Svolik2008p2979

8/7/2019 Svolik2008p2979

1/16

American Political Science Review Vol. 102, No. 2 May 2008

doi:10.1017/S0003055408080143

Authoritarian Reversals and Democratic ConsolidationMILAN SVOLIK University of Illinois at UrbanaChampaign

Ipresent a new empirical approach to the study of democratic consolidation. This approach leads tonew insights into the determinants of democratic consolidation that cannot be obtained with existingtechniques. I distinguish between democracies that survive because they are consolidated and those

democracies that are not consolidated but survive because of some favorable circumstances. As a result,

I can identify the determinants of two related yet distinct processes: the likelihood that a democracyconsolidates, and the timing of authoritarian reversals in democracies that are not consolidated. I findthat the level of economic development, type of democratic executive, and type of authoritarian pastdetermine whether a democracy consolidates, but have no effect on the timing of reversals in democraciesthat are not consolidated. That risk is only associated with economic recessions. I also find that existingstudies greatly underestimate the risk of early reversals while simultaneously overestimating the risk oflate reversals, and that a large number of existing democracies are in fact consolidated.

Why do some democracies survive for morethan a century, whereas others revert to dicta-torship after only a brief democratic period?

Academic debate and policy recommendations for newdemocracies frequently look to long-lived democra-cies such as the United States or Switzerland for cluesabout which institutional or economic factors may im-prove the survival of democracies after transition. Infact, a large amount of both theoretical and qualita-tive empirical research focuses precisely on such long-lived democracies and attempts to explain what distin-guishes them from new or failed democracies (see e.g.,Huntington 1991; Linz and Stepan 1996).

The premise underlying this focus on long-liveddemocracies is that their advanced age is an indica-tor of the enduring stability of democracy in thesecountriesthat they are consolidateddemocracies. Al-though substantial disagreement persists about the

exact causes or appropriate measures of democraticconsolidation, most research agrees that consolidateddemocracies face essentially no risk of an authoritar-ian reversal.1 But then even a long-lived democracymay be surviving for two different reasons: it may beeither a consolidated democracy whose odds of revert-ing to dictatorship are essentially zero (e.g., Swedenin 2001), or a democracy that is not consolidated, butsurvives because of some favorable circumstances (e.g.,Thailand in 2001).

However, the influential empirical literature on tran-sitions to democracy treats all existing democracies asa single group: after controlling for various covariates,

Milan Svolik is Assistant Professor, Department of PoliticalScience, University of Illinois at UrbanaChampaign. Address:361 Lincoln Hall, 702 South Wright Street, Urbana IL 61801([email protected]).

I would like to thank three anonymous referees, the editors, Jos eCheibub, Zach Elkins, Brian Gaines, Jude Hays, Woo Chang Kang,Anbal P erez-Lin an, Andrew Schrank, Jay Ulfelder, Bonnie Weir,the participants at the IPES Inaugural Conference at Princeton, theComparative Politics Workshop at the University of Illinois, and theMPSA Conference forhelpful comments. I am also grateful to CarlesBoix, Jos e Cheibub, and Jennifer Gandhi for sharing their data, and

to Seden Akcinaroglu and Tatiana Svolkova for research assistance.1 See, for example, Diamond 1999, Gasiorowski and Power 1998,ODonne and Sehmitter 1986, ODonnell 1996, Przeworski 1991,Rustow 1970, Schedler 1998, and Weingast 1997.

alldemocracies are expected to face the same risk of areversal (see e.g., Przeworski et al. 2000). This failureto account for how the potential heterogeneity amongdemocracies translates into observable data misses animportant dynamic in the process of democratic sur-vival. The observed survival of democracy may be theconsequence of two distinct causal mechanisms: demo-cratic consolidation, which practically eliminates therisk of an authoritarian reversal, or a separate mech-anism that prevents authoritarian reversals in thosedemocracies that are not consolidated.

