Selecting Starmer: The Nomination Preferences of Labour Parliamentarians in the 2020 Labour Party Leadership Election

ABSTRACT This article tests the nomination preferences of Labour parliamentarians in their 2020 leadership election against a range of individual, constituency and party-political based variables. From this our article produces the following three central findings. First, that the appeal of both Long-Bailey and Thornberry was narrow, and that they were competing for the support of Labour parliamentarians aligned to the Corbynista vote. Second, that the appeal of both Nandy and Phillips was based around an anti-Corbynista vote, with Nandy drawing support from parliamentarians from leave-voting constituencies. Third, that the unifying pitch of the Starmer candidature carries some validity: his nomination base was characterised by its breadth rather than any specific factional appeal.


Introduction
This article examines the nomination stage of the Labour Party leadership election of 2020 as a means of identifying any factional bases of support for the various candidates within the parliamentary Labour Party (PLP). We do this by identifying the nomination preferences of all members of the PLP and, by constructing a dataset on the PLP, we test to see what associations may exist between nomination choice and the following variables. Firstly, individually-based variables are considered, including age, gender, ethnicity, year of parliamentary entry, and insider/outside status, EU referendum position and levels of Euroscepticism. Secondly, we consider constituency-based variables including region, change in Labour's vote share under Jeremy Corbyn, estimated constituency leave sentiment and constituency Labour Party (CLP) endorsement. Finally, we consider wider party-political variables such as vote in the deputy leadership election; voting behaviour in previous Labour Party leadership elections; and membership of the following party-political groupsthe Socialist Campaign Group, the Tribune Group, Progress, and Labour Friends of Israel.
We justify our focus on the nomination stage on the grounds that the nominations candidates receive provide the clearest declaration of parliamentarians' real preferences amongst all candidates, whereas the votes cast in the actual party leadership ballot reflects the compromises that some parliamentarians will have to make when faced with a restricted pool of candidates. The nomination stage remains pivotal to the leadership selection process as the PLP use this privilege to act as gatekeepers, i.e. they can screen out candidates they do not wish to be considered for the leadership. 1 Table 1 below notes the support secured by each of the candidates across the three stages of the leadership election. Keir Starmer had a clear lead over the next placed candidate at the nomination and CLP stages, as well in the overall ballot, where he had majorities amongst members, registered and affiliated supporters.
Having selected Starmer, he will remain in position until he either voluntarily resigns, passes away, or if a formal challenge is initiated against him culminating in his removal. The decision to select him was made by members, alongside registered and affiliated supporters, but the decision to challenge him lies with parliamentarians alone. Establishing what motivated support for, or opposition to, Starmer amongst the PLP is the dilemma that this paper seeks to identify: i.e. can we establish patterns of opinion between Labour parliamentarians and the candidates that they nominated?
Our research findings will make a significant contribution to academic debates on factionalism within the Labour Party in the post-Corbyn era. They will also contribute to the *Phillips withdrew her candidature after the parliamentary nominations (stage one) **Thornberry was eliminated after the stage two (CLP nominations) Notes: Stage One nomination threshold: 22 nominations or 10 percent of the PLP plus the ten members of the European PLP (EPLP). Stage Two nomination threshold: five percent of CLPs, 33 in all, or three affiliate groups including two trade unions that together offered representation to over five percent of affiliated members. Thornberry had no support from affiliated trade unions, whereas Starmer was supported by fifteen, Long-Bailey seven and Nandy four out of a total of 32. Stage Three turnout: 62.6 percent Sources: Labour Party, 2020aParty, , 2020bNew Statesman, 2020a ongoing debates within the comparative leadership academic literature on party leadership elections and factionalismi.e. the extent to which party factions are exposed during leadership elections and can undermine parties (as implied by Bynander & 't Hart, 2008;Horiuchi, Laing, & t'Hart, 2015;Pedersen & Schumacher, 2015) or that leadership elections can aid the electoral prospects of parties by resolving conflicts and unifying parties after periods of factionalised conflict (as implied by Fernandez-Vasquez & Somer-Topcu, 2019;Somer-Topcu, 2017). 2 Our paper is broken down into three sections. The first section explains the selection of our variables and our hypotheses and provides details on how our data was collated. The second section will present our findings alongside a commentary on how these relate to our hypotheses. Our third section will identify the significance of our findings for ongoing academic research on leadership selection and internal cohesion within the Labour Party.

