Revisiting Shugart and Carey’s Relation of Executive Powers and Democratic Breakdown

The Original S&C Finding: An Inverse Relation of Presidential Powers and Democracy

The most central S&C result is based on a graph (Figure 8.1 in the original) that plots presidents’ legislative (LT) versus nonlegislative (NLT) powers. The range for these two axes is based on the authors’ careful coding of constitutional provisions, harkening the neoinstitutional era. Unlike older institutionalists (Blondel, 1987; Duverger, 1980; Neustadt, 1960), who generally listed constitutional provisions, here the authors operationalized their concepts and, while acknowledging the limitation of a small-n study, they provided a quantitative test of their hypothesis. To do this, they collected data on numerous powers, scaled them from weak to strong, and divided them into LT and NLT types. Then, to enable cross-system comparisons, they summed the individual powers for each dimension to generate a ranking. This ranking allowed evaluation of the impact of presidential powers on democratic breakdown. Their important finding was that countries with high scores on both LT and NLT rankings were associated with a higher propensity for breakdown.

This finding was built on an empirical test that lacked full theorizing. Without calling attention to the contradiction, Linz condemned presidentialism by arguing that both weak and strong presidents were problematic. Citing an inherent incongruity between presidentialism and democracy, he argued that presidents who were too weak to avoid gridlock could yield to breakdown, as could domineering presidents who engendered polarization and a winner-take-all approach to politics. Without addressing this potential contradiction directly, S&C roughly tested a linear hypothesis that posited that the strongest presidents were dangerous to democracy. Taken at face value, their tests imply that dominant presidents are problematic, while those too weak to overcome gridlock—the theme for which Linz is best known—are not.

In addition to the much-increased reflection that S&C generated about the relation of presidential power and breakdown, their development of power indices has inspired many expansions and revisions of their method (Frye, 2002; Hicken and Stoll, 2008; Payne et al., 2002; Siaroff, 2003; Taghiyev, 2006). Others have criticized the methodology, arguing that simply summing categorical variables produces inaccurate inferences (Doyle and Elgie, 2016; Fortin, 2013). A related problem is that the scales implicitly assume a linear and continuous distance among the points on the scale—that is, that the distance between a one and a two is the same as between a two and a three—while clearly the gain in power is greater in moving between some points than others. For example, the differences between no veto power, package veto power, and partial veto power are substantial and varied, and are not captured well by an interval continuum of 1,2,3.

Another crucial concern is whether S&C measured all of the pertinent variables. A potential source of omitted variable bias is their lack of the inclusion of informal presidential powers (i.e. the bully pulpit, or the president’s ability to command national media attention at any given moment). However, such criticisms do not take the book’s underlying hypothesis seriously. Certainly, the presence of informal powers may influence outcomes within contingent contexts. But the central point of S&C’s analysis is that formal institutions drive outcomes. Our goal, therefore, is to stick to their initial proposition and reconsider tests about the impact of institutions.

Alternative Approaches

The literature using and citing Presidents and Assemblies is expansive (with about 4000 citations on Google scholar), and we will not try to summarize it here. Instead, we will focus on how the recent literature has dealt with (or failed to deal with) their method for scaling powers and evaluating their thesis about the link between weak presidents and democratic survival.

A considerable cadre of scholars has followed S&C’s work with novel applications of their method as well as alternate indices of presidential power. Metcalf (2000: 663) notes that a central contribution of S&C was to provide a foundational structure for the measurement of formal constitutional powers that would come to be known as the “checklist method.” The checklist method consists of defining a list of powers in the constitution, scoring the extent to which each power can be found in different presidential systems, and then adding up the scores. The checklist method became a starting point for future scholars who would then debate the merits of including/excluding certain powers and whether to measure those powers continuously (such as S&C’s 0–4 scale) or dichotomously. Frye (2002) adopted the checklist method as well as most of S&C’s original institutional variables to score postcommunist executives, and retained the continuous scaling (0–3 in this case). Hicken and Stoll (2008) applied Frye’s revision to a study of the impact of executive institutions on presidential party systems.

Siaroff (2003) followed the checklist structure but broke the mold by opting for a dichotomous (0,1) scoring method, and replacing most of S&C’s original variables. Instead of focusing on constitutional provisions alone, Siaroff incorporated more wide-ranging, structural indicators such as the separation (or fusion) of symbolic head of state and executive roles, the direct or indirect election of the executive, and the accountability of the executive to the legislature. The advantages of Siaroff’s system are its parsimony relative to those who more closely followed S&C’s original design, as well as its incorporation of broader system-level variables that may situate the importance of narrower institutional factors.

Van Cranenburgh (2008) as well as Elgie and Schleiter (2011) adopted Siaroff’s (2003) method in tests of S&C’s hypothesis. Cranenburgh’s endorsement of Siaroff’s revision rests on a criticism of S&C’s lack of quantitative precision, arguing that a dichotomous method offers a less subjective, more generalizable, and more parsimonious approach. Yet Cranenburgh and Siaroff’s contention of parsimony glosses over Metcalf’s (2000) earlier defense of the use of continuous variables in the measurement of presidential powers. For example, in her review of Eastern European executives, Metcalf (2000: 664–666) points out that while Romanian and Polish Presidents both have the sole right to veto legislation, a two-thirds vote in the legislature is required to override the Polish president’s veto and only a simple majority is required to override the Romanian president’s veto, yet both would be scored as “1” in a dichotomous measure of presidential veto power.