As a result, the factors that determine whether ademocracy will consolidate may differ from those thatexplain the occurrence and timing of authoritarian re-versals in those democracies that are not consolidated.This distinction may seem subtle, but it is crucial toour understanding of democratic survival. A medicalanalogy may help highlight the importance of this dis-

tinction: consider an individual who survived one ofthe later waves of the Black Death in Europe (e.g., theplague wave of 1383 versus the original 1348 wave).She may have survived because (1) she developed animmunity to the plague during the prior wave(s) of theepidemic or because (2) she practiced careful hygieneand, in turn, minimized her exposure to the conta-gion.2 Although bothalternatives explain survival,theyare clearly two, distinct causal mechanisms. Similarly,the distinction between the two separate mechanismsthat may account for the survival of democracy is lostwhen we treat all existing democracies as a single, ho-mogenous group. Failure to distinguish between thesemechanisms not may only lead to incorrect statisticalestimates, but also confounds what is of central inter-est in the study of democratic survival: the causes ofdemocratic consolidation.

In this paper, I establish a new approach to the em-pirical study of democratic survival that is designedspecifically to addresses the concerns laid out in theabove discussion. I assume that the population of ex-isting democracies consists of democracies that aretransitionaland face a positive risk of an authoritarian

2 Cohn (2002) documents a decline in mortality over the BlackDeaths first 100 years, most likely due to natural or acquired im-munity.

153

8/7/2019 Svolik2008p2979

2/16

Authoritarian Reversals and Democratic Consolidation May 2008

reversal, and of democracies that are consolidated anddo not face the risk of an authoritarian reversal. Asa result, this approach is explicit about the differ-ence between consolidation and the observed survivalof democracy: consolidation amounts to being im-mune to the causes of democratic breakdowns, yetdemocracies that are not consolidatedtransitionaldemocraciesmay also survive because of some fa-

vorable circumstances.I approach the data on democratic survival realisti-cally anddo notassume that we canobservewhether anexisting democracy is transitional or consolidated. Thisunobservability implies a departure from the existingempirical literature on democratic transitions in oneimportant way: rather then being a single population,the existing democracies are a mixture of transitionaland consolidated democracies, and each group facesvery different odds of reverting to dictatorship. Alldemocracies that reverted to dictatorship were, by as-sumption, transitional. On the other hand, whether anexisting democracy is transitional or consolidated mustbe inferred from the data. Thus this approach neither

presumes that consolidation actually occurs nor does itimpose an arbitrary criterion on how old a democracymust be in order to be considered consolidated.

I argue that an empirical analysis based on thesesimple assumptions has several advantages over theexisting approach to the study of democratic survival. Ifind strong evidence that some democracies face qual-itatively different odds of survival: a large number ofexisting democracies are in fact consolidated. By con-trast, other democracies are transitional and may revertto dictatorship. Moreoverand quite intuitivelyourconfidence that an existing democracy is consolidatedincreases with its age. I also find that studies that do

not allow for this unobserved heterogeneity amongdemocracies greatly underestimate the risk of earlyauthoritarianreversals whilethey overestimate the riskof late reversals.

Importantly, I show that the factors that explainwhether a democracy is consolidated differ from thosethat explain the variation in the hazard of authoritar-ian reversals in transitional democracies. More specifi-cally, I find that low levels of economic development,a presidential executive, and a military authoritarianpast reduce the odds that a democracy consolidates.However, a transitional democracy experiences a re-versal primarily as a consequence of an economic re-cession. My results therefore strongly suggest that in

order to explain the survival of democracy, we shouldbuild separate theoretical models that account for theonset of authoritarian reversals, on the one hand, andwhy only some democracies consolidate, on the other.I now turn to a more detailed discussion of my methodsand findings.

MAIN FINDINGS AND THEIR CONTRIBUTIONTO EXISTING RESEARCH

The results in this paper are obtained using a split-population survival model. In the present context, akey feature of this technique is the assumption that

the population of existing democracies is a mixtureof those that will ultimately revert to dictatorship andthose that do not face the risk of a reversal. In contrast,standard survival models assume that each observa-tion ultimately experiences the event of interestinthis case, an authoritarian reversal.3 Although we can-not observe whether an existing democracy is transi-tional or consolidated, the split-population model ex-

ploits the fact that this unobserved heterogeneity im-plies differentexpectations about the survival of transi-tional and consolidated democracies. Importantly, thismethod does not rule out the possibility that there areno consolidated democracies in the data; whether anyamong the currently existing democracies are consol-idated is assessed using standard statistical tests. As Idemonstrate, however, the data on democratic survivalstrongly indicate that a substantial number of existingdemocracies are in fact consolidated.