Variables, Hypotheses and Data Collation
Below we provide details on the range of variablesindividual, constituency and partypoliticalfrom which we construct our dataset and hypotheses to test.

Model 1: Individual Hypotheses
We selected age as a variable given previous studies had found a positive relationship between age and support for Corbyn in the 2016 leadership election , and selected gender given the focus on Labour's historic failure to select a female party leader (Chakelian, 2020). Our hypotheses were as follows: H1 Age: Older Labour parliamentarians would be more likely to support the Corbynite candidate Long-Bailey compared to younger parliamentarians.
H2 Gender: Male Labour parliamentarians would be more likely to support Starmer than female parliamentarians.
We selected ethnicity as a significant proportion of the 2019 intake were from black, Asian or minority ethnic backgrounds. As the organisation representing BAME interests within the Labour Party, BAME Labour, recommended supporting Starmer (BBC, 2020) we hypothesised as follows: H3 Ethnicity: BAME Labour parliamentarians would be more likely to support Starmer than non-BAME parliamentarians.
We also wanted to assess whether those who were elected when Corbyn was leader would be more loyal to himand his preferred successor, Long-Bailey 3than those who were first elected before Corbyn became leader. We would expect that those who were first elected under Corbyn might attribute their success in part to his leadership. Therefore, we constructed the following hypotheses: H4 Year of parliamentary entry: Labour parliamentarians first elected under Corbyn would be more likely to support Long-Bailey than those who entered before Corbyn was leader.
We wanted to test the significance of the career status of Labour parliamentarians. Here we drew a distinction between insiders and outsiders. Outsiders were either permanent backbenchers or those who had served on the shadow frontbench before the mass resignations of June 2016, and had either self-excluded by refusing to serve under Corbyn or had been overlooked by Corbyn when allocating portfolios since June 2016. 4 Due to the high level of turnover amongst Corbyn's opposition, only those most deeply opposed to Corbyn's leadership would have been passed over. Reflecting the insider/outsider positions of the respective candidates we constructed the following hypothesis: H5 Career status: Labour parliamentarians who had been on the backbenches since June 2016 (outsiders) would be more likely to support fellow outsiders Nandy or Phillips than insiders.
However, there is an implicit relationship between when a parliamentarian was elected and their career status: those parliamentarians who were only elected in 2019 would simply not have had time to be appointed to serve under Corbyn and thus would be classed as an 'outsider' in this analysis. Thus, we have two classes of 'outsider': those who were elected under Corbyn in 2019 and had not had a chance to be appointed to an opposition role, and those who were elected before 2016 and were either passed over or refused to serve. To capture this distinction, we include a two-way interaction effect between 'elected under Corbyn' and 'outsider' status, creating the following hypothesis: H6 Newly-elected outsiders: Labour parliamentarians who were outsiders but were elected under Corbyn would be more likely to support Long-Bailey compared to those who were outsiders but not elected under Corbyn.
Our final individual-level hypothesis examines a parliamentarians' level of Euroscepticism. It is clear that the Labour Party was a remain party: the majority of Labour parliamentarians voted to remain in the 2016 referendum  and membership surveys consistently confirmed their pro-Europeanism (Bale, Webb, & Poletti, 2020). However, although all three candidates who proceeded to the full ballot did vote remain in the referendum, their subsequent positioning in the 2016 to 2019 period was illuminating. Starmer was a driving force in committing the Labour Party to adopting the policy of a confirmatory referendum on any future negotiated deal on leaving the EU, compared to Long-Bailey and Nandy, with Nandy indicating how problematic it could be in being seen to overturning the original referendum decision. 5 Given that Starmer was the most overtly pro-European of the candidates we hypothesise that: H7 Brexit (ideological): Labour parliamentarians who voted remain would be more likely to support Starmer.
However, due to the high number of Labour parliamentarians who backed remain in the 2019 PLP, we also identified a more nuanced method of addressing Euroscepticism. We utilised a dataset (explained in the next section) which ranked parliamentarians in terms of their 'Euroscepticism', as determined by their voting behaviour during the first round of indicative votes held in March 2019. From this we hypothesised as follows: H8 Euroscepticism: Labour parliamentarians with higher Euroscepticism scores will be more likely to support Nandy.