While there is not yet a conclusion to the dichotomous/continuous measurement debate, and disagreements about which variables to include are likely to rage on, both discussions may be premature if the checklist method itself is not well-founded. In reference to the checklist literature, Fortin (2013: 93) points out that “despite the general acceptance of such measures of presidential power and their widespread use, empirical investigations to ascertain the degree to which existing indices measure a single latent construct, and are valid and reliable, were never conducted.” Fortin (2013: 94) continues to argue that a consistent focus on composite indices may mask the disproportionate impact of individual features or specific combinations of certain features along one or more dimensions. Fortin’s criticism makes an incisive and fundamental point about understanding presidential power that seems to go unanswered in the institutionalist literature.

The other most prominent critiques of the checklist approach rebuke the strict constitutional focus. Cheibub (2007), for example, asserts that the failure of most third-wave presidential systems can be attributed to power of the militaries in the states within which they were established. Another example is Helmke and Levitsky (2006), who emphasize the role of informal institutions in democratic durability. These approaches, however, do not take S&C’s hypothesis seriously. As we noted, informal institutions and historical factors are indisputably important, but their emphasis detracts from the focus on the critical question of whether patterns of democratic success or failure are tied to variations in constitutional design.

Our goal, then, is to return to the original S&C thesis and propose a new test of their strict theory. In order to test their central finding that strong presidents are dangerous to democratic stability, we update their data and respond to the shortcomings that underlie their tests, by proposing new ways to measure their independent and dependent variables. As foreshadowed, our tests are not highly supportive of their theory. This may be attributable to data limitations and/or some of the factors other authors have emphasized. Rather than providing a definitive criticism, however, our goal is to the emphasize the value of the exercise, which is to move toward an understanding of the impact of constitutional design, which was S&C’s concern and is perhaps the key consideration in democratization.

Updating and Reevaluating S&C’s Evidence and Findings

In order to affirm S&C’s original findings and investigate whether they have held up in recent decades, in this section we verify their measurements of independent and dependent variables, and expand the sample both temporally and geographically. We find that the relationship does not appear as strong as S&C originally suggested, but rather than rejecting the hypothesis, in the next section we offer alternative ways to evaluate it.

With the advantage of 25 additional years of history we now examine the data. As noted, S&C combine their data on all presidential constitutions and democratic breakdowns into a plot (Figure 8.1) that uses the two dimensions of presidential power as axes. They then locate each of their 32 countries in the graph, highlighting those that have broken down. Summarizing the data in a table (Table 8.3), they find that those in the upper right (Region I) of the figure, where presidents are endowed with higher LT and NLT powers have a higher proclivity to have broken down.

To reconsider their results, Figure 1 recreates S&C’s figure (adding the change in polity scores since 1992 by each country, which we discuss later) and we summarize the data in Figure 2 (with more information in Appendix Table A1). Here, we only include countries for the time period 1992 and 2016 (hence we drop, for example, Weimar Germany and Chile, 1969). ¹ The number of regions follows a counterclockwise pattern: the upper right corner corresponds to Region I in the original Figure 8.1 in S&C (p. 156), followed by Regions II, III (empty upper left corner), IV, V, and VI (lower right corner).

OPEN IN VIEWER

Figure 1.

Legislative/Nonlegislative Powers, Updated, Plus Change in Polity Scores, 1992–2016.

OPEN IN VIEWER

Figure 2.

Original Predictions and Theoretical Adjustment.

Even before considering the methodological and theoretical concerns, it is necessary to note problems with the data themselves. First, S&C have two simple but serious errors in the counting and arithmetic in the two key regions of the original table. After eliminating several countries in the graph because they had either been authoritarian or the constitutions were too new in 1992 to consider the long-term effects of their constitutional framework, S&C code 6 of 12 cases with a breakdown in the top-right quadrant of the graph, Box I. This yields a very high breakdown rate of 50% in the area where they predict the most trouble for democracy. While it does not change their finding, a first error is that the box has 10 cases, with 5 breaking down. The second error, which does affect their empirical findings, is in Box VI. Here, S&C list three cases of breakdown out of nine in total, but they indicate that the breakdown rate was 25% instead of 33%. Since this is supposed to be the region with the fewest breakdowns, this higher rate works against their theory.

Turning to an evaluation of the data in the figure, for Region I of the graph, S&C predict the highest frequency of breakdowns because the countries are high in terms of both LT and NLT powers. In our reevaluation, we first consider the five countries from Region I that they eliminated because they were either new or nondemocracies at the time of their writing: Brazil (1988), Sri Lanka, Paraguay, Mexico, and Panama. They include in their analysis Chile (1989), but by their criterion it too should have been eliminated since the democracy was only 2 years old when their book was published. Finally, S&C coded Uruguay as a breakdown owing to the 1973 coup and ensuing dictatorship, but since it has been successful since 1985, we could logically count a success for its subsequent 35 years of stability.

