Introduction

That political parties are central to the functioning of democratic systems is beyond dispute (Lipset 2000) but the form of electoral system used, and the nature of inter-party competition, are also of importance (Boucek 2012; Ceron 2019). If there is a low number of parties competing within a given electoral system—for example, the dominance of the two-party system flowing from the first past the post electoral system used in the United Kingdom—we would expect the formation and maintenance of ideologically broad political parties. Political parties are thus coalitions in themselves and can exhibit internal party divisions. How those internal divisions contribute to the repositioning of political parties on the traditional left–right spectrum ‘has consequences for the opportunities, incentives and constraints’ for other parties, especially in two-party dominated systems (DiSalvo 2012, p. 7, and on the post-war UK party system, see Quinn 2013). As such, ‘the temperature of party competition cannot be taken by only looking at what happens between the two parties’, but ‘one must also account for what happens within them’ (DiSalvo 2012, p. 7).

The issue of internal divisions has long been central to the study of British political parties, and within this, research on the Labour Party (Shaw 1987; Randall 2018). Perceptions of division, be it between the Parliamentary Labour Party (PLP) and the membership (Watts and Bale 2019), or within the PLP itself (Crines et al 2018), became a feature of Labour Party politics between 2015 and 2020 under Jeremy Corbyn’s leadership.

This article seeks to update this debate and, in the context of the post-Corbyn Labour Party, asks the following questions:

  1. (1)

    How can we define ideological groupings within the post-Corbyn PLP?

  2. (2)

    How do these ideological groupings influence political behaviour?

By addressing these research questions, we can map out potential divisions within PLP under Starmer and assess the extent to which, as a legacy of the Corbyn years, Corbynite thinking is becoming embedded within the PLP (Roe-Crines 2021; Heppell et al. 2022). In so doing we will demonstrate that left-wing MPs are more likely to use collectivist rhetoric on Twitter/X, relative to other MPs.

Contextualising ‘division’

The main objective of our paper is to contribute to the ongoing debates around the divisions that exist within the PLP. We do this through the use of a cluster analysis, a statistical technique that groups together similar observations in a dataset, based on a set of given variables, into ‘clusters’. This methodology has not been applied to the analysis of internal party groupings in British political science. This approach enables us to move beyond the traditional academic debate, which has historically focussed on the problem of indiscipline and conflict within the PLP, be that behavioural (such as MPs voting against their whip) or attitudinal (such as MPs disagreeing over policy or criticising the leadership through various media outlets) (Shaw 1987; Randall 2018; see also Norton 1980; Cowley 2002, 2005; Cowley and Stuart 2003, 2008, 2014).

How we engage in defining Labour Party divisions, and the evolution of the left–right divide within the party, necessitates a brief overview of the literature. Richard Rose defined Labour as a ‘party of factions’, as they possessed groups of elites who organised around their shared attitudes and policy objectives in relation to nationalisation, defence (such as unilateralists versus multilateralists) and the Common Market. These were permanent groups as opposed to fluctuating alignment over specific attitudes or policy objectives (Rose 1964, p. 106). This description of the Labour Party as factionally divided captured the behavioural and attitudinal divisions during the party’s time in opposition in the 1950s and 1960s, when its electoral prospects had been disfigured by the ideological and personalised conflict between the Bevanite-inspired socialist left and the Gaitskellite social-democratic right (Haseler 1969; Crowcroft 2008).

However, this interpretation became too limited by the late 1970s as sub-divides within the socialist left and social democratic right emerged over the course of the decade (Shaw 1987). The left became fractured between the old or soft left and the new or hard left. Whilst both advocated nationalisation and unilateralism, the soft left were committed to parliamentarism and found a home within the Tribune Group inter alia, whereas the hard left emphasised extra-parliamentary movements, the Alternative Economic Strategy, industrial reform and withdrawal from the EEC. These new or hard-left MPs joined organisations such as the Campaign for Labour Party Democracy and/or the Socialist Campaign Group (Seyd 1987).