Model 2: Constituency Hypotheses
Our first constituency-based variable was region. This was selected due to media references to the fact that Long-Bailey and Nandy held northern constituencies, whilst the London-based Starmer had to withstand the media critique that he was part of the liberal establishment elite (Hardman, 2019). Our hypothesised as follows: H9 Region: Labour parliamentarians who held northern constituencies would be more likely to support Long-Bailey or Nandy.
We also examined whether changes in constituency vote share since 2015 had an impact on who parliamentarians backed, following the logic that parliamentarians who saw their vote share increase under Corbyn would be more favourable to continuing the Corbynite project, given most parliamentarians seek re-election and are thus attuned to opinion within their constituencies (Hanretty, 2017). For this we tested the following hypotheses: However, we also wanted to see the extent to which parliamentarians were sensitive to the views of their constituents in relation to the decision to leave the EU. For this we constructed the following hypotheses: H11 Brexit (constituency): Labour parliamentarians with lower levels of leave sentiment within their constituencies would be more likely to support Starmer.
Our final constituency-related hypotheses reflected an assumption that diverging from the endorsement of CLPs might be a cause for concern for parliamentarians on the basis that it could cause tension between the member and the CLP, thus increasing the risk of de-selection in the future. Our hypothesis for this was as follows: H12 CLP endorsements: Labour parliamentarians would be more likely to support the same candidate for leader as their CLP.

Model 3: Party-Political Hypotheses
Comparing parliamentarians' nominations for the leadership to their nomination for the deputy leadership could also reveal ideological alignments within the PLP. The deputy leadership election was contested by Rosena Allin-Khan, Richard Burgon, Dawn Butler, Ian Murray and Angela Rayner. Rayner held clear leads over her rivals at the nomination stage (on 41.5 percent support of the PLP with Murray on 16 percent; 56.4 percent at the CLP stage with Butler second on 12.7 percent) and she also had the most nominations amongst affiliates. She triumphed on the third round with 52.6 percent of the vote, over Allin-Khan on 26.1 percent and Burgon on 21.3 percent (Labour Party, 2020b). Given that Rayner nominated Long-Bailey for the party leadership our hypotheses was as follows: H13 Deputy leadership election vote: Labour parliamentarians who nominated Rayner would be more likely to nominate Long-Bailey for leader.
We also wanted to establish if any correlations existed between leadership preferences in the 2015 and 2016 Labour Party leadership elections and nomination choice in this contest. In seeking to determine whether there was any evidence of continuity we constructed the following hypotheses: H14 Previous Labour Party leadership election votes in 2015 and 2016: Labour parliamentarians who supported Corbyn in either or both of these leadership ballots would be more likely to support Long-Bailey.
Finally, we looked at the organisations that Labour parliamentarians choose to associate with, which could be used as a proxy for ideological positioning. We have identified hypotheses around the following membership groupsthe left-wing Socialist Campaign Group, the centre-left Tribune Group, and the Blairite-inspired Progress.
H15 Socialist Campaign Group: Labour parliamentarians who are members of the Socialist Campaign Group will be more likely to support Long-Bailey.
H16 Tribune Group: Labour parliamentarians who were members of the Tribune Group of Labour MPs will be more likely to support Starmer.
H17 Progress: Labour parliamentarians who were associated with Progress would be less likely to support Long-Bailey.
We also considered membership of Labour Friends of Israel given the issues Labour faced regarding antisemitism (Rich, 2018; see also Heppell, 2022). We assumed that those parliamentarians with closer ties to Israel, or who have Jewish heritage themselves, would oppose candidates aligned to the Corbyn project or members of the Corbyn shadow Cabinetand so we constructed the following hypotheses: H18 Labour Friends of Israel: Labour parliamentarians who were members of Labour Friends of Israel will be less likely to support Long-Bailey, Starmer and Thornberry for leader.

