Here Today, Gone Tomorrow - Political Ambition, Coalitions, and Accountability as Determinants of Ministerial Turnover in the Brazilian Multiparty Presidential System

The article aims at understanding the determinants of ministerial turnover in presidential systems. There are essentially three key factors that govern the propensity for ministerial turnover in a presidential system: 01. political ambition – reflected in the individual or collective strategies and tactics employed by the party-affiliated actors represented in the cabinet; 02. the presence or absence of a governing coalition – which has bearing on the degree of ideological distance between the president and such political parties as form [...]

The article aims at understanding the determinants of ministerial turnover in presidential systems. There are essentially three key factors that govern the propensity for ministerial turnover in a presidential system: 01. political ambition -reflected in the individual or collective strategies and tactics employed by the party-affiliated actors represented in the cabinet; 02. the presence or absence of a governing coalition -which has bearing on the degree of ideological distance between the president and such political parties as form part of any governing coalition, with consequences for ministerial scrutiny; and 03. accountability -a determinant in responding to such external pressures as may be brought to bear by public opinion. In order to investigate the impacts of these factors, this inquiry seeks to analyze the new and dynamic democracy of Brazil. Brazil's presidential system presents interesting features that make it an excellent laboratory in which to test hypotheses concerning every aspect of ministerial turnover. To this end, this paper employs a multivariate methodology, as well as descriptive and survival analyses, based on a comprehensive and original database of cabinet ministers that draws on three presidencies over five democratically elected governments from 1995 to 2014. Among other results, our most original finding was that there is an inverse correlation between the degree of ministerial politicization (political ambition) and ideological distance from the president during political scandals (interaction between the governing coalition and accountability) on the one hand, and the length of ministerial tenure on the other. Keywords: Cabinet appointments; presidential system; ministerial turnover; Brazil. In theory, there are essentially three key factors that govern the propensity for ministerial turnover in a presidential system: 01. political ambition -reflected in the individual or collective strategies and tactics employed by the party-affiliated actors represented in the cabinet; 02. the presence or absence of a governing coalition -which has bearing on the degree of ideological distance between the president and such political parties as form part of any governing coalition, with consequences for ministerial scrutiny; and 03. accountability -a determinant in responding to such external pressures as may be brought to bear by public opinion in response to unsatisfactory performance or media reporting of a corruption scandal. The first two of these three factors relate to principal-agent theory and the third to external shock adaptation; together they can explain key assumptions in regard to ministerial turnover in multiparty governments.
In order to investigate the impacts of these factors, this inquiry seeks to analyze the new and dynamic democracy of Brazil. Brazil's presidential system presents interesting features that make it an excellent laboratory in which to test hypotheses concerning every aspect of ministerial turnover.
First, there are the institutional arrangements of the Brazilian federative republic, with its three spheres of government, fragmented party system and biannual elections for the Executive and Legislative branches. These foster favorable conditions for any minister who may be inclined to resign and run for another office in a highly competitive political system. This being the case, political ambition may explain individuals or groups of ministers leaving office before a given presidential term has ended. T Pedro Lucas de Moura Palotti, Pedro Luiz Costa Cavalcante & Juliana Betini Fachini Gomes (2019) 13 (2) e0009 -3/37 The framework of Brazil's political system, so-called 'coalition presidentialism', exerts a powerful influence on the selection and potential dismissal of cabinet members, given that it involves deliberated calculations in a multifaceted environment. The literature on the matter makes it clear that the Chief of the Executive (i.e., the president) must take into consideration the proportionality of party representation in Congress, as well as the representation of social groups and regions (ABRANCHES, 1988;AMORIM NETO, 2006;CODATO and FRANZ, 2018;D'ARAUJO and RIBEIRO, 2018;FIGUEIREDO and LIMONGI, 1999;PALOTTI and CAVALCANTE, 2018). The scenario becomes even more complex and challenging in a multiparty system where there are some thirty parties represented in the Legislative branch and the proportion of seats held by the president's party seldom reaches 20%. The structure of Brazilian 'coalition presidentialism' is in effect supported by a range of parties with different ideological positions. This enables us to analyze whether the degree of ideological distance between the president and the parties forming the governing coalition entails consequences in terms of the scrutiny to which individual ministers are subjected.