The cohesiveness of the social democratic right of the PLP also fractured between the 1960s and 1980s, primarily over the issue of the Common Market (Meredith 2008) but also how best to resist the rise of the new left following the 1979 general election (Hayter 2005). This culminated in a splintering between the loyalist social democratic right, who remained within the Labour movement, and the defecting social democratic right who left to form the Social Democratic Party (SDP) in 1981 (Crewe and King 1995).

The evolution of the social democratic right in the 1980s and 1990s would lead to the emergence of the ‘old’ social democratic right, distinct from the Blairite modernisers. In contrast, the Blairite positioning could be defined as ‘new’ social democratic right and found a collective voice through groups such as Progress (Plant et al. 2004). Running parallel to the increasing dominance of the ‘new’ revisionist social democratic thinking within the Labour Party in the late 1990s and early 2000s was the erosion of much of the parliamentary left, and their side-lining within the party.

The apparent dominance of Blairite 'new' social democracy within the PLP in the 1997–2001 parliament was characterised by very low levels of backbench Labour parliamentary dissent (Cowley and Stuart 2003), although this did escalate in the post-Iraq era of the Blair government (Cowley and Stuart 2008). The neutering of the left was evident by the inability of their candidate—John McDonnell—to pass the parliamentary nomination threshold of 12.5% (or 45 MPs) needed to participate in their 2007 leadership election (Heppell 2010, pp. 187–188). Their numerical weakness continued in the leadership elections of 2010 and 2015, where Diane Abbott and Jeremy Corbyn only passed the threshold due to the supporters of other candidates lending their support in order to broaden the debate and choice (Dorey and Denham 2011, 2016; Denham et al. 2020).

However, in 2014, the Labour Party changed the process for electing a new leader, moving from an electoral college system where MPs/MEPs, Labour Party members and affiliated members had a third of the votes apiece, to a one member one vote (OMOV) system. The OMOV section was made up of three voter types: party members, affiliated supporters (trade unionists) and registered supporters, who could, as non-members or non-union members, pay £3 to participate. This new system diluted the importance of MPs—their role was effectively reduced to nominating candidates—and gave them as much power as a newly registered supporter. This system was tailor-made for an outsider like Corbyn (Dorey and Denham 2016): whereas Corbyn won 50% of the party membership vote and 58% of the affiliated supporter vote, he won a massive 84% of the registered supporter vote.

The subsequent election of Corbyn under the new leadership election rules served only to showcase the divergence between the left-leaning membership and the social democratic instincts (both the traditionalist and Blairite revisionist variants) of the PLP (Crines 2017).

We are also conscious of the comparative literature on political parties, which has identified various motivations for internal divisions. Ideology clearly matters, and these disagreements are often perceived to be a ‘struggle for the soul of the party’ (DiSalvo 2012, pp. 5–7). However, disagreements should not be seen as purely ideologically based. For example, personality, electoral performance and leadership ability play a role in determining support. Linked to this is a tendency of individuals to align themselves with specific leaders for personal and/or career aspirations (Hine 1982, p. 42).

Measuring Labour Party divisions: our new approach

Although the formation and reconfigurations between, and within, the socialist left and social democratic right are well studied, attempts to identify the numeric strength of these groupings remain underdeveloped.Footnote 1 Moreover, attempting to position MPs in relation to one another can rely too heavily on the logging of parliamentary voting records, which carry limitations due to collective responsibility binding frontbenchers, and the whipping system pressuring backbenchers into voting with the leadership.

The distinctive contribution of our paper is to take a more rounded approach to the identification of current internal divisions within the current PLP. Rather than rely solely on ideologically-motivated disagreements about policy, based extensively on behavioural indicators such as parliamentary voting records, we instead use variables related to leadership choice and group affiliations.

For leadership choices, we use the public nominations of PLP members in the 2020 leadership and deputy leadership elections. We decided against using publicly declared nominations from the 2015 and 2016 leadership elections because roughly one-third of MPs were not elected at that time and we sought to avoid variables related to time. Instead, we use the 2015 and 2016 nominations to check the face validity of our clusters. Leadership nominations are valuable as they indicate MPs’ perception of the ideological acceptability of the respective candidates at an early stage of the selection process.