Survival analysis techniques that model this form ofunobservable heterogeneity have been developed inbiostatistics (Farewell 1977) and are called cure ratemodels.4 In this literature, individuals that are not at

risk of experiencing the event of interest are referredto as cured, immune, or long-term survivors. Inthe social sciences, such models are referred to assplit-population models and have been applied to the studyof criminal recidivism (Schmidt and Witte 1989), addic-tion (Douglas 1998; Forster and Jones 2001), and long-term unemployment(Yamaguchi 1992). In political sci-ence, I am only aware of the use of this technique inthe study of campaign financing by Box-Steffensmeier,Radcliffe, and Bartels, (2005) and civil wars by Findleyand Teo (2006). Here I preserve the socialscientificnomenclature and refer to this class of models as split-population models.

The new empirical approach in this paper allows meto uncover patterns in the data on democratic sur-vival that methods typically employed in the relatedliterature do not reveal. My key substantive findingsconcern a relationship that is at the heart of a promi-nent debate in comparative politics: the dynamics ofdemocratic survival and its economic and institutionaldeterminants. Additionally, I argue that the presentapproach brings the quantitative study of democraticconsolidation closer to the qualitative empirical andtheoretical research and also provides a better statis-tical fit to the data on democratic survival. I brieflypresent these findings before turning to the empiricalanalysis itself.

A central debate in the literature on transitions todemocracy concerns the effect of economic conditionson the survival of new democracies (see e.g., Boixand Stokes 2003; Przeworski et al. 2000). In a sem-inal contribution, Przeworski et al. (Przeworski andLimongi 1997; 2000) make the influential claim thatthe level of economic development affects the survival

3 For an introduction to survival models, see, for example, Box-Steffensmeier and Jones 2004.4 See Maller and Zhou 1996 for an overview, Ibrahim, Chen, andSinha 2001 for a Bayesian treatment, and McLachlan and Peel 2000,Chapter 10, for a general overview of mixture survival models.

154

8/7/2019 Svolik2008p2979

3/16

American Political Science Review Vol. 102, No. 2

of democracy but not the transition from dictatorshipto democracy. The empirical model in this paper al-lows me to address the first part of this claim from anew direction: Do we observe a positive association be-tween economic development and democratic survivalprimarily because (1) wealth improves the odds that ademocracy consolidates, (2) a high level of economicdevelopment actually lowers the hazard of reversals in

transitional democracies, or (3) both effects hold?The analysis in this paper answers the above ques-tion in favor of the first alternative. Specifically, eco-nomic development affects consolidation, and there-fore whether a democracy faces a risk of a reversal,but it does not help us explain when a reversal mightoccur in transitional democracies. Instead, I find thatthe eventual timing of reversals is associated only witheconomic recessions. Conversely, recessions have noeffect on whether a democracy is consolidated. In otherwords, the level of economic development determinesthe extent to which a democracy is susceptible to therisk of a reversal, but the eventual timing of reversalsis only associated with economic recessions. Although

previous research finds that both the level of economicdevelopment and the economic growth are positivelyassociated with democratic survival,5 my approach al-lows me to disaggregate the impact of these economicfactors on two related yet distinct dynamics: the like-lihood of democratic consolidation and the survival oftransitional democracies.

How do political institutions affect the survival ofnew democracies? Another prominent debate in com-parative politics concerns the effect of past authoritar-ian institutions and the type of democratic executivelegislative relations on the survival of democracy (seee.g., Mainwaring and Shugart 1997; Cheibub 2007). In

particular, the effect of presidentialism on democraticsurvival remains controversial. In a series of influentialarguments, Linz (1994) elaborates on the many waysin which presidential systems are more prone to demo-cratic breakdown than parliamentary ones.

Yet existing empirical research is not unanimous re-garding the negative effect of presidentialism on demo-cratic survival. Although presidential democracies failat a higher rate than those with other types of execu-tives (Mainwaring and Shugart 1997; Przeworski et al.2000), this correlation may be due to the potentiallyconfounding effect of other economic and institutionalfactors. Boix (2003, 15055), for instance, finds thatthe effect of presidentialism on democratic survival

depends on an unfavorable type and distribution ofeconomic assets, but he finds no independent, neg-ative effect of presidentialism. Focusing directly onthe relationship between presidentialism and demo-cratic survival, Cheibub (2007) revisits the findings inPrzeworski et al. (2000, 12836) and after a series ofnew empirical tests concludes that what kills democ-

5 See, e.g., Bernhard, Nordstrom, and Reenock 2001, Boix 2003,Boix and Stokes 2003, Gasiorowski 1995, Gasiorowski and Power1998, Haggard and Kaufman 1995, Londregan and Poole 1996, andPrzeworski et al. 2000.

racies is not presidentialism but rather their militarylegacy (Cheibub 2007, 140).