The third factor, accountability, provides a particularly rich seam in the case of Brazil, with its compendious history of recurring political scandals, mostly related to corruption. The assumption underlying the presumed motive of accountability is that presidential decisions to fire ministers occur due to external pressures coming from public opinion or within the political system in response to unsatisfactory performance or media reporting of corruption scandals. As demonstrated by Araújo, Costa and Fittipaldi (2016), the greater a party's ideological divergence vis-à-vis the president, the likelier it was to be affected by dismissals in response to media reporting of corruption during the first Rousseff administration. Without question, coalition management in a government with a single-party cabinet would present an entirely different scenario. For one thing, the president's ability to exercise scrutiny over ministers would be less constrained, and this would have consequences for ministerial turnover.
In the more than two decades since the reestablishment of democracy in Brazil, political science has made significant progress in investigating Executive-Legislative relations. Nevertheless, little progress has been made on the influence of multiparty coalitions on the modus operandi of the Executive branch. governments. When confronted with sandals exposed by the media, presidents tend to fire ministers from parties whose ideologies diverge most from their own. This finding represents an original explanation for minister turnover in coalition governments faced with external shocks. Ministerial turnover is in the main explicable by reference to the political ambitions of ministers plus the president's decision to substitute a cabinet member, based on the factors outlined above.
Thirdly, this study provides a more extensive and multi-causal overview of Brazilian ministerial turnover than previous works in the field of political science.
In addition to this introduction, this paper is organized in four sections. The first discusses the literature on government formation in presidential systems.
The second presents the authors' hypotheses. The third provides some descriptive statistics regarding ministerial turnover in Brazil. The fourth section describes the regression models and their results. In the final remarks, the hypotheses are discussed and some conclusions are proposed.

The literature on cabinet formation in presidential systems
Ministers are key actors in the functioning of governments. They are political agents empowered to influence the legislative and executive agendas, and occupy positions of leadership in their organizations and policy areas. Therefore, when a president appoints or substitutes a minister, he or she sends an important message to a large number of actors. This has policy, political and symbolic implications, with consequences for public policy and governability.
Research on government formation in presidential systems focuses on three areas: coalitional presidentialism, minister recruitment and minister turnover (CAMERLO and MARTÍNEZ-GALLARDO, 2018). Coalitional presidentialism relates to the politics of forming a legislative coalition in order for the executive to be able to govern, and relies primarily on party linkages and institutional arrangements as sources of political support (AMORIM NETO, 2006;DRUCKMAN and WARWICK, 2005). Minister recruitment relates to ministers' profiles, and seeks to investigate recruitment strategies with reference to professional backgrounds and political or social linkages in executive formation. Non-party-affiliated ministers are the subject of considerable discussion in the literature, given the variety of interpretations of their roles as ministers (ESCOBAR-LEMMON and TAYLOR-ROBINSON, 2010;GROSSMAN and FRANÇOIS, 2013).
The third area of presidential government formation research is ministerial turnover, which is the object of two distinct analytical approaches: principal-agent theory and external shock adaptation (CAMERLO and MARTÍNEZ-GALLARDO, 2018). Initially, ministerial turnover was regarded as a political tool for leveraging alliances among the parties in a coalition government. The resignation of a minister may result from a policy difference with the president on the part of the minister's party or the minister him/herself; alternatively, resignation may be necessitated by the exigencies of the electoral calendar. A president may also realize his or her own error in making a specific ministerial appointment, especially in a complex political context, such as a multiparty presidential government or parliamentary system (DIERMEIER and MERLO, 2000;FISCHER and KAISER, 2008;HUBER and MARTÍNEZ-GALLARDO, 2008 Divergences between principals and agents are derived from their divergent preferences. Monitoring therefore becomes difficult, as agents use their asymmetric access to information and knowledge to gain benefits in the form of political advantages or to ignore the president's or party goals to pursue their own. This challenge becomes greater when ministers have multiple principals, such as the president and their political party (DIXIT, 1997;GAILMARD, 2009).