Our variables related to Labour-adjacent group affiliations include both parliamentary groups, such as the Socialist Campaign Group of left-wing MPs, Labour Friends of Israel (LFI), and Labour Friends of Palestine and the Middle East (LFPME), alongside extra-parliamentary groups including Progress, Tribune, Stop the War, The World Transformed, Novara Media and Momentum. These groups were selected because they are representative of current divides within the PLP.

The Socialist Campaign Group advanced the hard-left agendas associated with Benn(ism) in the late 1970s/early 1980s (see Seyd 1987, pp. 222–223) and is more recently aligned to the causes associated with Corbyn(ism) (Panitch and Leys 2020, p. 236). Similarly, Novara Media presented itself as an alternative to traditional media, and given it was heavily supportive of the Corbynite project it thus acted as a ‘party in the media’ (comparable to how the Daily Telegraph and Daily Mail act for the Conservative Party, see Bale 2023). Similarly, Momentum was often seen as an alternative to local Labour branches as the ‘party on the ground’ for the Corbynite wing of the Labour Party.

The Tribune Group is representative of the old left faction associated with Michael Foot (Heffernan and Marqusee 1992, pp. 21, 47), which although marginalised in the age of New Labour, has reasserted itself as a leftish alternative to Corbynism after 2015 (Panitch and Leys 2020, p. 230). Progress, which has rebranded itself as Progressive Britain, advances social democratic policy solutions, with strong associations with the politics of New Labour and Blairism (Rodgers 2021). We also include LFI, LFPME and Stop the War as indicators for opinion in relation to foreign policy, and both the ongoing antisemitism crisis that engulfed the Labour Party during the Corbyn leadership tenure (Shaw 2021), as well as views on the conflict between Israel and Hamas in Palestine.

Finally, we include data from a new study by Hanretty and Survation (2023), which asks local councillors to describe their local MPs’ left–right economic views on a scale of 0 (left) to 100 (right). In addition to the 2015 and 2016 leadership nominations, this variable also allows us to establish the face validity of our clusters. By doing so, we assemble a robust dataset of MPs positions within various factions and tendencies, thereby producing a comprehensive overview of the Labour Party under Starmer’s leadership ahead of a general election.

Research methods

The starting point for our study was the construction of a dataset for all members of the PLP returned following the 2019 general election. Of the 202 Labour candidates who were elected, we removed Rosie Winterton, given her role as Deputy Speaker of the House of Commons, and the six Labour MPs who were elected in 2019 but are no longer serving MPs, leaving us with 195 MPs.Footnote 2 The breakdown for these distinctions is provided in Table 1.

Table 1 Characteristics and the Membership of the 2019 PLP
Full size table

Nominations for the 2020 leadership and deputy leadership elections were based on lists published by the blogs Guido Fawkes (2020) and Labour List (2020). Just eleven MPs refused to publicly state who they nominated in the leadership election, and sixteen refused in the deputy leadership contest.

We used a variety of methods to determine affiliation with Labour-adjacent groups. For LFI and LFPME, we used the membership list provided on their respective websites (Labour Friends of Israel 2023; LFPME 2022). The Socialist Campaign Group and Tribune both provide membership lists online (Socialist Campaign Group 2020; Tribune Group 2023).

Determining the membership for Progress was more problematic as there is no official membership list. To overcome this, we accessed their article archive (10,155 articles between 2001 and 2018) (Progress Online 2020). We then counted the number of articles per author, which we cross-referenced against the 2019 PLP to give us a continuous variable representing the number of articles written by each MP over the seventeen years. MPs were classed as being associated with Progress if they had written more than five articles for the website. This is an understatement of the strength of the right of the party, but other similar organisations (Labour to Win, Labour First, etc.) do not provide membership or supporter lists either.

To measure whether an MP was associated with Stop the War, Novara Media, Momentum and The World Transformed, we looked at whether MPs tweeted approvingly of these groups. Given each of these organizations were somewhat controversial during the Corbyn era, we assume that MPs would be careful when it comes to engaging with or retweeting them if they were not supportive of their left-wing positions, given they would know what support would publicly signify.

