Abstract
Recent years have seen growing scholarly attention to the role of intra-coalition partisan change. Yet, Shomer et al. (2022) critiqued the common definition of government termination that regards any partisan change as a government termination, arguing that it inflates the number of governments and distorts scholarly understanding of government duration—particularly in certain countries. This research note examines whether, and to what extent, adopting Shomer et al.’s (2022) modified definition alters substantive conclusions about the determinants of government duration and stability. We follow their approach, redefining government termination based on: (1) a new prime minister, (2) an election, or (3) a crucial partisan loss that alters majority status. Using this revised definition and applying parametric event history models that incorporate established determinants of duration, we find evidence that the conventional measure can bias conclusions about what drives government stability, especially with regards to the impact of ENGP and parliamentary polarization.
Shomer et al. (2022) critique the common definition of government termination, arguing that it inflates the number of governments and distorts understanding of government duration. They specifically challenge the criterion that any partisan change in government composition signals termination. This research note examines whether adopting their modified definition—under which termination occurs only with a new prime minister, an election, or a partisan loss that changes majority status—alters substantive conclusions about the determinants of government duration. Using parametric event history models, we find that the conventional measure can bias conclusions about government stability.
To evaluate whether government duration measurement matters for substantive conclusions, we first review the main determinants identified in the literature. The literature identifies key determinants of government duration, providing a framework for evaluating bias in standard measurement approaches. We briefly outline country-level and institutional, structural, and political factors—such as preferences, bargaining costs, and critical events—cited as influencing government duration (Bergmann et al., 2015).
We begin with institutional factors. Positive parliamentarism, shapes government longevity. Under this system, failure to secure an investiture vote increases early termination risk (Bergmann et al., 2015; King et al., 1990), while passing the vote is expected to enhance durability (Bergmann et al., 2022) 1 . Yet evidence is mixed: some studies affirm this effect (Rubabshi-Shitrit and Hasson 2022; Warwick 1992), while others refute it (Maoz and Somer-Topcu 2010). Moreover, Bergmann et al. (2022) find that investiture procedures heighten the influence of extreme parties on non-electoral terminations, and Martin (2018) argues that such votes may increase, rather than decrease, instability.
The type of no-confidence vote (NCV) influences government duration, with constructive NCVs (CVNCs) making it harder for the opposition to oust governments and thus promoting stability (Diermeier et al., 2003; Walther and Hellström 2022). Studies show governments last up to 82% longer under CVNCs (Rubabshi-Shitrit and Hason 2022; Walther and Hellström 2022).
Bergmann et al. (2022) contend that head-of-state strength influences government duration and find that semi-presidential systems have longer government tenures than parliamentary ones in Western Europe (WE), due to increased risks of non-electoral replacement. However, this effect is not evident in Central and Eastern Europe (CEE). Martin (2018) argues that direct presidential elections raise termination risks, though other studies (e.g., Tzelgov 2011; Walther and Hellström 2022) dispute this.
Region is the main structural determinant to affect government duration. While some studies control for it, others advocate for separate models for WE and CEE (Grotz and Weber 2012; Tzelgov 2011). For instance, maximum constitutional inter-election periods (CIEP) influence WE but not CEE, and semi-presidentialism reduces durability only in WE.
Beyond country-level institutional and structural traits, scholars have identified numerous party-system and government-level level political factors to affect government duration. Party system fragmentation (ENPP) is thought to increase instability by enabling alternative coalitions (Bergmann et al., 2015; Grotz and Weber 2012; King et al., 1990; Krauss 2018; Saalfeld 2008). Yet findings are mixed: some studies find ENPP increases risk (Walther and Hellström 2022), others find no effect (Martin 2018; Somer-Topcu and Williams 2008).
Cabinet size and the number of government parties (ENGP) are also linked to early termination (Bergmann et al., 2022; Somer-Topcu and Williams 2008; Walther and Hellström 2022), although some studies find no effect (Bergmann et al., 2015; Martin 2018; Rubabshi-Shitrit and Hason 2022). In CEE countries, Savage (2013) finds more parties reduce risk, while Tzelgov (2011) finds the opposite.
