Research Design, Hypothesis, and Literature Review

Decisionmaking processes vary significantly among public policy areas. While some policies need legislative or collective approval, others can be implemented by presidential decree. These differences can make it difficult to compare different policy areas, such as foreign policy, health, education, and finance, by adding confounding factors to the analysis. Using ministerial survival solves this comparative difficulty by standardizing the decisionmaking locus: the presidency.

Presidents, as opposed to prime ministers, have individual rather than collective responsibility (Neto, Reference Amorim Neto and Malamud2020). This is of note when it comes to appointing or dismissing ministers. We can classify the decisionmaking process of appointing or dismissing a minister as the “predominant leader” type (Hermann et al. Reference Hermann, Preston, Korany and Shaw2001), characterized by the fact that no actors other than the president can veto the decision to implement a change of minister. In our sample and time series, the president had the exclusive right to appoint ministers. ^{Footnote 1} (A small exception is a recent period in Peru.) ^{Footnote 2} The countries’ constitutions delegate this responsibility to the president, although there is some variation in terms of the rules for dismissing ministers through the legislature. It is important to note that we did not register in our sample cases in which ministers were dismissed by any actor other than the president, with the exception of some recent cases in Peru. Thus, we isolate the possible effects of different decisionmaking arrangements among countries and, principally, among policy areas.

Data, Methods, and Results

When building our dataset, we included only permanently appointed ministers, excluding from the sample those who assumed the position on an interim basis. As ministerial appointments are in the remit of the executive, the survival of a minister is usually associated with the fate of their political leader. That is, in a scenario in which there is a departure from the political leadership of the government, the chance of ministerial change is very high (Quiroz Flores Reference Quiroz Flores2009). Therefore, we constructed a dependent variable recording failures when there was a change of minister during the same presidential term, accounting for 763 (62 percent) events. The ministers who remained in office until the end of the presidential term and those who left their post for reasons of health, death, or change of president were not counted as failures, nor were the ministers who still remained in office in December 2020. The dismissal of ministers was treated as a censoring event. Our database includes 1,229 ministers from the four portfolios during the 75 years, bringing together 131 presidents.

Table 1 presents descriptive statistics about the independent variables drawn from the specialized literature. The names of the variables are shown in the first column, followed by the number of observations (Obs), the mean, standard deviation (Std. Dev.), minimum value (Min), and maximum value (Max). To access the source of the data, see the supplementary document.

Table 1. Descriptive Statistics

As table 1 shows, just over a quarter (28 percent) of the cases analyzed took place during autocratic regimes. The president’s legislative seats average 38.7 percent with large variability. The variable Coalition indicates that more than half of the governments in the sample were coalitions. The economic growth of the countries has an annual average of 1.09 percent, with great variation in the time series. An average 43 percent of the ministers were part of the president’s political group, and the average effective number of parties was 4.04; some cases had more than 16 effective parties, and others had only one effective party, such as Mexico and the Institutional Revolutionary Party. GDP/capita, measured in thousands, shows an average of $8,000 and a low standard deviation (0.73).

The president’s legislative powers, in turn, have an average of 52.3 and a range that varies across the scale of the variable, from 1 to 100. Only 26 percent of ministers held elective office before arriving at their ministry. Technocrats make up precisely half of the sample. As expected, almost a quarter of the ministers (23 percent) hold a foreign policy portfolio. The average presidential approval in the four countries is 40.8 percent, with significant variation, ranging from presidents with 2.44 percent approval at minimum to 87.2 percent maximum.

The results begin with figure 1, which presents the Kaplan-Meier curve for ministerial survival disaggregated by portfolio. The red line represents the ministry of foreign affairs; the solid black line represents the ministry of finance; the dashed gray line represents education; and the dotted black line represents health. The horizontal axis displays months and the vertical axis the proportion of ministers who remained in office. For visual clarity, we have limited the horizontal axis to 100 months, including 99 percent of the ministries in our sample.