Finally, for our three face validity variables, we gathered 2016 leadership election nominations from the dataset constructed by Crines et al (2018), which was itself based upon evidence from personal webpages, social media posts and declared lists of supporters across various media outlets at that time. The 2015 leadership nomination data were taken directly from the Labour Party website (2015). MPs’ perceived left–right positioning was taken from the website http://mpsleftright.co.uk/ (Hanretty and Survation 2023).

To generate our clusters, which included categorical variables, we used Gower distance to create a dissimilarity matrix, and then used the partition around medoids (PAM) clustering algorithm. The number of clusters we chose was based on silhouette widths, an internal validation metric for cluster analysis (these are presented in Table 4). We looked at the silhouette widths for two to ten clusters (any more than this and the analysis becomes unwieldy) and found that the silhouette score peaked at two and five clusters.

To test the relevance of our clusters, we used a method called canonical correspondence analysis (CCA) to explore whether MPs used different languages based on the cluster they were in, controlling for whether they were on the opposition frontbench or not at the time of speaking. CCA was originally developed as a method for ecologists to assess the impact of environmental variables on the abundance of species present in a given ecosystem, but has application for researchers exploring the impact of a set of variables on the likelihood of a given occurrence—in this case, the impact of cluster membership on the ‘abundance’ of words used by any MP in a given medium.

Results

Stage one: initial cluster analysis

As noted above, we found two and five to be the optimal number of clusters. As shown in Table 2, we have labelled our groups in the two clusters model as “Mainstream” (N = 162) and “Left” (N = 33). Seventy-nine per cent of the Left cluster backed Long-Bailey for leader, but just 55% backed Burgon. Not a single MP from this cluster supported Phillips for leader nor Murray or Allin-Khan for deputy. No member of the cluster was associated with LFI—and interestingly, membership of LFPME amongst the Left cluster was lower than the PLP as a whole too (39% vs 47%). Unsurprisingly, just one member was associated with Progress and Tribune in the Left cluster.

Table 2 Characteristics and the membership of the 2019 PLP by cluster in the two-cluster model
Full size table

Instead, the Left was numerically concentrated in the Socialist Campaign Group (91% of the Left were members, and of the 34 MPs associated with it just four were part of the Mainstream cluster). The Left cluster was much more likely to be associated with Stop the War, The World Transformed, Novara and Momentum than the PLP as a whole.

In terms of our leadership variables testing face validity, just one MP in the Left cluster supported Owen Smith in the 2016 leadership election; the remainder of Left MPs backed Corbyn or were yet to be elected. MPs in the Left cluster also have an average left–right position of 24, compared to the Mainstream value of 35 (and a PLP average of 33), which is what we would expect (Tables 2, 3).

Table 3 Characteristics and the membership of the 2019 PLP by cluster in the five-cluster model
Full size table

1) Left: N = 32


Like with the two-cluster model, in the five-cluster model members of our Left cluster were more likely to support Long-Bailey for leader and Burgon for the deputy leadership, with none supporting Starmer or Phillips in the former contest or Allin-Khan or Murray in the latter. In terms of our ideological groupings, no Left MP was associated with Tribune or Labour Friends of Israel. Just one was associated with Progress, and under half were associated with LFPME—instead, MPs in this cluster preferred Stop the War as their foreign policy group of choice (63%), and there was a strong affiliation with the Socialist Campaign Group (91%) and Momentum (94%).

The majority of MPs (56%) were not elected when the 2016 leadership contest took place, but of those who were all but one backed Corbyn. This also suggests that the 2017 and 2019 elections served to provide the left of the party with fresh talent. This is especially important for longevity, given these new MPs are more likely to have a longer parliamentary career and could therefore represent the next generation of the left. In the 2015 leadership election, of those who were elected ten backed Corbyn and four backed Andy Burnham, seen as being on the soft left of the party. No member of this cluster backed Cooper or Kendall. The MPs who form this cluster are the most left-wing by Hanretty’s measure, with an average score of 24 against the PLP average of 33.


