Dying, surviving death, and reincarnating: differences in government replacements and their explanation

Abstract Not all governments survive until the next scheduled election. Some are replaced during their term in office by executives with a different party composition and/or portfolio distribution. Others are able to be ‘reborn’ as the successor government, undergoing only minimal changes. Such variation has to date received scant attention in studies on government durability. By classifying non-electoral replacements according to the degree of ministerial turnover, this article shows that new cabinets are often similar to their predecessors. It hypothesises that the likelihood of this pattern occurring is greater: when members of the current cabinet face bargaining problems in forming a very different cabinet, as in the case of surplus (unnecessary) parties in oversized coalitions; when the policy distance between the parliamentary median party and the current opposition widens; and when the executive’s economic performance discourages opposition parties from forming new coalitions with some incumbent parties. The risk of experiencing different types of replacement is estimated using data on Western European cabinets (1946–2021). Consistent with the hypotheses, the results indicate that governments are able to return to power almost untouched after their termination if they are oversized, if the opposition is far from the legislative median voter, and if inflation grows during a government’s tenure.

on 20 August 2019, the Conte I government was forced to resign by a no-confidence motion tabled by the League, a ruling party that a couple of months earlier had achieved an important electoral result in the European elections.A new executive, again headed by Giuseppe Conte, was installed two weeks later, with the Democratic Party replacing the League as a government partner of the Five Star Movement.Only one third of the ministries of the two Conte governments were distributed among the government parties in the same way.
The two events appear very different from each other, and they reveal different types of government replacement.Nevertheless, they would not be considered as distinct by many of the existing studies on cabinet duration and replacement.Extant works usually distinguish between government crises that lead to early elections and crises that lead to the replacement of one government by another within the same legislative term (Bergmann et al. 2022;Diermeier and Merlo 2000;Diermeier andStevenson 1999, 2000;Saalfeld 2008).To our knowledge, very few studies draw further distinctions among the latter cases in terms of the extent to which the new government is similar to the one it replaces (Damgaard 2008;Fernandes and Magalhães 2016;Schleiter and Evans 2022).Most of the literature assumes that the same explanatory variables are at work both when the people and the parties that controlled crucial policy areas of the old government are almost the same in the new one, and conversely when a government and its successor are very different.
This approach is potentially misleading.Certain factors may well predict political turbulence within a political alliance and a re-negotiation among the same parties that results in a new government very similar to the old one.However, these explanatory variables may be irrelevant when the fall of the government opens the way -in the same legislature -for the formation of an executive very different in terms of party composition and/or the distribution of ministries.Moreover, the decision to trigger new elections by bringing down a government is not necessarily explained by the same factors that account for a more or less important inter-electoral replacement.As Lupia and Strøm suggest (1995), these different 'deaths' correspond to different results in the expected utility calculations of the government parties when they have to decide about the demise and the continuation of the government to which they belong.
Distinguishing government termination through early elections from inter-electoral replacements, and then further differentiating among the latter according to the level of similarity between old and new government, is not just a conceptual refinement.Neither is it only important because they are different phenomena with potentially different explanations.Government durability has long been considered a significant factor in explaining crucial political phenomena, including public support for democratic regimes (Harmel and Robertson 1986), fiscal policies (Fortunato and Loftis 2018), and government effectiveness in policy implementation (Huber 1998).Explicitly distinguishing among types of government termination helps refine that received knowledge.Moreover, different types of government replacement have different implications for the functioning of a democracy.They are associated with different levels of policy stability, as well as with different ways in which the principal-agent chain of delegation and accountability reacts to possible or expected changes of voters' preferences in a parliamentary democracy.
As we show in the empirical section, the government replacements that can be observed most often correspond to what we call 'inter-electoral government replacements with high similarity': i.e. those replacements that install a new government which is very similar to its predecessor during the same legislature. 1This type of replacement prevails in some countries characterised by high levels of government instability.The key elements of the argument with which we seek to explain this empirical pattern concern the bargaining costs and the policy conflicts that prevent former government and opposition parties from forming a cabinet very different from the one that has just collapsed.We hypothesise that replacements with continuity between the government that terminates and its successor during the same legislature will be more likely under the following circumstances: when the governments include parties that are unnecessary to control a majority of seats in the parliament; when the median party in parliament is ideologically far from the opposition parties; and when the economic conditions are worsening.We use a comparative dataset on cabinets in 17 Western European countries during the period 1946-2021.
The article is organised as follows.The next section summarises the main explanations of different types of government replacement that can be found in the literature.The third section illustrates how different types of cabinet replacement are distributed across countries and decades.In the ensuing section we outline our theoretical argument and derive a set of testable hypotheses.In the fifth section we present the data and the research strategy we adopt to test them.The subsequent section presents the empirical results.The last section concludes by discussing the main implications of our findings.

