Construe as Mince

Craig Murray’s petition to bring in OSCE observers for the next Scottish referendum (which I hope you’ve all signed – DO IT NOW!) has resurrected the doubts about postal votes in 2014. These doubts solidify, for some people, into the claim that the 2014 referendum was rigged by the UK Government packing the postal vote with “No” ballots, as evidenced by the extraordinarily high postal turnout. This is a consoling narrative for those indy supporters who were gutted by the massive “No” wrecking-ball that crawled out of the woodwork on September 18th to steal the referendum (which many Yessers feel in their bones was in truth a clear victory for Yes) and demolish their dreams. So we need to look at that claim again now.

The claim

The detailed account of this rigging (the “Dunoon Report”) has gained some traction among the Indy community. It was authored by Andy Anderson, Education Officer of the Dunoon Unit of the Democratic Socialist foundation, and a video of Andy explaining the report to a group of activists shows him as a sober, thoughtful and concerned individual. We can summarise his report as follows:

1. Three public figures (John McTernan, political strategist and advisor to the Labour Party; Susan Dalgety, former Labour councillor and civil servant; and Ruth Davidson, leader of the Scottish Conservatives) said in public, before the count had started, that the postal ballot showed a clear lead for No. It’s illegal to count postal votes early, so these people couldn’t have acquired this information legitimately: there must have been government malpractice.

2. The reported postal-vote turnout is not believable as it stands, and broke world records in four local authority areas. To quote the Dunoon Report, “[The Electoral Management Board’s] claim that there was a spectacular world record 96.4% turnout in the PB in Argyll and similar spectacular results in the PB all over Scotland, needs to be carefully assessed, before it can be accepted as a fact.”

2. In Argyle and Bute, more people voted postally than were on the postal-ballot register. “We have a grand total of 743 people on the PB register who could not vote,” says the Report, when the count shows that only 526 people did not vote.

3. Areas with a higher proportion of postal votes returned a higher proportion of No votes; this correlation held across all 32 local authority areas. In the words of the Report, “[A]s the level of the [postal ballot] rises, so the No vote rises with it and takes a stronger and stronger lead.”

5. The fraud was committed by the UK Government replacing unused postal votes with fake votes for “No”. Clandestine government agencies will have access to local authority records, and therefore will have known how many postal ballots had been issued. They also knew, as the referendum approached, who had voted and who had not, and had copies of their signatures (which are on local government computers). So they printed postal ballot papers for electors who had not voted, and submitted them in the normal way through the mail or at local authority offices, reversing the result of the referendum. To quote the Dunoon Report, “[A] massive No vote was inserted into [the postal vote], ensuring an overall No vote majority at the Referendum.” This fraud left no trace in the counting process.

Evaluation

I found this account persuasive (and worrying!) when I read it in detail for the first time recently, so I decided to flesh it out with numbers. The data I used comes from Wikipedia, with postal ballot details from the Electoral Management Board, as quoted in the Dunoon Report. A copy of the Dunoon final report, and my extracted data-sets, are on my website here. Let’s take the Report’s five claims in order:

1. Public figures knew some postal results before the count. These politicians’ comments are consistent with there being government-sponsored fraud: they would have made the comments to soften up the public for the massive No that was about to engulf them. An alternative explanation is that some unscrupulous busybody had counted postal votes, and had primed three political idiots to talk about it in support of No.

2. The postal turnout was unbelievably high, and set world records. The postal-vote turnout was 93.7%, as against 82.5% for the in-person turnout. Postal-vote turnouts are always higher, but I could find no data that would show whether this particular turnout was unbelievably high or not. I could find no data at all on world records for postal-vote turnouts. The claims that these turnouts are unbelievably high and constitute world records are concrete and verifiable, so the absence from the Report of data to support them is worrying.

3. In Argyle and Bute, more people voted than was possible. The electoral register was a year old at the time of the referendum, and therefore included people who could not have voted. The Dunoon Report estimates their numbers as follows:

Those who had died since the register was updated: 125
Those who had moved away since the register was updated: 250
Those in prison: 6
Those with dementia: 362
TOTAL: 743

This total of 743 people who could not vote is clear evidence of fraud. 14,409 people were registered for postal votes, and 13,883 votes were received (13,926 in my data-set, due to rounding differences): so only 526 people did not vote. But 743 people could not have voted, so more people voted than was possible.

If we examine the data critically, however, more holes appear than we can be comfortable with. Firstly, “couldn’t have voted” – the form of words used consistently throughout this section of the Report – is not the same as “a ballot paper couldn’t have been sent in”, and the outstanding case of this is the group of people with dementia.

People with dementia can of course vote; in fact it is probably illegal to prevent a person from registering or voting on grounds of dementia. So care-home staff will take steps to ensure that such people are registered and vote; the care staff or a family member may even mark the ballot-paper for them (and there are statutory provisions for bypassing the signature, for persons who can’t sign their name). So it’s difficult to see the justification for including these 362 patients among those for whom a ballot paper could not have been sent in. Removing them reduces the number of impossible votes from 743 to 381, well within the count of 526 people who did not in fact vote.

Similar considerations apply to those who had died or moved away: the ballot papers were sent to their old addresses, and the current residents there could have sent in those votes, and no doubt in some cases did. This means that the Report’s third claim, that more people sent in ballot papers than was possible, doesn’t stand up.

