Marginal notes – 2

The story so far:

I looked at the size of Labour’s majority over the Conservatives – or vice versa – in the most marginal Labour/Conservative battleground seats, in general elections over the last twenty-odd years, i.e. going back to 1997 and New Labour. … Labour’s offensive battleground seems to be very much the same terrain as the area it needs to defend. In both cases, we’re looking at former safe Labour seats where a substantial majority was allowed to trickle away over successive elections – between 2001 and 2005, 2005 and 2010, 2010 and 2015; and in both cases, in 2017 two-thirds of these seats saw either a Labour gain or a substantial cut in the Tory majority.

All but two of the 40 marginals I looked at in that post were held by Labour in 1997; 28 went to the Tories between 2005 and 2015, of which 13 were regained in 2017. Moreover, in all but three of the 40 the Labour relative vote share fell in both 2005 and 2010; in 21 of them it fell in 2005, 2010 and 2015, then rose in 2017.

If these results generalise beyond the marginals, then we can conclude that

  1. Labour has had some bad elections – some elections that really cried out for a thorough rethink of the party’s goals, branding, resources and personnel.
  2. 2010 was definitely one of them, and you wouldn’t call 2005 or 2015 examples of best practice. (“He won three elections!” Yes, about that third one…)
  3. 2017, on the other hand, definitely wasn’t one of them. If you forget about the internal party politics and look at the results through an entirely pragmatic, vote-maximising lens – or view them from Mars, through a telescope which registers party names and vote numbers but nothing else – what leaps out is that 2017 was an astonishingly good result by the standards of the previous three elections; a result so good, you could say that the party of the 2005-15 elections didn’t really deserve it. (But then, it wasn’t the party of the 2005-15 elections that did it.)

But I’m getting ahead of myself. We can draw those conclusions, if these results apply generally. Do they?

A bit of methodology. First, I got hold of constituency-level election results for UK general elections from 1997 to 2017. What I’m interested in is the Labour/Conservative relative vote shares, so I limited my scope to England. Then I eliminated all seats which – across that 20-year period – had ever been held by a third or fourth party, or an independent: goodbye to the Speaker, to Richard Taylor and George Galloway, to Caroline Lucas and to the Liberal Democrats.

So far so straightforward. The next step was more of a leap in the dark: matching constituencies between the 2015 and 2017 results or between 2001 and 2005 was easy enough, but what to do about the 2008 boundary review? In the end I took the quick-and-dirty approach (political scientists, look away now) of treating every constituency with the same name as the same constituency. (Although when I say ‘the same name’… The 2008 reviewers had an infuriating habit of switching names around to make them more logical – main piece of information first – so goodbye West Loamshire, hello Loamshire West! That made for a fun evening’s work.) In addition to name-matching, I matched manually in a few cases where a post-2008 constituency was identified with a pre-2008 one by (I did say to look away) the Wikipedia entry on the Boundary Commission. This isn’t ideal; I’m sure there are constituencies out there with the same name pre- and post-2008 and vastly different boundaries, just as I’m sure that I’ve missed some renamed seats with more or less the same boundaries. If I were doing this for anything more enduring (or rewarding) than a blog post, I would do it properly and assess each of the 500+ constituencies individually. But I’m not, so I haven’t.

I ended up with 421 constituencies – English constituencies in contention between Labour and the Conservatives – which can be categorised as follows:

  • 142 were held by the Conservatives at every election from 1997 to 2017
  • 157 were held by Labour at every election from 1997 to 2017
  • 119 were held by Labour in 1997 but lost to the Conservatives at one of the next five elections
    • of these, 29 were then regained by Labour (one in 2010, eight in 2015, 20 in 2017)
  • 2 (Canterbury and Kensington) were held by the Conservatives from 1997 to 2015 but lost to Labour in 2017
  • one (South Dorset) was won from the Conservatives in 2001 and lost again in 2010

Discarding the last two oddball categories gives us three similar-sized groups to analyse, across a series of six elections.