Cabinet survival and non-electoral replacement of government
Investigating the different types of replacement that cabinets can experience between elections requires delving into the huge and ever-growing body of research on government stability and duration.Early studies on cabinet termination were split between two distinct approaches: those that focussed on 'structural attributes' -such as the characteristics of cabinets, parliaments and party systems -to explain the longevity of executives (e.g.Strøm 1985), and those that interpreted government termination as a consequence of random and unpredictable 'critical events' (e.g.Browne et al. 1984).These two approaches were merged into a 'unified model' of cabinet dissolution (King et al. 1990) which was later refined by Warwick (1994).Subsequently, Lupia and Strøm (1995) developed a game-theoretic model where cabinet dissolution was explained as a result of the bargaining between parties.Lupia and Strøm highlighted that party leaders can strategically choose between two different ways to terminate the current cabinet: either calling new elections or replacing the executive with a new one during the legislative term.By distinguishing between these two types of discretionary government termination, Diermeier andStevenson (1999, 2000) estimated distinct hazard rates and found different explanations for the risk of termination through early elections and for the risk of non-electoral termination.Thereafter, it became standard practice to study cabinet survival by analysing two competing risks: dissolution and replacement (Saalfeld 2008(Saalfeld , 2013;;Schleiter and Morgan-Jones 2009).In the meantime, a strand of the literature focussed on how institutions structure bargaining on government termination (Baron 1998;Druckman and Thies 2002;Strøm et al. 2003;Strøm and Swindle 2002).
Against this backdrop, very few studies have acknowledged and explored the possible variation in non-electoral replacements of governments.In his account of the forms that cabinet termination takes, Damgaard (2008) noted that not all new cabinets imply meaningful changes in the executive branch.In Western Europe , more than 20% of all new cabinets had exactly the same PM and party composition as the ones that they replaced.More recently, Fernandes and Magalhães (2016) and Schleiter and Evans (2022) have distinguished between two types of non-electoral replacement: those that imply a change in the party of the prime minister and those that do not.Using the competing risk framework, they argue that the two modes of replacement have different correlates.Fernandes and Magalhães (2016) focussed on European semi-presidential democracies and on the powers and partisan preferences of presidents.They found that the risk of non-electoral replacements involving changes in the cabinet leadership is higher when presidents enjoy the power to dismiss the government, while the risk of 'continuity' replacements increases when presidents lack dismissal powers and there is cohabitation between presidents and cabinets.In their analysis of government replacement risks in 20 European democracies, Schleiter and Evans (2022) emphasised that institutional instruments such as the power to call a vote of confidence can be used by PMs to manage the heterogeneous risks of government termination.They found that greater confidence powers reduce the risk of replacements that result in the ousting of the PM's party.Building on these studies, in this article we use a more refined measure of the degree of similarity between a government and its successor and analyse different risks of non-electoral government replacement.

Different ways to die (often never completely)
In Western Europe, the different types of inter-electoral government replacement are not rare and not necessarily less frequent than government collapses that lead to early elections.Even more importantly, they are not equally distributed over time and among countries.This can be seen by examining the different modes of discretionary cabinet termination in Western Europe since the end of World War II (see below for our country selection).A first type of discretionary termination is when cabinets end because early elections are called.A second and third type of discretionary termination occur when cabinets are replaced during the legislative term.
In particular, we define as 'high similarity' replacements those inter-electoral replacements where the new cabinet resembles the previous one by more than 50% in terms of how ministerial portfolios are distributed among the governing parties.Conversely, we define as 'low similarity' replacements those where the new cabinet resembles its predecessor by at most 50% in terms of the way in which portfolios are allocated among parties. 2 Generally speaking, the data shown in Figure 1 indicate that replacements that occur between elections -either with low or high similarity between a cabinet and its successor -are a majority in almost all the decades (the 1980s being the sole exception) and in half of the countries we consider.
In particular, the upper panel of Figure 1 displays the average number of discretionary government terminations -replacements with low similarity, replacements with high similarity, early elections -per 1000 days, by decade.We note that the inter-electoral replacements with high similarity have almost always constituted the largest portion of inter-electoral government replacements, with the partial exception of the 1990s.In the last two decades, high similarity replacements have even become the most frequent mode of discretionary termination, outstripping terminations driven by early elections.
The lower panel reports the number of discretionary government terminations per 1000 days, by country.The graph shows that some (albeit not all) of the countries with the highest level of government instability -Italy, Finland, France, and to a lesser extent Belgium -are characterised by new governments that are very similar to their predecessors.These are also countries where government replacements through early elections are quite rare.By contrast, countries like Denmark and Greece quite often experience government terminations, but mostly through early elections.The same is true for other countries with lower levels of government instability, such as Portugal, Spain and the United Kingdom.

A theoretical framework for government replacement analysis
A government crisis can be understood as the disruption of an equilibrium brought about by changes in the political environment that characterised the birth of the incumbent government.For at least one crucial actor, these changes can make preservation of the current government less convenient than alternative scenarios such as early elections or an inter-electoral government replacement.The equilibrium break does not depend only on the worsening of the government payoffs -in terms of policies and offices -for at least one of the ruling parties; it also depends on the expected payoffs associated with the alternative scenarios (Lupia and Strøm 1995).Therefore, a shock that lowers the current government payoffs will not necessarily cause a new equilibrium that is very different from the old one.This is because the same shock may also make the other possible outcomes less attractive and more costly.
The sequence of events that can lead to different types of discretionary termination of governments can be represented by means of a decisional tree (see Figure 2).In order to observe the (discretionary) termination of the current executive, one must first suppose that at least one government actor decides to break the alliance with the other government partners.If a majority in the parliament wants early elections and is sufficient to achieve them, then the current parliament will be dissolved and the new government will be accountable to a new parliament.If, on the contrary, there is no parliamentary majority supporting early elections, then the type of government replacement that will occur depends mostly on the behaviour of at least one opposition party and some (at least one) government parties. 3In our stylised account of cabinet termination, we distinguish between two types of inter-electoral replacement.One is when the new government includes at least one opposition party (or is supported by a parliamentary majority including at least one opposition party) and thus presents a distribution of ministerial offices among parties that is rather different from that of the previous cabinet.We call this a 'low similarity replacement' .The other type of inter-electoral replacement is when the new government is similar to its predecessor in terms of portfolio distribution, because no opposition party has joined the previous cabinet parties or the parliamentary majority.We call this a 'high similarity replacement' .