2) LFPME Soft Left: N  =  58


The second cluster—the LFPME Soft Left—is the largest of the clusters with 58 members. Majorities within this cluster backed Starmer (62%) and Rayner (57%), and support for other candidates was diffuse. All of the members of this group were in LFPME, and nearly a quarter were also in Labour Friends of Israel. 60% were associated with Tribune, but Progress and the various left-wing organisations were underrepresented in this cluster compared to the PLP as a whole.

Unlike the Tribune Soft Left cluster, just one member of this cluster joined parliament after the 2016 leadership election, and the vast majority backed Smith over Corbyn (44 to 11 MPs respectively). In 2015, the plurality of members backed Burnham, closely followed by Cooper (41 and 31% respectively), although interestingly more MPs supported Corbyn in 2016 than in 2015 (11 to 9, respectively). Economically, members of this cluster are representative of the party as a whole, with an average left–right placement of 33—the same as the average figure for the PLP.


3) Tribune Soft Left: N = 36


The third cluster has been labelled the Tribune Soft Left. 53% of the cluster backed Starmer and 31% backed Nandy for leader, whereas for deputy 47% backed Rayner, and roughly equal numbers backed Allin-Khan, Murray, and Butler (5—6 MPs each). None of these MPs backed Long-Bailey or Burgon.

Every member of this cluster is affiliated with Tribune, and this cluster accounts for just under half of the total number of MPs affiliated with Tribune. Labour Friends of Israel is well represented (33%) and Progress and—interestingly—the World Transformed both make up 17% of the cluster. No members of this cluster are associated with LFPME or Stop the War, just one is a member of the Socialist Campaign Group, and just two are affiliated with Momentum.

Of the MPs who were around for the 2016 leadership election not a single one backed Corbyn, and in the 2015 leadership election a plurality backed Burnham. Economically, this group is to the right of the party, with an average left–right placement of 35.


4) Unaligned Centrists: N = 42


The Unaligned Centrists cluster looks very similar to the Tribune Soft Left cluster, with the main differences being slightly more support for the left-wing candidates Long-Bailey and Burgon, as well as Thornberry, and not a single member being associated with Tribune. Left-wing organisations are significantly underrepresented in this cluster, and Labour Friends of Israel is slightly over-represented. A plurality of MPs was not yet elected by the time of the 2016 leadership election (45%) but of those who were, a majority backed Smith (43% of the cluster, compared to 2.4%—1 MP—supporting Corbyn). In the 2015 leadership election, fewer MPs backed Burnham compared to the Tribune Soft Left, and more refused to make a public nomination, but none backed Corbyn. On the left–right axis, the Unaligned Centrists cluster’s average position was also, like the Tribune Soft Left, 35.


5) Right: N = 27


Our final cluster is labelled the Right. This cluster is notable for providing the bulk of support for Phillips for leader and Murray for deputy leader (71 and 56% of their support, respectively). Sixty-three per cent of MPs in this cluster are associated with Progress, and interestingly both LFPME and LFI are over-represented in this group: 85% are associated LFI and 74% with LFPME. Conversely, not a single MP is from the Socialist Campaign Group nor affiliated with Momentum, and the other left-wing groups are heavily underrepresented too.

Like with the LFPME Soft Left cluster, all but one MP in the Right cluster were elected by 2016, and these MPs overwhelmingly backed Smith (78%). In 2015, the vote was split between Kendall (30%) and Cooper (26%), with Corbyn and Burnham tied on 15% apiece (4 MPs). The cluster is also the most right-wing on the right-left axis, with an average score of 38.

Stage two: cluster validation

Now we have shown how our clusters are structured, we can test whether MPs within our clusters do indeed act differently from one another when it comes to language used in the House of Commons or on Twitter/X.

