Stonecot byelection: Anatomy of a dodgy Tory bar chart

There’s an election on Thursday (6 Dec) for a vacant council seat for the Stonecot ward in the London Borough of Sutton. The Tory candidate is Graham Jarvis, who promises to be a “fresh faced councillor with new ideas”. Unfortunately, Jarvis isn’t above using some old and disreputable tactics to gain votes. Step forward my old friend, the dodgy bar chart.

While all parties have used them, bar charts (dodgy and otherwise) are more commonly seen on Lib Dem election leaflets (“It’s a two-horse race!”), usually to encourage tactical voting in first past the post elections. The Tory bar chart here is no different. While I’d greatly prefer candidates to stick solely to campaigning on the issues, there is such a thing as an acceptable bar chart. I laid out the criteria when I wrote about Lib Dem candidate John Dixon’s shameful effort in the 2010 general election.

That’s not setting the bar impossibly high, if you’ll excuse the pun. It’s just what you’d need to pass a high school maths exam. No PhD in barchartology required.

What does the Tory chart say?

For the moment, let’s take the Tory chart at face value and read what it says.

Now let’s see how the Tory chart measures up against our four criteria for good bar charts.

1. Is the chart adequately labelled?

In short, no. The chart doesn’t tell you explicitly what it’s showing. Where did this data come from? How could I go back to the source if I wanted to check it? And most importantly of all, if I don’t know exactly what this chart is showing, how can I judge whether it’s relevant? The two captions, “How Stonecot voted last time…” and “Stonecot Ward, May 2012” aren’t literally untrue but they’re incredibly vague. Why not just say which election this data comes from?

This fooled me for a while. The big caption “How Stonecot voted last time…” immediately made me think it was about the last council election in Stonecot ward. This would be an obvious data set to use. But if you know that the last council election was held in May 2010 (and most people won’t know this without looking it up), that’s contradicted by the small caption that says these figures come from May 2012.

You’ll have to use memory or Google to work out what people voted for in May 2012. It was the London mayoral election and the elections for the London Assembly.

Political geeks but no-one else will know that those London elections give you three votes. While none of them elect a candidate to solely represent a single council ward, ward-level data is available for the votes cast in each of these ballots. There’s the vote for the mayor, where people have a first and second preference choice of candidates. There’s a vote for a constituency member of the London Assembly where you vote for one candidate that represents a large area (the whole of Croydon and Sutton boroughs in this case). To top it all you also vote for extra London Assembly members which are drawn in a proportional way from a party list according to the London-wide share of the vote for each party.

So if you search for the 2012 London mayoral and assembly election results and you download the ward-level data spreadsheet and you find the right table from the 15 in the file, and then you navigate its 57 rows and 60 columns and then you convert the raw vote tallies for each of the three ballots to percentages you can discover, if you’ve done this accurately, that the data used is from the Stonecot ward in the Croydon and Sutton constituency member ballot, not the mayoral vote or the party list.

Not only does this require pretty sound knowledge of British elections but it’s also a considerable amount of time and legwork. There’s no effort made to be transparent about the source of the data.

2. Is the data relevant?

Relevance is a make or break issue for any data visualisation. No matter how accurate or beautiful your charts, if they don’t address the key issue of your subject they’re worthless at best and misleading at worst. You don’t show the performance graph for one investment while selling another, nor do you put someone else’s grades on your CV.

So how is the London Assembly constituency member vote relevant to a Sutton council byelection to elect a new local councillor?

Firstly, it’s recent. As the Tories say, it is how Stonecot voted “last time” in the sense of “in the most recent ballot for any political body”, or at least, in one of the three held on the same day. And it’s not just the latest data, it’s only six months old. That’s pretty fresh in local election terms where we don’t have the benefit of opinion polling or the good fortune of a more recent council election in this case.

Secondly, the data is from the same electoral area as the forthcoming election. While the London Assembly vote was to elect a constituency member for the whole of Sutton and Croydon boroughs (42 wards in total), the data here is just from the voting in Stonecot ward.