What makes the above-described behaviour of parties more or less likely?How can we account for different types of inter-electoral replacement?In what follows, we primarily focus on the factors that can explain why cabinets often terminate and are replaced by a very similar executive during the same legislative term.In particular, we hypothesise that patterns of replacement with high similarity between the outgoing and incoming cabinet are more likely to occur under three conditions.
A first factor concerns the type of government that is in office when the game starts.For a new government to form during the same legislative term, key actors inside and outside the incumbent cabinet must have enough seats (a majority) to support a new government in parliament, as well as the will to do so.In this regard, distinct types of cabinet offer different incentives to their members.When the government still in office is a minimal winning one, any party that is a member of the coalition can, at least in theory, build a new government with the opposition parties.In other terms, in a minimal winning government, each coalition party is pivotal in the current government for preserving a majority of seats in parliament, and at the same time it is sufficient for the creation of a new majority with the opposition parties.Each member of a minimal winning cabinet does not necessarily need to reach an agreement with other parties of the same government to create a new government together with the opposition.This would imply that, when a minimal winning cabinet collapses and is replaced during the legislative term, the new government is unlikely to resemble the previous one.
The transaction costs associated with forming a new cabinet with the opposition are much higher for the members of oversized coalitions.This is because, in an oversized government, each surplus (unnecessary) party that wants to build a new government coalition must reach an agreement not only with the opposition, but also with at least one other party of the same government (see Carrubba andVolden 2000, 2004 for a similar argument).Moreover, when an 'unnecessary' party leaves a government coalition, its ministerial spoils could be redistributed among the remaining government parties.Each of the latter can gain something in terms of ministerial offices.Of course, also the government party(ies) with which the 'leaver' party wants to create a new government (together with the opposition) may gain something if the leaver actually leaves the cabinet.From the point of view of this party (these parties), the uncertain benefits of leaving the other government parties can be lower than the certain and increased benefit of remaining with them.In an oversized government, the 'unnecessary' government members are therefore trapped in a prisoner dilemma and the transaction costs they would pay to form a new government with the opposition can be very high.Because their threats are not very credible, the bargaining power they can deploy -not only to obtain a different and more convenient distribution of ministerial offices, but also to preserve what they have already obtained -is very weak.This weakness makes them easy victims of conflicts arising within the government, even when they are not directly involved.A conflict between the 'necessary' parties can be resolved by offloading the costs of pacification on one 'unnecessary' party, for example through a redistribution of ministerial spoils.Whether this redistribution is accepted or whether, on the contrary, it induces the 'unnecessary' party to leave the government, the outcome will still be a new government very similar to the previous one precisely because of the low weight of the 'sacrificial victim' .Hence, we expect that, when an oversized cabinet terminates and is replaced during the legislative term, the new government is likely to resemble its predecessor.We thus put forward the following hypothesis: Hypothesis 1: Compared to minimal winning governments, oversized governments have a higher risk of being replaced by a government that is very similar. 4  A second factor that can make inter-electoral replacements with high similarity between the outgoing cabinet and the next one particularly likely relates to key actors' anticipation of policy making under a possible new government.Obviously, governments are not only a set of ministerial offices to be distributed.They are also associated with policies to be implemented.As is well known from comparative studies on multi-party governments (e.g.Tsebelis 2002), conflict within the government can prevent efficient policy making.In this respect, the policy distance between the main actors that can form a new government, different from the current one, is crucial because it makes it possible to predict the probability and the strength of intra-cabinet conflict in the future.If this distance is too great, then the prospect of a new government becomes much less attractive to its potential members.For example, all else being equal, a large opposition party will probably refrain from entering a new cabinet during the legislative term if it is far in policy terms from the parliamentary median voter (the latter being crucial for passing legislation).In these circumstances, when the incumbent cabinet terminates and is replaced, the new government is likely to resemble its predecessor.Therefore we hypothesise that: Hypothesis 2: An increase in the policy distance between the parliamentary median party and the current opposition increases the risk of an inter-electoral government replacement with high similarity.The executive's economic track record is a third factor that can increase the risk of an inter-electoral replacement where the new cabinet is very similar to its predecessor.This is related to the fact that voters are known to care about economic trends.From the point of view of opposition parties, the worsening of a cabinet's economic performance makes the continuation value of the legislature under a new government lower than the expected benefits from early elections.This may even hold for opposition parties that are offered cabinet membership.The opposite is true for the incumbent parties: these would not benefit much from early elections, but need to convey a message of change to address the difficult situation.Moreover, opposition parties will find it very risky to share government responsibilities with some former government parties during an economic crisis.These conflicting preferences about the continuation of the legislature and government membership make the following hypothesis plausible: Hypothesis 3: A worsening of the government's economic performance will increase the risk of an inter-electoral replacement with high similarity.

Data and methods
In order to test the hypotheses put forward in the previous section, we used a dataset that includes information on 17 Western European democracies: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden and the United Kingdom.Although the data encompass the whole post-war period until 2021, the timespan covered for some countries is shorter due to the limited availability of information on cabinet composition (data start in 1966 for France) or due to their more recent democratic history (data start in 1977 for Greece, in 1980 for Portugal, and in 1979 for Spain).