Caretaker governments are generally short-lived. Some scholars omit them (Bergmann et al., 2015; Savage 2013), while others include them, finding they increase termination risk (Martin 2018). Cabinet format also matters. Single-party governments, which lack coalition conflicts, are more stable than coalitions (Budge and Keman 1990; Grotz and Weber 2012; Lijphart 1984).
A related line of research focuses on cabinet type. Debate persists over which cabinet type promotes longevity. Though most agree majority governments are more stable (Diermeier and Merlo 2000; Walther and Hellström 2022; Warwick 1992), evidence is mixed on whether minimum winning coalitions (MWCs) outperform surplus coalitions. Some studies favor MWCs (Baltz 2022; Bergmann et al., 2015; Krauss 2018), others find no advantage (Lijphart 1999; Shomer et al., 2022). Still others argue that surplus coalitions may offer stability by tolerating defections (Volden and Carrubba 2004).
Parliament polarization may enhance stability by reducing viable coalition alternatives (Lupia and Strøm 1995; Maoz and Somer-Topcu 2010; Savage 2013) 2 . It is measured via seat-weighted ideological distances or ranges (Savage 2013; Walther and Hellström 2022). However, findings are mixed: Bergmann et al. (2015) and Walther and Hellström (2022) find no effect in WE or CEE; Walther et al. (2019) link polarization to early elections but not replacements, while Bergmann et al. (2022) suggest investiture votes mitigate its effects.
Government heterogeneity is thought to raise internal policy conflict and termination risk (Grotz and Weber 2012; Saalfeld 2011). Savage (2013) finds support, but others do not (Bergmann et al., 2015; Saalfeld 2011; Walther and Hellström 2022), though Bergmann et al. report a non-significant effect in the expected direction.
King et al. (1990) argue that longer formation periods indicate coalition difficulties and predict instability. Others disagree, suggesting such delays may not reflect conflict (Lupia and Strøm 2008; Savage 2013). Strøm (1985) finds longer negotiations may enhance duration by resolving early obstacles.
Lastly, critical events such as inflation and unemployment have been theorized to increase termination risk (Strøm et al., 1988), though findings are mixed. Several studies report that unemployment reduces government stability, including Bergmann et al. (2015), Saalfeld (2008), and Walther and Hellström (2022), while Krauss (2018) finds no such effect. In contrast, inflation is identified as a destabilizing factor by Somer-Topcu and Williams (2008) and Warwick and Easton (1992), though this is not supported by Walther et al. (2019) or Walther and Hellström (2022).
Research design
Against this backdrop, we now examine whether the choice of government definition alters conclusions about these determinants. We conduct a cross-national analysis using two datasets: the first adopts the conventional definition, terminating governments upon elections, a change in prime minister, or any partisan shift; the second applies Shomer et al.’s revised definition 3 . Both draw from ParlGov data (Döring et al., 2022). The sample consists of the 29 democracies included in ParlGov, covering the period from 1945 to 2014, or from the first year of democratization in cases that democratized later. These are parliamentary or semi-parliamentary systems, in which governments may end before the regular electoral term through no-confidence votes, coalition breakdown, or early elections. Presidential systems, by contrast, are not directly comparable because executives serve fixed terms.
The key explanatory variables are measured in the following way: positive parliamentarism is a dummy variable indicating whether governments require an investiture vote. Party system fragmentation (ENPP) and government party fragmentation (ENGP) are measured using Laakso and Taagepera’s (1979) index; under the modified definition, ENGP is averaged across all partisan compositions during a government’s tenure. Cabinet size (including the PM, excluding non-partisan ministers) is similarly averaged. Polarization is operationalized as the ideological range between the most extreme left and right parties in parliament, using ParlGov’s 0–10 scale 4 . Government heterogeneity reflects the ideological range among governing parties, averaged in the modified dataset. We control for caretaker status (1 = limited mandate cabinets; Döring et al., 2022), democracy level (Polity IV), and the constitutional interelection period (CIEP; Williams and Seki 2016).
