The transformation of European socialism

Party-society relations

The second set of explanations for the ideological transformation of European socialism has focussed on parties’ connection to society. The central assumption here is that party elites respond to electoral pressure by shifting their ideological commitments in order to win votes (e.g., Downs Reference Downs1957; Kitschelt Reference Kitschelt1994). The secular decline in left-wing parties’ vote share over time (see Figure 1 and Benedetto et al. Reference Benedetto, Hix and Mastrorocco2020) makes it difficult to assess whether left-wing parties are really winning over voters in this way and, in fact, there is limited evidence to support the idea that rightward ideological shifts lead to durable electoral gains (Loxbo et al. Reference Loxbo, Hinnfors, Hagevi, Blombäck and Demker2021; Polacko Reference Polacko2022). But the argument that shifts in parties’ ideologies can be explained as responses to structural changes in society remains extremely powerful.

Figure 1. The decline of left-wing parties’ vote share (Western Europe, 1945–2021).

Notes: Data from the Comparative Manifesto Project. The top panel displays the election results of the UK Labour Party. In the bottom panel, each point shows the combined electoral strength of all left-wing parties (“Social Democratic,” and “Socialist and other left”) in each election in western Europe since 1945. Falling turnout means that the downward trend would be even more marked if the data tracked share of the electorate, rather than share of voters.

One version of this argument focusses on the numerical decline of the manual working class and the fragmentation of that group in terms of lifestyles and political identity (Hobsbawm Reference Hobsbawm1978; Przeworski and Sprague Reference Przeworski and Sprague1986). In this account, as the power of that voting block (or the party funding apparatus associated with it, Ferguson Reference Ferguson1995) declined, left-wing parties were forced to turn to other social groups for electoral support. They then evolved into “catch-all” parties that attempted to represent working- and middle-class groups simultaneously and transformed themselves ideologically to do so (Kirchheimer Reference Kirchheimer, La Palombara and Weiner1966).

An alternative version of the electoral-responsiveness argument focuses not on shifts in the economic base, but on the massive expansion of education over the twentieth century. This has led to the phenomenon of the “Brahmin Left,” where support for left-wing parties increasingly comes from highly educated voters and where parties are increasingly polarized along socio-cultural, rather than economic, lines (Gethin et al. Reference Gethin, Martínez-Toledano and Piketty2022). One explanation for this tendency is that, while party elites have often been dominated by the highly educated, it was only as education expanded to the masses that appeals to the kind of socio-cultural liberalism traditionally correlated with education (Cavaille and Marshall Reference Cavaille and Marshall2019; d’Hombres and Nunziata Reference d’Hombres and Nunziata2016) became a viable electoral strategy (Shor Reference Shor2020).

Both versions of the party-society explanation contain powerful insights. But, as Dylan Riley’s (2015: 183) “neo-Gramscian research program on parties” makes clear, this unidirectional model cannot do justice to the dialectical and reciprocal evolution of classes, parties, and civil society. More narrowly, it also does not explain why party elites interpreted structural changes in society in the particular way they did. According to Stephanie Mudge (Reference Mudge2018), the key actors here are the “experts” who do the interpretive and diagnostic work of translating social shifts into new political strategies. From this insight, Mudge then develops an account of the evolution of left-wing party experts: from the “party theoreticians” of the interwar period with their backgrounds in journalism, agitation and party organizing; to the “economist theoreticians” of the post-war settlement who emerged out of the world of professional economics; and finally the “trasnationalized, finance-oriented economists, strategists and policy specialists” of the 1990s (Mudge Reference Mudge2018: 1–43). Each new cohort of experts had a different institutional background and so brought with it a different “ethic” or “habitus.” And it was these new ethics that led to particular interpretations of structural change and to the emergence of new ideological positions.

But Mudge’s (Reference Mudge2018) work remains centered on the party-society dyad and does not address the fact that socialization also takes place within the institutions of the state. This is where I propose turning to Miliband (Reference Miliband1964, Reference Miliband1969) for a more nuanced and less deterministic account of party-state relationships, and for insight into the workings of a crucial mechanism in the transformation of western European socialism.

