Has Labour Really Lost the Working Class?

In the aftermath of Labour’s general election defeat, a new conventional wisdom has rapidly taken hold: Labour, increasingly middle class and out of touch, has been abandoned by its traditional working class base.

This narrative is reinforced by influential academics who argue that working class support for Labour collapsed between 2010 and 2019. The claim is made forcefully in a recent paper by David Cutts, Matthew Goodwin, Oliver Heath and Paula Surridge, forthcoming in Political Quarterly, and in a presentation to the Parliamentary Labour Party by Goodwin and Heath.

A key piece of evidence presented in support of this claim is the “falling ladder” figure, which is reproduced below.

The horizontal axes measure the strength of blue-collar populations in parliamentary constituencies in England and Wales, while the vertical axes measure the difference between Labour and Conservative vote shares across these constituencies. The sequence of plots shows the evolving relationship between these variables over the last four general elections. The black line, summarising the strength of the relationship between blue collar populations and Labour-Tory electoral margins, has steadily flattened, implying – so it is claimed – a significant weakening in working class support for Labour relative to the Tories in England and Wales.

As Cutts, Heath, Goodwin and Surridge argue in their paper:

In 2010, despite losing the election, Labour still enjoyed a healthy lead over the Conservatives in seats with a large working-class population. But in each successive election this advantage gradually dissolved. In 2019, Labour lost its competitive edge in its blue-collar heartlands and its advantage is now not statistically different from zero. This is a watershed moment for Labour. It is one thing to lose an election but it is quite another to lose your advantage in the very working-class communities which the Labour movement was founded to represent.

One reaction to the claim of collapsing working-class support is to take issue with the definition of working-class used in exercises like the falling ladder. Equating class with a particular type of occupation or employment contract, it is argued, fails to consider the shifting composition of the UK labour market or changes in the age distribution of asset ownership. While this argument has merit, we focus on a narrower question: taken on its own terms, does the “falling ladder” plot demonstrate that blue-collar support for Labour collapsed between 2010 and 2019? We argue that it does not.

Look closely at the “falling ladder” figure. If the dark grey summary lines were not imposed on the plots, could you identify a clear shifting relationship – a “collapse” among the points to the right of the plot in particular? What statisticians call the joint distribution – effectively the overall shape of the data – does not noticeably change between 2010 and 2019, other than at the very top end of blue collar employment shares, where there are some reductions in the Labour-Conservative margin between 2017 and 2019.

This is illustrated in the figure below which plots estimated kernel densities – representations of the shape of the data – for Conservative vote shares, Labour vote shares, and the Labour-Conservative margin against the proportion of blue-collar workers in Parliamentary constituencies. (The Labour-Conservative margin is the Labour vote share minus the Conservative vote share: we include this because it is the measure used in the “falling ladder” graph.)

Lighter coloured areas represent high concentrations of points: for example, the light coloured areas at the top left of the plots of Conservative vote shares show that there is a substantial cluster of parliamentary constituencies for which the proportion of blue-collar workers is between 20% and 25% and where the Conservatives achieved vote shares of around 50%.

The joint distributions between Labour’s share of the vote and blue-collar occupations are plotted in the middle four panels and are similar over the four elections: the major differences are the upward shifts in Labour’s vote share in 2015 and 2017, which occur in constituencies with both low and high blue-collar employment shares. These upwards shifts were reversed in the 2019 election, with a more pronounced reduction in constituencies with high blue-collar employment shares – but the overall shape of the distribution in 2019 is not noticeably different to that in 2015.

There is, however, a further problem with the falling ladder: the plot uses vote shares from the 2010, 2015, 2017 and 2019 general elections, but only uses data on blue-collar occupation shares from the 2011 census. The figure therefore ignores any changes in the populations of blue-collar workers since 2011. If we want to measure changes in the relationship between Labour voting and occupational characteristics, we ought to use data that are roughly contemporaneous. At the very least, data on occupational characteristics from more than one year should be used.

Unfortunately, the data used in the falling ladder plot is only available in census years. Cutts et al. use the NS-SEC (National Statistics Socio-economic Classification) system, which includes broad classes of “routine” and “semi-routine” occupations that are usually interpreted as representing blue-collar jobs. But while we do not have access to job characteristics using the NS-SEC system after 2011, we do have access to data on job characteristics using the 2010 Standard Occupational Classification (SOC) system, which are gathered in the Annual Population Survey. The two methods of classifying jobs are not the same, but are quite closely related and attempt to capture similar things.