Hierarchical Weibull model: Government duration using the conventional measure.
Hazard ratios (Standard errors in parentheses), *p < .1, **p < .05, ***p < 0.
Hierarchical Weibull model: Government duration using Shomer et al.’s (2022) modified measure.
Hazard ratios (Standard errors in parentheses), *p < .1, **p < .05, ***p < .01.
We estimate event history models using cabinet-level data from 29 democracies. To account for intra-country dependence, we employ multilevel random-intercept mixed-effects parametric survival models with a Weibull distribution. The main analysis assumes proportional hazards (PH); Table 3 in the appendix reports robustness using an accelerated failure time (AFT) model.
To aid interpretation, hazard ratios above 1 indicate a higher risk that a government will terminate at any given point in time, and therefore a shorter expected duration, whereas hazard ratios below 1 indicate a lower risk of termination and thus greater durability. A hazard ratio of 1 indicates no association. For example, a hazard ratio of 1.20 corresponds to roughly a 20% higher risk of termination, while a hazard ratio of 0.80 corresponds to roughly a 20% lower risk, holding other variables constant.
Analysis
We compare results from Table 1, based on the conventional definition, with Table 2, which uses Shomer et al.’s (2022) modified measure. Model 1 in both tables presents the baseline specification, while Models 2–9 introduce different covariates. Model 2 adds a control for semi-presidential systems; Model 3, for the seat share of the largest party; and Model 4, for the type of NCV. Model 5 includes regional controls, while Models 6 and 7 interact region with parliamentary polarization and government heterogeneity, respectively. Model 8 interacts polarization with investiture vote, and Model 9 excludes caretaker governments 5 .
The results in Table 1 align with key findings in the literature. A one-unit increase in parliamentary fragmentation (ENPP) raises the risk of termination by 10.7% (Model 9) to 21% (Model 3). Similarly, heterogeneous governments are more short-lived: a one-unit increase in heterogeneity raises termination risk by 8.9% (Model 1), holding other variables constant. Caretaker governments are less durable, and investiture votes increase breakdown risk (significant at the 0.1 level). The conventional definition supports the hypothesis that MWCs are more stable than other government types. Surplus and other coalition types are less durable, with hazard ratios above one (p < .05); for instance, surplus governments increase termination risk by 31.3% (Model 3) to 36.9% (Model 8). In line with previous mixed findings, we do not observe a significant effect of parliamentary polarization on government duration. Notably, our results on ENGP contrast with much of the literature: as ENGP increases, the hazard rate declines. We also find no evidence that the seat share of the largest governing party, NCV type, region, or its interaction with government heterogeneity influence instability. Additionally, the interaction between region and polarization yields overlapping confidence intervals, suggesting no significant regional differences at any level of polarization (see Figure 1, left panel). Marginal means from model 6 (region interact with polarization) for conventional and modified measures.
By contrast, Table 2 shows what changes once the revised definition proposed by Shomer et al. (2022) is used. Several factors continue to influence government longevity similarly to the conventional measure. Parliamentary fragmentation remains associated with shorter duration: a one-unit increase in ENPP raises termination risk by 8.7% (Model 1, p < .1). Government heterogeneity also increases instability, with a one-unit increase raising the risk by 6.7% to 12.1% (Models 4 and 7). Caretaker governments are significantly shorter-lived. Other variables, including cabinet size, polity, and CIEP, remain insignificant—consistent with prior results and those presented in Table 1.