Overall, the updated empirics for Region I are not supportive of the S&C thesis, because contrary to their expectations, countries in that region show no democratic breakdowns since 1973. The new countries that we have included generally work against the theory. Of these, Brazil and Mexico have not experienced democratic breakdowns, though Brazil has suffered from constitutional crises and impeachments. Panama was a new democracy in 1992, forced to build a constitutional structure after the 1991 US invasion. Since then it has modified its constitution several times, but the newest version (2004) does not seem to have increased the president’s powers, as there are still no abilities to decree laws or control the budget, and there are no provisions to allow exclusive introduction of bills. Sri Lanka, finally, was mired in a horrific civil war that ended in 2009. S&C eliminate the case because it was nondemocratic; it could also merit a label of a nonfunctioning regime.

The cases that S&C did evaluate also fail to support their hypothesis. Furthermore, of the countries that did break down in earlier years (Brazil, Chile, Korea, and Uruguay), democracy has generally consolidated. Supporting their thesis, however, is that Chile’s and Korea’s post-transition constitutions lowered the powers of their president. Uruguay’s constitution did not change, however, and the 1988 Brazilian constitution increased the president’s powers. Of these countries, Brazil has suffered the most constitutional crises, with two presidents having been impeached and removed from office. Finally, as we discuss below, Argentina could arguably be in Box I, though S&C place it in Box VI. It too works against the thesis, as it has not suffered democratic breakdowns since the 1983 democratization.

Revisiting Region VI, with low LT powers but high NLT powers, also throws caution to the S&C thesis. This box is “off the diagonal,” so perhaps the expectations are not particularly strong, but the authors do argue that those with the lowest LT powers may be better able to “weather severe crises that have not been allowed to become clashes between the two elected branches of government over constitutional powers, in part because the assembly is clearly the dominant branch” (p. 158). In that box are the United States and Costa Rica, which may help sustain the S&C argument, but the box also includes Nicaragua, El Salvador, and Nigeria, which do not. Even more damning to the theory is that it includes Venezuela, whose president scores a 0 on LT powers, but its democratic standing has collapsed (and they suffered two attempted coups, one just prior to the book’s publication). Likewise, despite their exclusion from Figure 8.1, Honduras and Bolivia (which they code for presidential powers but do not plot in the figure) remain problematic cases because their lower scores in LT powers at the time the book was published did not prevent the deterioration of democratic quality in these countries or the ousting of Honduran President Manuel Zelaya in 2009. In spite of those low LT powers, two Bolivian presidents resigned their offices after clashing with the legislature (and the public). Since those resignations, the Bolivians have redesigned their constitution, and their current President has consolidated his power. This consolidation, which challenges the country’s democratic label according to some indices and analysis, could be the result of additional power afforded to the president by their 2005 constitution. That document explicitly recognizes the president’s ability to issue “supreme decrees,” (Article 94) which should increase the power over the 1994 constitution, which S&C coded as 0 on that item. The earlier constitution, however, did have several provisions that implicitly recognized decree authority. If the new constitution does increase the president’s power, then the harm to democracy might be attributed to the increased LT power. However, it would also be reasonable to argue that the older constitution failed to prevent the situation, which gave rise to the increased power.

Next, Box V has two very different sets of cases, Peru and Namibia versus Finland and Iceland (plus Korea 1948). Peru suffered a breakdown just as the S&C book was published, and Namibia, which gained independence (from South Africa) in 1990, remained at 6 on the polity scale for the whole 24-year period under study (indicative of a statically fragile democratic aspirant). On the other hand, Finland, Iceland, and South Korea have all remained stable, robust democracies.

The final box that we reevaluate is IV, which S&C suggest represents the best conditions for democracy given the weak presidential powers on both LT and NLT dimensions. They note that most of their “premier-presidential” systems fall into this box, and code four of these countries as remaining stable in the period observed. Among premier-presidential systems, however, is also Haiti, which they ignore as too new a case to evaluate. Haiti, however, has not prospered, with the country suffering coups in 1991 and 2004, both times under President Aristide. More in favor of S&C’s thesis, the box also includes two other countries that the authors did not include in their analysis but have seen their democracies survive: Romania and Bulgaria. They also put in that box the proposed constitution for Argentina, which was approved in 1994. The country has had continuous democracy, but as noted, we follow others who have recoded that constitution such that it fits better in Region VI. Overall, then, we see one case of breakdown out of seven in the box that has the weakest presidents.

This result begs a comment on the debate surrounding semipresidential systems and regime stability. This literature was influenced by S&C’s disaggregation of semipresidential systems into premier-presidential and president-parliamentary systems on the basis of presidential cabinet power and separation of survival. For example, research has observed the diversity found within these regimes as it pertains to NLT powers to better understand instability (Roper, 2002). And while some scholars have analyzed the extent to which the differences between these two semipresidential types have generated varying degrees of intraexecutive conflict (Sedelius and Mashtaler, 2013), others have emphasized premier-presidential systems’ successful democratic record relative to presidential and parliamentary regimes (Sedelius and Linde, 2018). This debate, coupled with the success (excepting Haiti) of semipresidential regimes in Box IV, is suggestive of the importance of institutional design.