In this paper, principal-agent theory is employed to explain how political ambition and government coalitions contribute to ministerial turnover. Since 1988, multiparty coalition has been a strategy used by Brazilian presidents to form In general, when there is uncertainty as to whether a politician will be in office in the next term, politicians do not fully internalize the costs associated with their choices for public policies [...] and become unable to credibly commit themselves to inter-temporal arrangements. In terms of cabinet policy, a certain degree of stability is also needed to promote long-term policies and to support the implementation of programs and policies in their completeness [...]. In the literature, cabinet instability has also been associated with the breakdown of relations of accountability between politicians and bureaucrats, with weak accumulation of experience and instability and inconsistent public policies (MARTÍNEZ-GALLARDO, 2010, pp. 137-138). In Germany, individual substitutions tend to be more associated to personal enrichment or traffic of influence than to managerial motivations, such as poor performance or misconduct (FISCHER and KAISER, 2009). In Portugal, during more than thirty years (from 1976 to 2005), less than 10% of replaced ministers had had no prior political careers. It can be assumed that the autonomy of prime ministers was not constrained by pressures from political parties to replace ministers. Thus 'independents' became so important to the 'quality' of cabinets that the prime minister had to think twice before dismissing them (PINTO and ALMEIDA, 2009, p. 157).

Camerlo and Pérez-Liñán (2015a) used a database covering 12 Latin
American countries over the period 1979 to 2007 to investigate the effects of critical events and election calendars on ministerial turnover 1 . Their findings indicated that political clashes -such as protests against ministers and involvement in corruption scandals -had an effect on cabinet turnover, but that such effects were tempered by electoral projections and the desire to avoid the lame duck effect 2 . In the Argentinean case, Camerlo and Pérez-Liñán (2013b) developed an analysis based on three ministerial profiles: politicians, technocrats, and loyal friends (cronies).
Ministerial stability was linked with the popularity of the president and the existence or absence of economic recession. In scenarios of declining presidential ______________________________________________________________________________________________ 1 The case of Latin America highlights key reasons for ministerial turnover. Focusing on Latin America from 1990to 2003, Martínez-Gallardo (2010 argues that ministerial stability is lower than in other regions. Despite some differences between these systems, she demonstrates that in most of the region's countries, ministers retain their posts, on average, for approximately one third of a presidential term; the exceptions are Chile, Mexico, Costa Rica, and Uruguay, where ministers stay in their posts for over half a presidential term. Another relevant aspect is the average number of appointments per ministry. The countries with the lowest average number of ministers were Chile, Mexico and Uruguay, with approximately 04 to 06 ministers per ministry. In Ecuador, Peru, Colombia and Bolivia there were more than 10 ministers, on average, per portfolio, during the period analyzed. Ministerial instability and high turnover rates impede the effective implementation of public policies in the region. 2 Presidents are characterized in the literature as lame ducks when they are likely to be out of office within a short timeframe, either because they were not re-elected, chose not to run again or were limited by some institutional constraint, such as ineligibility to run in new elections. In these situations, presidents are expected to have less political influence over other politicians, such as those making up a legislative coalition, and devote themselves to unilateral and sometimes controversial or unpopular acts with a view to their long-term legacy (CAMERLO and PÉREZ-LIÑÁN, 2015a). popularity, a minister's ideological alignment with his or her president had no bearing on tenure. Conversely, in cases of rising presidential popularity, more aligned ministers are dismissed more frequently than less aligned ones. These results show that a cabinet reshuffle can be a proactive tool and is not only used as a reactive strategy when faced with external difficulties.