The unit of analysis in our dataset is the cabinet.A new cabinet was recorded when any of the following events occurred: a change in PM, a change in the set of parties holding cabinet portfolios, or a new general election.We considered as cabinet members only those parties that had designated representatives with voting rights in the council of ministers (Müller et al. 2008).
Because our purpose is to explore how different factors influence distinct types of inter-electoral cabinet replacement, technocratic cabinets were dropped from our dataset.As a result, the total number of observations in our dataset is 511.Roughly 68% of these cabinets (347) terminated for discretionary reasons: 139 cabinets terminated because early elections were held, and 208 cabinets fell and were replaced by a new cabinet without new elections.Among the latter, 152 cabinets (i.e. more than those terminated through early elections) were replaced by a very similar cabinet, while 56 were replaced by a very dissimilar one.The remaining 32% of observations in the dataset correspond to cabinets that terminated for technical reasons (i.e.constitutionally mandated elections or death of the PM).
The dependent variable in our analyses is cabinet duration, measured as the number of days between the start and the end of a cabinet.Adopting a widely-accepted conceptualisation of cabinet duration (Müller and Strøm 2000;Strøm et al. 2003Strøm et al. , 2008)), the first day in a cabinet's life was defined according to official criteria found in a country's constitution: the date when the PM or cabinet was appointed by the head of state or the date of the formal vote of investiture.A cabinet was considered to terminate when the PM or the cabinet formally resigned, when general elections were held, or when a change took place in the party composition of the cabinet or the identity of the PM.
We analysed duration in the case of different types of cabinet termination using Cox's partial likelihood survival regressions.The central element in survival analysis is the hazard rate h(t), which is the probability that an event occurs at a particular point in time, conditional on the fact that it has not yet occurred.Here, government termination is the event of interest.The hazard rate has two components.The first one is the set of covariates that are hypothesised to affect systematically the timing of an event.The second one is the baseline hazard rate h 0 (t), which indicates the underlying probability of the event to occur over time when the vector of all covariates is zero.
We distinguished between two types of inter-electoral government replacement and employed a latent survivor time approach to competing risks.Our classification of replacements was based on a comparison between the composition of the cabinet that terminated (i.e. the observation in the dataset) and the composition of the cabinet that replaced it after its termination.Such a comparison was made at the portfolio level.We compared the distribution of parties across portfolios in the government with the distribution of parties across portfolios in the following government. 5Information regarding the allocation of cabinet portfolios among parties in each government was taken from the Who Governs dataset (Casal Bértoa 2021).We computed a similarity index ranging from 0 (when no party controlled the same portfolio in the current and following government) to 1 (when the same parties held the same portfolios in the current and following government).We then created two categories: those governments which, compared to the ones that followed them, had a similarity index between 0 and 0.5 (low similarity replacements), and those governments which, compared to the ones that followed them, had a similarity index higher than 0.5 (high similarity replacements). 6 Figure 3 plots non-parametric Kaplan-Meier estimates for the survivor functions of the cabinets that terminate and are replaced with a very dissimilar or a similar government.The curves -which indicate the unconditional probability that a cabinet will survive beyond a given time point (without including any covariates) -show that the two risks of termination differ especially during the initial three years of a cabinet's life. 7 A crucial part of survival analysis is the censoring regime adopted.Following previous studies (Diermeier andStevenson 1999, 2000), in our competing risk analysis of government duration, events other than those of theoretical interest were assumed to be randomly censored.In particular, we adopted the censoring strategy that follows.(a) When estimating the general risk of discretionary termination, the right-censored observations were the records of all cabinets terminated for technical reasons (regular elections or death of the PM).(b) When estimating the risk of early elections, we right-censored the records of cabinets terminated for technical reasons and those that failed due to an inter-electoral replacement.(c) When estimating the risk of cabinet replacement, we right-censored the records of cabinets terminated for technical reasons, and those terminated by early elections.In addition, we estimated separate models for distinct types of non-electoral transfers of power.More precisely, (d) for the estimation of the risk of cabinet replacement by a very different government (low similarity), we right-censored the records of cabinets terminated for technical reasons, those terminated through early elections, and those terminated due to replacement by a very similar government.Finally, (e) to estimate the risk of cabinet replacement by a very similar government (high similarity), we right-censored the records of cabinets terminated for technical reasons, those terminated by early elections and those terminated due to replacement by a very different government. 8 Having discussed our regression strategy, we now describe the operationalisation of the variables included in our analyses as covariates.Most of the variables used were taken from the European Representative Democracy Data Archive/Party Government in Europe Database (ERDDA/ PAGED) (Andersson et al. 2014;Bergman et al. 2021).One of these variables is the type of government.In line with our first hypothesis, governments were classified into three categories: minimal winning, minority and oversized.We included the two dummies 'minority' and 'oversized' in the analysis, with minimal winning governments as the omitted reference category.We expect to find that the risk of inter-electoral government replacement with high similarity is greater with oversized governments than with minimal winning ones.
Following our second hypothesis, we calculated the policy distance between the position of the median party in parliament and the biggest opposition party ('median-opposition distance').The party positions in the policy space were taken from the ParlGov database (Döring and Manow 2021), where the policy scores of parties on a 0-10 scale are inferred from standardised expert surveys data. 9We computed the distance between the parliamentary median and the largest opposition party in two ways: in a one-dimensional policy space (i.e.considering party positions on the general left-right scale) and in a bidimensional policy space (i.e.considering jointly the dimensions 'State-Market' and 'Liberty-Authority'). 10 Our expectation is that, as this policy distance increases, so too does the risk of inter-electoral government replacement with high similarity.