To summarize the findings, once accounting for S&Cs coding errors and political developments over the past 25 years, Table 1 compares the breakdown rates for each of the six boxes in the graph (see Appendix Table A1 for details). These results put in stark relief the conclusions of S&C. Their data show that those regimes with the highest scores in both LT and NLT powers, which fall into Region I, would be at the greatest risk. The revised results show, however, that Region VI (low LT and high NLT) represents the highest risk for democratic stability and Region I is only moderately worse than Region V. We exclude the 1994 proposal for the Argentine constitution in our analysis, but if we were to include it in Box I as another example of stability, the result would further undermine S&C’s conclusions. To summarize the differing results, Figure 2 includes the original and updated empirical results and summary labels. Most important, where S&C found Box I to be the most troubled, that label now fits Box VI. Box II has only three cases, but it and Box IV remain the most stable systems, respectively, with Haiti being the only country reporting a breakdown (Box III is empty). And although Region V remains safer relative to Regions I and VI, it also has a worrying breakdown rate.

OPEN IN VIEWER

Table 1.

Breakdown Rates: S&C Versus Updated Data.

Box #	S&C data	Updated
I	50	33
II	0	0
III	NA	NA
IV	0	14
V	25	20
VI	25	45

Rethinking the Dependent Variable: Breakdown Versus Democratic Quality

In this section we critique S&C’s key variable, “Democratic Breakdown,” and offer an alternative, “Democratic Quality,” which better captures the relationship between presidential power and democratic stability. S&C’s original dichotomous variable (democratic breakdown/no breakdown) was a reasonable choice, given that they were studying soon after many Latin America countries had transitioned to democracy, and it improved Linz’s analysis by considering a broader temporal frame. Rather than basing their analysis upon single-time “snapshots” of different democracies, S&C consider a failure as “a breakdown that led to a replacement of the regime by a non-democratic regime” (Shugart and Carey, 1992: 38). This means that countries which experienced temporary periods of authoritarianism or weak democracy did not count as failures. Instead, democracies had to institutionally transform into nondemocracies in order to be coded as failures.

This approach carries advantages and disadvantages. One may consider any democracy that is capable of backsliding into authoritarian-style leadership a failure. However, from an institutional perspective, it seems more prudent to take a longer term view. Historical events such as world wars or powerful political movements may subject democracies to periods of characteristically authoritarian politics that do not carry the institutional reforms necessary for outright dictatorship or anarchy. Should we treat these periods as democratic failures? Unless such periods were able to bring about a change in regime type, the possibility for a return to a more democratic politics remains available through institutional means. What S&C call “cases of reequilibriation,” such as Chile in 1925 and 1933, France in 1958, and India in 1972 may suggest the resiliency of democratic institutions rather than their failure (Shugart and Carey, 1992: 38–39). Another example is Uruguay; while its 1967 constitution is coded as a failure due to the breakdown in 1973, the country’s successful democracy since 1985 does not yield a positive label. This is problematic when compared with Argentina or others that implemented new constitutions after redemocratization because the method labels Uruguay, with an equally or longer lasting democracy, as worse than Argentina which has a failed constitution (of 1949) and a successful one (from 1994). Long success of a constitution, in short, should get weight in the model, and the dichotomous nature of the “democratic failure” variable that S&C use is too blunt an instrument to capture this type of nuance.

Another concern with the S&C’s study is that they fail to delineate a metric by which to measure success in terms of the time that a country survives as democratic. Thus, they have cases of poor-quality democracies that have survived just a few years as well as higher quality democracies that have avoided dictatorial interludes for an extended period of time. We reevaluate their important hypothesis, that the institutional framework of executive—LT relations affects democratic stability, by changing the dependent variable to the level of and change in the quality of democracy. In our analysis, therefore, we consider levels and changes in the degree of democracy (based on polity scores; Integrated Network for Societal Conflict Research (INSCR), 2016) to gain leverage in identifying which configurations of presidential powers deliver positive or negative tendencies in democratic conditions. Using polity scores to measure changes in “democratic quality” ultimately offers a far more sensitive output variable with which to test S&C’s original thesis.

Overall, our analysis demonstrates that countries’ democratic quality and stability varies considerably even when presidential powers remain constant. The empirical findings cast doubt on S&Cs initial hypothesis, but also show the need to rethink the dependent variable in a manner that considers levels and changes in the quality or stability of democracy.

To show the results, recall that Figure 1 includes the change in each country’s polity scores from 1992 to 2016, which we detail in Table 2. Polity scores range from −10 (most authoritarian) to 10 (most democratic). In just 14 years, for example, Mexico made significant improvements in its democracy score. On the other hand, Haiti, which started at a much lower baseline, has improved its democracy score, yet it is still lagging behind in its democracy score compared to other countries in the region. This country’s generally low polity score is problematic and highly volatile, reflecting Haiti’s democratic fragility. Polity also allows us to trace the deterioration of democracy in Venezuela, which already was in a trouble zone according to S&C. Honduras offers another interesting insight: although we decided to code the coup in 2006 as breakdown, it has managed to retain, if not improve, its democratic score (though the post-election conflicts of 2017 will likely reduce that score).

OPEN IN VIEWER

Table 2.

Summary of Polity Scores: 1992–2016.