Miliband and the experience of governing hypothesis

Miliband’s work on the Labour Party was part of a broader movement within the British New Left, which was concerned by the party’s lack of radicalism and keen to evaluate the prospects for more progressive forces working within it. Many of these accounts focussed on Labour’s idiosyncrasies: the gradualism and empiricism it inherited from the British trade union movement (Nairn Reference Nairn1964a, Reference Nairn1964b), its commitment to parliamentarianism and repudiation of any form of direct action (Miliband Reference Miliband1964), the defensive concern to protect the institution of the party at all costs (Panitch Reference Panitch1979), and the party’s peculiar constitution (Minkin Reference Minkin1978). But Miliband (Reference Miliband1964, Reference Miliband1969) also drew attention to a more universal and abstract process: the way the experience of governing led to institutional socialization, which in turn led to deradicalization.

The clearest exposition of this mechanism comes in Miliband’s account of the Labour Party’s early history. This part of his narrative begins with the party joining the wartime government in 1915, as junior partners in a Liberal-Tory coalition (Miliband Reference Miliband1964: 47). As Miliband points out, this had a profound ideological effect because Labour politicians did not “lose the habits of mind engendered by this experience [of government] when the war came to an end” (Miliband Reference Miliband1964: 48). Indeed, this legacy became clear in 1924 when Ramsay MacDonald formed the first minority Labour government. The backbench Labour MP, George Lansbury, during a debate over the Labour government’s reversal of its previous policy on soldier-officer relations in the military, said “I think one of the faults of the system under which affairs are managed in this House is that men, when they accept office, are expected immediately to change their attitude towards great public questions” (quoted in Miliband Reference Miliband1964: 111). A founder of the immensely influential Fabian Society, Beatrice Webb, similarly noted in her diary from March of 1924 that “one of the most unpleasant features of this Government has been the willingness of convinced and even fanatical pacifists to go back on their words once they are on the Treasury Bench as Under-Secretaries for the War Services” (quoted in Miliband Reference Miliband1964: 111).

Despite the ignominious collapse of the first MacDonald government after only ten months in power, by 1929 Labour had become the largest party in Parliament and formed a second minority administration. Drawing on his earlier experience in government, MacDonald ignored a Labour Party conference decision that had given the party the power to appoint Cabinet and asserted his right as Prime Minister to pick his own. But whatever plans he might have had; events soon overtook them. A few months after Labour took office, news of the Wall Street Crash reached Britain and a global depression quickly ensued. There are many different accounts of Labour’s turn to austerity in this moment, and the eventual decision of MacDonald and his allies to abandon the Labour Party for a Tory-Liberal “National Government” (e.g., Howell Reference Howell2002; Marquand Reference Marquand1977). But Miliband’s (Reference Miliband1964: 163–79) distinctive contribution is to draw attention to the ways in which their experience of governing and socialization into the institutions of the state shaped their actions.

Alongside MacDonald in this moment was his key ally Philip Snowden, and Miliband is careful to show how both men embraced the new institutions they found themselves connected to through the offices of Prime Minister and Chancellor. As Miliband describes, Labour ministers did not “lack informed [economic] advice from friendly sources… [The party had established] a National Economic Committee, which would be the ‘eyes and ears’ of the Prime Minister on economic questions… and it included [the economist John Maynard] Keynes, [trade union leader Ernie] Bevin, [Fabian economist G. D. H.] Cole and [socialist economist R. H.] Tawney… [But] responsible Ministers were at all times more ready to listen to advice from industrialists and Treasury officials than from their own friends” (Miliband Reference Miliband1964: 163). Rather than a simple story of “betrayal,” what Miliband is describing here is a complex process of transformation. Many Labour figures were deeply affected by the institutions of government in which they found themselves. They were transformed by the experience of governing, and they then fought to remake the party in their own image.

There are several important points to draw out of Miliband’s experience of governing hypothesis. The first is that it focusses on governing. This marks an important difference with earlier writers who had focussed on the ways socialist politicians were absorbed into various elite social scenes (for the British case, see Owen Reference Owen2007; Webb Reference Webb1930). But entering government is a very particular experience (Miliband Reference Miliband1964: 106). Crucially, it involves socialist party elites moving out of one set of institutions (parties, trade unions, socialist organizations) and embedding themselves in the alien institutions of the executive (the civil service, central banks, government committees, the military, international trade boards, etc). This exposes them to new sources of information, new ways of thinking and, ultimately, a new habitus.