In this paper, the theory of external shock adaptation corresponds to the accountability factor. Brazil has 30 years of interrupted democracy with a free press, as well as independent auditing and control agencies and a vigilant public (MELO and PEREIRA, 2013). This creates a propitious environment for ministerial turnover, especially when ministers are accused by media of unsatisfactory performance or involvement in corruption. Figure   While coalition formation and its consequences for the relationship between executive and legislative powers has been extensively investigated, few studies have been devoted to understanding the durability of ministers that test for both principal-agent theory (here characterized as both political ambition and government coalition) and external shock adaptation (here as accountability). Over the following paragraphs, our research hypotheses, grounded in these three main explanatory assumptions mentioned above, are presented and discussed.

Government coalitions
The existence of broad governmental coalitions directly influences agency relations between the president and party leaders and the chief of the Executive and ministers (agents), mainly in their selection and the monitoring of their performance. Therefore, as in multiparty parliamentarism, in presidential systems, the president negotiates posts and scrutinizes the profiles under consideration for appointment.
The rationale underlying this process is that presidents will attempt to examine the profiles so to avoid subsequent problems derived from poor selection.
However, in a game that involves leaders of political parties who are more ideologically distant from the president, there is always the possibility of moral hazard associated with ministerial performance. In the expression used by Huber and Martínez-Gallardo (2008, p. 172), more distant ministers can cause "public policy damage" by their performance, which causes them to be more carefully monitored by the chief of the Executive vis-à-vis those ideologically closer to the president.
Considering the features of Brazilian coalition presidentialism, the following research hypothesis is tested: H2: The greater the ideological distance between the president's party and the party of the minister, the shorter their tenure in the ministry. In presidential systems with broad multiparty coalitions, like Brazil's, it is assumed that presidents will be more likely to dismiss ministers appointed by their government partners when they are involved in media-reported scandals. Because they face fewer constraints on the formation of their cabinet than chiefs of the Executive in parliamentary systems, presidents can more openly punish deviant behavior by representatives of allied parties (DOWDING and DUMONT, 2009). They choose to tolerate or find other ways out (ministerial substitutions or other possible appointments) for ministers from their own party and those without party affiliation, who are in general ideologically closer to the president.

Accountability
From the beginning of the first Rousseff administration, Araújo, Costa and Fittipaldi (2016) identified the president's responsiveness to corruption scandals.
However, she kept ministers from her own party in cabinet and dismissed those from allied parties. The greater the ideological distance of these parties from the president's party, the greater the likelihood of dismissal. Thus, it is assumed that Brazilian presidents are sensitive to corruption scandals. Nevertheless, they tend to spare their closest allies and punish more ideologically distant partners more severely, that is, the presence of government coalitions influences the pattern of presidential responsiveness.
H3: The involvement of a minister in media-reported scandals reduces his/her tenure.
There are, however, ministers who are closer and others who are more distant from the president and their party. This leads to differential treatment in their appointment and, where necessary, their substitution. The hypothesis below traces an interactive effect between coalitions and the occurrence of scandals.
H3.1: In cases of involvement in media-reported scandals, more ideologically distant ministers tend to stay for less time in their positions than ministers who are ideologically closer to the president; Before testing these hypotheses, the next section will describe the Brazilian case, based on the main descriptive numbers concerning ministerial turnover and its causes.

Here today, gone tomorrow: ministerial turnover in Brazil
This section addresses the following questions: do ministers in Brazil remain in office across successive governments? Is there variability that needs to be better understood? If turnover rates are high, are they symptomatic of conditions within a specific administration or do they reflect a broader phenomenon?
Based on data on all Brazilian presidential appointments, the findings discussed by Martínez-Gallardo (2010) were confirmed: there is a significant variance in the turnover of ministers both over and between presidential terms.
The variable used is the length (in days) of the presidential terms of office between 1995 and 2014. This descriptive estimate was carried out with the use of survival analysis to model determinants of the end of a ministerial appointment prior to the end of a presidential term. This is the event to be modeled.   respectively. This behavior provides evidence of significant differences in presidential style in contexts of varied adversity, given the noticeable differences between administrations under the same president 5 .