One final methodological note: the measure being used here is relative vote share, a phrase which here means “Labour vote % minus Conservative vote %”. Since my dataset excludes Lib Dem and minor-party seats, this is usually the same figure as the majority expressed as a percentage (or the majority multiplied by -1 for a Conservative seat). Usually, but not invariably: although none of these seats has ever gone to a third party, a number of them have had either the Lib Dems or UKIP in second place at some of these elections. If I was doing a professional job, I could have addressed this complication by adding a new dimension to the analysis, cutting down the dataset or a combination of both. As I’m not, I turned a blind eye and simply measured the Labour-Conservative difference in all cases.

Now for some charts. First, here are those 119 Labour losses, and when they were lost. In this chart – and most of the others – I’m adopting the convention of treating Labour gains from the Tories as positive numbers and Tory gains from Labour as negatives. A bit partisan, perhaps, but I am specifically looking at gains and losses as between those two parties, and this makes it easier to see what’s happening.

Labour seat gains and losses, 1997-2017

Every time I see this chart I think I’ve accidentally deleted the label on the 2010 ‘loss’ bar. Scroll down… oh, there it is. Basically 2001 saw a bit of slippage compared to 1997, and 2005 was a bad result – but 2010 was an appalling result. There was a bit of fightback in 2015 and a lot of fightback in 2017, but we’re still a long way short of where we were, thanks largely to those losses in 2005 and 2010 – substantial losses and huge losses, respectively.

The next series of charts shows loss and gain in relative vote share. The bars represent the numbers of seats in which Labour’s vote share relative to the Conservatives went up or down (by any amount) at each election. Since the total number of seats doesn’t change from one election to the next, the bars in each chart stay the same overall length, but with larger or smaller portions above the origin line.

All seats:

Just look at those first three blue bars. Up and down the country, Labour threw away vote share in 2001; then we did it again even more widely in 2005, and then again in 2010 – with the (cumulative) results we’ve just seen. Again, 2010 stands out as a disaster, with near-universal vote share losses and almost no increases, even after the reductions in vote share over the previous two elections. (Curiously, while there were 31 seats showing an increased vote share in each of the 2005 and 2010 elections, there’s only one where vote share increased in both 2005 and 2010 – and it’s a safe Tory seat where Labour was in third place both times.) But then things look up in 2015 (with 92 more seats with increased vote share than decreased), and even more so in 2017 (222 more)

Here’s the same data for the “Labour losses” group of seats – the 119 seats featured in the first chart, including those that were retaken by Labour.

There isn’t much to say here, except “here’s that trend again” – and perhaps “no wonder they were former Labour seats”. The 2015 recovery is (proportionately) weaker here, but the 2017 rally is just as strong.

Here are the safe Conservative seats.

This is quite interesting. Naively, I wouldn’t have expected very much variation in the Labour vote in safe Tory seats, what with them being… well, safe Tory seats. Far from it: there were quite a few seats where Labour saw losses in vote share between 2001 and 2010, and many more where Labour’s vote share increased in 2015; as for 2017, in that year there were Labour increases in getting on for 90% of Tory seats. These are all seats that were Tory in 1997 and have been Tory ever since, so I wouldn’t want to read too much into this, but it is a strong trend; it suggests that there may be a substantial suppressed Labour vote out there, released by Corbyn’s – and, to be fair, Ed Miliband’s – new direction(s) for the party. (Perhaps the trouble with trying to poach Conservative votes by moving Right is that you end up giving Conservative voters no particular reason to switch.) One, two, many Canterburys!

To complete the set, here are the safe Labour seats, where the trends are a bit different.

Oddly, 2010 isn’t the nadir now, but represents a bit of an improvement on 2005 in terms of the numbers of seats showing vote share gains and losses. Nor is 2017 the peak fightback year; that would be 2015. I don’t know if the post-Iraq tactical voting campaign – or the Lib Dems’ anti-war positioning – had a huge effect on the 2005 vote, but if they did these are the kind of seats where you’d expect to see it. As for 2015 and 2017, from this chart we can already see that there were 30-something safe Labour seats where vote share went up in 2015 – Ed Miliband, hurrah! – and went down in 2017 – Jeremy Corbyn, ugh! As with 2005, these are perhaps the kind of seats where issues and debates within Labour are most likely to make themselves felt (albeit without any immediate effect on the results).