The second key point is that the experience of governing hypothesis focusses on institutional practices: the ways-of-doing and taken-for-granted assumptions that are reproduced within particular institutions (Miliband Reference Miliband1969: 119–145; for the famous “Treasury View” see Davis Reference Davis2022; Peden Reference Peden2000; Pliatzky Reference Pliatzky1989). These habits are not just sustained by individuals, but also by departmental procedures, models, and working assumptions. They form institutional practices that often outlast the individuals who implement them. (This does not mean that individuals are totally irrelevant to Miliband’s theory. Most importantly, it is assumed that party elites have the power to reshape the ideology of the party as a whole and, in the quantitative test below, it is assumed that they exert influence over what ends up in their election manifestos. In that sense, Miliband’s is a top-down theory of party change and one that privileges party elites and their response to the experience of governing.)

The third point of clarification is that Miliband’s account of deradicalization is domain-specific, with each branch of the state nurturing a potentially unique worldview and habitus. His view of deradicalization is therefore a more nuanced and multi-dimensional than in most of the social science literature (where it is normally operationalized as movement along a single left-right dimension). Instead, Miliband (Reference Miliband1964, Reference Miliband1969) draws attention to the different interests, priorities, and ways of working that are embedded in different institutions. The state is therefore not seen as a monolithic entity but one that is divided and fragmented, with diverging (and at times even opposing) traditions inculcated in the Treasury, the Home Office, the military, and other branches of the state.

The fourth and final point concerns the scope restrictions of this experience of governing hypothesis. Miliband’s example is that of the British Labour Party, and the details of his account make it clear that it cannot be easily generalized beyond the world of left-wing, socialist, and social democratic parties. There are several reasons for this. At the level of ideology, liberal, conservative, and Christian democratic parties often explicitly work within the existing social and political framework. This is very different to socialism that was, at least in theory, committed to moving beyond capitalism and bourgeois democracy. In terms of personnel, those other parties also often recruit leaders who already have connections to the institutions of government, or at least do so at a greater rate than socialist parties (see Alexiadou Reference Alexiadou2022). The effects of governing on other smaller, party families (green, regional/separatist, far-right) may be pronounced, but they would need separate theorization and testing, which unfortunately lies beyond the remit of this essay.

In sum, Miliband’s work represents a distinctive contribution to our understanding of European socialism. He argues that the experience of governing poses a unique challenge for socialist parties. It embeds left-wing party elites into new institutions, where they are gradually socialized into new ways of thinking and acting. And, ultimately, it leads them to reshape and deradicalize their own parties. As a quantitative proposition, this can be formalized in two hypotheses:

The challenge taken up in the rest of this essay is to subject these two hypotheses to systematic, cross-national testing, demonstrating the importance of Miliband’s work to scholars outside of the narrow confines of British political history.

Data

Quantitative tests of deradicalization require systematic measures of party ideology to serve as the dependent variable. At present, there are four main sources of such data: surveys of party members or supporters, surveys of party elites, expert surveys, and content analysis of manifestos. Each of these methods has its own advantages and disadvantages, and much work has been done to evaluate how well the measures produced in these different ways correlate with one another (Ecker et al. Reference Ecker, Jenny, Müller and Praprotnik2021; Norris Reference Norris2020). But in my case, the options are somewhat limited. Surveys of party members and supporters can be discounted on theoretical grounds: the experience of governing hypothesis is explicitly top-down as it is party elites who are exposed to the practice of government and who then reshape party ideology, not grassroots members. Of the remaining options (surveys of party elites, expert surveys, and manifesto data), my choice was guided by the variables they record and their temporal/geographic coverage. Unfortunately, surveys of party elites have tended to be country-specific with limited temporal coverage, while no expert surveys include measures of parties’ embeddedness in the institutions of state (instead they tend to include more general measures of left-right, libertarian-authoritarian, and populist ideology). This leaves us with content analysis of manifestos, of which by far the most significant is the Comparative Manifesto Project (CMP) (Volkens et al. Reference Volkens, Burst, Krause, Lehmann, Regel, Theres, Weßels and Zehnter2021).

The CMP is a large and very well-established dataset within political science, covering 67 countries, 849 elections, 1373 parties, and 5089 manifestos. The variables are produced by breaking down party manifestos into discrete statements, which are then sorted into various semantic categories. For example, the following sentence from the Democratic Party platform from 2012 can be broken down into three statements: “This approach includes tough spending cuts that will bring annual domestic spending to its lowest level as a share of the economy in 50 years [Economic Orthodoxy] // while still allowing us to make investments that benefit the middle class now [Middle Class and Professional Groups] // and reduce our deficit over a decade [Economic Orthodoxy]” (Manifesto Project Team 2018: np). Across a given manifesto, it is then possible to count the number of phrases in support of the Economic Orthodoxy, Freedom and Human Rights, Traditional Morality, and other variables. Important nuance is no doubt lost in this process. But it has the central advantage of creating a systematic, cross-national measure of party ideology, allowing researchers to easily test, for example, whether a party has made more statements in support of Welfare State Expansion than it used to or than its rivals have done.