Reasons for leaving cabinet can be grouped into four main possibilities, according to the division presented by Escobar-Lemmon and Taylor-Robinson (2010). Based on their taxonomy, in the period from 1995 to 2014, as shown in Figure 03, the number of ministers who remained in office until the end of their presidential terms amounted to 47% of all appointments. This group is denominated the 'Survivors'. Figure 03 shows the ministers that were dismissed by their presidents, mainly due conflicts or disasters involving public policies, personal scandals or corruption allegations. These accounted for 10% of the total number and are known as the 'Tragic Ending' cohort. 11% of total appointments corresponded to ministers who were moved from one ministry to another, the socalled 'Migrators'. Lastly, a more heterogeneous set of reasons, ranging from the abolition of certain ministries, substantive changes and/or disruptions to coalition arrangements, resignation and/or departure from office to run in an election accounted for the remaining 30% of cases ('Exit by choice or convenience'). Approximately one third of such exits occurred because of election or re-election to a public position. Less than 1% were either not identified, or due to death or health reasons ('other reasons/unidentified').
One advantage of adopting the classification proposed by Escobar-Lemmon and Taylor-Robinson (2010) is the possibility of using the aggregated data, collected by the authors for other presidential systems, for comparison with the Brazilian case.   Source: Authors' own elaboration from data collected for Brazil, and from data collected by Escobar-Lemmon and Taylor  Other reasons for exit from office accounted for more than 30% of cases in Brazil, which is above the average of the other countries and close to that of Chile. Our analysis found that one third of exits were associated with resignation to run in an election, and another third due to changes in cabinet relating to intra-coalition politics. These findings are similar to those for Chile. Thus, from a comparative perspective, Brazil's ministerial turnover trends do not constitute especially disruptive characteristics. Brazilian ministerial turnover is high in comparison to parliamentary systems 6 , but similar to those found in other presidential governments in Latin America.
These data on ministerial turnover must, however, be carefully interpreted. Pinto and Almeida (2009, p. 155) warn about the limitations of mapping the reasons for ministerial tenure ending. According to them, "the real motives are often hidden from the public". For instance, unsatisfactory performance may be presented as a difference of opinion. In the Brazilian case, it is reasonable to assume that the low rate of tragic endings may be the result of agreements between ministers and presidents to treat them as resignations so as to avoid further damage to the reputation of the erstwhile minister and/or his party or the government in the court of public opinion.
This paper's statistical analyses therefore address the set of motivations that led ministers to leave office before the end of their presidential term. In this sense we hope to avoid subjective or inferred classifications based on limited publicly-available information.

The determinants of ministerial turnover Data and methods
To address these hypotheses, this paper uses a comprehensive and original database of cabinet appointments that covers three recent presidencies over five different democratically-elected administrations from 1995 to 2014 7 . Considering the dependent variable description, survival analysis will be used as a statistical tool, with minister tenure expressed in days. As Huber and Martínez-Gallardo (2008, p. 174) have pointed out, these models "are very useful precisely when the dependent variable of interest is the time for the occurrence of a terminal event (or a fault) -in this case the exit of a cabinet minister".
The terminal event (or fault) is therefore a ministerial departure from office before the relevant presidential term expires. Migration of a minister from one ministry to another will also be considered a terminal event (or fault). The objective is to arrive at an understanding of all of the possible reasons for ministerial turnover. We have not differentiated between dismissals and resignations, since the officially released documents concerning such cases are unreliable as sources. It is commonplace, after having been pressured into resignation by president or party, for a minister to announce publicly that his or her exit was made for personal reasons. Further study of this area, requiring a qualitative approach, presents enticing prospects for future research.
The longest a minister can stay in office is 1461 days, being the full four years of a presidential term. Observations on terms without failures have been censored. The same individual can be appointed to more than one role. This allows the survival model to be applied without needing to consider the effects of covariates on tenure. A similar strategy was applied by González-Bustamante and Olivares (2016).