Data Collation
We were able to construct a robust list of supporters for each of the candidates using the following methods. First, we exploited lists of declared nominations from various sources (Labour List, 2020; Order-Order, 2020) and for parliamentarians who were not included in nomination list we examined social media and personal webpages. By using these methods our dataset identified 83 who nominated Starmer, 33 for Long-Bailey, 31 for Nandy, 22 for Thornberry and 21 for Phillips, and the remaining 12 who did not make a public declaration. The robustness of our dataset is evident given that we identified the nomination preferences of 94.1 percent of the PLP which compares favourably with similar studies on the leadership preferences of Conservative parliamentarians (Jeffery, Heppell, Hayton, & Crines, 2018, 2021. We used the same method to determine parliamentarians' deputy leadership nominations. In terms of our independent variables we collated and coded as follows. We were able to gather our data on age, gender, year of parliamentary entry, insider/outside or frontbench status, region, and change in constituency vote share from the UK Parliament website (2020). Age was coded as a continuous variable, ranging from 24 to 80 years old. There were ten MPs for whom we could only find partial age data. In cases where we could only find a year of birth, we used 1st January as a dummy date, and for parliamentarians where would only find a month and year, we used 1st as a dummy date.
We coded change in vote share as a continuous variable, which ranged from −17 to 29. On region we differentiated between whether a constituency was in the North of England (North West, North East, and Yorkshire and the Humber) or not. We coded year of entry into 'elected under Corbyn' or not, and for outsider status we coded parliamentarians as 'outsider' or 'insider'. We were able to collect information on which Labour parliamentarians were from BAME communities by utilising a list complied by Uberoi & Lees, 2020 For the leave vote share in each constituency, we exploited Hanretty's estimates which were accessed via Norris' (2020) database. CLP leadership endorsements were made available by Labour List (2020). Data for the 2015 and 2016 Labour Party leadership elections and the 2016 referendum vote came from Crines et al. (2018). For newly elected parliamentarians we gathered their 2016 referendum position from their social media profiles. To assess Euroscepticism, we utilised a dataset which assigns each parliamentarian a Euroscepticism score, based on which options they backed during the first round of indicative votes in March 2019 (Afonso, 2019). For those elected after this date, we assign to them the mean value for Labour parliamentarians as a whole.
For membership of the Socialist Campaign Group, we have used the membership list provided by the Group on Twitter (Socialist Campaign Group, 2020). For membership of the Tribune Group (2020) we accessed the most recent online listing available from October 2020. Determining the membership for Progress was more problematic as no official membership list was available. To overcome this, we accessed their archive of articles which comprises 10,155 articles between 2001 and 2018 (Progress Online, 2020). We then scraped the details of every article author in that period, cleaned up the data (i.e. splitting jointly-authored articles, merging duplicates) and counted the number of articles per author. We then cross-referenced this list against the 2019 PLP to give us a continuous variable representing the number of articles written by each Labour parliamentarian for Progress' website over the seventeen-year period, which we use as a measure of closeness to the organisation. Finally, Labour Friends of Israel provide a membership on their website (Labour Friends of Israel, 2020).

Research Findings
Our research findings are presented as a series of tables showing the output of logistic regression models, where the dependent variables are dummy variables of support for a given candidate (nominated the candidate = 1) and the coefficients are shown as log odds. Thus, in Table 2 below, the first column should be interpreted as 'for every oneunit increase in age, the log odds (likelihood) of a parliamentarian supporting Starmer relative to not supporting Starmer increased by 0.03 when holding all other variables constant'.