Hansard data were taken from They Work For You’s repository of Hansard xml files, covering debates from 17th December 2019 to 19th September 2023 (They Work For You 2023). Data collection from Twitter was slightly more complex. MPs’ Twitter handles were taken from the Politics Social (2021) website, and tweets were scraped using the R package academictwitteR (Barrie and Ho 2021). The dataset covers the period from the start of the 2019 parliament to 1st February 2023 (this differs from the Hansard dataset due to changes in the Twitter API which made further data gathering impractical). In terms of coverage, every MP is recorded as having spoken in parliament and, of the 195 MPs in our study, 185 have Twitter accounts, all of which have tweeted.Footnote 3

For both Hansard and Twitter, text was grouped by speaker and then cleaned. We removed URLs, hashtags, punctuation, numbers, emojis and stopwords. For Hansard, we removed the names of MPs and constituencies. For Twitter, we removed all usernames (i.e. anything following the @ symbol). The text was then stemmed. In the case of Twitter, we ran the analysis for three sets of tweets: all tweets (tweets, replies, quote tweets, and retweets), all tweets except retweets (tweets, replies, and quote tweets—which we label the ‘own words’ dataset), and standard tweets (no replies, retweets, or quote tweets—which we label the ‘original’ dataset). We report the results from the ‘all tweets’ dataset below but will include insights from the other analyses where they differ from this.

Following the lead of Sältzer (2022), we then removed very uncommon words—we required words to be used by at least 25% of MPs to be included in the analysis. We also controlled for whether an MP was on the opposition frontbench or not, as this would influence the type of language they use in public. This means that, for MPs who have been on the frontbenches and backbenches, we will have an entry for their speech/tweets as a backbencher and as a frontbencher.

Hansard

Both our two-cluster and five-cluster models are statistically significant (p = 0.012 and 0.001). In the two-cluster model, the marginal effect of an MP’s frontbench status is significant in structuring the type of language they use in the Commons, as expected (p = 0.001), but their assigned cluster is not (p = 0.061). However, for the five-cluster model, the marginal effects for both variables are statistically significant (p = 0.001 for both cases).

When undertaking an analysis using CCA, we are interested in the constrained axes—a statistically significant constrained axis suggests that, for our case, the variation in words used along that axis is significantly related to the variables we provided as constraints (in this case, cluster and opposition status). In terms of the constrained axes, none of them are statistically significant in our analysis of Hansard data.

The fact that both models are statistically significant suggests there is a relationship between MP and words used, and the fact that the marginal variable effect of opposition frontbench status, and in the five-cluster model, the cluster variable, suggests these variables do contribute to the variation in language used by MPs in the House of Commons.

Twitter/X

Comparing Hansard records to Twitter/X is necessary because speech in the Commons is constrained: topics are pre-selected, the language that can be used is regulated, and the time members have to speak is also very limited. In contrast, MPs can—and often do—tweet much more freely.

Reassuringly, both the two- and five-cluster models are statistically significant (p = 0.001 for both, across all datasets). In terms of the marginal effects of each variable, we find both of our variables—the opposition status and our two- and five-cluster variable—are statistically significant at the p ≤ 0.05 level across all samples of text. As with our Hansard analysis, we have shown that the language used by MPs on Twitter/X varies based on their opposition status and their assigned cluster.

Finally, we can analyse the constrained axes. For the sake of brevity, we will explore the results for the full dataset of tweets only. For our two-cluster model, the CCA analysis has extracted two statistically significant axes: axis 1 has a p-value of 0.004 and axis 2 has a p-value of 0.002. In terms of the words associated with each axis, those with the strongest loadings on the first axis were associated with the socialist wing of the party. Amongst the top 35 words we find “picket”, “socialist”, “#solidar[ity]” (and indeed “full_solidar[ity]” and @suppor[t]and_solidar[ity]”), “billionair[e]”, “working-class”, “public_ownership”, “proper_pay_ris[e]” and “better_pay”, alongside the names of prominent left-wing figures like “berni[e sanders]”, “rebecca” [Long-Bailey], “jeremy_corbyn”, “mcdonnel[l]” [John, the MP for Hayes and Harlington] and “nadia” [Whittome, young left-wing Labour MP for Nottingham East].

There were also hashtags associated with left-wing campaigns. “#policingbil[l]” refers to a campaign to oppose the Conservative government’s plans to give police additional powers to stop protests. "green_new_d[eal]" shows support for a report by the New Economics Foundation which seeks government action to tackle global warming, the financial system, and oil use. “#endfireandrehir[e]” and “fireandrehir[e]”, whereby employees fire workers and rehire them, often on worse conditions.