According to our third hypothesis, worsening economic circumstances increase the risk of inter-electoral government replacement with high similarity.Consistently with existing studies on the impact of the economy on cabinet stability (e.g.Pinto 2018;Robertson 1983a;Saalfeld 2008Saalfeld , 2013;;Warwick 1994), we used the rates of unemployment and inflation as key economic indicators.In particular, following Hellström and Walther (2019) we calculated the difference between the inflation rate in the year when the government ended and the inflation rate in the year when the government started ('inflation difference').Similarly, 'unemployment difference' is the difference between the unemployment rate in the year when the government ended and the unemployment rate in the year when the government started. 11 In accordance with several comparative studies on cabinet survival, our models incorporated as control variables a number of cabinet attributes, characteristics of the bargaining context in the parliament, and measures that captured the institutional features of the countries considered.Most of these factors have been proposed to explain government stability in general or government termination through early elections, since the literature has rarely addressed the differences between distinct types of inter-electoral government replacement.
Starting from the attributes of cabinets, we included 'cabinet core' .We calculated this variable both in one dimension (as the absolute distance between the two most extreme cabinet parties on the general left-right scale), and in a bi-dimensional policy space (as the average of the two absolute distances between the two most extreme cabinet parties on the dimensions 'State-Market' and 'Liberty-Authority'; see Curini 2011;Tsebelis and Chang 2004).An increase of the cabinet core should increase the general risk of government replacement (Tsebelis 2002;Warwick 1994).As suggested by Lupia and Strøm (1995), in order to consider the continuation value of the legislature we calculated a logged function of the so-called maximum duration ('logmaxdur').The latter corresponds to the maximum possible cabinet duration (in days) and it was obtained as the difference between the last day of the legislative term (according to the constitution) and the first day in a cabinet's life.The expected finding is that as the maximum duration of a government decreases, so the risk of government termination through early elections increases.Following Schleiter and Morgan-Jones (2009), we also expect to find that non-electoral replacements do not become more likely as a government's term elapses.
Concerning the properties of the parliamentary bargaining environment that can influence government survival, we incorporated the effective number of parliamentary parties in the lower chamber ('ENPP') -calculated using Laakso and Taagepera (1979) formula -as a measure of parliament's fractionalisation (Laver and Schofield 1990).The expected finding is that higher levels of fragmentation increase the general risk of government replacement. 12 With regard to country-level institutional features, democracies characterised by bicameralism and semi-presidentialism are often supposed to have further veto players in addition to the cabinet parties.The enlargement of the de facto core of the legislative veto players (Diermeier et al. 2007;Druckman and Thies 2002;Schleiter and Morgan-Jones 2009) is supposed to lower the policy payoffs associated with governing and to increase the risk of general government termination.Our analyses incorporated a dummy for bicameralism that is equal to 1 for cabinets in countries featuring an upper parliamentary house able to impact on legislative decision making.In the ERDDA/PAGED dataset, France, Finland until 2000, Greece until 1985 and Portugal until 1982 are coded as semi-presidential.Accordingly, we included a dummy for semi-presidentialism, which equals 1 for those cabinets in the abovementioned countries/periods.
The requirement of an investiture vote to be passed before a cabinet takes office is expected to make executives relatively durable, because the ruling parties have publicly committed to the government (Bergmann et al. 2022).In addition, by enhancing the visibility of the parliamentary support for a government, the existence of an explicit investiture vote can raise for the opposition parties the political cost of supporting a new government together with parties that previously have been opponents.Therefore, in the presence of positive parliamentarism (a dummy in our analysis), one should expect not only a low risk of cabinet dissolution in general, but also a low risk of inter-electoral government replacement with low similarity.
When, in order to dismiss a government, a parliamentary majority has to vote in favour of an alternative cabinet (constructive no-confidence vote), then the risk of any type of termination should decrease (Bergmann et al. 2022;Diermeier et al. 2003;Lento and Hazan 2022).The government can still stay in office without majority support in parliament by taking advantage of the ideological divisions among the opposition parties.Accordingly, we created a dummy which equals 1 in Belgium since 1995, Germany and Spain ('constructive no confidence').
When the country's constitution leaves the final decision about new elections to the Prime Minister, cabinet termination through early election can be less difficult (Kayser 2005;Schleiter and Morgan-Jones 2009;Strøm and Swindle 2002).Moreover, during the bargaining to form a new government, the PM's dissolution powers can strengthen the position of her/ his party by discouraging the formation of governments too different from their predecessors (Schleiter and Evans 2022).Therefore, when a premier enjoys dissolution power (included as a dummy in our analyses), one should expect a lower risk of replacement with low similarity.
Finally, all the models were estimated by employing shared frailties to control for unobserved characteristics at the country level.The latter may systematically affect the duration of cabinets.Table 1 provides descriptive statistics for the variables included in our analysis.