There are two important issues with the CMP data for the purposes of this research. First, parties may say one thing in a manifesto but then go on to do something different in government. This represents a substantial limitation to my argument and cannot be discounted, although there is good evidence that most parties in mature democracies do fulfill their promises (Thomson et al. Reference Thomson, Royed, Naurin, Artés, Costello, Ennser-Jedenastik, Ferguson, Kostadinova, Moury, François and Praprotnik2017). Second, what parties say in a manifesto may be driven by what they believe voters want to hear and so not accurately reflect their real ideological position. This would only be problematic if parties who were in government in the previous election cycle were more likely to misrepresent their ideologies in their manifestos (so that the measurement error was correlated with the treatment). To evaluate this possibility, I repeat the matching procedure described below but use the absolute difference between each party’s score and that of the average voter ^{Footnote 2} as the dependent variable. I find no statistically significant differences between treated and untreated groups, suggesting that parties that have recently been in government are no closer to the average voter than their rivals from the opposition (Supplementary Material A9). ^{Footnote 3}

Using the CMP to measure the ideological affinity of political parties with respect to key state institutions (the dependent variable), I operationalize my two hypotheses as follows:

Full descriptions of each variable and the CMP’s coding procedures are available in Supplementary Material A2.

The independent variable is a binary indicator of whether a party has been in government at any point since the last election (thus capturing their “experience of governing”). As this is not recorded in the CMP, I linked the CMP data to two other datasets: ParlGov (Döring and Manow Reference Döring and Manow2021) and Party Facts (Döring and Regel Reference Döring and Regel2019). There are a handful of inconsistencies between these datasets (largely driven by decisions about how to code new parties formed out of splits), all of which were resolved manually. The governing experience of western European left-wing parties captured in this data is depicted in Figure 2.

Figure 2. Left-wing governments (Western Europe, 1900–2021).

Notes: Data from ParlGov. The shading of the tiles indicates the number of days spent under left-wing cabinets per year, weighted by the proportion of seats held by left-wing parties.

Finally, two variables are used as controls: the percentage of the vote attained by the party in that election, and their overall left-right scoreFootnote ⁵ at the previous election. These are both drawn from the CMP. Density plots for all key variables are displayed in Figure 3.

Figure 3. Density plots for key variables (n = 473 party-election observations).

Notes: Data from the Comparative Manifesto Project, ParlGov, and Party Facts. Sample includes all western European left-wing parties that have ever been in government (n = 37) resulting in 473 party-election observations between 1945 and 2021. Plots display density curves for each variable.

The argument in this essay focusses on left-wing parties in western Europe.Footnote ⁶ However, I also limit the quantitative analysis to left-wing parties that have ever been in government (n = 37, listed in Supplementary Material A1). This proviso is designed to better facilitate comparison across different socialist parties and assumes that small, fringe parties that have never entered government are subject to fundamentally different pressures than large, electorally competitive parties. (In any case, the main results are robust to dropping this condition and using all the data available in the CMP, see Supplementary Material A8.) The final dataset is therefore composed of 473 party-election observations between 1945 and 2021.

Methodology

Until very recently, the standard tools for quantitative analysis of this sort were two-way fixed effect (TWFE) models. However, recent methodological advances have revealed some serious problems with this approach (see de Chaisemartin and D’Haultfoeuille Reference de Chaisemartin and D’Haultfoeuille2022; Goodman-Bacon Reference Goodman-Bacon2021; Imai and Kim Reference Imai and Kim2021). Crucially, the coefficients generated by TWFE regressions are not robust to heterogeneous effects over time or across units and can be easily contaminated by other treatments. The fact that TWFE models can contain negative regression weights also makes it very difficult to untangle the implicit comparison being made by the model.