However, using the modified definition alters our understanding of the determinants of government instability. Some variables lose significance once we adjust for the overestimation of instability in the conventional measure. Specifically, we find no evidence for the effect of positive parliamentarism in eight of the nine models. Likewise, ENGP no longer influences government stability, with consistent results across all models in Table 2. While Table 1 did not support an effect of parliamentary polarization on government duration, the modified measure analysis aligns with Maoz and Somer-Topcu (2010) and Lupia and Strøm (1995), suggesting that greater polarization prolongs government survival. Model 1 in Table 2 shows that a one-unit increase in polarization reduces termination risk by 7.6%, controlling for other variables. Given the literature’s mixed findings, we ran two interaction models. Model 6 allows polarization’s effect to vary by region (Savage 2013), and Model 8 tests whether investiture vote moderates this effect (Bergmann et al., 2022). Figure 1 presents marginal means from Model 6: the left panel uses the conventional measure (Table 1), and the right panel uses the modified one (Table 2). In both, polarization appears stabilizing in WE countries and destabilizing in CEE, but these differences are not statistically significant at the 95% confidence interval across the full polarization range.
Robustness checks
To assess whether these patterns are robust, we conducted several additional tests. First, we relaxed the proportional hazards assumption and estimated accelerated failure time models, addressing Warwick’s (1994) claim that hazards increase with a government’s age. Results, presented in Table 3 (appendix), are substantively consistent with earlier models. Second, we controlled for single-party majority governments, often considered more stable than coalitions. Table 4 shows that results remain substantively unchanged. Third, Table 5 (appendix) controls for bicameralism, but finds no evidence that it affects government duration (Walther and Hellström 2022), leaving our main conclusions unchanged. Fourth, Table 6 adds economic indicators (e.g., inflation, unemployment). Fifth, Table 7 introduces a four-category operationalization of government type and confirms that the bias identified persists: surplus governments are not less stable than MWCs. Log-transforming CIEP also did not alter results. Lastly, Table 8 presents a right-censoring permutation, redefining governments lasting beyond CIEP minus 30 days as non-failures. These results remain substantively consistent with previous findings.
In sum, across all robustness checks, we consistently find evidence of bias in inferences about the effect of government type on stability. The commonly held view that MWCs are more stable than surplus coalitions appears to stem from a government definition that overstates the number of terminations in certain countries. Treating every partisan change as a new government inflates instability and skews conclusions about the role of government type in explaining duration.
Conclusions
Taken together, the main analyses and robustness checks point to a common conclusion. Equating any change in a government’s partisan composition with its termination artificially inflates instability, particularly in countries already considered unstable under the conventional definition (Shomer et al., 2022). This paper tests whether this inflation biases our understanding of factors affecting government duration. Our findings indicate potential bias. While variables such as ENPP, government heterogeneity, seat share of the largest party, and caretaker status maintain their effect across both definitions, others do not. For example, ENGP appears to reduce instability under the conventional definition but has no effect using the modified measure. Similarly, positive parliamentarism loses significance, while parliamentary polarization, insignificant in the conventional analysis, emerges as a stabilizing force under the revised measure.
We further find no support for the prevailing view that MWCs are more stable than surplus coalitions. Across all nine models in Table 2 and multiple robustness checks, there is no significant difference in duration between the two. This discrepancy arises because the conventional measure classifies many surplus governments with nonconsequential partisan shifts as terminated, thus exaggerating their instability.
Of course, any partisan change can matter—particularly when stability is conceptualized as policy or personnel continuity. In such cases, Shomer et al.’s (2022) modified measure may not be suitable. However, if stability is defined as a government’s capacity to survive without real threats to its tenure, the conventional definition inflates instability and biases inference. The appropriate definition thus depends on the research question and conceptualization of (in)stability.
In an era of democratic backsliding, understanding what drives government instability is essential. Our study highlights biases in dominant scholarly conclusions and underscores the need for future research on the consequences of instability.
Supplemental material
Supplemental material - Determinants of government duration: Correcting for hyper inflating measures
Supplemental material for Determinants of government duration: Correcting for hyper inflating measures by Yael Shomer and Osnat Akirav in Party Politics
Footnotes
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the ISRAEL SCIENCE FOUNDATION (Grant No. 1543/19).
Supplemental material
Supplemental material for this article is available online.