Results
Table 2 reports the results of the Cox regressions we estimated to subject our hypotheses to a multivariate test.In each model, the coefficients are expressed as hazard ratios, with standard errors in parentheses.Hazard ratios indicate the multiplicative effect that a one-unit change in a covariate has on the baseline hazard.A hazard ratio larger than 1 implies that a covariate increases the failure risk of governments (i.e.reduces cabinet duration), while a hazard ratio smaller than 1 corresponds to a lower failure risk (i.e.longer life of cabinets).A hazard ratio equal to 1 suggests no effect.In models 1-5, policy-related variables ('median-opposition distance' and 'cabinet core') were calculated considering the parties' positions on the left-right axis.In models 6-10, the same variables were computed in a bi-dimensional space as discussed above.Generally speaking, our three hypotheses are confirmed, irrespective of the number of policy dimensions taken into consideration.The hazard of inter-electoral government replacement with high similarity is about 2.7 times higher when the current government is oversized rather than minimal winning (H1).In this regard, Figure 4 (right panel) plots the probability that oversized and minimal winning cabinets survive the risk of being replaced with a very similar executive.Since the early days of their lives, oversized cabinets are less likely to survive the risk of being replaced by a similar executive, and the difference with respect to minimal winning cabinets increases over time.The same does not happen with the risk of being replaced by a very dissimilar government (left panel), as the survival functions of the two types of cabinets almost overlap.
With regard to H2, for a one-unit increase in 'median-opposition distance' we observe an increase in the hazard of inter-electoral government replacement with high similarity of between 22% (left-right dimension) and 12% (multidimensional space).The effects of the economic performance of the cabinets are less clear-cut, as H3 is supported in the case of inflation but not in the case of unemployment.More precisely, growth of the inflation rate during a government's tenure significantly increases its risk of being replaced with a very similar government in the same legislative term.Changes in the level of unemployment during a government's term in office are instead not associated with its premature dissolution through a non-electoral replacement.The growth of unemployment seems to increase only the risk of early elections (models 2 and 7).These findings resonate with those of Hellström and Walther (2019), who find that -unlike inflation growth -unemployment growth does not affect the risk of non-electoral government replacement. 13 The variables behind our hypotheses significantly affect only the specific type of government replacement for which the hypotheses were proposed.This implies that a generic approach to inter-electoral replacement as a single pooled phenomenon would fail to identify these specific mechanisms.As regards the other variables, we note that only minority status systematically affects all types of government termination.Compared to majority cabinets, minority cabinets face increased hazards of any kind, including inter-electoral replacement with a very similar cabinet.The rest of the control variables differently affect the distinct types of government replacement and none plays a role in explaining the inter-electoral government replacements with high similarity.In line with the literature's predictions (Lupia and Strøm 1995;Saalfeld 2008;Schleiter and Morgan-Jones 2009), the maximum cabinet duration is significantly and negatively associated with the risk of early elections (models 2 and 7).The same applies to the constructive vote of no confidence, an institutional feature that is able to decrease the risk of early elections but has no effect on any other type of discretionary termination.Countries characterised by semi-presidentialism seem to suffer a higher risk of inter-electoral government replacement with low similarity (models 4 and 9). 14 No other institutional features seem to play a systematic role. 15Interestingly, the level of fractionalisation in the parliament -which has been indicated in the literature as a general source of government instability -is associated positively and significantly only with the inter-electoral government replacements with low similarity.This seems to suggest that, when complexity and fragmentation are particularly high in parliament, there are several possible opposition parties that can be included in the cabinet when the incumbent one collapses.Our finding is consistent with those of existing works on different types of replacement (Fernandes and Magalhães 2016;Schleiter and Evans 2022), where parliamentary fragmentation turns out to increase the risk that cabinets terminate in a replacement of the PM's party.
Our analysis of different types of cabinet terminations relies upon comparison among pairs of successive cabinets, classified as similar or dissimilar according to how ministerial positions are distributed across coalition parties.The distinction we propose between similar and dissimilar cabinets raises a number of issues and certainly requires some robustness tests.First, not all ministerial portfolios are equally important.As a robustness check, we re-calculated our index of similarity by focussing on a limited set of key portfolios (Prime Minister, Finance, Labour and Welfare, Foreign Affairs, Internal Affairs).The results were very similar to those shown in Table 2 (see Online appendix F).
Second, it can be argued that the most substantial change between a government and its successor occurs especially when the PM's party changes.Indeed, the few existing studies that distinguish between different types of non-electoral replacements employ a classification based on whether there is a change or not in the PM's party (Fernandes and Magalhães 2016;Schleiter and Evans 2022).Conceptually, the classification we propose is more precise because it measures dissimilarity by considering the entire portfolio distribution, and not just which party controls the prime ministership.In empirical terms, the overlap between the two classifications is large, but not complete.If we focus on the 208 inter-electoral replacements in our sample, we note that 28 of them are coded differently in the two classification schemes.In particular, 20 replacements that occurred with no change in the PM's party -and which would hence be coded as 'continuity replacements' by Fernandes and Magalhães -in fact involved a remarkable change, since more than 50% of portfolios were re-distributed among government parties.Conversely, eight replacements involved a change in the PM's party -and hence would be coded as 'change replacements' -but were characterised by close similarity between the outgoing government and its successor.We can thus argue that our index improves the classification of replacements in 14% of the cases in our sample.We would add that our findings remain almost unaltered if we employ a classification of replacements based on changes in the PM's party (see Online appendix G).
Third, although our index of similarity is based on portfolio allocation among parties, several cabinets in certain countries feature a number of non-partisan ministers -a phenomenon that seems to increase over time in countries like Italy (Cotta 2018;Zucchini and Pedrazzani 2021).When calculating our index, a portfolio that was controlled by a given party in the current government and then assigned to a non-partisan minister in the next one (and vice versa) was coded as a change, while a portfolio that was controlled by a non-partisan figure in the two governments was coded as no change.To check the robustness of our findings, we again calculated the index of similarity, by considering non-partisan ministers as members of the PM's party.This new coding slightly increased the number of high similarity replacements from 152 to 157 in our sample.Our regression results remain the same if we re-run the analyses considering non-partisan ministers as members of the PM's party (see Online appendix H).