Unlike in other studies in this area, we did not find any justification for the application of fragility 8 . Our model does not compare countries with each other, as did those of Huber and Martínez-Gallardo (2008) and Escobar-Lemmon and Taylor-Robinson (2010), therefore we needed to adjust estimates coming from a cluster, in this case from the same country. None of the variables identified in the literature as having an impact on ministerial turnover, such as the locus of an appointment, media-reported scandals, protests, presidential popularity, or macroeconomic ______________________________________________________________________________________________ not replaced more quickly than men. They also find that ministers appointed during the initial cabinet tend to spend less time in their positions than others, due to problems deriving from poor selection. Berlinski et al. (2009) note that younger ministers are more likely than older ministers to exit cabinet positions of their own volition, since, for the latter, such a move may coincide with the end of their careers.
The relevance of the regional dimension is emphasized in the literature (D'ARAUJO and RIBEIRO, 2018;DOWDING and DUMONT, 2009). More especially, the distinct effect of a minister coming from São Paulo seems interesting, not only because it is the country's richest state, but also due to the fact that the parties whose candidates ______________________________________________________________________________________________ 9 To confirm this assumption, the likelihood ratio test after the application of fragility does not reject the null hypothesis that the variance of the fragilities is zero (chibar2 (01) = 0.00; Prob>=chibar2 = 1.000).
(2019) 13 (2) e0009 -21/37 Finally, variables that measure the president's popularity, unemployment and inflation were also employed. As the president is expected to react to these indicators, the values for the three months leading up to the dismissal of a minister were taken into consideration. Popular presidents tend to have more alternatives to exhaust before reshuffling their cabinets, and therefore use this strategy less than their unpopular counterparts (MARTINEZ-GALLARDO, 2011) 10 . However, when faced with media-reported scandals, dismissals may be necessary to improve the popularity of the chief of the Executive (DOWDING and DUMONT, 2009). It is assumed that increased unemployment and inflation rates, both provided by the Brazilian Institute of Geography and Statistics (IBGE), will correlate to greater ministerial turnover. Internal and external shocks that cause economic crises tend to be countered by cabinet reshuffles, and can indicate a change in policy direction 11 .  Source: Authors' own elaboration from data collected for Brazil.

Descriptive analysis
Initially some exploratory analyses were performed in order to map the effects of independent variables on survival time. From the analyses developed, and taking into account survival analysis, we gave due consideration to the IMP, different areas of government and involvement in media-reported scandals. Because they are categorical, these variables can be graphically visualized in survival curves using the Kaplan-Meier estimator. In Figure 04, seven distinct curves relating to the IMP are depicted, one for each level, from zero to six. To facilitate visualization, the greater the politicization of a ministerial appointment, the darker the line.
It is observed that the dark lines, representing greater politicization, begin their terms presenting similar behavior. Over time (specifically, just before the 500-day mark), the dark lines reach their nadir -indicating minimum probability of survival. This preliminary result shows that the higher the degree of politicization, the greater the likelihood of leaving cabinet, especially after a certain time in office. The Wilcoxon test points out the existence of (chi2 (6) = 13.83; p = 0.0315).  Next, we can observe the behavior of the survival function when mediareported scandals occur during a minister's time in office. As can be seen from Figure 06, early in the presidential term, ministers with or without involvement in scandals tend to remain in cabinet for equal amounts of time, and in the short term those involved in scandals tend to survive more than those who are not. Thereafter, the curves tend to separate gradually. Ministers not involved in media-reported scandals tend to survive longer in cabinet.   At around the 1000-day mark in a presidential term, the curves get closer again, although they do not change places, and this can indicate a loss of statistical significance for the difference between them. The log -rank test presents statistically significant evidence on the distinction between curves, confirming that this effect is probably not isolated from other intervening factors, given its lower statistical significance (chi2 (1) = 3.10; p = 0.078).
From these initial descriptive statistics and the research hypotheses previously outlined, a survival analysis will be performed for the dependent variable that points to the event which ends a ministerial appointment before the end of the presidential term.   president and kinship with politicians -did not substantively alter the results.