The present model allows me to investigate the rela-tionship between political institutions and democraticsurvival in a novel way: is the effect of political institu-tions (1) direct, so as to raise the hazardof authoritarianreversals in transitional democracies; (2) indirect, soas to make democracies more or less susceptible to

otherfactors that will eventually lead to a democraticbreakdown; or (3) is it both?I find that both a military past and presidential ex-

ecutive have a large, negative, and independent effecton a democracys susceptibility to reversalsthat is,on a democracys chances of being consolidated ratherthan transitional. However, neither a military past nora presidential executive have any direct effect on thehazard of authoritarian reversals faced by transitionaldemocracies. As mentioned earlier, the primary factorassociated with the timing of authoritarian reversals iseconomic recession. Therefore, if a democracy revertsto dictatorship, its death certificate will most likelyrecord an economic recession as the immediate cause

of thereversal, despite thefact that political institutionsmay have had the key, ifindirect, effect on its survival.

Building on these results, I address a question that isat theheartof thestudyof democratic survival: supposea country has been democratic for T years. Can weconclude that this country is consolidated?6 Below, Ishow that the probability that an existing democracy isconsolidated indeed increases with its age. Importantly,however, the economic and institutional history of ademocracy considerably moderates the positive effectthat age alone has on our confidence that it is consol-idated. Both an unfortunate or a lucky combinationof economic and institutional factors overwhelms any

information conveyed by the continuing survival of ademocracy. Thus in contrast to Epstein et al. (2006)and Przeworski et al. (2000), I find that the age ofa democracy is in fact associated with an increase inthe odds of its survival. As a result, qualitative studiesof long-lived democracies searching for clues that willimprove the survival of new democracies are condi-tionally warranted.

To further illustrate how the model here differsfrom existing approaches to democratic consolidation,consider the following result of my analysis: an ex-isting democracy that survived an economic recessionis more likely to be consolidated than a democracythat survived an expansion. This may seem rather

straightforwardif a democracy overcame a crisis, ourbelief about its enduring stability grows more than ifit experienced good times only. But note that exist-ing empirical models imply only that recessions leadto democratic breakdowns (see e.g., Bernhard, Nord-strom, and Reenock 2001, Gasiorowski 1995) and havenothing to say about what to infer from the fact thata democracy has in fact survived a recession. In con-trast, the present approach obtains this intuitive resultwithin a unified empirical model that is explicit about

6 See, e.g. Gasiorowski and Power 1998, Huntington 1991, Rustow1970, and Schedler 1998.

155

8/7/2019 Svolik2008p2979

4/16


the different odds of survival faced by transitional andconsolidated democracies.

Although I have focused on my substantive findingsso far, I argue throughout the paper that the empiricalapproach here provides a notably better statistical fitto the data on democratic survival than do existingmodels. For instance, I show that new democraciesface a risk of an early reversal that is much greater

than would be expected using existing methods. At thesame time,thesemethods overestimatethe risk of a latereversal. This is because an analysis that ignores thepotential but unobservable presence of consolidateddemocracies assumes that long-liveddemocracies, suchas the United Kingdom, will eventually revert to dicta-torship. By contrast, the present model allows for thepossibility that some democracies will not revert andtherefore leads to a more informative view of the dataon democratic survival.

For instance, summary statistics based on this modelprovide information about the expected lifetime ofdemocracies that actually are at risk of a reversal andallow us to predict which countries belong to that

group. Specifically, my analysis indicates that at themedian levels of the economic and institutional covari-ates, about one-half of the transitional democracies willrevert to dictatorship by their 14th year. In contrast,for these same democracies, an analysis that does notaccount for the existence of consolidated democraciespredicts a median survival age of approximately 57years!7 Thus existing models exaggerate the longevityof transitional democracies by more than fourfold andare consequently too optimistic about new democra-cies odds of survival.