Model 1: Individual Hypotheses
When it comes to support for Starmer, the regression output for model 1 shows the only statistically significant relationship present was parliamentarians' Euroscepticism scores. Those who high a higher level of Euroscepticismi.e. they were willing to support 'harder' forms of Brexit during the first round of indicative voteswere less likely to support Starmer. This relationship is unsurprising, given Starmer's role in advocating for a second referendum and also being seen as generally more anti-Brexit than his main rivals, Long-Bailey and Nandy.
Long-Bailey gains support from numerous groups. Whilst demographic variables are not strong predictors of support, the rest of the model does show some interesting results. Firstly, relative to insiders, outsiders are less likely to support Long-Bailey's candidacy for leader, when controlling for all other variables in the model. This is unsurprising given those outsiders had also refused to serve under Corbyn, so it follows that they would be less likely to support the continuity-Corbyn candidate. Furthermore, the interaction effect between whether and a parliamentarian was elected under Corbyn and if they were an outsider is both positive and statistically significant. This means that those who were elected under Corbyn and were outsiders were more likely to support Long-Bailey's candidacy than those who were elected under Corbyn but were insiders. Since only the 2017 cohortnot the 2019 cohortwere in the Commons long enough to make it onto the opposition frontbench, this suggests that the 2019 cohort were more Corbynite than the 2017 cohort, and the party as a whole. Finally, we see that both Brexit-related variables point in the same direction: the higher the Euroscepticism score that a parliamentarian had, the greater the likelihood of them supporting Long-Bailey, whilst, those who voted Remain (compared to those who voted leave or who would not say) were also less likely to have supported Long-Bailey, suggesting that there was some element of a 'Lexiteer' support base for the Corbynite project. The model shows that support for Nandy came from those with a higher Eurosceptic score, reflecting the media narrative that those who were willing to accept a harder Brexit were more likely to support Nandy, whose pitch was centred around winning back leavevoting areas. The coefficient is larger than that for Long-Bailey's support, and the level of statistical significance is higher too. On the other hand, Phillips' support was drawn from younger parliamentarians (the only demographic relationship to be statistically significant), and from outsiders. She also drew support from those with a lower Euroscepticism score, suggesting she was the standard-bearer for a group of Corbyn-sceptic remainers dissatisfied with anyone associated with the Corbynite project. Support for Thornberry was found amongst insiders, i.e. those who had served under Corbyn. This suggests that Thornberry's support was amongst those who wanted to continue the Corbynite project but did not rate Long-Bailey.
Overall, we see a few key variables in Model 1 structure support amongst the PLP: Starmer and Phillips drew support from the less Eurosceptic parts of the party, with Phillips' support also coming from Corbyn-sceptics. Nandy drew on Eurosceptic support, as did Long-Bailey, but who complemented that support with those who were either new to the PLP or had served under Corbyn. Thornberry seemed to be a Corbynite alternative to Long-Bailey.
By comparing the McFadden adjusted r-squared values, we see that Model 1 has the greatest explanatory power for Phillipsshowing outsider and Remain status were important driver of her supportand for Long-Bailey, which also confirms her as the continuity-Corbyn candidate. However, for the other three candidates, and for those who made no nomination, the very low adjusted r-squared values suggest the model is not very useful for predicting support.

Model 2: Constituency Hypotheses
Model 2 focuses on constituency elements. Firstly, for Starmer, we can see that the only variable which is statistically significant comes from CLP nominations: parliamentarians with CLPs who backed Long-Bailey or Nandy for leader were less likely to support Starmer's candidacy, relative to those with CLPs who backed Starmer. This is not surprising and does provide some evidence that parliamentarians and their CLPs may share ideological positions, or at least take cues from one another (Table 3). Support for Long-Bailey came from those who represented seats where Labour had actually lost votes under Corbynthe greater the increase in vote share between 2015 and 2019, the less likely a parliamentarian was to support Long-Bailey. The reason for this is not immediately clear and certainly needs further investigation. Long-Bailey also drew support from MPs in seats with lower levels of leave sentiment, which is ironic given that on an individual level her support was drawn from more Eurosceptic parliamentarians. Finally, those with CLPs who were more likely to support Long-Bailey were themselves more likely to choose her over Starmer, as well those with CLPs who made no choice. It is the CLPs which could hold the key to the apparent tensions in Long-Bailey's support, with parliamentarians more beholden to a Corbynite CLP than perhaps the status of the constituency at large. Or, perhaps more likely, these Labour parliamentarians had such big majorities that a drop of a few percentage points might not worry them. For Long-Bailey's rival for the Corbynite vote, Thornberry, none of the constituency variables were statistically significant determinants of support.
We see an interesting reverse of the conundrum of a parliamentarian's constituency not looking like the profile of the parliamentarian supporting a candidatewhereas Long-Bailey's support is more Corbynite and less Remain, from those in areas which have lost support under Corbyn and are more remain, for Phillips the reverse is true. Phillips' supports tend to come from areas where their vote share increased under Corbyn, and which had greater support for leaving the EUin contrast to her support base, who are Corbyn-sceptic remainers. For Nandy, support was driven by those in constituencies with higher leave vote shares, and from those with CLPs who backed Nandy, relative to those with CLPs backing Starmer. The higher R-squared value for Nandy's support in Model 2 compared to Model 1 lends weight to the view that Nandy's campaign to give a voice to forgotten Labour leave areas was actually successful when it came to securing nominations.
Overall, the McFadden adjusted r-squared values from Model 2 are higher than for Model 1 in the case of Starmer and Nandy, suggesting our constituency model has greater explanatory power for support for these candidates compared to our individual-level model, whilst the reverse is true for Long-Bailey and Phillips.