Results -survival analysis
The fourth model, which adds socio-demographic variables, did not alter any of the observed results. The IMP remained statistically significant at the level of 1%.
The variables of media-reported scandals, protests and ideological differences still did not present statistical significance. The variable of 'political family' became statistically significant in two models. This indicates that belonging to traditional political families can contribute to increasing the probability of survival in a ministerial post. However, adding the variables of the last two models results in the loss of statistical significance for this variable. In any case, the statistical significance of 'political family' was already less robust, at 10%.
The fifth model adds the additional controls of macroeconomic indicators, rates of inflation and unemployment, as well as presidential popularity. All of these variables were measured for the third month prior to the dismissal of the minister, to provide an idea of the lagged effects of these factors. The variables of the IMP and the political articulation and State functions areas remain statistically significant. Media-reported scandals, protests and ideological differences remain without statistical significance. Presidential popularity has a positive impact on the survival of ministers, being statistically significant at the level of 1%. The unemployment rate is also statistically relevant, contributing negatively to the probability of ministerial survival. Thus, with each increase in presidential popularity, ministers are more likely to survive in office; conversely, there is an inverse correlation between the unemployment rate and the length of ministerial tenure.
The sixth and last model observed in Table 03  It is observed that, in none of the models, were gender and race variables statistically significant. The initial cabinet also does not present itself as a satisfactory explanatory factor to understand individual ministerial turnover. Age at nomination becomes statistically significant, pointing to a lower probability of staying in cabinet with an increase in the age, with statistical significance at the level of 10%. In all models, party affiliation was not statistically significant.
Withdrawal of party affiliation did not substantially alter the results obtained and previously reported.  Thus, at the beginning of the interaction, when the ideological distance is zero or close to it, the existence of media-reported scandals has no effect on ministerial tenure. The straight line and the confidence interval pass through zero. The effect, however, changes with the widening of the ideological difference between the president and the minister. After the ideological difference exceeds the value 01, as this distance increases, coefficients become statistically significant and negative. In this sense, in the presence of media scandals, the greater the ideological distance in relation to the chief of the Executive, the less likely the survival of ministerial appointments.
______________________________________________________________________________________________ 13 Royston and Sauerbrei (2009) present the solution of multivariate interaction of fractional polynomials for cases involving an interaction between a continuous variable and a categorical one, as in medical experiments. This method is a more suitable alternative to the traditional approach of categorizing the continuous factors for the application of standardized tests of interaction.

Final remarks
This paper tested the usual explanations for the exit of ministers before the end of a four-year presidential term, using the case of the Brazilian presidential system from 1995 to 2014. For this purpose, we considered the variables of profile, professional career, political and socio-demographic features, composition of specific ministries and the occurrence of external shocks, such as scandals and protests covered by the media, as well as macroeconomic and presidential popularity indicators.
The models' results confirm the H1 hypothesis. Consistently, in all models, with robust statistical significance, the greater the politicization, the shorter the time in cabinet. This finding is consonant with that identified by Cavalcante and Palotti (2015), who use two independent indicators to measure the technical and political aspects of the ministers, including two other administrations - Collor (1990Collor ( to 1992 and Franco (1992Franco ( to 1995. In this paper, the degree of politicization was measured using the Index of Ministerial Politicization (IMP). It is possible to infer that greater scrutiny prior to an appointment by the president is a condition of ministerial selection. More popular chiefs of the Executive have greater autonomy to vet and propose names, and can take lead roles as coalition builders. Likewise, unpopularity and recurring conflicts with coalition partners may force the president to accept indications that will later prove unsatisfactory. Problems derived from poor selection may therefore depend on a combination of factors that interfere with the observation of the expected relationship between ideological distance and ministerial turnover. Data collected on future administrations should help unravel this relationship.