The second axis is also related to economics and campaigning. The phrases with the largest positive loading include “#stopfireandrehir[e], “energy_cost”, “strike_today”, “picket”, “picket_lin[e]”, “proper_pay_ris[e]”, and “#solidar[ity]”. In terms of negative loadings, we see words associated with Labour Party campaign slogans (for instance, “government_must_tak[e]”, “labour’s_plan”, “british_people_deserv[e]”, “sticking_plast[er]”, “worst_economic_crisi[s]”, “labour_leader_keir”, “starmer_say” and “shadow_education_secretari”, “shadow_foreign_secretari”, “shadow_minist[er]”.

As such, we see a divide on the first axis between left-wing politics and not, and on the second axis we see a divide between extra-parliamentary action and mainstream Labour Party campaigns. This is a divide which would not have come out in an analysis of voting behaviour, given the whipping system and lack of opportunities to vote along the extra-parliamentary/party campaign divide.

For the five-cluster model, only the first axis is statistically significant. This is highly correlated with the first axis in the two-cluster model, outlined above, with a Pearson's correlation coefficient of 0.988, and so we see even with the introduction of more clusters the left-wing/centrist divide remains.

Retweets

We can also use our data to explore whether MPs are more or less likely to retweet MPs from within their own clusters, and explore the patterns of retweets across clusters. This may give us some idea about ideological positioning, based on the assumption that MPs will retweet content they agree with but not seek public conflict with colleagues they disagree with.

Figure 1 shows two sets of bar charts. Each bar chart shows the proportion of retweets for a given cluster coming from each cluster, with the solid black line representing the proportion of retweets for that cluster as a whole. The graph shows that 64% of Left MPs’ retweets are of Left MPs, whereas just 10% of Mainstream MPs’ retweets are of Left MPs (this is against a total of 19% of all retweets being of Left MPs). This clearly shows that MPs are more likely to retweet colleagues from within their cluster, when controlling for the size of the clusters.

Fig. 1
figure 1

Proportion of retweets by cluster members for each cluster in the two-cluster model

Full size image

Amongst the five-cluster model, shown in Fig. 2, we see very clearly that a disproportionate amount of the Left cluster’s retweets come from Left MPs themselves, and the same is true for MPs in the right cluster. Furthermore, MPs in the Right cluster are less likely to retweet MPs in the Left cluster, relative to all other clusters, and vice versa. Members of the Unaligned Centrists cluster are more likely to retweet colleagues from their own cluster or the Right, and are much less likely to retweet the Left. Both of our Soft Left clusters—LFPME and Tribune—are more likely to retweet one another than other clusters, and less likely to retweet the Right and the Left.

Fig. 2
figure 2

Proportion of retweets by cluster members for each cluster in the five-cluster model

Full size image

As such, when it comes to Twitter, these clusters do reflect real-world behaviours. Our clusters are useful for categorizing not just what MPs themselves share online, but also the type of content they wish to boost to their followers.

Analysis and conclusions

Our analysis has shown that our clusters are important and meaningful, and each canonical correspondence analysis model has reported that our cluster variables are statistically significant when distinguishing between the type of language used by MPs. On Twitter, where speech is less constricted, we have shown how certain clusters are more likely to use terms associated with traditional left-wing values and campaigns associated with the left of the Labour Party. Overall, this study provides evidence that the main divide within the PLP is between the left of the party and the rest of the party.

From the comparative literature on political parties, we know that understanding the internal divisions within a party is central to understanding how parties engage in change (Harmel and Tan 2003) and the extent to which there are constraints upon the position of the incumbent party leader (Ceron 2012). Our findings will aid future studies of the Labour Party by helping us better understand continuity and change from Corbyn to Starmer, and help to reveal the constraints under which Starmer is operating at the level of the PLP.