Model 3: Party-Political Hypotheses
Model 3 shows variables reflecting parliamentarians' positions on internal Labour Party politics. For Starmer, we see that those who nominated Ian Murray as deputy, or made no nomination, were less likely to support his candidate relative to those who backed the eventual winner, Rayner. We also find that, relative to those who never nominated Corbyn, those who became Corbynites between 2015 and 2016 were less likely to support Starmer, as were those who were part of the Socialist Campaign Group. Interestingly, Starmer did not gain a statistically significant level of support amongst Tribune parliamentarians, despite being a member of that group himself. Overall, given all of these variables are negative, we can interpret these results as Starmer winning support from across the party except for a core group of Corbynites and a core group of Corbyn-sceptics (Table 4). Phillips drew support from those who backed Murray, relative to those who backed Rayner, and from those associated with Progress. Thus, we can somewhat characterise Phillips' support as a New Labour or Blairite movement. Interestingly, the model provides no statistically significant variables for support for Nandy, suggesting that a) her support base did not also cohere around a deputy leadership candidate, and b) that the support she gained worked across established party factions. However, instead of this being a strengthin the sense that it allowed her to speak across existing boundaries what it actually meant was that she had no real base of support to draw from. This is in direct contrast to Long-Bailey. Model 3 once again shows that Long-Bailey drew support from the Corbynite core of Labour parliamentarians: those who had nominated Corbyn twice or in 2016, and who were part of the Socialist Campaign Group. Indeed, the strength of this relationship is highlighted by the very high McFadden Rsquared value of 0.8. Perhaps the most interesting detail is regarding the deputy leadership nomination. Firstly, there is no statistically significant relationship between support for Richard Burgonthe Socialist Campaign Group-backed deputy leadership candidate and support for Long-Bailey. This may be because Burgon is somewhat of a ridiculed figure in the media, and is not seen as a serious candidate by many within the PLP or media. Secondly, those who supported Butler were much less likely to support Long-Bailey.
However, those who supported Butler were more likely to support Thornberry. Thornberry also drew support from those who nominated Allin-Khan, and relative to those who never nominated Corbyn for leader, those who nominated Corbyn twice, or just in 2016, or were not elected all were more likely to support Thornberry. Thornberry, however, did not gain more support from members of the Socialist Labour Group. This further suggests that Thornberry's support base was Corbynite parliamentarians who were not part of the core Corbynite grouping and who were not happy with Long-Bailey as leader.
Overall then the McFadden adjusted R-squared values suggest that Model 3 provides the best explanation for support for Long-Baileyindeed an adjusted R-squared value of 0.658 is very good for analyses of this type.
Our analysis suggests that we can see three broad clusters of support when it comes to leadership nomination amongst the PLP. Firstly, there are two Corbynite clusterswhich perhaps can be termed the socialist cluster and the soft-left cluster. Socialist Corbynites supported the Corbyn-backed candidate Long-Bailey. The soft-left rejected Long-Bailey and instead supported Thornberry, along with Allin-Khan and Butler for deputy. The Corbyn-sceptics were represented by Phillips, with Murray as deputy.
Both Starmer and Nandy did not draw from a specific cluster of support: for Starmer, this was because he was supported by parliamentarians from across the party. For Nandy, this was because she found limited support for her message amongst parliamentarians, apart from those in more leave voting areas who perhaps had a greater self-interest in listening to a candidate with a message of listening to left-behind leave voters.
Beyond these clusters, the McFadden adjusted R-squared also adds to our understanding of the goodness-of-fit our models provide. For example, whilst the individual-level model provided the most explanatory power for Phillips, support for Starmer and Nandy was best explained by the constituency model. For Long-Bailey and Thornberry, the party-political model provided the best fit, but for Thornberry this was very much a case of the best of a bad bunch. Indeed, what is striking is how poorly any model explained support for Thornberry or Starmer.
To counter this, the final part of this paper utilises a stepwise case selection process to identify the variables which best explain the support base of each candidate, bringing together the best of each of our models. This approach involves iteratively adding and removing predictor variables from the model in order to find the optimal subset of variables which result in the best performing modelthat is a model that lowers prediction error, based on the Akaike information criterion (AIC) value. This method is utilised as a complement to our literature-informed models above. Before running these models, we also turned our categorical variables into binary variables (e.g. in the case of deputy leadership vote, instead of being relative to Rayner, we used a dummy variable of whether an MP supported a candidate or not). The outputs of the stepwise model are presented in appendix 1.
The stepwise model produced for predicting support for Starmer has a McFadden adjusted r-squared value of 0.265, which is significantly higher than any of our hypothesised models presented above. Interestingly, apart from whether a parliamentarian nominated Rayner, the direction of all statistically significant variables is negative, pointing to the fact that Starmer had a good range of support across the party, and only lost support parliamentarians from certain groupstypically Corbynites or MPs with CLPs backing Nandy.
For Long-Bailey, two variables were removed before the stepwise regression model was ran due to issues of multicollinearity: constituency leave vote and membership of the Socialist Campaign Group. The resultant model had a slightly lower McFadden rsquared value to our party-political model (0.585 vs 0.658), but did find that supporting Corbyn in 2016 and voting for Burgon in the deputy leadership elections (both highly correlated with being a member of the Socialist Campaign Group) were key drivers of support for Long-Bailey, as was supporting Rayner for deputy, but also showed that BAME Labour MPs were less likely to support Long-Bailey.
For Nandy, the model shows again that having a CLP which nominated Nandy was statistically significant, as was a parliamentarian's Eurosceptic score. Interestingly, association with Progress was negatively related to supporting Nandy, reflecting Nandy's more communitarian pitch to Labour voters. The McFadden adjusted r-squared for this model was 0.259, again well above the best model presented above (the constituency model, 0.111).
The previously identified link between age, Labour's changing vote share under Corbyn and constituency leave vote all remain statistically significant in the stepwise model for Phillips, as does nominating Murray for deputy leader. However, this new model also has a much better goodness of fit compared to our previous modelsthis model's adjusted r-squared is 0.376, compared to our individual model, which was 0.172.
Finally, for Thornberry the model again confirms that support for Corbyn in the 2016 leadership election was related to supporting Thornberry, as was being elected under Corbyn. Support for Allin-Khan or Butler for deputy are also statistically significant, meaning this model further supports our finding that Thornberry was the soft-left Corbynite option. This model also has a greater McFadden adjusted r-squared value of 0.113, compared to the other model's values of −0.041, −0.053 and 0.034).