The area of the presidency being responsible for political articulation and State functions tends to shelter more stable ministers compared to the more politically attractive core areas of government. Although not initially considered, it is reasonable to suppose that the explanation is close to the logic proposed for the H2 hypothesis. Poor selection problems are avoided for this area of government by not appointing allied ministers; this results in more scrutiny prior to appointment and greater ministerial stability thereafter. Coalition dynamics remain relevant in influencing presidential responses to media-reported scandals involving ministers.
The H3 hypothesis, when observing in isolation the occurrence of mediareported scandals involving ministers, is not supported by empirical evidence. However, it is supported by observing the derived hypothesis, H3.1. Presidents care about public opinion, and are permanently exposed to it. A president's accountability in response to media reports, especially of personal and corruption scandals, is conditioned by ideological distance from ministers. The strengthening of control bodies, in combination with a free press, as in Brazil, result in several checks that limit the action of strong presidents (MELO and PEREIRA, 2013).
Scandals act as alarm mechanisms for presidents, advising them of the existence of unwanted deviations coming from ideologically distant auxiliaries.
Therefore, ministers involved in scandals who are closer to the ideological mainstream of a coalition are less likely to be replaced than those who are more distant. Although at the macro level presidents react to scandals, it is observed that these responses are not equal, and depend on the affinity of the staff involved. These findings corroborate the evidence presented by Araújo, Costa, and Fittipaldi (2016) for the first two years of the Rousseff government, tested here for the 20-year period beginning with the first Cardoso administration. They also confirm that presidents apply differential treatment to their most ideologically distant coalition partners, which in other studies have been identified as being less likely to participate in decision-making processes (GAYLORD and RENNÓ, 2015;SILVA, 2014).
The results obtained for the variables unemployment rate and presidential popularity are consistent with the presidential responsiveness argument, in this case due to unsatisfactory performance rather than episodic events such as media scandals. Chiefs of the Executive in presidential regimes are directly exposed to evaluation via public opinion and appear to use the ministerial reshuffles to produce changes that reflect governmental responsiveness.
In the last model, it is observed that age on taking office was statistically significant; the older a minister, the shorter their time in office. This result goes against the assumption that younger ministers are more likely to leave the cabinet in order to preserve their political careers, while more experienced ministers, upon leaving the cabinet, are likely to end their working lives (BERLINSKI et al., 2009).
Future research should test the extent of this and investigate which factors may influence ministerial tenure according to age. It is also worth noting the lack of statistical significance relating to gender, race and the initial composition of the cabinet, which indicates that these factors do not interfere with decisions to retain, change or dismiss state ministers in Brazil.
Thus, it is argued that all three sets of explanations initially presented are relevant for understanding ministerial turnover. Ministers, by entering in the world of politics, create links and situations that constrain their tenure, and are mainly driven by political ambition. Government coalitions, in turn, involve multiple principals and asymmetrical treatment by presidents of their auxiliaries. Finally, media-reported scandals constitute constraints and alarms that mobilize presidential attention. Presidents tends, when faced with media reporting of scandals, to fire ministers who are ideologically more distant from them. This finding represents an original explanation for ministerial turnover. It comes primarily from a presidential decision influenced by the factors of multiparty coalition government and external shocks.
In sum, the paper's findings contribute to the advancement of the field of study insofar as it explores the main theories of the literature on ministerial turnover applied to a new and dynamic democracy. Brazil is most certainly a suitable laboratory for this type of analysis, due to its coalition presidentialism, its multiparty system and its politicians' constant involvement in scandals, and this article sought to analyze ministerial stability with a comprehensive and unprecedented review of the local literature on the subject. This study gives an extensive and multi-causal answer to minister turnover in multiparty systems as Brazil.
The emphasis of political science studies has been on understanding the joint composition of cabinets, rather than the choice or replacement of individual ministers. The explanatory factors for a cabinet's initial composition, however, do not explain its later conformation. Observing the durability of cabinets allows progress on knowledge about the macro functioning of governments, while focusing on ministerial tenure helps understand how they function at the micro level. These approaches generate implications for the accountability and effectiveness of governmental actions.