4. A high postal vote correlated with a high No vote across all 32 local authority areas. This at least looks sound: the chart below shows the relationship between postal votes in the local authority areas (ranked by the proportion of the vote that was postal, and divided into four groups) and the No vote in each group:

dunoon-chart-1“R-squared” is a statistic that shows how closely two data-sets match: 0.0 is no relationship at all, 1.0 is a complete match. A value of 0.98 is extremely good for any human-society data, perhaps even too good to be true. So let’s look at it more closely.

I first asked myself why, and on what basis, the 32 areas had been divided into four groups, but the Report is silent on this. Then I looked at the figures for each of the 32 separate areas, and found that the postal-vote levels do not in fact fall into natural groups at all (red bars on the chart below, which show what percentage of that area’s vote was postal). Similarly, the percentage of No votes (blue bars on the chart below) shows no observable relationship to the size of the postal vote (red bars), and in Group Two and Three goes down as the postal vote goes up:

dunoon-chart-2Deeply suspicious now, I inserted the 32 data-lines for the separate local authority areas into a scatter-plot like the first chart above, plotting the full data, rather than just the four group-averages. The scatter-plot looked like this:

dunoon-chart-3A statistician’s professional verdict on this correlation (R-squared = 0.14) would be that it’s a bit below piss-poor. What the Dunoon Report has done here is to hide the detailed data of the 32 local authorities by reducing it to four averages, which give a spectacularly different result. My own qualifications in statistics are nothing to write home about, but I do know that this is elementary data-rigging, arrived at by ignoring the principle of data integrity. Data integrity says that the investigator should look at all the data, not just the bits she likes; next time you meet a statistician, ask them whether data integrity matters, and then hide under a rock until they’ve finished exploding. So no cigar for the Dunoon Report on this one: the full data shows that the level of postal votes has no correlation with the level of No votes, and blows its fourth claim right out of the water.

5. A clandestine government agency replaced unused postal votes with fake No votes, compromising the result. The clandestine tactic, it is claimed, was to hoover up unused postal votes and replace them with fake No votes. So how many votes would have been available to these purveyors of state-sponsored skulduggery? We’ll have the answer if we can estimate how many postal votes were genuine.

We know the electorate size and the votes cast for all voters, and the electorate size and the votes cast in the postal vote, so by subtracting one set from the other we can get the turnout for the in-person vote. It’s 82.5%, slightly lower than the 84.6% overall turnout, which was inflated by the high postal vote. Nobody thinks that the genuine postal-vote turnout could have been lower than the in-person turnout, so let’s take that figure of 82.5% as our baseline, our threshold for genuine postal votes. The possible number of fake votes is the difference between that number, 82.5% of the postal electorate, and the number of postal ballots received, 93.7% of the postal electorate.

A turnout of 82.5% of the postal electorate gives us 657,410 genuine postal votes; the number of postal ballots received was 746,308. The difference between the two – the window for fraud – is 88,898 votes. However, “Yes” lost the referendum by four-and-a-half times that number, 400,000 votes. These numbers are not shown in the Dunoon Report; but, presented or not, they mean that the Report’s fifth claim, that a clandestine government agency changed the result of the referendum by faking postal votes, can only be construed as mince.

Working through these calculations to verify the Dunoon Report’s claims has made me angry. Not because my comfort blanket – that the referendum was stolen from us by some malign agency – has been stripped away, but because of the egregiously dishonest presentation of data. The Dunoon Report says the postal-vote turnout was unbelievably high and broke world records, but shows no evidence for this; it claims that more people voted than was possible, but the choice of words is sloppy and the figures do not stand up; it says that a high postal vote, area by area, always gives a high No vote, but it does not, and the Report hides the data that show it does not; and the Report says that packing the postal votes “ensured an overall No majority”, when the putative packing amounted to less than one quarter of the majority, and this data is not even referred to. That this dishonest Report should have been going the rounds for five years, consoling Yessers with its unexamined data and false claims, annoys me.

Lessons

What can we learn from this? I think there are two take-aways for us. The first is fairly obvious: if a document comes your way that makes a data-based case, examine the data carefully, compare it with other data that you know or can easily acquire, and verify that it all makes sense. The second relates to how we expect to win IndyRef2. We should not console ourselves with comforting narratives, false expectations and wishful thinking. Instead we need to concentrate on clear, factual arguments, presented soberly and straightforwardly; at the end of the day, we want an IndyRef2 that has been properly thought through by the electorate and reflects the considered consent of most voters. That’s not an impossibility; all the facts are on our side, and we just need to put them over properly.

Advertisements

3 thoughts on “Construe as Mince

  1. Thank you for this; when I read the report some time ago, I also construed it as mince, though I didn’t do so systematically as you have done.

    Re your Point 1: it is standard practice for registered campaigners to send scrutineers to the postal vote opening sessions. These people attempt to get a sneak preview of the votes while the council staff try to hide them. So the votes are not “counted”, but it’s easy enough (especially with only two options) to get a feel for the balance. I attended such a session in 2014; I was watching the signatures being ratified, but I had colleagues who could see that No was ahead in our area. I believe there has been some tightening of the rules, but at that time it was perfectly legal to make this assessment, but not legal to tell anyone what had been observed.

    So the comments by Davidson et al need have been no more than the illicit passing-on of ballpark observations; they don’t require any actual unauthorized counting. (And yes, the people who told me No was ahead in our postal votes should not have done that, either!)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s