Taking onboard these critiques, I borrow from and extend Imai’s et al. (Reference Imai, Kim and Wang2023) matching approach for time-series cross-sectional data. Their work builds on the framework of causal identification, in which the central challenge is to identify a suitable “control group” against which we can compare those who receive the “treatment.” (The language used in the causal identification literature has its origins in experimental science but is now commonly used for observational data. In essence, the “treatment” refers to the independent variable and the “outcome” to the dependent variable.) Imai et al.’s (Reference Imai, Kim and Wang2023) contribution is to suggest a standardized way of identifying a control group in settings with many units and repeated observations over time (time-series cross-sectional, or panel, data). Their central intuition is that each treated unit can be matched to a control group that share a similar trajectory in the independent variable up to the moment where one of them receives the “treatment.” For example, their method suggests that a suitable comparator for a country that was run by center-left governments from 1970 before switching to a far-right government in 1985 (received the “treatment”) would be a different country that was also governed by the center-left from 1970 to 1985 but that did not then switch to the far-right (control).

More formally, there are three stages to Imai et al.’s (Reference Imai, Kim and Wang2023) approach. In the first, each treated unit is matched to a control group of untreated units that otherwise share an identical treatment history over the recent past. Adapting one of their original examples, country A that, in 1970, transitioned to democracy after ten years of authoritarian rule (received the “treatment”) would be matched to a control group of countries X, Y, and Z that share a history of authoritarian rule from 1960 to 1970 and have not yet made the transition to democracy (not received the “treatment”). Imai et al. recommend using a series of control variables to further refine the control group and ensure that it matches the treated unit as closely as possible. Continuing the same example, countries from the control group (countries X, Y, and Z) that have a similar Gross Domestic Product and population size to the treated unit (country A) could be given greater weight in the subsequent analysis. With each treated unit matched to a control group, the final stage is to estimate the overall effect of the treatment through a difference-in-difference calculation, which compares the trajectory in the dependent variable for the treated unit with that of the control group. Developing the same example, a researcher might calculate whether the change in economic growth (outcome) after country A transitioned to democracy was greater than the change in growth for those similar countries that did not make the transition (control group). (Difference-in-difference calculations implicitly include fixed effects for time and unit and are therefore often seen as analogous to TWFE, but Imai et al.’s approach is much more robust to the kinds of problems outlined above.) Across many treated units and many control groups, Imai et al. then take a simple average of the various difference-in-difference coefficients with bootstrapped standard errors, producing a robust, non-parametric estimate of the causal impact of the treatment.

Imai et al. (Reference Imai, Kim and Wang2023) present their approach in the context of regular and balanced panel data (e.g., observations for every year and every country). But their framework is clearly generalizable to other settings. In the case of socialist parties competing in elections, elections rarely take place in the same year across countries, while parties are continually emerging and disappearing. I have therefore adapted their framework to accommodate irregular panel data and developed an R package to allow other researchers to follow this more general approach in other empirical settings (Tiratelli Reference Tiratelli2024). The R package implements the following four-stage procedure:

1. 1. Defining the contemporary time-period: This is the only addition to Imai et al.’s (Reference Imai, Kim and Wang2023) original approach and involves matching each treated observation to all untreated observations occurring within a user-specified period. In this case, I focus on socialist parties competing in an election within a five-year window.Footnote ⁷

2. 2. Exact matching: Find a subset of those contemporary, untreated observations that have the same treatment history over the last n observations. In this case, I match parties who been out of government over the last three elections cycle.Footnote ⁸

3. 3. Refinement matching: To control for potential confounders, Imai et al. recommend further matching using propensity scores, covariate balancing propensity scores or Mahalanobis distance. In the first two cases, they can be used to produce weights that are used to calculate a weighted mean for the control group (giving more weight to more similar cases). All three can also be used to limit the size of the control group to the n most similar observations. In the analysis below I control for two variables (standardized vote share this election and standardized overall right-left position in the previous election) and set n to five.

4. 4. Estimate treatment effects: For each matched and refined set j: $\widehat {{b_j}} = \Delta T\; - \overline {\Delta U} $ ., where $\Delta T$ is the change in outcome for the treated unit and $\overline {\Delta U} $ is the (weighted) mean change in outcome for the control group. The R package then takes the average of those difference-in-difference coefficients as the final estimand: $\hat \beta = \mathop \sum \limits_{j = 1}^n \widehat {{b_j}}\;/n$ . Standard errors are calculated using a block bootstrap procedure (resampling across parties with 1000 iterations).