Analysis and Conclusions
The aim of this study was to contribute to wider academic debates on leadership selection within British politics, by identifying if any patterns of opinion existed between members of the PLP, and the candidates that they nominated for the party leadership.
By providing a quantitatively-driven study on the choice of Labour parliamentarians in the Labour Party leadership election of 2020 we have provided different insights into the leadership-follower relationship, at the parliamentary level, than have tended to emerge from the qualitative dominated studies on previous Labour Party leadership elections, see for example, Alderman & Carter, 1993, 1995Dorey & Denham, 2011, 2016Quinn, 2016;Heppell, 2021; see also Heppell, 2010. By choosing a quantitively-driven study we have embraced methods which have been traditionally deployed when identifying what drives leadership preference in Conservative Party leadership selectionsee, for example, Cowley & Garry, 1998;Jeffery et al., 2018Jeffery et al., , 2021Roe-Crines et al., 2021. Moreover, with the exception of the study by Crines et al. (2018), academic research on the recent conflicts within the Labour Party have tended to focus on patterns of opinion within the membership vis-à-vis the leadershipon why membership grew under Corbyn, see Whitely, Poletti, Webb, & Bale, 2019 and why it might contract under Starmer, see Barnfield & Bale, 2021. Although academically significant, we need to place the leadership-membership dynamic alongside the leadership-parliamentarians dynamic: the members may elect the party leader, but it is the parliamentarians who choose the candidates which are put to the members, and it is the parliamentarians alone who can initiate a challenge to the party leader.
The value of our study is that is can provide us with clearer insights into the composition of different clusters of opinion and circumstances amongst the PLP in relation to the party leadership. We can summarise the nomination preferences of Labour parliamentarians in terms of the following key findings:

The fragmentation of the Corbyn vote between Long-Bailey and Thornberry
Support for Long-Bailey was driven by factors relating to loyalty to Corbyn. Indeed, her support base is relatively well explained by model 3, which includes our politically-based hypotheses of who parliamentarians nominated in the deputy leadership election, whether they voted for Corbyn, and whether they were first elected when Corbyn was party leader. Long-Bailey and Thornberry were competing amongst those who were generally supportive of Corbyn. Support for both was driven by being an insider, supporting Corbyn in both leadership elections or in 2016 only, or being elected whilst Corbyn was leader. However, they were not being driven by holding a northern constituency, metrics surrounding constituency vote share, backing remain or membership of Labour Friends of Israel. The Long-Bailey/Thornberry support base fractured the Corbynite loyalist bloc and this was most evident in relation to the deputy leadership nominations: Long-Bailey overwhelmingly secured the support of those who backed Burgon, whilst Thornberry had a nomination base linked to nominations for either Allin-Khan or Butler.