I now turn to a brief summary of the data on demo-cratic survival. Next, I discuss the assumptions under-

lying split-population models. I then present my em-pirical analysis and detailed results. I conclude with areview of my central findings and suggestions for futureresearch.

DATA ON DEMOCRATIC SURVIVAL

My data on democratic survival are based on theregime type data compiled by Boix and Rosato (2001).Although there are other widely used datasets onregime type (see e.g., Marshall and Jaggers 2003; Prze-worski et al. 2000), the Boix and Rosato dataset coversan extensive period(18001994) andcodes regime typeas a binary variable based on explicit institutional cri-teria.8 I extended the temporal coverage of these data

7 This comparison is based on an analysis with covariates that usesthe log-logistic parameterization. By median values of covariates,I mean a country with median values for GDP per capita and GDP

growth and modal values for the type of democratic executive andpast authoritarian regime.8 Boix and Rosato 2001 define a country as a democracy if it meetsthree conditions: (1) the legislature is elected in free, multipartyelections; (2) the executive is directly or indirectly elected in popularelectionsand is responsible either directly to votersor to a legislatureelected according to the first condition; (3) at least 50% of adult menhave the right to vote (Boix 2003, 66). I used these criteria whenextending the data. For a detailed discussion of regime type data see

FIGURE 1. Distribution of Survival Time ofCurrently Existing and Failed Democracies,17892001

0

10

20

30

40

0

10

20

30

40

0 25 50 75 100 125 150 175 200

Currently existing democracies

Failed democracies

Frequency

Time in years

backward to 1789 and forward to 2001 in order to coveressentially the entire existence of modern democracy.

In order to study democratic consolidation and au-thoritarian reversals, I reshaped the data to consist ofdemocratic spells only. That is, the data contain alldemocracies from the year of their democratic tran-sition to the year of their authoritarian reversal, if oneoccurred. If no reversal occurred in a country, then thelast year in the dataset is the last recorded year for thatcountry.9 The resulting data consist of 193 democraticspells in 133 countries.

As I emphasized earlier, an important feature of thedata on democratic survival is that we cannot observe

whether currently existing democracies are transitionalor consolidated. By assumption, spells that have endedin a reversal are transitional. In terms of the vocab-ulary to be used later, these observations are uncen-sored. On the other hand, democracies that have notreverted by their last observed year are right-censoredobservations. These right-censored observations maybe either transitional democracies which, if observedlong enough, would eventually revert, or consolidateddemocracies that will never revert to dictatorship.

In Figure 1, I separately plot the distribution of sur-vival time of currently existing and failed democra-cies. Of the 193 democratic spells in the data, a 62%majority are currently existing democracies. Of these,a substantial number remain democratic after morethan 30 years. This long-lived group makes up almost

Boix 2003, Boix and Stokes 2003, Elkins 2000, Epstein et al. 2006,Marshall and Jaggers 2003, and Przeworski et al. 2000.9 Insomecases, a short-termlossof sovereignty due to a war leads tothe splitting of a spell. In order to avoid conflating an authoritarianreversal with the termination of democracy due to a temporary lossof sovereignty, I code such periods as democratic if there was acontinuation of democracy after sovereignty was regained and theperiod was not longer than five years. For instance, the Netherlandslost sovereignty during the years 19401944 due to World War II.Rather than creating two spells, 18971939 and 19452001, I recorda single spell, 18972001.

156

8/7/2019 Svolik2008p2979

5/16


one-third of all existing democracies, including manyobservations that have been democratic for more thana century, such as Canada or New Zealand. In con-trast, only three among the74 failed democracies lastedlonger than 32 years. Thus an initial inspection of thepattern of democratic survival reveals that a substantialnumber of currently existing democracies are survivingfor a remarkably long period of time.

A Split-Population Model of DemocraticSurvival

Before presenting my estimation results, I briefly sum-marize the key assumptions and derivations underly-ing split-population models. Maller and Zhou (1996)provide a detailed discussion of the large-sample prop-erties of these models.

Denote by C {0, 1} whether a democracy is con-solidated or transitional, and let C= 1 whenever ademocracy is consolidated, whereas C= 0 whenever ademocracy is transitional. Because we do not observe

whether the right-censored observations in the data aretransitional or consolidated democracies, Cis an unob-servable variable for the right-censored observations.Denote theprobability that a democracyis consolidatedby . Note that is constant with respect to time. ThenPr(C= 1) = and Pr(C= 0) = 1 .