Region	Country	Polity-1992	Polity-2016	Change in polity
I	Brazil	8	8	0
	Paraguay	7	9	2
	Chile	8	10	2
	Uruguay	10	10	0
	Mexico	0	8	8
	Guatemala	3	8	5
	Philippines	8	8	0
	Germany	10	10	0
	Sri Lanka	5	6	1
	Panama	8	9	1
II	Korea	6	8	2
	Ecuador	9	5	−4
	Colombia	9	7	−2
III	N/A
IV	Romania	5	9	4
	Portugal	10	10	0
	France	9	9	0
	Austria	10	10	0
	Haiti	−7	0	7
	Bulgaria	8	9	1
	Ireland	10	10	0
V	Finland	10	10	0
	Peru	−3	9	12
	Namibia	6	6	0
VI	Nicaragua	6	6	0
	United States	10	8	−2
	Cuba	−7	−7	0
	Venezuela	8	4	−4
	Bolivia	9	7	−2
	Honduras	6	7	1
	El Salvador	7	8	1
	Argentina	7	9	2
	Nigeria	−5	7	12
	Costa Rica	10	10	0
	Dom Republic	6	7	1

How do the predictions of S&C perform against the quality of democracy of their cases? The countries in the top-right (Region I) corner of Figure 1 were the ones at greatest risk according to S&C. When the change in quality of democracy is compared against that of the lower right corner (Region VI), we observe a pattern in the lines of the analysis in Figure 2. Region VI exhibited some democratic decline (with an average change of +.81, glacial democratic growth for a 24-year period that is dragged down by the backsliding of Bolivia, Venezuela, and the United States. Proportionally, however, it is the center-top (Region II) that reports the greatest fall in democracy (with an average change of −1.33), which is an additional insight we cannot observe when looking at “collapse” alone. We are not claiming that a declining polity score is always followed by breakdown. Instead, we propose a new perspective that allows a more dynamic process of democratic evolution and backsliding and calibrates our understanding of the processes immediately preceding breakdown.

A potential criticism is that we have plotted the change in democracy scores for two different time periods (1992 and 2016) without considering the possible changes in the LT and NLT powers of these presidential regimes over time. We have done so because our intention was not to update the original model’s prediction based on changes in the domestic institutions of the countries observed, but rather as a cautionary tale: we cannot take the original predictions as fixed over time.

Rethinking the Independent Variable

Moving from concerns about the dependent variable, we now consider methodological criticisms regarding how the authors indexed the independent variable: presidential powers. S&C chose five LT and four NLT powers, coded the president’s authority in each on a scale from 0 to 4, and then summed the different variables to provide summary power levels on the two dimensions. While an important advance in detailing the powers, the coding system has multiple flaws.

First, by summing categories to generate overall power levels, the authors clearly violated statistical premises. This procedure implicitly assumes that the categories have equal weight, but some of the powers are clearly more important to a president than others—a president who scores the top mark on decree powers and 0 on exclusive LT introduction is much stronger than one who has the reverse set of powers. A related problem is that this system implicitly means that the difference between, say, a 1 and a 2 is the same as between a 2 and 3 for any measure. This is clearly not correct, moving from the absence of a power to the presence, even if circumscribed, is different than changing the details of the circumscription.

The authors do recognize the problems of implicit equal weighing among and between the powers but argue that an unweighted measure is “preferable to a purely nonquantitative impressionistic ranking” (p. 149). This, however, is only true if the quantitative measure provides correct inferences. If they are misleading, then the quantitative measures would not be an improvement. This should not lead to the conclusion that quantifying powers is not a useful exercise; au contraire, by doing so, we have the opportunity to examine and compare specific powers and the combination thereof, which is crucial to the comparative enterprise. But because the indices are intended for rankings and hypothesis testing, they must be specified clearly and subjected to testing. A specific benefit of quantification is that it lays bare the presumptions, thus providing clear opportunities for criticism and improvement.

A second (and related) problem with the additive S&C index is that presidential powers could be substitutes for one another, such that a high level of one obviates the need for another (Morgenstern et al., 2013). As such, while presidents who score high on all powers would be strong (and the reverse) it may not be the case that presidents who rate high on a few powers and low on others would be weak. Decree powers again provide a clear example: a president whose total score was 4 could be stronger than one who scores 8, if the former scored the top value on decree powers and 0 on the others with the latter scoring 0 on decree power but high or medium values with relation to initiative rights, veto powers, or others.

While data limitations could hinder analysts’ ability to empirically separate the potential independent and interactive role of each power, we can consider some theoretical fixes. S&C provide one such solution in their very highly cited paper on electoral systems (Carey and Shugart, 1995). In the paper that came out 3 years after the book, the authors ranked combinations of variables, instead of summing the traits that they had coded. This led to 13 combinations of the three variables that they coded. They then ranked those codes according to a theoretical consideration of the combinations. Importantly, this yielded a very different ranking than they would have generated by summing the traits. Table 3 shows, for example, that the summation of powers would have yielded a large group of countries with a ranking of 4, some of which S&C conclude as having more incentives to cultivate a personal vote than others that would score 5 by summing the categories.

OPEN IN VIEWER

Table 3.

Theoretical and Additive Rankings of S&C’s Coding of Electoral Systems.