The fragmentation of the anti-Corbyn vote between Nandy and Phillips
If the Corbynite faction fractured between supporting either Long-Bailey or Thornberry then the same could said about those who were the Corbyn-critical bloc.
Anti-Corbyn sentiment drove the nomination base for Phillips. Whereas Nandy only drew increased support from parliamentarians representing seats with a higher share of leave voters (as well as those seats where the CLP backed Nandy, although this is not a useful predictive tool due to the issue of timing), Phillips drew support from outsiders and those not elected under Corbyn, as well as those with smaller majorities, those who saw their vote share increase in 2019 and members of Labour Friends of Israel. Unlike Nandy, who did not have a clear parallel in the deputy leadership election, those who supported Murray were more likely to support Phillips. 3. The breadth of the nomination base of Starmer The lead Starmer held at the nomination stage was significant. The size of one's parliamentary nomination base does not necessarily determine success in the leadership contest as a whole, as Corbyn himself had demonstrated, but it could not have hurt Long-Bailey to have had a larger parliamentary nomination base going into the full leadership ballot. This might have been possible had Thornberry not stood (or vice versa), thus enabling the parliamentary Corbynites to coalesce around an agreed candidate. The lead that Starmer held over Nandy was broadly similar, but her problem was that she failed to appeal to any specific group beyond parliamentarians in leavevoting areas. The appeal of Phillips was narrow and her candidature appealed only to the most anti-Corbynite parliamentarians. These factors are significant because our research findings illuminate one key factor: whereas Long-Bailey (and Thornberry) and Phillips had nomination bases in which patterns of support are identifiable, the same cannot be said of Starmer (aside from him being exclusively remain based, but this observation is largely nullified by the remarkably low levels of leave sentiment within the PLP). His nomination base came from across most areas of the PLP, reflected in the fact that our research findings suggest there were few obvious variables which could predict support for Starmer. It is clear that decisions made by the CLPs did have some predictive power, but generally speaking low pseudo R-squared values suggest that no individual or multi-nominal model provided much in the way of predictive power, especially when compared to other candidates.
Overall, our findings will contribute to the emerging academic literature on the Starmer eraand they will drive ongoing debates on factionalism and positional change in the post-Corbyn era (Thompson, Pitts, & Ingold, 2021). When we position our findings within the context of the wider debates within the comparative academic literature on leadership transitions, we can note the following. The behaviour of the PLP in the nomination stage of the Labour Party leadership election of 2020 did not escalate factional conflict, even if the Long-Bailey nomination base was narrow. Rather, the eventual selection of Starmer, and the breath of his nomination base was more consistent with the Fernandez-Vasquez and Somer-Topcu (2019) argumenti.e. it could symbolise a reduction in factionalised conflict and may enable voters to 'reconsider' their perceptions of the Labour Party.
Notes 1. Gaming in nominations is conceivable, but we assume that the experience of the small number of parliamentarians who 'lent' their support to Corbyn at the nomination stage of the 2015 Labour Party leadership election (see Crines et al., 2018) to broaden the debate, will have made this unlikely in this nomination process. 2. On party leadership selection from a comparative perspective see Cross & Blais, 2012;Pilet & Cross, 2014; Aylott & Bolin, 2021for debates on leadership selection rules and democratisation, see LeDuc, 2001;Kenig, 2009a;Cross & Blais, 2010; or leadership election competitiveness, see Kenig, 2009b. 3. Long-Bailey had served as shadow Business Secretary in the Corbyn shadow Cabinet. 4. Nandy had been in the first Corbyn shadow Cabinet, covering Energy and Climate Change, but she partook in the mass frontbench resignations in July 2016 and thereafter remained on the backbenchers. Starmer had participated in the mass resignations (from his shadow junior Home Office role) but returned months later to the high-profile position of Shadow Brexit Secretary. Thornberry served as shadow Foreign Secretary in the Corbyn shadow Cabinet. Phillips had no frontbench experience. 5. Nandy was one of nineteen Labour parliamentarians who had voted for the Withdrawal Agreement (HC Deb,Vol. 666,22 October 2019).

Disclosure Statement
No potential conflict of interest was reported by the author(s).