Suppose that the probability that a transitionaldemocracy reverts by year t 0 follows a cumula-tive distribution function F(t|C= 0). Denote the cor-responding density function by f (t|C= 0) and thesurvival function by S(t|C= 0) = 1 F(t|C= 0). Notethat F(t|C= 0), f (t|C= 0), and S(t|C= 0) are condi-tional on C= 0, that is, on the democracy being transi-

tional.Let Tdenote the total number of years a democracyis observed in the sample. Let r= 1 indicate that ademocracy reverts by the last observed year; these ob-servations are uncensored. Otherwise, r= 0 and theseobservations are right-censored. When r= 1, a democ-racy must be transitional and C= 0. The likelihood ofthese observations is Pr(C= 0)f (t|C= 0), or equiva-lently, (1 )f (t|C= 0).

On the other hand, when an observation is right-censored, then a democracy may be either transi-tional or consolidated. The likelihood of these right-censored observations is then a combination of thelikelihood that an observation is a consolidated democ-

racy and the likelihood that an observation is a transi-tional democracy that has not reverted by the last ob-served year T, Pr(C= 1) + Pr(C= 0) Pr(t> T|C= 0),or equivalently, + (1 )S(T|C= 0).

Denote observations of democratic spells by i =1, 2, . . . , N. The joint likelihood of all N observationsin the sample is

N

i=1

[(1 )f (ti|Ci = 0)]ri

[ + (1 )S(Ti|Ci = 0)]1ri ,

and the joint log-likelihood of all Nobservations in thesample is

N

i=1

{ri[ln(1 ) + lnf (ti|Ci = 0)]

+ (1 ri)ln[ + (1 )S(Ti|Ci = 0)]}.

This log-likelihoodis maximizednumerically. Notethat, the probability that a democracy is consolidated, isestimated jointly with the parameters of the distribu-tion function F(t|C= 0).

In order to ensure that my results are not sensitiveto the choice of functional form for the survival distri-bution F(t|C= 0), I employ two alternative parame-terizations of the survival distribution: the Weibull andthe log-logistic. Both the Weibull and the log-logisticdistribution are two-parameter distributions; I denotethese parameters by and . I refer to as the scaleparameter because it determines the rate at which re-versals occur; I refer to as the shape parameter be-cause it determines the shape of the hazard rate. Thedistribution function F(t|C= 0) is

1 e(t)

for t 0, > 0, > 0

for the Weibull and

(t)

1+ (t)for t 0, > 0, > 0

for the log-logistic parameterization. The key distinc-tion between the two parameterizations is that the log-logistic distribution allows for a nonmonotonic haz-ard rate, whereas the hazard rate is monotonic in theWeibull parametrization. Later in the paper, I examinethe goodness-of-fit of alternative model specificationsand find that the log-logistic parameterization providesthe best fit to the data. Examples in the remainder ofthe paper are therefore based on this parameterization,unless otherwise noted.

ESTIMATION AND RESULTS

I begin by estimating a split-population model with-out covariates. I do this because the key differencesbetween the split-population model used here and thediscrete choice and survival models typically employedin the literature on democratic transitions are best

understood when we consider the survival of democ-racy only, without any covariates.10 In subsequentsections, I will examine the effects of key economicand institutional covariates on both the timing of au-thoritarian reversals and the likelihood of democraticconsolidation.

10 The estimates of the continuous-time, split-population model ex-amined here and discrete choice survival models are not directlycomparable, but the key points raised in this paper apply to discretechoice survival models as well. In further comparisons between themodel presented here and approachestypically usedin the literature,I therefore restrict attention to continuous-time survival models.

157

8/7/2019 Svolik2008p2979

6/16


TABLE 1. Estimation Results for a Simple and a Split-Population Survival Model withoutCovariates

Weibull Log-Logistic

Parameter estimates Simple Survival Split-Population Simple Survival Split-Population

Scale parameter 0.014 0.055 0.027 0.080

(0.003) (0.008) (0.005) (0.012)Shape parameter 0.758 1.250 1.016 1.720

(0.050) (0.106) (0.084) (0.154)Probability consolidated 0.428 0.420

(0.055) (0.059)

Log-likelihood value 373.693 357.232 366.457 354.500LR statistic for H0: = 0

a 23.918 32.923

Note:193 observations,133 countries, 74 reversals. Robust standard errors in parentheses. Significance levels 10%, 5%,1%.aSignificance levels are based on the 12

20 +

12

21 likelihood ratio test statistic.