Figure 1. Kaplan-Meier Survival Estimation by Ministry (1945–2020)

Figure 1 partially confirms the propensity for higher survival of ministers of foreign affairs when compared to other portfolios. After six months, health and foreign affairs make up around 15 percent of the ministers dismissed, while finance and education make up around 25 percent. After one year, education and finance show 40 percent laid off while health displays 30 percent dismissed and foreign policy shows approximately 25 percent. In two years, the foreign policy line splits from health with 45 percent and 55 percent, respectively, of ministers replaced. In the same two years, education and finance lose 60 percent of their ministers. Nearly 36 months later, health and foreign affairs converge again, with just over 30 percent of ministers still in office. Education, at this same level, has approximately 80 percent of ministers dismissed, emerging as the most unstable ministry in the region, closely followed by finance.

After four years, finance is more clearly delineated from education, with 20 percent of ministers kept in office. Foreign policy converges again with health with close to 30 percent of ministers held in office. Martínez-Gallardo (Reference Martínez-Gallardo2014) finds a similar result in a comparative sample of Latin American countries, suggesting that the constant economic crises in the region are a possible explanation for ministerial instability in finance portfolios. The author also cites the ministry of foreign affairs as an opposite example of stability (Martínez-Gallardo Reference Martínez-Gallardo2014), concurring with our results.

Table 2 presents the Cox regression models for our longitudinal time-to-event data presented earlier. In this type of estimation, the dependent variable is transformed into a hazard rate, interpreted in our case as the instantaneous probability for a minister being dismissed from a ministry, assuming that it has not happened already. The advantage of using Cox models is their flexibility when dealing with time-dependent coefficients, not requiring very strong parametric assumptions about the distribution of probabilities for the event to occur (Therneau and Grambsch Reference Therneau and Grambsch2000).

Table 2. Results

***p<0 .001, **p<0.01, *p<0.05.

Note: Hazard ratios with standard errors in parentheses (conditional on frailty of the six models; grouped by country). Efron Method.

The Cox regression models were adjusted using the Efron method, due to its suitability for the analysis of adjacent events observed at the same time, such as ministerial rotation in different countries (Box-Steffensmeier and Jones Reference Box-Steffensmeier and Jones2004). Another relevant aspect of the Cox regression estimation performed is the use of shared frailty (Cox shared-frailty model), which allows us to detect how latent variables can affect the survival function. This type of estimation is analogous to the use of random effects in a linear regression model. The logic behind this model is to assume that some ministers may be more likely to leave office because of unobserved country attributes. Data were organized into groups represented by the seven countries, whose observations share a Theta weakness (Therneau and Grambsch Reference Therneau and Grambsch2000). The proportional hazards model is adjusted for fragility and estimated with the logarithm of maximum likelihood (Log likelihood) weighted by Theta, conditioning the calculation of coefficients and standard error to the assumed weakness.

To determine the robustness of the estimated models, we checked the nonviolation of the proportionality assumption, based on the constant effect of the variables in the analysis. In addition, we evaluated the fit of the model using Cox-Senell residuals, which indicate a good fit when the cumulative hazard function has an exponential distribution with a hazard rate equal to 1 (Box-Steffensmeier and Jones Reference Box-Steffensmeier and Jones2004). We also applied the variance inflation factor (VIF) test on the variables and models to check for multicollinearity. The estimated models, with the exception of model 3 (m3), showed adequacy in relation to the proportionality assumption, model adjustment and absence of multicollinearity. The VIF test indicated the necessity of excluding some variables, like GDP/capita, from models 4, 5, and 6.

Table 2 displays the six Cox regression models estimated using relevant variables from the literature described earlier. The hazard ratio coefficients shown in the table represent the relative risk corresponding to a change of one unit in the independent variable. When the coefficient is less than 1, the variable is characterized as predicting survival, since it reduces the chance of failure. Coefficients greater than 1, on the contrary, represent a risk factor, which increases the probability of dismissal. Model 1 includes only the dichotomous variable Foreign policy, whose relative risk coefficient is less than 1 and statistically significant. It is important to note that the same six models estimated for foreign policy were also estimated for finance, education, and health, which results are reported in the supplementary document.

As the intention of this section is to observe the relevance of the foreign policy portfolio in the face of important controls in the literature, model 1 serves to analyze the suitability of the central variable of analysis in the survival model.