Conclusions
The dynamics leading to cabinet replacement can vary a great deal and produce very different outcomes.During the legislative term, certain cabinets are replaced by others with a different party composition and/ or with a very different distribution of ministerial portfolios among parties.Other cabinets, on average the majority, are instead able to be 'reborn' , undergoing only minimal changes.The literature on government termination has extensively investigated the distinction between inter-electoral replacements and replacements through early elections.Less attention has been paid to analysing the different types of cabinet replacement that occur between elections.In an attempt to fill this gap we classified different types of inter-electoral government replacement, from no or minimal turnover (i.e. a government returns almost untouched after its termination) to a complete change (a government is displaced by a different set of parties taking office in ministries previously 'handled' by other parties).We used competing-risk survival analysis to estimate the risk of these different types of government replacement.
We have proposed and tested a set of hypotheses to explain the risk that governments are replaced by other executives which are very similar.These hypotheses share as crucial explanatory factors the bargaining costs and the potential policy conflicts that prevent former government and opposition parties forming a new cabinet together during the same legislature.The nature of the government that is going to collapse can increase the transaction costs of some of its members if they try to create an alternative government with opposition parties (H1).Moreover, the expectation that the new cabinet will be internally heterogeneous can prevent an extensive change of government (H2).Last but not least, the attractiveness for opposition parties of becoming government members without new elections cannot be taken for granted, above all when economic conditions worsen (H3).
Our findings have some interesting implications for further research on government formation and replacement, and more in general for the study of policy change in democracies.
The confirmation of H1 sheds new light on the rationale of oversized governments, a non-trivial phenomenon that has generated a number of models and hypotheses.One of the most convincing explanations relies upon the difficulties encountered by 'unnecessary' coalition partners in reaching agreement with other government parties to break the coalition.The main party in a multi-party government proto-coalition would build up an oversized government with unnecessary parties precisely in order to increase the government's stability.We show that such an effect on stability is even stronger.When the government that collapses is oversized, substantial stability from one government to its successor is usually achieved through a very similar distribution of ministerial offices among the parties.Related to this, our findings suggest reconsidering at least in part the idea that minimal winning cabinets are the most 'durable' ones (e.g.Robertson 1983b).Minimal winning cabinets are less subject than other executives to the risk of being replaced by a very similar cabinet during the legislative term.However, minimal winning cabinets are no less vulnerable than oversized ones to the risk of being replaced by a very different executive.Assuming that negotiators can choose attributes like the type of government, as well as anticipate the duration of the government while bargaining over government formation (Diermeier and Merlo 2000), one can argue that coalition builders can pursue government stability in two ways: either by forming a minimal winning cabinet, or by building an oversized coalition that during the legislative term will probably be replaced by a very similar cabinet.
Our distinction between government replacement types (with low and high similarity) is supported by the results of the empirical analysis.Except for the variable capturing the minority status of cabinets, the covariates that appear significant for the general risk of inter-electoral government replacement are never significant for both the components of such a risk, but only for one of the two.Variables that are explanatory factors of the risk of early elections almost never are of the other types of government replacement.They are definitely different phenomena.Counting the number of government replacements is not necessarily the best way to assess the level of political instability unless one takes into account the rate of similarity between one government and its successor.The level of government instability of some countries should be considerably scaled back.Moreover, also the role played by some variables as sources of instability needs a new evaluation.Ideological polarisation, which we measured as the distance between the parliamentary median position and the position of the largest opposition party, increases the risk of a government being replaced -but with another government very similar to the previous one.On the contrary, fractionalisation in parliament, which we operationalised as the number of effective parties, seems to be an overlooked source of policy instability.It significantly and positively affects the risk of replacement with governments that are quite different.
Finally, a likely fall in the public support for the government cancounterintuitively -cause substantial political stability.An increase in the inflation rate is associated with an increase of the risk of inter-electoral government replacement with high similarity, instead of inducing more substantial changes in the government's composition.In these circumstances, demands for a substantial change in the executive branch may arise from society.However, these demands seem to be almost ignored at the institutional level, because the interaction among the political actors often leads to the formation of new cabinets that are very similar to the old ones.

Notes
1.The fact that governments frequently resemble their immediate predecessors is well-known in the literature on government formation and can be traced back to the so-called 'incumbency advantage' (Martin and Stevenson 2010).