To make my procedure more concrete: when the Socialist Workers’ Party of Luxembourg competed in the 1989 election, it had just come out of a five-year period in government (i.e., it had received the “treatment”). I therefore match it to a control group of other socialist parties who competed in elections between 1987 and 1991 from opposition, and that otherwise share a similar history of being in and out of government over the previous three elections (in this case, the matches are: the Finnish People’s Democratic Union, the Icelandic People’s Alliance, the Icelandic Social Democratic Party, and the Portuguese Socialist Party). This matching process is then repeated for every “treated” unit, producing 59 matched sets. I then refine the control groups by controlling for vote share and overall right-left position and, finally, calculate the change in the dependent variables for the treated and control groups from the last election to this one. For example, I compare the change in the economic orthodoxy score (H1e) from 1984 to 1989 for the Socialist Workers’ Party of Luxembourg, to the (weighted) average change for the four parties in the control group.

The average trajectories for positive mentions of the economic orthodoxy (H1e) are depicted in Figure 4. In the first period (pre-treatment), none of the parties had been in government in the previous period and the treated and control groups both have very similar scores. But by the second period (post-treatment), there is a clear divergence, with parties who have been in government on average making far more positive mentions of the economic orthodoxy than those who remained in opposition. The difference between these two average trajectories is captured by the final difference-in-difference coefficient, which gives a robust estimate of the impact of the experience of governing on party ideology.

Figure 4. Average economic orthodoxy score for treated and control groups.

Notes: Data from the Comparative Manifesto Project, ParlGov, and Party Facts. Sample includes all western European socialist parties that have ever been in government (37) resulting in 473 party-election observations between 1945 and 2021. The analysis follows Imai et al. (Reference Imai, Kim and Wang2023): I match each treated observation with untreated observations that were also involved in an election within a 5-year time window and that have the same treatment history over the last 3 election cycles. This reduces the effective sample size to n = 234 with 59 matched sets. Further matching was then conducted to control for standardized vote share this election and their overall right-left position (lagged and standardized), using covariate balancing propensity score weights to produce weighted averages for the control group.

Alternative mechanisms and limitations

Miliband’s (Reference Miliband1964, Reference Miliband1969) experience of governing hypothesis foregrounds institutional socialization as the key causal mechanism linking serving in government with deradicalization. But scholars of populism have suggested an alternative explanation: the necessity of compromising as part of a coalition (e.g., Bernhard Reference Bernhard2020; Capaul and Ewert Reference Capaul and Ewert2021). If true, this alternative theory would imply that the causal effect will be smaller for parties that have less need to compromise. To test this, I mirror the analysis in Figure 6 and Table 1 and compare the individual difference-in-difference estimates for (a) parties ruling on their own vs those in coalition, and (b) parties that hold the position of prime minister vs those that don’t. This analysis produces no statistically significant results and so supports the original theory that institutional socialization is driving the associations seen in Figure 5 (Supplementary Material A15–16).

Another alternative explanation could be extrapolated from work on “cartelization.” If governing parties are more dependent on the state for resources than opposition parties, the former might also be more likely to adopt the ideologies of the state for purely instrumental reasons, rather than because of socialization. However, this argument runs against a central tenet of Katz and Mair’s (1995) original thesis, which was that the spoils of power are shared out more widely in a cartelized system than in a genuinely competitive one. Cartelization should therefore have reduced the impact of governing on party ideology, whereas my results suggest that this is only true for one dimension of deradicalization (parties’ relationship with the constitution). This is not decisive evidence against a focus on state resources but, on balance, the argument for socialization seems more strongly supported.

There are, however, important limitations to the results presented here. First, the analysis focusses on the effect across one electoral cycle, and it is not easy to determine how long those effects persist. In part, this is a question of methodology. Difference-in-difference designs (in fact most causal identification strategies) tend to be better at detecting credible causal effects over shorter time periods. This might therefore be an area that benefits from careful historical work and in-depth case studies. Second, I was unable to detect any variation by institutional context. As I argued above, this supports my focus on socialization rather than coalition-building as the key causal mechanism. But scholars may in future want to examine whether the experience of governing is the same for presidential and parliamentary executives, or for proportional representation and first-past-the-post systems. Third, a related area for future research concerns parties’ internal structures. One hypothesis, which follows Seymour Martin Lipset et al.’s (Reference Lipset, Trow and Coleman1956) suggestion that robust internal democracy allows organizations to evade the “iron law of oligarchy,” is that the effect would be smaller in more democratic party structures. But there are also other organization-level variables that could be examined, such as funding structures and the degree of party members’ control over elected officials.