The last rows of table 2 show the number of observations: 1,229 ministers is the maximum number in the sample. The number of groups used to estimate the frailty survival model includes 7 countries in our case. ^{Footnote 4} The number of dismissals and ministerial changes in the same presidential term is shown as failures, and the probability of obtaining the chi-square statistic (8.41 in model 1) is shown, given that the null hypothesis is true (χ²). The log likelihood is of the model that serves to compare models. Theta, as explained above, represents the shared weakness of the frailty model of the seven countries being used to weight the calculations of coefficients and standard error (in parentheses). The PH assumption reveals the values of the tests of violation or not of the proportionality assumption, whose non-significant values indicate adequacy to the assumption. ^{Footnote 5} The average VIF (variance inflation factor) of the model detects multicollinearity.

Model 2 (m2) includes all the independent variables drawn from our literature review that do not have missing data, keeping to 1,229 observations. Here we can already observe the effect of foreign policy, controlling for economic growth, GDP/ capita, political regime, coalition government, technocrat minister, political group of the president, and political experience of the minister. The foreign policy coefficient remains stable and significant in all six estimated models, giving robustness to the conclusion of a positive effect of foreign policy on the probability of ministers’ remaining in office.

A very useful way of interpreting hazard ratio coefficients (HR) is to convert them into probabilities: p = HR/(1 + HR). In model 2, keeping the aforementioned variables constant, the foreign policy portfolio is 42.5 percent more likely to retain ministers when compared to the education, finance, and health portfolios. Regarding the independent variables used as controls in model 2, some confirm their associated theoretical expectation. Technocratic ministers have an informational advantage over nontechnocrats and tend to survive longer in their position. The same occurs with ministers who belong to the president’s political group, a factor that reduces the chance of dismissal. The ideological proximity between the minister and the presidency, prior knowledge, and greater scrutiny are elements that provide stability in the position. The economic variables—Economic growth and GDP/capita—are significant ministerial stability factors in model 2, indicating that economic stability allows the president to maintain the ministerial team for a longer time. The variables Coalition government and Politician were not significant.

Model 3 includes all the variables collected except for the president’s popularity. The total number of observations drops to 1,001 ministers, due to missing data. This is the reason to run the model separately. As can be seen at the bottom of table 2, model 3 violates the proportionality assumption, indicating inadequacy. Also, we found in model 3 variables with high VIF: GDP/capita, Effective number of parties, and the Percent of president’s legislative seats. This is the reason we ran model 4 without those three variables. With better accuracy than model 3, model 4 confirms the relevance of foreign policy, economic growth, autocracy, belonging to the president‘s political group, the legislative powers of the president, and the technocrat minister as survival factors in our analysis.

Model 5 includes all the variables in model 4 plus Presidential popularity, which drops the number of observations to 443, due to the availability of popularity data from 1979 on only. Therefore, for example, the variable Autocracy loses statistical relevance in model 5 but maintains its relevance in models 2, 3, 4, and 6, in which observations cover the entire authoritarian period in the countries of the sample.

The popularity variable, which was the central motivation for estimating model 5, was not significant. However, for this reduced sample the variables Politician and Percentage of the president’s legislative seats were statistically significant. The more seats the president’s party has, the greater the propensity for ministers to be dismissed, a result contrary to what is expected in the literature. This was observed mostly in the postdemocratization period. While in model 5 coalition governments have a coefficient that suggests a risk factor for survival, there is no statistical significance.

The greater propensity of authoritarian regimes to keep ministers in their positions is an interesting result that can contribute to an important discussion regarding political appointments and ministerial stability in the research agenda about autocracies (Buckley and Reuter Reference Buckley and Reuter2019). To further explore the relevance of political regime in our database, we ran interaction terms between foreign policy and the significant variables in models 2 and 4. Only autocracy revealed significant interaction with foreign policy. We also tested interactions between autocracy and the finance, education, and health portfolios. None of those were significant (see supplementary document).

