invalidly cast votes for two candidates in the same column (rather than in two columns), the 'multiple vote' problem.

What's most curious, however, is the category 'uncertain or blank', amounting to 304,686 rejected votes in 2000. This category includes both cases where there is a valid undervote - that is, the voter casts a vote for first choice but finds nobody appealing among the second choices - AND invalid votes (for either first or second choices), such as cases of unclear marks, voting for the same candidate for both choices and so on.

The vast majority of this figure is made up of second choices. So what are we to think? Maybe many voters validly didn't vote for a second choice. Maybe lots of confused people invalidly voted for the same candidate twice (something the ballot paper does not advise against). Maybe there was a contingent who drew cartoons in the second choice column to while away the time. We just don't know how many of these so-called 'rejected votes' were in fact valid, and how many were invalid.

Considering that these so-called 'uncertain or blank' rejected votes amounted to some 82% of all rejected votes in 2000, representing more than 8.6% of all votes cast in the mayoral election, would it not make sense to distinguish valid undervotes (blank) from known invalid votes?

According to London Elects, the stastical categories are exactly the same for 2004, hence the Standard's headline. I'm assured by the press office that some are pushing for a better statistical breakdown, though no assurances are being provided that this will happen.

Voting is a secret process in a democracy, so it's difficult to find realistic, large-scale data on voter behaviour that is useful to the design process. Useful information, such as distinguishing how many (invalidly) voted for the same candidate twice, how many in some way spoilt the second choice column, and how many didn't vote for a second choice at all, would help those designing ballots in future. Without this kind of data, it's next to impossible to know what went on.