Ballot	Pool	Votes	Sum	S&C rank	Rank by sum
0	0	0	0	1	1
0	0	1	1	2	2
1	0	1	2	3	3
1	1	1	3	4	4
1	0	2	3	5	4
1	2	1	4	6	5
2	0	1	3	7	4
2	1	1	4	8	5
2	0	2	4	9	5
2	2	1	4	10	5
1	2	2	5	11	4
2	1	2	5	12	4
2	2	2	6	13	6

The next issue with regard to the powers index is S&C’s decision to divide the powers into two types. Here, our quarrel is more with authors who have followed S&C who have sometimes combined the two scales into a single dimension. S&C justify these two dimensions by explaining that LT powers provide a view of control over legislation while NLT powers define the separation of powers. As such, the latter helps to separate pure presidentialism from hybrid or parliamentary systems. ²

This is a useful and important theoretical approach, which others sometimes ignore. For example, while Hicken and Stoll (2008, 2011) include a variable that indicates whether regimes are presidential, mixed, or parliamentary, they combine all of the powers in the S&C index (plus one other) into a single scale.

Presumably, S&C created the two dimensions because they were worried, at least implicitly, that the two scales were not correlated. Following this direction, plus the concern with adding measures, Fortin (2013) created an index based on the correlated variables. She argues that if the powers are related, then there is validity in adding them up. It also then justifies her factor analysis, which reduces the number of indicators. The latter of these methods seems unnecessary, since it essentially provides a summary measure, as the additive measure does. The former issue is a more important critique, but it ignores the substitutability argument—maybe presidents are not more powerful when they possess two correlated powers, if one of them is sufficient to allow the president to direct legislation.

Another issue regarding the power index is that S&C did not fully justify which powers to study. They seem to imply that the list is comprehensive, but while S&C chose five LT powers for their study, Hicken and Stoll include six and Fish and Kroenig (2009) include dozens. What is the proper set of powers?

A final problem is the difficulty in coding practice versus parchment. This concern has become evident in the S&C coding of Argentina, where the president in charge at the time they were writing (Menem) was a serial-user of decrees, yet the constitution did not recognize such a power. In this case the breech between the constitution and practice was the result of a court decision (which Menen had packed) that validated the use of decrees by saying that the constitution did not explicitly prohibit the action.

While these problems are easy to identify, multiple authors have continued to use some sort of additive ranking. Table 4 summarizes a few prominent studies.

OPEN IN VIEWER

Table 4.

Theoretical and Additive Rankings of S&C’s Coding of Electoral Systems.

Author (year)	Measure	Additive/combinatorial	Dependent variable
Clark and Wittrock (2005: 182–183)	Scored 0–3 (cabinet formation/dismissal, assembly survival, presidential survival)	Additive	Effective number of parties
Doyle and Elgie (2016)		Meta-analysis; factor analysis
Fish and Kroenig (2009)	Expert survey based on impressions of powers	Additive
McGregor (1994)	List of dichotomous variables based on constitution	Additive	None—legal study
Frye (2002)	Scored 0–3, same as S&C but adopted for postcommunist systems	Additive	Economic growth
Siaroff (2003)	Nine dichotomous variables {0,1}	Additive	None—typology
Hicken and Stoll (2008) (and Stoll and Hicken)	10 variables, scored 0–4	Additive	Number of presidential candidates
Metcalf (2000)	11 variables, scored 0–4	Additive	Democratic consolidation
Taghiyev (2006)	15 variables, scored 0–4	Additive	Democracy score, Freedom House
Frye et al. (2000)	10 variables, 0–4	Additive	None—database.

Among these, Doyle and Elgie (D&E; 2016) do a meta-analysis of over 30 studies to develop their ranking, but this results in their summing of other authors’ errors. They also make the error of combining the two dimensions into one. Their scores are highly correlated with S&C, but there are some significant outliers, as shown in Figure 3.

OPEN IN VIEWER

Figure 3.

Correlation of Doyle and Elgie Rankings with those of Shugart and Carey.

This is based on ranking of S&C cases, versus ranking of Doyle. Specifically, a Spearman’s (rank) test shows that S&C LT scores are correlated with the D&E (normalized) rankings, but tests with neither the NLT powers nor the summation of powers yield a correlation.

There is clearly a trend in the cloud of points, but especially at the ranks of 4 and 6 for S&C, there is a wide variance in how the cases are coded by D&E. As an example, while S&C code both the United States and Argentina a 4, D&E find that the former is a 10 and the latter a 30. Only four countries earn a 6 ranking for S&C, but these range from a D&E ranking of 5 for Guatemala to 32 for Portugal.

An Alternative Multivariate Approach

A different approach to the problem is to use a multivariate analysis to assess the weighting of the variables. Such an approach would not only provide a direct means for testing the thesis, it would also deal with the problem of substitutability among variables and the presumption of similar weights on the different powers. Theoretically this is not a very difficult problem, but it is a challenge to estimate the model due to data limitations. Yet, in what follows we develop the theoretical model and give a first attempt to resolve some of the empirical challenges.

To begin with, for simplicity of explanation we assume a model of presidential powers, X₁, X₂, and X₃. The S&C thesis implicitly tests a model that first sums the variables into a combined power index, X_c, and then tests a model where democracy is a function of that index, or

Y D = a + B 1 X c + e

where X_c = X₁ + X₂ + X₃ and where Y_D = “democratic quality” and X_c = executive powers.

This is a highly restricted hypothesis, as it presumes a single coefficient for three variables with an unexplored relationship among them.