Thirdly, the London Assembly constituency member vote is carried out under the same first past the post system as the local council election this Thursday. If the data used had been from a ballot under a different system there’d be a likelihood that it’d be less relevant due to the effect of that voting system on voters' choices. Comparing a system like FPP that rewards tactical voting with another system that doesn’t is always tricky.

So what’s the problem? The implicit claim in the Tories' bar chart is that it gives a reasonable guide to the political strengths of the parties locally and broadly suggests how voters might intend to vote in this week’s election. Past results aren’t always a good indicator of the future but huge shifts in political support are rare. Parties in fifth place one election are unlikely to take the crown in the next.

Let’s recap on the Tory chart: Tories way ahead by 13 points; Lib Dems and Labour slugging it out for second place with three points separating them; UKIP and the Greens very distantly in fourth and fifth.

If anything similar to that result were achieved this week it’d be a significant upset. Stonecot is a solidly Lib Dem ward when it comes to the council. Here’s how the parties have fared in council elections back to 2002:

You don’t need to be a professional statistician to see that the council seats in Stonecot ward probably aren’t a safe bet for the Tories. I haven’t researched further back than 2002 but at the very least, the last time Stonecot had a Tory councillor some of this year’s voters were still in primary school. How did Stonecot swing from an 18-point clean sweep for the Lib Dems in the 2010 council elections to a 13-point Tory win, of sorts, in the 2012 London Assembly elections? The central question is: Can that London Assembly result be used as a reliable guide to voting intention in this council byelection?

So let’s look at the relationship between council votes and London Assembly constituency votes. Again I’m going back to 2002.

Stonecot ward votes, Sutton Council & London Assembly constituency, 2002-12. Other parties omitted.

The result of the 2004 London Assembly election predicted the result of the 2006 council election fairly well at least in terms of the winner, two leading parties and the margin between them. The Lib Dems won the ward in the London Assembly election by two points and followed that up two years later by winning the council ward by three points. On this one comparison, the correlation between London Assembly and council votes looks fairly close.

But then it starts to diverge. By the 2008 London Assembly election the Tories have pushed ahead, winning the ward by five points over the Lib Dems. But that victory wasn’t repeated two years later in the council elections, where the Lib Dems scooped all three Stonecot seats by a whopping 18 point margin.

Finally we end up in 2012 with the data from the Tories' bar chart and that high-margin Tory win in the London Assembly constituency.

On the evidence of the past two council elections, we can’t say that the previous London Assembly result is a reliable indicator of the vote. In one case the two results were similar, in the next they were wildly different.

Things were going well enough for the Lib Dems in the Stonecot London Assembly vote, for what that’s worth, until Boris came along. Boris is that rarest of creatures, not just a likeable politician but a likeable Tory politician. When he stood for election as mayor for the first time in 2008, the Boris halo effect rubbed off on the London Assembly vote, delivering a five-point “win” for the Tories in Stonecot.

But Boris was no help in 2010, that unusual year when the general election and the full council election were held on the same day. The Lib Dems have long been the strongest parliamentary party in Sutton and Cheam, which contains Stonecot ward. Paul Burstow has held the seat for the Lib Dems since 1997 when he took it from the Tories. The area has had a Lib Dem council since 1990. Although Burstow was re-elected to parliament with a reduced majority, when it came to re-electing Lib Dem councillors, Stonecot voters were more than happy to oblige. They gave the Lib Dems a 18-point victory over the Tories and reduced the Tory vote by a full 10 points.

Fast forward to this year’s London Assembly elections and the Boris effect was in full force again, with voters being unwilling in many cases to make the Boris-or-Ken choice for Boris and then place a Lib Dem vote in the constituency ballot. We also see an apparent resurgence of the long-dormant Stonecot Labour vote, rising from 9% in the 2008 assembly elections to 21% in 2012, though this can be most easily explained by the absence of support for a wider slate of smaller parties. In 2008 the “big three” parties took just 82% of the overall vote. In 2012 it was 95%.