The parameter estimates for the split-populationmodel without covariates are presented in Table 1. Theestimates are = 0.055 (0.008) and = 1.250 (0.106)

for the Weibull, and = 0.080(0.012) and =1.720(0.154) for the log-logistic model. Robust stan-dard errors are in parentheses. All estimates are signif-icant at the 1% level.

What are the implications of the split-populationmodel for the survival of democracies? How do theydiffer from the implications of a model that ignoresthe potentialunobserved heterogeneity among existingdemocracies?

Theparameter of the log-logistic model determinesthe shape of the hazard rate of authoritarian reversalsover time. For > 1, the hazard is first increasing andthen decreasing, while the hazard is strictly decreasing

for 0 < 1. Jointly, the estimates of and implythat in transitional democracies, the hazard rate ofan authoritarian reversal is sharply increasing in thefirst 10 years and declines thereafter. In contrast, theestimate of the shape parameter is 1.016 for a log-logistic model that ignores the existence of consoli-dated democracies, and we cannot reject the hypothesis 1. For that model, the estimate of is indistin-guishable from 1 at the 95% confidence level, and theestimated hazard rate of an authoritarian reversal istherefore greatest in the first year and declines after-ward. The split-population model therefore implies avery different hazard dynamic than a simple survivalmodel.

A second notable difference between thetwo modelsis that they lead to very different expectations aboutthe survival of new democracies over time. The split-population model implies that about one-half of thetransitional democracies will revert to dictatorship bytheir 13th year. In contrast, a log-logistic model thatignores the existence of consolidated democracies pre-dicts that the median reversal time will instead beabout 37 years. In other words, a model that ignoresthe potential unobserved heterogeneity among exist-ing democracies underestimates the risk of an earlyreversal while simultaneously overestimating the riskof a late reversal.

FIGURE 2. Empirical Distribution ofUncensored Survival Time and the Estimated

Density of Survival Time of TransitionalDemocracies According to theSplit-Population and the Simple Log-LogisticModels

0

.02

.04

.06

Density

0 25 50 75 100Time in years

Splitpopulation loglogistic model

Simple loglogistic model

Empirical distribution of uncensored survival time

Thus the two models differ not only in their implica-tions for hazard dynamics but also in their estimate ofthe timing of the risk of a reversal faced by new democ-racies. Thisdifference is evident in Figure 2, which plotsthe estimated probability density of survival time ac-

cording to the two models against a histogram of theobserved survival time of democracies that revertedto dictatorship.11 Clearly, when compared to a simplesurvival model, the split-population model predicts adistribution of survival time that is much closer theactual distribution of the uncensored survival time ofdemocracies. In other words, estimates based on thesplit-population model employed here provide a much

11 A similar result holds forthe comparison of a split-populationanda simple Weibull model. The median survival time of the transitionaldemocracies is 14 years according to the split-population model, butit is 45 years according to the simple model.

158

8/7/2019 Svolik2008p2979

7/16


better fit to the data on democratic survival than domodels typically employed in the literature.

Now consider the estimate of , the probabilitythat a democracy is consolidated. The estimates are = 0.428(0.055) for the Weibull and = 0.420(0.059)for the log-logistic model, with robust standard er-rors in parentheses. The corresponding 95% confi-dence intervals are (0.321, 0.535) for the Weibull and

(0.305, 0.536) for the log-logistic model.Can we reject the hypothesis that there are noconsolidated democracies among currently existingdemocracies? In order to do so, I test the hypothesisthat assumes the boundary value of = 0. The appro-priate test statistic forthis hypothesis is a 5050 mixtureof20 and

21 random variables (Maller and Zhou 1995).

The likelihood ratio statistics are 23.918 and 32.923 forthe Weibull and the log-logistic models, respectively,whereas the critical value of the test statistic is 5.41 atthe 1% significance level. Thus both parameterizationsstrongly suggest that a substantial number of currentlyexisting democracies are consolidated democracies.