Model 6 shows the results of the model with an interaction term between autocracy and foreign policy and the variables that had statistical relevance in the other models and at the same time did not present a VIF higher than 10. Before interpreting the estimate of the interaction between autocracy and foreign policy, we tested its statistical significance. We estimated model 6 without the interaction and then with the interaction. With that result, we applied the likelihood ratio test that suggests evidence of interaction (p = 0.01). ^{Footnote 6} In other words, there is evidence that the effect of the political regime on greater ministerial survival is not the same in foreign and domestic policy portfolios. The detection of a statistically significant interaction means that the risk rate of ministerial change depends on the value of another independent variable. In model 6, the hazard ratio of the interaction is greater than 1, meaning an increased risk of removal. Political regime is mediating the difference in stability between foreign and domestic portfolios, indicating that in autocratic regimes, foreign ministers have a greater risk of removal than domestic ministers. We further explore this result with some disaggregated descriptive statistics to better understand the relationship between political regime, foreign and domestic policies, and ministerial survival. Figure 2 shows the Kaplan-Meier curve for ministerial survival disaggregated by political regime.

Figure 2. Kaplan-Meier Survival Estimation by Regime Type (1945–2020)

Figure 2 clearly exposes the greater ministerial stability in the autocratic regimes, graphically corroborating the results of the models shown in table 2. Autocratic regimes have a higher proportion of ministers remaining in office compared to democratic regimes throughout the series of months displayed on the horizontal axis.

To compare the greater stability in autocracies, taking into account the difference between foreign and domestic policies, figure 3 presents a descriptive comparison of ministers’ survival by portfolio and political regime. It places the incidence rate of ministers’ changes on the vertical axis. The incidence rate is the number of observations of ministers leaving office divided by the risk measured in months. The risk, in turn, is calculated through the frequency of ministers in each portfolio and the sum of the time they have been in office. The higher the incidence rate, the greater the ministerial volatility.

Figure 3. Incidence Rate of Ministerial Dismissal by Portfolio and Political Regime (1945–2020)

Three main empirical regularities can be extracted from the results in figure 3. First, chancelleries in autocracies are more stable than chancelleries in democracies. Second, in autocracies, the foreign policy portfolio is a little more unstable than finance, a little more stable than education, and more unstable than health. Third, in democracies, foreign policy is more stable than the three domestic portfolios. While democratic presidents make fewer changes in ministries of foreign affairs when compared with domestic portfolios, this difference is not observed in autocratic presidents.

The hypothesis of the specificity of foreign policy that we have outlined is valid only in democratic regimes. Due to the greater ministerial stability of autocratic regimes in all portfolios analyzed, the international audience’s incentive for stability in the chancellery is diluted in comparison with domestic policy. As autocratic chancelleries are more stable than democratic ones, it does not seem reasonable to conclude that foreign policy is a risk factor in the ministerial survival of autocratic regimes. The autocracies in the present sample signaled ministerial stability to international partners, including chancelleries. Democracies, on the contrary, show high instability in relevant domestic portfolios. In this scenario of domestic instability in democracies, the incentives arising from the international audience affect the greater stability of the foreign policy ministry when compared to the domestic portfolios.

Final Considerations

The fundamental finding of this empirical research is that foreign policy ministers in democratic regimes are more likely to stay in their posts for a longer time than their peers in finance, education, and health. The comparison is valid, as it controls for confounding factors characteristic to each policy area, focusing the analysis on the same type of presidential decision. Furthermore, the Cox regression models include relevant explanations for ministerial survival, identified through the extensive literature on the subject, lending robustness to the finding. It was possible to identify the political regime as an important intervenient variable to the relationship between ministerial survival and the foreign and domestic portfolios. The foreign policy tendency toward less risk of ministerial removal is valid only for democratic regimes.

The analysis of foreign policy, which is essentially interdisciplinary, substantiates its specialization by drawing on the unique characteristics of this policy area when compared to domestic policies. The incentives coming from the international audience are the main difference highlighted between foreign policy and domestic policy. In the international arena, a change of the main figure responsible for diplomacy below the president can harm informal arrangements and agreements between states. The fulfillment of agreements depends on trust between parties. A change of minister can generate instability in negotiations and arrangements in progress, especially those of an informal nature carried out directly between the chancellors of two countries. There is greater credibility in the political compromise between two nations if the negotiators who formulate the informal arrangement remain in their positions. Foreign policy specificity can be observed in the differential behavior of decisionmakers in democracies, highlighting its difference from other public policy areas in that institutional context.