To look at those trends in a bit more detail, here are a couple of charts which need a bit more of an introduction. As we’ve seen there’s an overall tendency for the Labour vote share to drop between 1997 and 2001, then again in 2005 and (mostly) in 2010, before going up in 2015 and (mostly) in 2017. But how many seats actually follow this pattern – down, down, down, up, up – and how many are exceptions? If there are exceptions, what pattern do they follow? Can we distinguish between Tory, Labour and ex-Labour seats, or do the same trends apply generally?

Following a qualitative comparative analysis approach, I translated vote share change into a letter – D for (Labour relative vote share) down, U for up – giving a string of Ds and Us for each seat based on that seat’s successive changes in relative vote share. Since there are six elections overall, each seat has five letters, corresponding to the vote share changes in 2001, 2005, 2010, 2015 and 2017: DDDDD, DDDDU, DDDUD and so on. An ordered series of characters each of which can only take two values is just asking to be translated into binary digits, so that was the next step: DDDDD=0, DDDDU=1, DDDUD=2, and so on up to UUUUU=31. This meant that I could easily calculate frequency tables for the dataset and for each of the three main groups of seats, which in turn made it possible to visualise different patterns and their frequency.

And that’s what you see here, albeit in slightly cut-down form; for simplicity I left the 1997-2001 period out of these charts. The four letters you see here thus correspond to up/down vote share changes in 2005, 2010, 2015 and 2017.

I’ve singled out five patterns – DDDD, DDDU, DDUD, DDUU and DUUU – for the sole reason that these were the only ones which occurred in the data in significant numbers. You’ll notice the prevalence of Ds in the first two positions (loss of vote share in 2005 and 2010) and Us in the fourth (increased vote share in 2017). The way that this first chart is arranged, the first four blocks reading up from the origin – from dark red up to mid-blue – represent all the seats in which the Labour relative vote share went down in both 2005 and 2010. That is, 84% of them: five out of six.

Here’s the same data ordered differently:

In this version the first five blocks reading up from the bottom – i.e. the red blocks – represent all the seats in which Labour’s relative vote share went up in 2017. Which is to say, 72% of them – nearly three quarters. The first two red blocks represent the DDxU pattern, i.e. “down in 2005 and 2010, up in 2017”: 65% of the total, 69% of former Labour seats and 81% of Tory seats. The exception – and the reason why that total isn’t higher than 65% – is the “safe Labour” group, where this pattern only applies to 48% of seats.

The message of the data is pretty clear. While there is some variation between different seats – and regional variation can’t be ruled out (see below) – across England there are some fairly consistent trends. Where 2017 is concerned, the only realistic conclusion is “we’ve had a terrible election, but this wasn’t it” (apologies to Groucho Marx). 2017 was better than 2015 – and 2015 was better than 2010, in much the same sense that vitamin C is better for you than cyanide. We on the Left have a great deal to be proud of and nothing to apologise for – except, perhaps, letting the culprits for the 2010 disaster off the hook, and not moving against them harder and more decisively. (This isn’t sectarianism; this isn’t a quest for ideological purity. We want a party that can win back vote share and gain seats, like the party did in 2017 – not one that loses vote share everywhere and loses seats by the dozen, like the party did in 2005 and 2010.)

Some generalisations about the different categories of seats are also possible – and about Labour seats in particular. Reading from the bottom of the chart:

  • In the Tory and Ex-Labour groups around 40% of seats fit the DDDU pattern, compared to less than 5% of the Labour group
  • In the Tory and Labour groups around 40% of seats fit the DDUU pattern, compared to around 25% of the ex-Labour group
  • In the Labour group around 15% of seats fit the DUUU pattern, compared to less than 5% of the Tory and Ex-Labour groups
  • In the Labour group around 20% of seats fit the DDUD pattern, compared to around 5% of the Tory and Ex-Labour groups
  • In the ex-Labour group around 20% of seats fit the DDDD pattern, compared to around 5% of the Tory and Labour groups

Translated into English, Labour relative vote share in has gone up at some point in 95% of Tory and Labour safe seats in England, and 80% of ex-Labour seats. Around 40% of Labour and Tory seats, and 25% of ex-Labour seats, showed an increased Labour vote share in 2015 and 2017 (only); around 40% of Tory and ex-Labour seats showed an increased vote share in 2017 (only). Among the Labour seats, smaller groups of seats showed increases either in 2015 alone or in 2010 as well as 2015 and 2017.