The figure below reproduces the falling ladder plot, replacing the blue-collar 2011 census data with 2010, 2015, 2017 and 2019 SOC data on “elementary”, “process”, and “sales and customer service” occupations (the results don’t change substantially if “caring, leisure and other service occupations” are added or “sales and customer service occupations” removed).

The “falling ladder” is now much less apparent. In fact, on this measure, the relationship between Labour voting and occupational characteristics was quite stable between 2010 and 2019, with a small decrease in the strength of association between 2017 and 2019. The relative stability of this overall relationship does not, however, imply the absence of significant changes in either Labour votes share or blue-collar population shares in individual parliamentary constituencies. Labour vote shares have increased substantially in many metropolitan areas with high proportions of university graduates, and have fallen substantially in some constituencies with high shares of blue-collar workers, low shares of graduates and older, whiter populations; this pattern has been widely noted and discussed.

But Labour’s vote share has also increased substantially in other areas with high blue-collar population shares, particularly in constituencies in cities such as Birmingham, Bradford, Liverpool and Manchester.

In contrast, the relationship between Conservative voting and occupational characteristics does change, becoming markedly weaker between 2010 and 2019. In turn, this means that the relationship between the Labour-Conservative margin and occupational characteristics became weaker between 2010 and 2019, but this tells us very little about working class support for Labour.

If Labour’s working-class support has not “collapsed” since 2010, what is the underlying story?

It cannot be reduced to a snappy one liner using these data, but the forthcoming British Election Study post-mortem will provide more concrete evidence. In the meantime, we can observe that overall Labour and Tory blue-collar voting patterns were essentially unchanged between 2010 and 2015, when the Liberal Democrats slumped and UKIP surged. Tory voting increased in blue-collar areas when it was UKIP’s turn to collapse in 2017, while Labour’s vote share increased across the occupational spectrum. Finally, in 2019 Labour’s vote share fell across the board. It seems likely that Labour lost blue-collar voters to the Tories and white-collar voters to the Liberal Democrats, but we will have to wait for survey data to be sure.

What, therefore, does the “falling ladder” plot demonstrate, if not a collapse in working-class support for Labour? One possibility is that information on employment characteristics from 2011 becomes an increasingly poor predictor of election results as time goes by. Another is that the relationship between blue-collar employment and political preference is very sensitive to the occupational definition used, and a third is simply that linear regression summary lines are very sensitive to outliers.

The relationship between occupational class and Conservative-Labour vote shares has been heavily studied, and subject to extensive debate since at least the 1970s. There is no doubt that the relationship between occupational characteristics and voting behaviour changed substantially over the course of the twentieth and twenty-first centuries – as has the structure of employment – and that the question of class has been central to Labour’s electoral strategies and fortunes since Gaitskell led the party.

It is not clear, however, that the relationship between occupational characteristics and Labour voting has changed dramatically since 2010. This observation should be borne in mind when considering the strategic conclusion of Cutts, Heath, Goodwin and Surridge, i.e.:

Labour urgently need to find a way to reconnect with left behind areas that do not instinctively share the more socially liberal outlook of its activist and parliamentary base.

If the electoral fortunes of Labour were not, in fact, driven by dramatic changes in the relationship between occupational class and voting behaviour, then this conclusion loses much of its strength.

Notes on Annual Population Survey Data

Annual Population Survey data at Parliamentary constituency level were downloaded from NOMIS. The APS data by SOC2010 classification omits occupation counts where there are fewer than 3, 10 or 500 observations. We set the first two cases to zero, and the single instance of the third case to 250. While NS-SEC classifications can be derived from SOC2010 data in conjunction with other sources of information, which would be a worthwhile exercise, it would be difficult to ensure consistency with the 2011 census observations. In any case, if we ignore information on employer size, we can note that over 90% of the ‘elementary’ occupations in SOC2010 correspond to ‘semi-routine’ or ‘routine’ jobs in NS-SEC, just under 80% of the ‘process’ occupations in SOC2010 correspond to ‘semi-routine’ or ‘routine’ jobs in NS-SEC, and just over 60% of the ‘sales and customer service’ occupations correspond to ‘semi-routine’ or ‘routine’ jobs in NS-SEC. See the derivation tables accessible here.

This article was originally published on OpenDemocracy