The S&C model is slightly more complicated, in that their thesis suggests that the interaction of the total LT and NLT powers threatens democracy. That is

Y D = a + B 1 X LT + B 2 x NLT + B 3 X LT X NLT + e

Again, however, there is a single coefficient (or weight) for each independent power within the index (plus another for the interaction of the interacted totals). We do not know, therefore, if decree powers and exclusive bill initiation rights have the same impact on democratic performance.

A simple solution, at least theoretically, is to develop an unrestricted model that considers the independent role of each variable, plus the ways that those variables interact. The implied general model, assuming just three powers, would be based on the following

Y D = a _ + B 1 X 1 + B 2 X 2 + B 3 X 3 + B 4 X 1 X 2 + B 5 X 2 X 3 + B 6 X 1 X 3 + B 7 X 1 X 2 X 3 + e

where Y_D = “democratic quality” and X_c = executive powers and where each X is a different LT or NLT power. Under such a model, each coefficient would provide a weight with which to judge the value of each variable and interaction. As such, the model would allow a general means to assess the impact of presidents’ powers both individually and collectively.

There are several ways that this could be specified, based on theories of substitutability, interactions, and independence of the variables. The key implication is that these models have multiple terms that have potential independent and/or correlated impacts on democratic breakdown (or other dependent variables). A multivariate regression, then, should determine the size of those impacts, helping to sort out whether some powers are more critical than others, whether combinations of the variables is critical, and whether the powers are substitutes for one another.

One potential model, in simplified form, would be

Y D = ( 1 − θ 1 X 2 ) ( 1 − θ 2 X 2 ) ( 1 − θ 3 X 3 )

where Y_D = “democratic quality” and θ = presence/absence of X_c = executive powers.

Such a model would produce coefficients that would allow empirical evaluation of, for example, whether a president who has strong decree power but lacks independent initiative is as strong as one who has the reverse set of powers.

While this model has “only” 7 parameters, the full model would be very difficult to estimate and test, since S&C created their indices out of nine separate variables (and they provided only about 30 cases for the test). We can, however, specify the model, and begin the empirical testing. In what follows we provide some limited results and provide more in an online Appendix (which also details the definition of variables). To conduct the tests, we first run the basic model implied in S&C, where the interaction of LT and NLT powers explains the proclivity to breakdown. We then test the impact of the individual powers on breakdown, and test the Linzian hypothesis that both strong and weak presidents are dangerous to democracy. After this, we test other dependent variables (such as the change in the Polity index) and begin testing the impact of individual presidential powers.

Our tests largely fail, which is in line with the graphical findings we discussed earlier. We are not, however, ready to fully discount the relation of presidential powers and democratic stability, because the data are too limited for a full test. Our regressions provide a roadmap for testing, but because they are based on just a few dozen cases, we are unable to implement the full model. Thus, while these initial tests lend weight to an alternative hypothesis—that stronger presidents are not more dangerous to democracy than weaker ones—we are not yet ready to discard the S&C proposal.

An alternative approach, given the limited data, might pursue a qualitative comparative analysis (Ragin, 2008). We have not attempted that here, primarily because we are interested in identifying the precise impact of each individual variable, for which the configurational approach of qualitative comparative analysis (QCA) is not well-equipped. In addition, the high number of conditions relative to cases would damage the leverage of a QCA. According to Berg-Schlosser and de Meur (2009), the appropriate number of conditions for an “intermediate-n” analysis (10–40 cases) such as ours is six, yet a proper test of S&C requires far more given that they test both LT and NLT executive powers. This being said, we recognize the potential benefits of a QCA approach for evaluating our argument, which is rooted in the substitutability of institutional conditions. Therefore, we invite future scholars to move in this direction.

To show the implied multivariate regression from S&C, the first logistic regression in Table 5 regresses the two power indices and their interaction against a dummy variable indicating breakdown. The second regression changes the dependent variable by recoding a few cases of breakdown to which S&C gave a questionable coding, and the third updates the data with new cases. The final two regressions change the dependent variable to capture the quality of democracy by using the Polity index and its change (capped at 8 to prevent bias by exceptional cases). Each of these (linear) tests fails to show statistically significant coefficients. In Table 6, we then test the Linzian hypothesis that the relationship is not linear by modifying the independent variables such that the extremes have equal weight and the middle is considered low. These alternative tests also fail.

OPEN IN VIEWER

Table 5.

Total Legislative and Nonlegislative Powers and Democratic Breakdown: S&C.

	(1) a	(2) a	(3) a	(4)	(5)
	Breakdown S&C	BreakdownUpdate 1	BreakdownUpdate 2	Polity 2Original	Polity 2Change
Total legislative	0.22	−0.35	−0.47	−1.28	−1.28
Total nonlegislative	0.26	0.08	0.15	−0.13	−0.23
Legislative × nonlegislative	−0.02	0.03	0.03	0.10	0.12 +
Constant	−4.06	−1.43	−1.65	3.04	4.09**
N	43	45	45	33	33
R 2	0.07	0.03	0.07	0.06	0.11

a

Models (1), (2), and (3) are logistic regressions.

+

p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001.

OPEN IN VIEWER

Table 6.

Total Legislative and Nonlegislative Powers and Democratic Breakdown: Linz.