Without any evidence of a reliable correlation between the London Assembly constituency and Sutton council votes, I’ll indulge one last Tory argument for the relevance of their bar chart. Here’s the Tory agent and former Sutton Tory group leader Paul Scully making his case on Twitter:

We’ve seen Nick Clegg fall from messiah in 2010 to pariah in 2012. The “protest vote party” going into government as junior coalition partners and austerity outriders to the Tories has hit them hard. Their promise to oppose any rise in university tuiton fees, such a strong Lib Dem vote winner in the 2010 general election, became their Waterloo (in the French sense) by 2011. Clegg said sorry, but that didn’t stop national support for the Lib Dems falling away until UKIP could claim with considerable chutzpah but some plausibility to be the third strongest national party.

But would any of this make an impact on the seemingly-impregnable Lib Dem wards on Sutton Council? Let’s have another bar chart:

Worcester Park ward is adjacent to Stonecot in Sutton. Demographically it’s similar to Stonecot. Here’s what a post-tuition fees, post-Cleggmania, pre-apology Lib Dem performance looks like in a council byelection, the same type of election we’ve got this Thursday. And this is in a ward where the Tories actually won a seat in 2010 (the Lib Dems got the other two). Persuade all those UKIPpers to pull a tactical vote for the Tories and you still get a Lib Dem victory by six points.

I’m not calling the Stonecot election for anyone. It could go either way. But I’m very certain that the Tories' bar chart doesn’t show any data relevant to predicting the likely result. Council elections have a logic of their own and there’s no use in grabbing any set of nice-looking figures just because they show your party in a good light. If there’s a case to be made for the relevance of the Tories' bar chart, I can’t see it.

3. Is the data accurate?

Remember that parenthetical aside in the small print at the bottom of the Tories' chart? It says “with rounding”. Rounding is a good thing. With data this far spaced it’s perfectly fine to round to the nearest whole numbers.

But I don’t know a system of rounding that lets you round the Lib Dems' 26.94% down to 26% and at the same time round Labour’s 21.15% up to — no, not 22% but a whole 23%! The Lib Dems' figure should be 27% and Labour’s should be 21%.

This narrows the margin between the Lib Dems and Labour from six points to just three. It’s simply wrong. While the far bigger problem is the sheer irrelevance of the whole dataset, this fundamental inaccuracy really does paint the Lib Dem/Labour race in a different light.

4. Is the chart drawn accurately?

I don’t know which school teaches wonky bar chart drawing but it can’t be one of those strict, rigorous old-fashioned places that the Tories prefer. Perhaps some woolly liberal express-yourself types infiltrated the team.

The toppling bar style isn’t a Stonecot innovation. I’ve seen them before on Tory leaflets from other parts of the country. Perhaps they come from central office.

Either way, we can’t have a bar chart where the length of the bar isn’t proportionate to the value being displayed. The wonkiness in itself kills any pretence at accuracy. We can also see that Labour’s “23%” bar is actually taller than the Lib Dems' “26%” bar. This is shockingly poor work and curiously to the Tories' advantage in diminishing the apparent Lib Dem vote.

Conclusion

It’s impossible to tell to what extent the problems with the Tories' bar chart are due to mischief and how much is plain incompetence, but either way, don’t trust it. The data has been chosen based on unsupportable assumptions, the figures themselves are literally wrong and it’s inaccurately drawn to boot. No effort has been made to help people make their own judgements about the data by being transparent about its source. Statistically it’s a total car crash.

The result in Stonecot on Thursday isn’t a foregone conclusion. Historically it’s very safe for the Lib Dems. That safety might slip a little making a Tory win just conceivably possible but that would be a significant and wholly unexpected coup. There’s no good reason to think that the Tories' bar chart bears any meaningful resemblance to the likely final result. You couldn’t sell an investment on this kind of prospectus. Why accept it from someone looking for your vote?

Adrian Short works to get people the information they need, when they need it, in a way that they can understand.

adrian@adrianshort.co.uk