A number of studies also highlight that coalition bargaining often takes place among those parties that have closer affinities and prior experience of governing together, rather than among all parties in parliament (Tavits 2008).2. The way in which we measured the degree of similarity between pairs of cabinets is illustrated in the empirical section below.3.For ease of exposition, here we do not consider the different rules that govern parliament dissolution.For a comprehensive account of the institutions regulating assembly dissolution and their impact on cabinet termination, see Schleiter and Morgan-Jones (2009); Strøm and Swindle (2002).See also Goplerud and Schleiter (2016).4. We also expect that a government's risk of being replaced by a very similar government increases as the proportion of surplus (unnecessary) parties in the government grows.In ancillary analyses, we tested and found support for this hypothesis (see the Online appendix B). 5. A portfolio that was controlled by the same party both in the current government and in the government that replaced it was coded as 1.A portfolio that was controlled by party A in the current government and by party B in the next one was coded as 0. We then calculated the average of these scores across all portfolios.We did the same for each couple of successive cabinets in our dataset.To compare pairs of portfolio distributions in a meaningful way, we matched the portfolios in the current and following government.For example, if the current government had an 'Environment and Energy' portfolio while the following government had two distinct portfolios -one for 'Environment' and another for 'Energy' -we duplicated the 'Environment and Energy' portfolio in the current government.Regarding the political dynamics of portfolio re-design, which we do not address in this article, see Indridason and Kam (2008), Meyer et al. (2023), andSieberer et al. (2021).6.Even if plausible, the 0.5 threshold remains arbitrary.To check for the robustness of our findings, we re-ran all the regression models by considering thresholds of any value in a range from 0.4 to 0.7.Estimated coefficients were quite robust to changes in the threshold (see Online appendix A).We also acknowledge that using the 0.5 threshold to create two categories implies wasting variation in the data.However, the density distribution of the similarity index is quite bimodal: almost 90% of cases are lower than 0.4 or higher than 0.6 (Figure A1).
7. A log-rank test indicates that, up until 1000 days, the probability that the two functions are the same is p < 0.05.8.Because we are interested in explaining different types of cabinet replacements, in this study we modelled our dependent variable as government duration and employed survival analysis.In ancillary analyses, for robustness we addressed a slightly different research question: Why are some governments more similar to their successor than others?More precisely, we focused on inter-electoral replacements and used fractional logit regression, with the similarity index as dependent variable.Despite the limited number of observations (N = 208), some results were consistent with the findings set out in this article.In particular, if compared to minimal winning cabinets, oversized ones tend to be followed by a more similar government (results available upon request).9.The Manifesto Project (Volkens et al. 2021) data represent an important alternative source for party positions.However, we chose not to use manifesto data because of the reliability problems that affect the placements of parties in several of the countries included in our dataset.The literature has repeatedly emphasized the problems associated with the party positions derived from Italian manifestos (e.g.Pelizzo 2003;Zulianello 2014), and a large portion of our data (as many as 12.5% of cabinets) is constituted by Italian cabinets.The same problems have been highlighted also in other countries (Dinas and Gemenis 2010;Gemenis 2013).We also point out that these studies (especially those on Italy) mainly cover the post-war period before the 2000s.If we consider the three most recent decades and run our analysis using party manifesto data, the results (available upon request) are similar to those presented in this article.10.In the latter case, median-opposition distance was calculated by the Pythagorean Theorem as SM LA 2 2

+
. In this formula, SM is the distance on the 'State-Market' dimension and LA is the distance on the 'Liberty-Authority' dimension.11.Real GDP growth is undoubtedly another important -albeit less frequently used -indicator of economic performance.After calculating the difference between the growth rate in the year when the government ended and the growth rate in the year when the government started, we included the resulting variable in our analyses.Our main findings were not affected, while the difference in GDP growth -taken alone or included with the other two economic indicators -had no systematic impact on the different risks of cabinet termination (see Online appendix D). 12.In the literature, other attributes of parliaments such as polarization have been theorized to increase the risk of cabinet failure, although the relative findings are mixed (Bergmann et al. 2022;King et al. 1990;Laver and Schofield 1990;Maoz and Somer-Topcu 2010;Saalfeld 2008;Warwick 1994).In analyses that we do not report here, we included a number of measures of polarization, none of which had a systematic impact on cabinet replacements.13.In ancillary analyses we explored the possibility of asymmetric impacts of positive and negative economic developments.In particular, we created a dummy variable for decrease in unemployment (1 when 'unemployment difference' was negative, and 0 when it was positive or equal to zero) and found that any reductions in the rate of unemployment make cabinets less subject to all kinds of risk of termination (Online appendix E).
14.This is consistent with Fernandes and Magalhães' (2016) finding that, in semi-presidential systems where presidents are granted the power to dismiss the government, cabinets are more prone to the risk of non-electoral replacements that involve changes in the PM's party.However, it should be borne in mind that Fernandes and Magalhães focused on semi-presidential democracies, while the data we used include both semi-presidential and parliamentary regimes.Moreover, unlike Fernandes and Magalhães we did not consider the different prerogatives that presidents have in semi-presidential systems.15.A central assumption underlying the Cox model is that the effect of the covariates is proportional over time.Individual tests of scaled Schoenfeld residuals revealed that the proportional hazard assumption was violated for two of our control variables: 'bicameralism' (in model 4) and 'positive parliamentarism' (in models 4 and 5).Adopting a standard solution to this problem, we interacted these two variables with a logged function of time.The results indicated that the effect of our key explanatory variables did not change, while both bicameralism and positive parliamentarism were found to initially raise the risk of cabinet replacement, and then to decrease it over time.See Online appendix C.
Political Research, West European Politics, Political Science Research and Methods and European Union Politics. [francesco.zucchini@unimi.it]

Figure 1 .
Figure 1.number of discretionary cabinet terminations per 1000 days, by decade and country.

Figure 4 .
Figure 4. estimated survival functions for oversized and minimal winning cabinets (inter-electoral replacements with low and high similarity).note: survival functions in the left panel are estimated from model 4, while those in the right panel are estimated from model 5. the other variables are taken as the following values: 'cabinet core' , 'median-opposition distance' and 'enpp' at their mean, 'bicameralism' and 'positive parliamentarism' as equal to 1, and all the other covariates as equal to 0.

Table 2 .
Determinants of cabinets survival.cox proportional hazard models.competing risks.