Our findings confirm that the divisions within the PLP can be understood based on the existence of clusters, whether two clusters—the Left and the Mainstream—or a more varied five-cluster model, which sees a Left and Right cluster, an Unaligned Centrist cluster, and two Soft Left clusters. The significance of these findings is as follows:

  1. (1)

    Corbyn and the PLP: Corbynite Cohort Effects in 2017 and 2019

    Our research identifies that the 54 Labour MPs first elected after the 2016 leadership election were disproportionately found within the Left cluster—of those elected after 2016, 35% were assigned to the left cluster, compared to just 10% of MPs who were elected before 2016. This means that new MPs were more likely to be left wing than the party as a whole, and shows a generational shift in the composition of the PLP under Corbyn. To what extent this will be reversed in the next general election remains to be seen, given the control Starmer’s Labour HQ have exercised over selections, as will the extent to which these new Left MPs can make their presence felt.

  2. (2)

    Starmer and the PLP: Limits to his unifying capability?

    Our research identifies how there are limits to the argument that Starmer can be a unifying figure. He secured just one nomination preference from within the Left cluster in our two-cluster model and none from the Left cluster in our five-cluster model. Additionally, he received under 20% of support from the Right cluster, but did win a majority of support from the two Soft Left clusters. The right of the party backed Jess Phillips (56%), whilst the Left backed Long-Bailey (81%). Against the backdrop of Starmer winning just 41% of the nominations for leader, we should not assume that his capacity to unify the party is limitless.

  3. (3)

    Rayner and the PLP: More of a unifier than Starmer?

    If Starmer’s ability to reach out across the party is perhaps more limited than is commonly assumed, the same cannot be said for Rayner. Like Starmer, she won a majority of the LFPME Soft Left cluster and a plurality of the Tribune Soft Left and Unaligned Centrists cluster. Unlike Starmer, however, she also won 31% of the Left cluster but did do worse than Starmer amongst MPs on the Right (11% versus 19%). However, having a relatively strong support base within the left of the Labour Party puts Rayner in a useful position within Starmer’s frontbench team. Rayner possesses a breadth to her mandate from the PLP that is stronger than Starmer’s. She is strong amongst the new intakes of 2017 and 2019, and her claims to be a unifying figure are thus more credible than his.

These findings showcase the conundrum that those on the left of the PLP face. Our findings demonstrate the evidence of a Corbynite/left cohort effect based on the strength of Corbynite support amongst new entrants in 2017 and 2019. Despite the improvement in their numerical strength, they perceive themselves to be marginalised by Starmer. Not one of the old or new Corbynites nominated Starmer for the party leadership, and they see little evidence of Starmer seeking to accommodate ideological diversity within the shadow cabinet or the PLP.

This perception flows from two developments. First, how Starmer dismissed Long-Bailey from the shadow cabinet (in June 2020) in response to her retweeting comments which Starmer viewed as being potentially antisemitic (Burton-Cartledge 2021, p. 193). And, second, the suspension of Corbyn from the Labour Party given his questioning of the findings of the Equality and Human Rights Commission report into antisemitism in the Labour Party, findings which Starmer had already stated must be fully accepted and not questioned (Burton-Cartledge 2021, p. 193). Their sense of marginalisation is compounded by the knowledge that their options are limited. They recognise that given they are located on the left of spectrum, there are fewer avenues for departure than existed for those social democrats who felt marginalised under Corbyn. This sense is further compounded by the disquiet over the general election candidate selection process, whereby Labour HQ is accused of side-lining left-wing voices.

Beyond understanding the current PLP, our study provides a new and distinctive method for identifying clusters within British and international political parties that is not dependent on measures influenced by collective responsibility or whipping, such as voting records, by one’s backbencher/frontbencher status, or by ideologically irrelevant distinctions such as year of entry to the Commons. By using data from leadership elections and membership of, or affiliation with, party-aligned organisations, we have introduced a new method for analysing inter-party clusters that travels well across parliaments and other legislatures across the democratic world. The appropriateness of using cluster analysis as a way of understanding the PLP is further supported through our analysis of Hansard and Twitter text data, which shows MPs from different clusters do use different language to one another in the Commons or on Twitter. The value of our study lies not just in what it tells us about the contemporary PLP under Starmer, which is significant, but in how it can drive forward further research on intra-party division more generally, nationally and internationally.