In short, if we compare Labour seats to all other seats, as well as a lot of commonality there are some significant differences: there is

  • a sizeable group of Labour seats (and very few others) where 2015 was the only recent election with an increased Labour vote share
  • a very small number of Labour seats (but sizeable numbers of others) where 2017 was the only election with an increased vote share
  • a sizeable group of Labour seats (and very few others) where 2010, as well as 2015 and 2017, saw increased vote share

This tends to suggest that – while most of them are living in the same world as the rest of us – non-negligible numbers of Labour MPs are living in a world where Corbyn and the 2017 campaign didn’t deliver the goods; or a world where Miliband and the 2015 campaign did; or a world where the disastrous result of 2010 wasn’t actually all that bad. The effect that these perceptions are likely to have had on their view of the Corbyn leadership – and their retrospective view of life before Corbyn – doesn’t need to be spelt out. These MPs can – and often do – speak eloquently about their own experiences and the threat that Labour faces in their locality, but they are not reliable sources on Labour’s situation nationally.

There’s also a sizeable number of ex-Labour seats – and not very many others – where Labour’s relative vote share has gone down at every one of the last four elections; this suggests that the loss of the seat to the Tories was part of a long-term trend in those areas, and one which hasn’t yet been reversed. To be precise, this pattern applies to 6 seats held by the Tories throughout the period, 8 held by Labour and 24 which went Tory at some point between 2001 and 2017. This would be worth investigating. A quick scan of the 24 seats and their former MPs on Wikipedia gives few pointers, other than to remind me that the 1997 wave swept some truly awful placemen and -women into the Commons: some are noted only for their loyalty (to Tony Blair); others made headlines in the local press during the expenses scandal; one became head of a local NHS trust on leaving Parliament, having continued to practice as a GP throughout. (I guess time weighs heavy when your only responsibility is being an MP.)

Geographically, it may be worth noting that the 32 1997-Labour seats in this group include

  • 4 in the North East
  • 5 in east Yorkshire
  • 7 in the east Midlands, and
  • 6 in Staffordshire

All of which are, perhaps, areas where Labour MPs had grown accustomed to weighing the vote rather than counting it; where weak local parties made for soft targets for incoming Blairites; and where, after five or ten years of New Labour, there just didn’t seem to be that much of a reason to keep up the old habit of voting for the red rosette, whoever wore it. That’s speculation; all I can say is that if I were one of the MPs for the eight seats in this group where Labour hung on in 2017 – Ronnie Campbell, John Woodcock(!), Helen Goodman (seat inherited from Derek Foster), Paul Farrelly (heir to Llin and before her John Golding), Ian Lavery (heir to Denis Murphy), Catherine McKinnell (heir to Doug Henderson), Ruth Smeeth (heir to Joan Walley) or Gareth Snell (heir to Tristram Hunt and before him Mark Fisher) – I wouldn’t be placing the blame for my 2017 performance on things that have changed since 2015. There’s a downward trend in those constituencies which was clearly established long before that – and the great majority of Labour seats, along with the great majority of English constituencies generally, broke that trend in 2017, if they hadn’t already broken it in 2015. It’s not him, it’s you.

One Comment

  1. Blissex
    Posted 24 October 2019 at 19:52 | Permalink | Reply

    «Labour has had some bad elections – some elections that really cried out for a thorough rethink of the party’s goals, branding, resources and personnel.
    2010 was definitely one of them, and you wouldn’t call 2005 or 2015 examples of best practice.»

    In 2010 Labour had just made southern property prices crash over the previous 2-3 years, and that has nothing to do with “a thorough rethink of the party’s goals, branding, resources and personnel”. When property prices do well, voters return the government party, when property prices crash, voters throw the bums out.
    In 2001 and 2005 the government was widely detested or even despised, but property prices were doing well, so many voters abstained or voted LibDem, rather than voting for the party that had crashed property prices in the 1990s.
    In 2015 southern property prices were booming, so many voters stopped voting LibDem and rewarded the government party that delivered the big money.

    Your post, while technically interesting, does not mention “property” or “housing” once, it is is by far the single most important political issue, economic benefit or cost to most voters, nearly doubling the after tax income of many rentiers, or taking half or more of the after tax income of those who are not rentiers.

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

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: