Reduced incentives,reduced party unity: Evidence from parliamentary speeches

Abstract

This article examines how reductions in individual legislators’ political incentives affect their relationship with their party in a party-centered environment. We develop a principal-agent framework to argue that the absence of re-election incentives and diminishing prospects of attaining higher legislative offices increase the likelihood of agency loss, manifested in weaker adherence to party positions in parliamentary speeches. We hypothesize that when MPs are unable to run for re-election and face increasingly limited career advancement opportunities over the course of the electoral period, the mechanisms party leaders rely on to foster party unity become less effective, making these MPs more likely to deviate from the average party positions. To test these expectations, we leverage the case of the Five Star Movement’s party-imposed term limits in Italy. We draw on an original dataset of speeches delivered in the Italian Chamber of Deputies between 2013 and 2022. The results support our hypotheses, showing that reduced individual political incentives are associated with systematic increases in intra-party deviation in parliamentary speech.

Introduction

One of the most common assumptions in legislative studies is that elected representatives are primarily motivated by re-election, which becomes a necessary precondition for pursuing other political aspirations, such as advancing policy agendas. To achieve re-election, representatives strategically engage in parliamentary activities (Mayhew, 1974; Strøm, 2012). Among these activities, parliamentary speeches stand out as among the most visible and electorally relevant tools in an MP’s repertoire. Indeed, Proksch and Slapin (2014: 1) describe plenary speeches as “a forum for public communication that parties and their Members of Parliament (MPs) utilize for electoral purposes.” Consequently, because the prospects of securing re-election or attaining higher legislative office largely depend on party leaders and the electorate, ambitious legislators are expected to convey messages that align with the preferences of these key gatekeepers (Müller, 2000; Saalfeld, 2000).

Building on these assumptions, this article examines how MPs use parliamentary speeches when re-election under the same party label is no longer an available option and parties’ capacity to shape legislators’ future career prospects in parliament progressively declines. We focus on legislative term limits, a legal constraint that bars incumbents from seeking re-election after serving a fixed number of terms (Mooney, 2009). Our intuition, grounded in principal-agent theory and the last-period problem literature (e.g., Zupan, 1990), is that when party leaders have limited opportunities to leverage electoral and office incentives to provide desirable perks to their representatives, these representatives become less likely to promote party positions in their speeches. Specifically, we expect that reductions in re-election and office incentives increase the likelihood of agency loss, as term-limited parliamentarians prioritize their own interests and preferences–and/or those of alternative principals–over party unity.

We formulate two hypotheses to assess the extent to which agency loss manifests in parliamentary speeches. First, we hypothesize that term-limited MPs are more likely to deviate from their party’s average positions in their speeches. Second, we expect that, as the legislative term progresses, term-limited MPs become increasingly likely to deviate from average party positions, as office-related incentives, such as the prospect of securing committee chairs or ministerial appointments, progressively contract over time.

To test these expectations, we leverage the case of Italy between 2013 and 2022, a party-centered legislative environment characterized by weak incentives to cultivate personal vote and deviate from party positions. Our analysis focuses on the Five Star Movement’s (M5S) party-imposed two-term limit, a rule that effectively barred parliamentarians elected in both the 2013 and 2018 elections from seeking re-election or holding a governmental office at the regional, national, or European level under the M5S party label. For the empirical analysis, we compiled an original dataset of plenary speeches delivered by members of the Italian Chamber of Deputies between 2013 and 2022. We assess party unity by applying a supervised scaling method to estimate legislators’ ideological positions and computing their distance from the party average. Our empirical strategy compares changes in party unity among term-limited M5S MPs with corresponding changes among their non-term-limited peers. The results support our hypotheses, showing that declining individual political incentives are associated with systematic increases in intra-party deviation in parliamentary speech.

By examining how reduced re-election and office incentives affect the relationship between individual legislators and their party in a party-centered environment, this article makes several contributions. We add to recent literature (e.g., Motolinia, 2025) by quantifying how changes in political incentives affect party unity and legislators’ propensity to adhere to party positions in parliamentary speeches. In addition, we isolate the effect of declining office-based incentives over the course of the parliamentary term by examining how party unity evolves among legislators who already lack re-election incentives, an aspect that has received limited attention in the literature on term limits and the last-period problem. We also provide direct evidence in support of the assumption that plenary speeches serve electoral purposes, a core premise in research on parliamentary debates (e.g., Proksch and Slapin, 2014). Finally, although our analysis focuses on a relatively extreme case of incentive reduction–legislative term limits–our findings speak more broadly to contexts in which representatives’ political incentives fluctuate, as adverse shifts in re-election, office, or policy incentives, such as declining electoral support or institutional reforms, occur relatively frequently (e.g., Bøggild and Pedersen, 2024).

The remainder of the article is structured as follows. We first present our theoretical framework and derive our hypotheses. We then describe the research design, including the case study, dependent and independent variables, and statistical approach. Next, we present the results and discuss limitations, alternative explanations, and robustness checks. Finally, we conclude with a discussion of the implications of our findings and outline avenues for future research.

Political incentives, agency loss, and Party unity in parliamentary speech

In any organizational context, a fundamental challenge lies in ensuring members’ cooperation and compliance in the pursuit of collective goals. As Olson (1971) famously argued in The Logic of Collective Action, self-interested individuals will not necessarily–or automatically–act to advance group interests, even when all members would benefit from doing so. This collective action problem is particularly acute when individuals can free-ride, enjoying non-excludable benefits without bearing their share of the costs or efforts required to obtain them. To mitigate this problem and the associated risk of agency loss, organizational leaders, acting as principals, typically rely on selective incentives and coercive mechanisms to align agents’ behavior with the group’s collective interests.

The principal-agent framework is frequently used to understand the functioning of organizations, and parliamentary party groups (PPGs) are no exception (Müller, 2000; Saalfeld, 2000; Strøm, 2000). In political parties, legislators act as agents, while party leaders–especially in party-centered environments–serve as principals. Agency loss occurs when legislators fail to promote party positions or advance the party’s interests through their parliamentary activities. Such misalignment dilutes the party label, weakens executive control when the party is in government, and reduces its overall influence on policy outcomes (Greene and Haber, 2015). Undermining the party’s overarching objectives–winning votes, securing office, and implementing preferred policies (Müller and Strøm, 1999)–also harms individual MPs, as a less influential or successful party reduces their chances of re-election, access to higher legislative offices, and the advancement of their policy preferences (Strøm, 2012). For these reasons, a party’s collective influence and success constitute a critical public good from which all its members benefit.

However, the goals and priorities of principals (party leaders) and the actions of agents (legislators) can diverge, giving rise to agency loss. While parties attempt to mitigate this risk ex ante by selecting candidates who share their policy inclinations and have a record of proven loyalty (Carroll and Nalepa, 2020; Galasso and Nannicini, 2015), MPs may nonetheless find themselves at odds with the party platform once in office. To discourage deviations from the party line, principals can rely on their capacity to allocate or withhold desirable perks. Assuming, as we do, that the pursuit of re-election is a necessary condition for MPs’ political goals (Mayhew, 1974), the primary instrument for curbing agency loss lies in the control parties exert over legislators’ re-election prospects, as party leaders typically play a central role in candidate selection. In addition, by controlling access to other coveted perks, such as spokesperson and committee assignments (Mickler et al., 2025), parties can offer opportunities for policy influence and career advancement. This creates a do-ut-des relationship, in which parties use these incentives to align legislators’ behavior with collective partisan interests (Müller, 2000). Party leaders can thus allocate benefits to reward cooperation or withdraw them to sanction members who break the whip, that is, those who deviate from the party’s collective interests. In essence, parties as principals foster party unity through “carrot-and-stick” mechanisms.

However, when institutional constraints such as legislative term limits are enforced, the risk of agency loss increases. By analogy with the literature on the last-period problem among retiring MPs (Bailer and Ohmura, 2018; Zupan, 1990), term-limited legislators are no longer responsive to the renomination incentives normally provided by their parties. Without the need to satisfy the principal who holds the key to their re-election, these representatives may engage in “shirking” and “slacking” (Frech et al., 2021; Willumsen and Goetz, 2017; Zupan, 1990). Notably, in the former case, MPs may ideologically deviate from the principal’s preferred position either to pursue personal interests or to appeal to a new principal, thereby generating agency loss vis-à-vis the original principal. ¹

Against this backdrop, term limits (or, more generally, the lack of re-election prospects) have been shown to affect several political outcomes, including the provision of resources and particularistic legislation (Aidt and Shvets, 2012; Motolinia, 2021), constituency responsiveness (Carey et al., 2006), parliamentary productivity (Bromo et al., 2025; Fiva and Nedregård, 2024; Fouirnaies and Hall, 2022), and party unity, especially in roll-call votes (Frech et al., 2021; Motolinia, 2025). In this study, we measure agency loss through parliamentary speeches, a setting in which collective action problems manifest (Proksch and Slapin, 2012: 522). Speeches allow legislators to signal individual policy positions to constituents or alternative principals, potentially diverging from the party line. By publicly articulating individual positions or priorities that do not fully align with the party’s agenda, legislators can dilute the distinctiveness of the party brand. As a result, voters may struggle to infer the party’s policy stance (Van Kleef et al., 2024), while coalition partners may question the party’s reliability in light of internal disagreement (Martin and Vanberg, 2008). These dynamics undermine the party’s strategic capacity and generate agency loss. Figure A1 in the Appendix visually summarizes our theoretical framework.

Compared to roll-call votes, parliamentary speeches provide a more reliable basis for estimating legislators’ positions (Bäck et al., 2021; Laver, 2021; Schwarz et al., 2017). Party unity in roll-call votes is often inflated because, in parliamentary systems, rebel votes are rare and typically represent ”the ultimate act of defiance” by an MP (Proksch and Slapin, 2014: 26). Parliamentary party group leaders may also strategically call roll-call votes to discipline dissenting members and suppress internal disagreement (Carrubba and Gabel, 2008; Hug, 2010). By contrast, parliamentary speeches are relatively less constrained, as party leaders are less likely to sanction MPs for expressing dissenting views “as long as they vote with the party line” (Schwarz et al., 2017: 522). As a result, speeches can function as an “institutional safety valve” (Maltzman and Sigelman, 1996), channeling intra-party disagreement into rhetoric rather than more consequential acts of defection by legislators with weaker incentives to promote the principal’s positions. In this sense, as emphasized by Laver (2021: 24–25), the analytical value of parliamentary speeches lies in their capacity to reveal “heterogeneous personal views of legislators” alongside otherwise “homogeneous voting behavior.” Accordingly, the analysis of parliamentary speeches can uncover internal party disunity precisely where roll-call votes might misleadingly suggest unity.

Building on this theoretical framework, our central expectation is that reductions in political incentives affect individual legislators’ propensity to deliver speeches that deviate from average party positions. Prior work on the last-term problem suggests that MPs serving their last term, for example, because they are about to retire, are more likely to prioritize their own interests (Geys and Mause, 2016). Similarly, term-limited legislators whose future prospects no longer depend on party renomination face weaker constraints to align their speeches with the party line. As a result, the suppression of re-election incentives increases the likelihood that these legislators stray from average party positions in their speeches, prioritizing their own interests and preferences and/or those of alternative principals over collective partisan interests.

Hypothesis 1

(H1): Term-limited MPs are more likely to deviate from the average party positions in their parliamentary speeches.

While H1 posits a general increase in deviation among term-limited parliamentarians across the entire legislative term, we now adopt a more dynamic perspective and examine how the risk of agency loss evolves within a parliamentary cycle. To account for this temporal variation, it is important to recognize that the incentives principals use to curb agency loss, most notably the prospect of re-election and the allocation of higher legislative offices, vary in their effectiveness over time. On the one hand, re-election incentives become most salient toward the end of the term, when party leaders typically make decisions about candidate selection for the subsequent general election (Fernandes et al., 2018; Greene and Haber, 2015). On the other hand, the effectiveness of office-based incentives, such as committee chairs or ministerial posts, declines over time, as the pool of available offices that principals can distribute shrinks during the term, with most prestigious appointments occurring earlier in the parliamentary cycle or only at infrequent moments, such as committee renewals or the formation of a new government (Martin, 2014).

Consequently, MPs are expected to adjust their behavior in response to these temporal changes in electoral and office-based incentives. The positions expressed in parliamentary speeches should therefore not be static; rather, they are expected to shift as a function of the perceived effectiveness and availability of both electoral and office incentives. Importantly, these temporal patterns of responsiveness are likely to differ markedly between legislators subject to term limits and those not. While the decline in office-based incentives affects all parliamentarians, MPs who remain eligible for re-election continue to face strong incentives to promote the party agenda in their speeches, particularly as the end of the legislative term approaches and candidate selection for the next general election looms (Greene and Haber, 2015). By contrast, term-limited legislators may initially align closely with the party line to maximize access to offices earlier in the term, but are expected to become progressively less responsive to party leaders as the electoral period draws to a close. As a result, for term-limited MPs, the costs of deviating from party positions decrease over time, while the returns from upholding collective partisan interests likewise sink. Conversely, we should not observe a comparable decline in party unity among non-term-limited parliamentarians. This argument leads to our second hypothesis:

Hypothesis 2

(H2): As the electoral term progresses, term-limited MPs become increasingly likely to deviate from average party positions in their parliamentary speeches.

Research design

Case study

We test our hypotheses in the context of the Italian Chamber of Deputies (Camera dei Deputati), the lower chamber of the Italian Parliament, focusing on the party-imposed legislative term limits adopted by the Five Star Movement (Movimento 5 Stelle). During the period under study (2013–2022), the M5S–a populist party founded in 2009 (Mosca and Tronconi, 2019)–was a central actor in Italian politics, competing in the 2013, 2018, and 2022 general elections. In 2013, the party’s first electoral contest, the M5S won 25.6% of the vote and remained in opposition throughout the 17^th legislative term (2013–2018). In 2018, the M5S emerged as the largest party, securing 32.7% of the vote and participating in three coalition governments (Conte I, Conte II, and Draghi). In 2022, the party won 15.5% of the vote and returned to the opposition.

The rule on legislative term limits–unique to the M5S during the period under study–restricted party members to a maximum of two terms (consecutive or non-consecutive) in any governmental office at the European, national, or regional level. ² This rule has been a flagship organizational principle of the Five Star Movement from the very beginning and remained a credible institutional constraint throughout the period covered by our study, until it was ultimately repealed by party members in 2024. ³ These restrictions, reflecting widespread distrust of professional politicians, were embraced by the party as a symbol of rupture between M5S candidates and career politicians from mainstream parties (Casaleggio and Grillo, 2011), thereby reinforcing its identity as a challenger party.

As the term-limit rule was in place for less than three electoral periods, it effectively prevented candidates from running for office only in the 2022 general election. Legislators elected in 2013 and re-elected in the subsequent 18^th legislative term (2018–2022) were therefore barred from running again under the M5S label in 2022. As a result, this cohort–predominantly drawn from low- and medium-income occupations and without prior political experience before 2013 (Tronconi and Verzichelli, 2014)–entered the 18^th legislative term knowing from the outset that it would be their final term with the M5S. ⁴ Therefore, while M5S party leaders could initially offer both re-election and office incentives, the principal reward they could still offer to MPs serving their second term was access to higher legislative offices. Such rewards were largely administered by the party, given its influence over ministerial appointments and its authority over committee assignments and leadership positions (e.g., committee chairs), which are renewed every two years in the Italian Chamber of Deputies.

According to our theoretical framework, this institutional configuration implies that MPs who cannot stand for re-election and whose prospects of securing higher legislative offices diminish over time should be less inclined to use parliamentary speeches to promote their party’s positions. In the party-centered Italian context of the 18^th legislative term, promoting the party line through speeches constitutes a “strategic activity” (Strøm, 2012), as electoral institutions provide limited incentives for cultivating personal vote and strong incentives to rely on the party brand (Viganò, 2025). The 2013 general election was held under a closed-list proportional representation (CLPR) system, which was replaced in 2017 by a mixed electoral system allocating 61% of seats through CLPR and 37% through first-past-the-post in single-member districts. Under both systems, re-election prospects depended primarily on party performance and candidates’ placement on party lists (Pedrazzani and Pinto, 2018). ⁵ Although the mixed system could, in principle, have encouraged personal vote-seeking strategies, competition in single-member districts remained limited, and individual candidates’ contributions to parties’ electoral performance were marginal.

A similar pattern emerges in candidate selection. Although the M5S formally held online primaries for parliamentary candidates, the process remained highly centralized and tightly controlled by party leaders in practice. For the 2013 election, M5S candidates were selected online by party members from a pool of pre-approved individuals screened by the party leadership. Notably, MPs elected through these primaries did not differ in their propensity to pursue personal vote-seeking strategies (Rombi and Seddone, 2017). This absence of differences plausibly reflects an ex ante screening process by the M5S leadership, which restricted access to the primaries to candidates ideologically closer to the party platform. Such a strategy allows parties to reduce the risk of adverse selection and reduces the need to rely on “carrot-and-stick” mechanisms during the parliamentary term (see Carroll and Nalepa, 2020). In 2018, the same rules applied to the selection of candidates for the proportional tier, while candidates for single-member districts were chosen exclusively by party leaders.

Finally, parliamentary speeches constitute a suitable venue for gauging party unity in the Italian Parliament. Particularly in sessions where roll-call votes are not scheduled, constraints on speech are relatively low, affording MPs considerable freedom to express their positions. According to the Chamber of Deputies’ Standing Orders, all representatives have the right to speak on topics of their choosing on specific occasions, such as debates on daily motions (ordini del giorno), oral questions, and statements delivered at the close of a plenary sitting (interventi di fine seduta). Moreover, during debates on the ratification of decree-laws–which occur frequently in plenary sessions (e.g., Bromo et al., 2023)–PPG leaders exercise more limited control over speaking opportunities (Giannetti and Pedrazzani, 2016). Taken together, this institutional setting, combined with the M5S-imposed term limits, provides a well-suited context for examining how the relationship between parties and individual legislators evolves as re-election incentives disappear and office-based incentives wane. In an environment where high party unity and limited deviation from collective partisan interests would otherwise be expected, the M5S rule thus creates a unique opportunity to study agency loss through parliamentary speeches.

Dependent variables

Our goal is to assess the extent to which MPs deviate from average party positions in response to changes in electoral and office incentives. To measure such deviation, we use plenary speeches delivered by elected members of the Chamber of Deputies during the 17^th and 18^th legislative terms (Scopelliti et al., 2025). Following the pre-processing procedure (see Appendix), we obtained two corpora comprising 90,345 speeches in total, whose characteristics are summarized in Table A1 in the Appendix.

A variety of methods have been developed to estimate legislators’ positions from parliamentary speeches. Among these, the Wordshoal text model (Lauderdale and Herzog, 2016)–an unsupervised scaling approach derived from Wordfish (Slapin and Proksch, 2008)–was developed specifically for this purpose. Owing to its unsupervised nature, however, careful validation of Wordshoal estimates is essential. In our case, the model did not recover meaningful ideological variation across parliamentary party groups. Instead, it primarily captured a relatively sharp government versus opposition divide. Because our primary interest lies in comparing deviations from party positions between legislators subject to term limits and those without such constraints, we instead rely on Wordscores (Laver et al., 2003). This supervised approach allows us to estimate the positions of members of the Italian Chamber of Deputies in a manner that more directly reflects ideological differences at the party level and is therefore better suited to our empirical objectives.

Wordscores is a supervised scaling method that uses reference texts with known reference scores, specified by the researcher, to estimate the positions of virgin texts, that is, texts whose latent positions are unknown and are the object of inference. The application of Wordscores to estimate positions from legislative speeches is well established. For example, Laver and Benoit (2002) employed the method to scale MPs’ positions in the Irish Dáil Éireann along a pro- versus anti-government dimension, using speeches by the prime minister and the opposition leader as reference texts. Subsequent studies have adopted similar strategies while focusing on ideological dimensions. For instance, Bäck and Debus (2016) used electoral manifestos as reference texts to scale legislators’ speeches along an economic left-right dimension. Bernauer and Bräuninger (2009) extended this approach by applying Wordscores to estimate German MPs’ ideological positions, using party leaders’ speeches as reference texts. Party leaders’ speeches provide broad and stable vocabulary coverage over the legislative term, accommodating variation arising from temporal dynamics and agenda changes. In this case, reference scores for the leaders’ speeches are derived from expert surveys.

We follow an approach similar to that of Bernauer and Bräuninger (2009). Specifically, we use speeches delivered by PPG leaders between 2013 and 2022 as reference texts. Because PPG leadership can change within a legislative term, we include only speeches delivered while an individual formally held a leadership position. Reference scores are drawn from expert survey data (Gambacciani et al., 2025). ⁶ Table A2 in the Appendix summarizes the PPG leaders and the expert survey reference scores assigned to the reference texts by legislative term and PPG. We estimate two separate Wordscores models, one for each legislative term. After computing scores of words from the reference texts, we scaled each speech and normalized the Wordscore estimates into z-scores to ensure comparability across the two periods. ⁷ Finally, to obtain the legislator’s ideological position, we averaged the estimated positions across their speeches. We include additional information on the methodology in the Appendix, together with Figure A2, which plots the average and individual estimated positions by PPG and legislative term.

To test our first hypothesis, the dependent variable must capture the extent to which each legislator deviates from their parliamentary party group. We operationalize this concept as the absolute difference between an individual MP’s estimated ideological position and the mean position of their PPG, which we interpret as reflecting the party’s collective partisan interest. ⁸ To compute the PPG average, we take the mean of the estimated positions of all legislators affiliated with that PPG. We rely on each representative’s initial PPG affiliation. Because party switching is not uncommon in the Italian Parliament, this approach may not fully capture all underlying dynamics. To address this concern, we include a control for party switchers in our main models and conduct additional robustness checks. We discuss these alternative specifications in greater detail below. Higher values of the dependent variable (Deviation from PPG Average) indicate a greater divergence of a parliamentarian’s speech from the PPG average.

To test our second hypothesis, we shift the unit of analysis from individual MPs to parliamentary party groups and focus exclusively on the 18^th legislative term. We calculated the average level of deviation within each PPG over 3-month intervals, from the convening of the new parliament in March 2018 to its dissolution in July 2022. Starting from the estimated positions of individual representatives, we first computed the PPG average for each trimester. We then calculated the mean absolute distance between each MP’s position and the corresponding PPG average within each trimester, yielding a measure of average intra-party divergence at the PPG-trimester level. For the M5S, we compute two separate sets of estimates: one for MPs not constrained by term limits (i.e., serving their first term) and one for those constrained by term limits (i.e., serving their second term). This distinction is essential, as we expect MPs who are unable to run for office in the subsequent electoral period to deviate increasingly from average party positions as the legislative term progresses.

In the case of H2, party switching is addressed by construction. Because the analysis relies on a time-series structure, when calculating both individual legislators’ positions and the corresponding PPG averages, we include only speeches delivered while MPs were affiliated with a given PPG. Accordingly, if a representative switched party groups during a particular trimester, their speeches are assigned to the PPG to which they belonged at the time of delivery. This approach ensures that the calculated PPG averages and levels of intra-party deviation accurately reflect each group’s composition and activity within each time interval.

We validate our measurement strategy and estimates in three ways. First, as shown in Figure A3 in the Appendix, party-level averages of our estimated positions closely align with substantive expectations derived from expert surveys, which are distinct from the data used to assign reference scores. Second, following Motolinia (2025), we show that individual legislators’ estimated positions cluster within parliamentary party groups (Figure A2). Third, we demonstrate–both visually and through a t-test–that legislators are closer to their own party’s average position than to the averages of other parties (Figure A4).

Independent variables

For H1, our main independent variables include a binary indicator identifying M5S legislators (Five Star Movement) and a binary indicator identifying the 18^th legislative term (18 ^th Legislative Term). For H2, the key independent variables are a binary indicator identifying M5S MPs serving their second term during the 18^th legislative term (No Re-Election) and a time indicator capturing the progression of the term, measured in trimesters (Trimester (18 ^th Legislative Term)).

Several factors beyond changes in political incentives induced by term limits may affect individual parliamentarians’ degree of deviation from party positions; these factors are therefore included as controls in our models. For H1, we control for: (1) the number of speeches delivered by a given MP (Total Number of Speeches), as more active legislators may address a broader range of issues and thus have greater scope for deviation from the PPG average; (2) MPs’ electoral vulnerability, capturing how “safe” an MP’s seat is (Candidate Vulnerability), as more vulnerable legislators may display greater loyalty to the party due to heightened concerns about renomination and re-election (e.g., Jastramskis et al., 2024); (3) a binary indicator identifying legislators elected in single-member districts (SMD Election), who may, in principle, face stronger incentives to cultivate personal vote; ⁹ (4) a dummy variable identifying party switchers (Party Switching), to account for deviation associated with changes in party affiliation; (5-6) dummy variables identifying MPs who held higher legislative offices (Legislative Office) or ministerial positions, including junior posts (Government Membership), as these roles are typically associated with higher party unity due to collective responsibility, leadership rewards, and involvement in the executive agenda; (7) the number of terms served (Tenure), capturing potential experience effects, whereby more seasoned parliamentarians may either accumulate political capital that affords them greater autonomy or become more deeply embedded in party norms; and (8-9) legislators’ age at the time of election (Age) and gender (Gender). We collected all the relevant data from the official data repository of the Italian Chamber of Deputies. ¹⁰

For our second hypothesis, we incorporate controls to account for the following factors: (1) changes in public opinion between trimesters (ΔOpinion Polls), which may generate fluctuations in political incentives beyond those associated with term-limit constraints (e.g., Bøggild and Pedersen, 2024); (2) parliamentary party group opposition status (Opposition), as party unity tends to be higher among opposition parties (Sieberer, 2006); (3) the net balance of PPG membership within each trimester (PPG Size Balance), since gains or losses in membership may affect a group’s average level of deviation by altering the pool of speakers; and (4) government alternation (Government Alternation), capturing potential dynamics related to changes in the governing coalition.

Statistical approach

For H1, we consider a population of MPs who were elected to the Italian Chamber of Deputies in 2013 and re-elected in 2018. ¹¹ Our sample consists of 183 representatives observed across the 17^th and 18^th legislative terms. Of these, 59 were members of the Five Star Movement who served their first term between 2013 and 2018 and their second term between 2018 and 2022. Due to the party’s term-limit rule, these MPs were ineligible to run for office under the M5S label in the 2022 election. The data are structured at the individual legislator (i)-legislative term (t) level. We estimate the following model:

$\begin{array}{c}\hfill {\text{Deviation\hspace{0.17em}from\hspace{0.17em}PPG\hspace{0.17em}Average}}_{it}={\beta }_0+{\beta }_1{\text{Five\hspace{0.17em}Star\hspace{0.17em}Movement}}_{it}+{\beta }_2{18}^{\text{th}}{\text{\hspace{0.17em}Electoral\hspace{0.17em}Period}}_{it}\\ \hfill +{\beta }_3{(\text{Five\hspace{0.17em}Star\hspace{0.17em}Movement}\times {18}^{\text{th}}\text{\hspace{0.17em}Electoral\hspace{0.17em}Period})}_{it}+\mathit{\beta }\mathbf{Z}+\mathit{ϵ},\end{array}$

(1)

where β₃ is the coefficient of primary interest, capturing the difference between changes in behavior among M5S legislators as they transition from their first term in the 17^th legislative term to their second term in the 18^th legislative term, and the corresponding changes in behavior among MPs from other parliamentary party groups over the same period. Z denotes a vector of control variables with associated coefficients β .

We estimate several specifications: a baseline model without controls (column 1); a model including controls (column 2); a model with controls and legislator random effects (column 3); a model with controls and parliamentary party group random effects (column 4); and a model with controls and both legislator and PPG random effects (column 5). In addition, we estimate two robustness specifications that exclude party switchers from the sample, one with controls only (column 6), and one with controls and legislator random effects (column 7) (Table 1). Random effects are included to account for within-unit correlation in unobserved characteristics. Table A3 in the Appendix reports summary statistics for the variables included in this model.

Table 1.

H1 estimates (Deviation from PPG average).

	(1)	(2)	(3)	(4)	(5)	(6)ExcludingSwitchers	(7)ExcludingSwitchers
Intercept	0.33*** (0.02)	0.31*** (0.10)	0.31*** (0.10)	0.30*** (0.11)	0.29** (0.11)	0.27*** (0.10)	0.27*** (0.10)
Five Star Movement (Ref. Other PPGs)	−0.19*** (0.04)	−0.15*** (0.05)	−0.15*** (0.05)	−0.15 (0.07)	−0.15 (0.08)	−0.12** (0.05)	−0.12** (0.05)
18th electoral period (Ref. 17th electoral period)	−0.00 (0.03)	−0.02 (0.04)	−0.02 (0.04)	−0.03 (0.04)	−0.03 (0.04)	−0.04 (0.14)	0.03 (0.04)
Five Star Movement×18th electoral period	0.27*** (0.06)	0.22*** (0.07)	0.22*** (0.07)	0.23*** (0.07)	0.23*** (0.07)	0.13* (0.08)	0.13* (0.07)
Total number of speeches		0.00(0.00)	0.00(0.00)	0.00(0.00)	0.00(0.00)	−0.00(0.00)	−0.00(0.00)
Candidate vulnerability		−0.26(0.17)	−0.26(0.17)	−0.24(0.17)	−0.24(0.17)	−0.27(0.17)	−0.27(0.17)
SMD election (Ref. PR election)		0.01(0.14)	0.01(0.14)	0.02(0.14)	0.02(0.14)	−0.04(0.14)	−0.04(0.14)
Vulnerability × SMD		0.22(0.24)	0.22(0.24)	0.20(0.24)	0.19(0.24)	0.28(0.25)	0.27(0.25)
Party switching (Ref. No switching)		0.06(0.04)	0.06(0.04)	0.06(0.04)	0.06(0.04)
Legislative office (Ref. No office)		−0.04(0.03)	−0.03(0.03)	−0.03(0.03)	−0.03(0.03)	−0.05(0.03)	−0.05(0.03)
Government membership (Ref. No membership)		−0.01(0.04)	−0.01(0.04)	−0.01(0.04)	−0.01(0.04)	−0.01(0.05)	−0.00(0.05)
Tenure		0.01(0.02)	0.01(0.01)	0.01(0.02)	0.01 (0.02)	−0.01 (0.01)	−0.01 (0.01)
Age		0.001(0.00)	0.001(0.00)	0.001(0.00)	0.001(0.00)	0.00(0.00)	0.003(0.00)
Gender (Ref. Male)		−0.01(0.03)	−0.01(0.03)	−0.00(0.03)	−0.00(0.03)	0.00(0.03)	0.00(0.03)
Legislator random effects			✓		✓		✓
PPG random effects				✓	✓
Observations	366	366	366	366	366	273	273
R2/Conditional R2	0.08	0.10	0.14	0.11	0.17	0.12	0.15

Note. Ordinary least squares (OLS) estimates in columns 1-2 and 6. Mixed-effects (restricted maximum likelihood) estimates in columns 3-5 and 7. Standard errors in parentheses. ***p < .01, **p < .05, *p < .10.

For H2, we consider the eight parliamentary party groups represented in the 18^th legislative term (2018–2022), observed at 3-month intervals. The data are structured at the parliamentary party group (i)-trimester (t) level. We estimate the following model:

$\begin{array}{l}{\text{Deviation\hspace{0.17em}from\hspace{0.17em}PPG\hspace{0.17em}Average}}_{it}={\beta }_0+{\beta }_1\text{No }\mathrm{R}\mathrm{e}{\text{-Election}}_{it}+{\beta }_2\text{Trimester }{({18}^{\text{th}} \text{Electoral\hspace{0.17em}Period})}_{it}\\ +{\beta }_3{(\text{No }\mathrm{R}\mathrm{e}\text{-Election}\times \text{Trimester \hspace{0.17em}}({18}^{\text{th}} \text{Electoral\hspace{0.17em}Period}))}_{it}+\mathit{\beta }\mathbf{Z}+\mathit{ϵ},\end{array}$

(2)

where β₃ is the coefficient of primary interest, capturing the differential change over time in average deviation among M5S legislators serving their second term relative to both M5S legislators serving their first term and members of other parliamentary party groups. As before, Z denotes a vector of control variables with associated coefficients β .

We estimate three specifications: a baseline model without controls (column 1), a model including controls (column 2), and a model including controls and parliamentary party group fixed effects (column 4) (Table A9 in the Appendix). Summary statistics for the variables included in this model are reported in Table A4 in the Appendix. ¹²

Results and discussion

The estimation of Model (1) yields the results reported in Table 1. Turning first to the two constitutive terms, the coefficient for the M5S indicator is negative and statistically significant at conventional levels in columns 1–3 and 6–7. This suggests that during the 17^th legislative term (2013–2018)–when the party first entered the Italian Parliament–M5S MPs, on average, deviated less from their parliamentary party group’s average position than legislators affiliated with other PPGs in the Chamber of Deputies.

This finding is important for interpreting the results that follow. Because our analysis tracks the same cohort of M5S legislators across two legislative terms, any positive and statistically significant increase in deviation observed in the 18^th term compared to other groups would be consistent with changes in political incentives associated with term limits rather than compositional changes within the parliamentary party group. In other words, the shift cannot be explained by the entry of new, potentially more nonconforming MPs. Moreover, since M5S legislators did not initially exhibit higher levels of deviation than members of other PPGs, and in fact exhibited comparatively greater unity during their first term, any subsequent increase in deviation represents a meaningful behavioral change.

Moving on to our central interaction term, the coefficient is positive and statistically significant across all specifications, as hypothesized. Substantively, this suggests that M5S legislators who started their second term in 2018 (at which point, they would have been constrained by the term-limit rules) deviated from the party’s average positions to a greater extent than other party groups in parliament. Again, the pool of M5S MPs is the same as the previous electoral period, and we know they did not always exhibit a greater tendency to deviate than representatives affiliated with other parties. This change is therefore consistent with our explanation: the loss of re-election incentives prompts these legislators to disregard collective interest on the floor, resulting in greater divergence from the party average. The results are consistent across all specifications, including in the models where party switchers are dropped from the sample. These results increase our confidence that the observed change in M5S legislators’ behavior is not driven by party switchers who may align their speeches with a new or prospective parliamentary party group and might therefore be qualitatively different from legislators who remain with their original party group. We show marginal effects in Figure 1. ¹³

Figure 1.

Marginal Effect of 18^th Electoral Term (M5S vs. Other Parties).

The figure shows a positive and significant change in the M5S group, in line with our expectation that deviation from average party positions would increase as political incentives decline. Conversely, we do not see any significant changes in the groups unaffected by term limits. As anticipated, the results remain consistent when party switchers are excluded from the sample. We plot linear predictions in Figure A10 in the Appendix.

Model (2) estimates are shown in Table A9. Our main coefficient of interest (β₃) is positive and statistically significant across all specifications, as expected. Here, we plot the marginal effects of term limits on deviation from the PPG average by trimester (Figure 2). We posited that the party leaders’ carrot-and-stick mechanism would become less effective for term-limited MPs as the end of the electoral period approaches due to the progressive reduction in office incentives, meaning that term-limited M5S members’ deviation from the party agenda in their speeches should increase over time. That is what our results suggest. Average deviation for the group of M5S affiliates who cannot run for office in the subsequent term increased throughout the 18^th period, compared to their non-term-limited peers. Because term-limited legislators already lack electoral incentives, this evidence is consistent with our hypothesized additional effect of the decline in office incentives on party unity. Furthermore, we can see that party unity peaks at the beginning of the electoral period, namely when MPs can still chase office incentives. By contrast, average deviation in the group of M5S legislators serving their first term and other PPGs decreased, suggesting that the salience of electoral incentives peaked as the election got closer (see linear predictions in Figure A11 in the Appendix). With respect to controls, we find that when PPGs experience a decrease in support in opinion polls, their members become more prone to deviating from the average party positions, in line with our broader theory about negative shocks to political incentives.

Figure 2.

Marginal Effect of Term Limits by Trimester (18^th Electoral Period).

Limitations, alternative explanations, and robustness checks

We now turn to limitations of our analysis and alternative explanations and present robustness checks and additional tests to support our claims. One limitation of our case study is that assignment to treatment is, of course, not random (or as-if random). MPs are not randomly assigned to the M5S; term limits affect all M5S MPs serving their second mandate; all M5S MPs are aware of these constraints when they choose to run for election or re-election in affiliation with the M5S in 2018. That said, this awareness does not automatically imply that term-limited legislators will not alter their behavior once they start their second term. We observe significant changes despite anticipation.

Ours is a difference-in-differences (DiD)-style approach; however, we deliberately avoid referring to it as such because of the DiD requirements. The key difference-in-differences identifying assumption is parallel trends: in the absence of treatment, the treated (term-limited M5S legislators) and the control group (non-M5S legislators) would have experienced similar changes in average deviation from the PPG average over time. However, since the Five Star Movement made its debut in parliament in 2013 and the term-limit rule was enforced from the outset, establishing parallel trends is not straightforward. We never observe (a subset of) M5S MPs serving their second mandate who can run for re-election under the Five Star Movement label, which would have given us a glimpse into the M5S deviation trend in the absence of treatment. Again, all M5S MPs are subject to term limits, and there has never been a time when the party was in parliament without the term-limit rule.

With this caveat in mind, we tested the difference in deviation between M5S MPs serving their first mandate in the 18^th electoral period and M5S MPs serving their second mandate in the 18^th electoral period (those who cannot run for re-election). These have the same PPG leaders and the same PPG average positions. The idea here would be that M5S MPs serving their first mandate in the 18^th electoral period behave more closely to how their colleagues serving their second mandate in the 18^th electoral period would have behaved, were their incentives not affected by term limits. We observe a statistically significant difference (p-value=.00), with the latter group (those who cannot run for office in the following term) deviating significantly more than the former group. This is not definitive evidence, but it is a positive sign regarding parallel trends. In any case, given the limitations of our case study and research design, we purposely tone down our language in discussing our results.