	(1) a	(2) a	(3) a	(4)	(5)
	Breakdown S&C	BreakdownUpdate 1	BreakdownUpdate 2	Polity 2Original	Polity 2Change
U-shaped total legislative	0.17	0.08	0.18	0.64	0.49
U-shaped total nonlegislative	0.66	0.28	0.46	1.43	1.25
U-shaped legislative × nonlegislative	−0.08	−0.02	−0.04	−0.13	−0.11
Constant	−2.62	−1.93	−2.74	−5.17	−3.49
N	43	45	45	33	33
R 2	0.05	0.01	0.02	0.07	0.09

a

Models (1), (2), and (3) are logistic regressions.

+

p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001.

Next, Tables 7 and 8 modify the model to test the impact of individual presidential powers. An ideal regression would include all of the powers individually and the interactions among them, as we indicated earlier. The dataset of just 45 observations, however, required that we test just one or two variables at a time. We present the results from what we thought would be the most theoretically important variables for presidential systems, decree and dissolution powers, first with the linear (S&C model and then with the U-shaped Linzean formulation of the variables. These tests fail, too.

OPEN IN VIEWER

Table 7.

Decree and Dissolution Powers and Democratic Breakdown: S&C.

	(1) a	(2) a	(3) a	(4)	(5)
	Breakdown S&C	BreakdownUpdate 1	BreakdownUpdate 2	Polity 2Original	Polity 2Change
Decree	−0.93	−1.32	−1.15	−0.78	−0.49
Dissolution	−0.60	−0.79 +	−1.19 +	−0.30	0.03
Decree × dissolution	0.58	0.71 +	0.84*	0.29	0.09
Constant	−0.61	−0.13	0.19	1.95	2.07**
N	43	45	45	33	33
R 2	0.09	0.13	0.17	0.03	0.02

a

Models (1), (2), and (3) are logistic regressions.

+

p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001.

OPEN IN VIEWER

Table 8.

Decree and Dissolution Powers and Democratic Breakdown: Linz.

	(1) a	(2) a	(3) a	(4)	(5)
	Breakdown S&C	BreakdownUpdate 1	BreakdownUpdate 2	Polity 2Original	Polity 2Change
U-shaped decree	30.53		27.33	3.62	0.69
U-shaped dissolution	29.94		27.51	2.43	−1.03
U-shaped decree × dissolution	−15.29		−13.48	−1.52	0.09
Constant	−61.09		−56.18	4.532	−0.24
N	43		45	33	33
R 2	0.03		0.03	0.02	0.06

a

Models (1), (2), and (3) are logistic regressions. Model (2) failed to converge.

+

p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001.

The null results suggest weaknesses in the S&C findings that weaker presidents are better for democratic survival, and they also fail to support the Linzian hypothesis that all presidential systems are problematic. Our results are insufficient to fully discard these authors’ conclusions, because our tests are based on limited data, and coding rules for the data are not straightforward. Especially given the small size of the dataset, changing the case selection or the coding rules can yield different results. Furthermore, as we have argued, there are important left out variables in these models. Yet, our findings, which are based on multivariate tests that the other authors sidestepped, do show the lack of a clear and strong relationship between presidentialism and democratic breakdown or quality. The ball is back, therefore, in the court of the critics of presidentialism.

Conclusion

After Linz developed a damning critique of presidentialism, S&C provided a provocative correction, explaining that presidential systems come in a variety of styles, and that only one particular type—those with particularly strong executives—was dangerous to the survival of democracy. In this article we ask if their findings hold up, given data updated by a quarter of a century plus a reconsideration of their methodology and empirical strategy. In so doing, we try to stay true to their main project, by focusing on the relation between formal constitutional powers and the survival of democracy. As such, we thus avoid entering into discussion with the recent literature that emphasizes the importance of informal institutions, international context, economic resources, or other factors not found in the constitutions.

Using their measures but updated data, we do not find evidence to support the S&C original conclusions that systems with stronger presidents are more likely to break down. This finding, however, is not a full refutation of the S&C hypothesis, because there are several empirical and methodological problems with the analysis.

First, the tests might have failed due to data limitations. The universe of presidential systems is relatively small, preventing a test of the complete theoretical model. Second, perhaps Linz was right, that the relation of presidential powers and democratic breakdown is nonlinear; that is, both strong and weak presidents are problematic; our tests of this thesis also fail, but we do not attempt a comparison with parliamentary systems. Third, Linz could be incorrect with respect to both ends of the scale, and presidential systems are either immaterial to breakdown or breakdown among presidential democracies might be contingent on other factors, such as the level of development (Przeworski et al., 2000), the party system (Mainwaring and Shugart, 1997), or informal institutions (Helmke and Levitsky, 2006). Next the S&C powers index might fail to capture meaningful differences in presidential powers. Are presidents who score low on this index truly weak? As we have pointed out, a president with decree powers but no formal ability to propose new legislation is more powerful than one with the reverse powers. They also break methodological rules by summing different scales into a total power index. We thus propose, but are unable to test, a new technique.

While our updated tests of S&C’s findings do not provide support for their thesis, we are not ready to reject their critical theory. Instead, we hope that their influential thesis combined without our empirical and methodological challenges will lead to a full reevaluation of their theory. That reevaluation is important not only to the debate over presidentialism, but to the larger critical questions of constitutional design and the role of institutions.

Supplemental Material