In terms of alternative explanations, the existence of term limits imposed by the M5S does overlap with the electoral reform approved in 2017. One potential challenge to our empirical investigation is that changes in legislators’ behavior induced by term limit restrictions might be confounded by changes in incentives stemming from the electoral reform. The reform had little impact on MPs in the 17^th electoral period, as it was approved in November and parliament was dissolved in December. As for the 18^th electoral period, MPs who cannot run for office are unlikely to significantly alter their behavior in response to changes in their election tier or potential changes in their electoral vulnerability resulting from the reform, precisely because they need not worry about re-election. It follows that, within our population, the electoral reform would primarily concern non-M5S legislators. We believe, however, that this does not pose a threat to our analysis. Our research design focuses on how the difference in parliamentarians’ behavior between the treatment group (term-limited MPs) and the control group (non-term-limited MPs) changes across two terms. This implies that elected officials from all parties were exposed to the same changes in electoral rules. Since our sample is restricted to MPs elected in 2013 and re-elected in 2018, any differential impact of candidate-selection procedures should be minimal: incumbents are typically re-selected through procedures that differ little across parties, regardless of the introduction of the new tier. Moreover, existing evidence indicates that previous electoral reforms had limited effects on legislative behavior in the Italian context (Viganò, 2024) and, in our results, we find no significant change in deviation in non-M5S legislators between the 17^th and 18^th electoral period (see Figure A10 in the Appendix). In our models, we control for members transitioning from the PR to the SMD and for candidate vulnerability, which might have changed in the aftermath of the reform, and do not find any meaningful differences. Finally, we replicated the model in Table 1, column 5 and excluded the new SMD tier from our sample (Table A5 in the Appendix). The results remain consistent.

As mentioned earlier, another concern is that the effect we observe might be at least partly driven by party switchers who align their speeches with their new or prospective PPG. As anticipated, we address this concern in multiple ways. First, we explicitly control for party switchers in our main model. The control is meant to capture the variation that is potentially induced by party switching. Second, in Table 1, we presented two models in which switchers were excluded from the sample altogether with consistent results. Third, we present models with an alternative treatment of party switching in Table A7 in the Appendix. In this case, we account for party switching by allowing multiple observations for MPs who changed party affiliation, estimating their positions solely on the basis of speeches delivered while they were affiliated with each respective parliamentary party group. The results are again consistent.

Finally, we run a series of robustness checks. In Table A6 in the Appendix, we present H1 models where random effects are replaced by legislator and PPG fixed effects. Fixed effects (FE) are not optimal in our case because we have only two time points per legislator. We also have one time-invariant variable (gender) and one that increases by one for all legislators (tenure). These are excluded from the FE analysis. The results are consistent with those displayed in Table 1. To attenuate concerns about covariate imbalance between the treatment and control groups, we explore balance in Figure A7 (H1) and Figure A9 in the Appendix. With respect to H1, two imbalanced covariates that stand out are age and tenure. This indicates that M5S MPs tend to be younger than those of other parties. And, of course, given that all M5S MPs served their first term ever in the 17^th electoral period, they all served two terms each, as opposed to non-M5S MPs who might be a lot more experienced. In Table A8, we thus included models that feature an interaction between age and time and an interaction between tenure and time (column 1). As an additional check, we also interacted all the variables whose absolute standardized mean difference exceeds 0.10 with time (column 2). The results remain consistent. As for H2, we have three imbalanced covariates (opposition, opinion polls, and PPG size balance). We included interactions between these three variables and time in column 3 in Table A9 in the Appendix. The results remain consistent once again. We kept these covariates in the main models for precision.

Conclusion

In this article, we examined how the loss of electoral and office incentives affects party unity as expressed in parliamentary speeches. Our findings provide strong support for the expectations derived from our theoretical framework, reinforcing the idea that legislators’ behavior is fundamentally shaped by the incentives provided by party leaders. When representatives can expect electoral or office-related rewards, they contribute to the party’s collective objectives by promoting party positions in their speeches. In doing so, they help advance the party’s goals of increasing electoral success or participation in government coalitions, both collective goods that directly benefit their individual MPs’ electoral and career ambitions. By contrast, when legislators can no longer benefit from these incentives, their motivation to promote average party positions diminishes. Consistent with our framework, they become more likely to prioritize their own interests and preferences and/or those of alternative principals in their parliamentary speeches.

Empirically, our analysis of term-limited M5S MPs in the Italian Chamber of Deputies provides support for our hypotheses. We show that legislative term limits are associated with changes in legislators’ behavior, manifested in an increased likelihood of agency loss in parliamentary speeches from one term to the next. Moreover, the risk of agency loss grows over time among legislators subject to term limits, as their opportunities to access higher legislative offices progressively diminish. By contrast, legislators who remain eligible for re-election increasingly align their speeches with the party agenda as the end of the legislative term approaches.

From a comparative perspective, the M5S term-limit rule represents a particularly strict application of legislative term limits, as it applies to all elected offices and is lifetime-based rather than limited to consecutive terms. After serving two terms, M5S members were barred from continuing a political career within the party, either as elected representatives or as holders of a governmental office at the regional, national, or European level. As a result, the effects identified in this study are potentially stronger than those that would arise under more permissive term-limit regimes. Future research could evaluate this expectation by examining cases characterized by less restrictive rules, such as those in place in several countries in the Americas and Southeast Asia.

While our analysis focuses on a single case study, the findings shed light on a broader mechanism of agency loss in parliamentary speech that extends beyond the Italian context and beyond cases characterized by legislative term limits. Similar dynamics are likely to emerge whenever parties lose the capacity to credibly offer career prospects to their members (Bøggild and Pedersen, 2024; Jastramskis et al., 2024). Finally, although our study shows that MPs are more likely to move away from average party positions in the absence of political incentives, it does not directly address where they move to as a result of this electoral disconnection. One possibility is that term-limited legislators prioritize their own interests and preferences by adopting a trustee-style mode of representation. Another, non-mutually exclusive, possibility is that they reorient their behavior toward alternative principals, such as other parties or organized interests. We encourage future research to further investigate these dynamics, for example, by collecting data on the post-parliamentary careers of former M5S MPs.

Supplemental material

Supplemental material - Reduced incentives, reduced party unity: Evidence from parliamentary speeches

Supplemental Material for Reduced incentives, reduced party unity: Evidence from parliamentary speeches by Francesco Bromo, Paride Carrara, Paolo Gambacciani, Edoardo A. Viganò in Party Politics.