To summarize, this preliminary analysis of the split-

population model leads to very different conclusionsabout both the susceptibility to and the timing of therisk of a reversal faced by new democracies than wouldbe obtained by methods typically employed in the lit-erature on democratic transitions. I find that transi-tional democracies face a much greater risk of a re-versal and a much shorter median survival time thanwould be estimated by using a simple survival model.At the same time, both the Weibull and the log-logisticparameterizations strongly suggest that a substantialnumber of currently existing democracies are consoli-dated democracies, and thus are not at risk of revertingto dictatorship. Finally, when compared to the actual

distribution of survival time among democracies thatreverted to dictatorship, the split-population modelprovides a much better fit tothe data than does a simplesurvival model.

The Effects of Covariates

Employing data on the survival of democracy only,the empirical analysis so far strongly suggests that asignificant number of currently existing democraciesare consolidated. I will now extend the analysis andinvestigate the effects of covariates on the two relatedyet distinct mechanisms that I have identified as key

to understanding the dynamics of democratic survival:the likelihood that a country becomes a consolidateddemocracy and the timing of authoritarian reversals intransitional democracies. As I discuss in greater detailbelow, I examine the effects on democratic survival oftwo economic and two institutional covariates: levelof economic development, economic growth, type ofdemocratic executive, and past authoritarian institu-tions.

The introduction of covariates implies that the pa-rameters of the survival distribution for transitionaldemocracies are now conditional on these covariates,as is the probability that a democracy is consolidated.

In other words, these covariates are a source of addi-tional, observable heterogeneity in the population ofdemocracies. As I have emphasized, only transitionaldemocracies revert to dictatorship, whereas consoli-dated democracies are not at risk of a reversal. Thusin transitional democracies, covariates affect the occur-rence and the timing of reversals. Meanwhile, I estimatehow the same covariates may affect the likelihood that

a democracy is consolidated rather than transitional.The inclusion of covariates is also central to ad-dressing the existing research on democratic consoli-dation. This research associates consolidation with thehypothesis that the risk of an authoritarian reversaldeclines with the age of a democracy but finds no em-pirical support for this proposition (Epstein et al. 2006;Przeworski et al. 2000). Additionally, that literatureemphasizes the need to control for economic and insti-tutional covariates because higher survival ratesamongolder democracies may be confounded with trends inthose covariates, such as an increase in the level ofeconomic development (Przeworski et al. 2000, 103).

The present empirical model highlights why the ex-

isting approach to the study of democratic consolida-tion is inadequate: the effect of a covariate on thetiming of authoritarian reversals among transitionaldemocracies cannot be separated from the effect ofthesame covariate on their likelihood of consolidation.But as I emphasized previously, these are two distinctmechanisms by which a democracy may survive. Thisdistinction is key to identifying the determinants ofdemocratic consolidation and cannot be made withmodels typically employed in the literature on demo-cratic survival.

The main complication arising from the inclusionof covariates is the lack of reliable and comparable

covariate data for the entire period 17892001. Mosteconomic and institutional covariates are only avail-able for the post-World War II period. However, ananalysis based on such historically limited data may notbe representative of the relationship between those co-variates and democratic survival throughout the entirehistory of modern democracy. Przeworski et al. (2000),for instance, do not find that economic developmentaffects transitions to democracy when using data thatstart in 1950, while Boix and Stokes (2003) do detectsuch an effect when using data that go back to 1850. Inorder to best ensure that my results are representativeof the entire history of modern democracy, I proceedby using the limited economic and institutional covari-

ates available for both the pre- and post-World War IIperiod.

I use two covariates, the level of economic devel-opment and annual economic growth, to study the ef-fect of economic conditions on democratic survival.Maddisons Historical Statistics (2003) is the most ex-tensive source of historical economic data. The twocovariates based on Maddisons data are annual GDPper capita and annual GDP growth.

In order to study the effect of political institutions,I employ one measure of the institutional characteris-tics of the democratic regime and one measure of theinstitutional characteristics of the authoritarian regime

159

8/7/2019 Svolik2008p2979

8/16


prior to transition to democracy. The dummy variablesPresidential, Parliamentary, and Mixed code presiden-tial, parliamentary, and mixed democratic executive,respectively. On the other hand, the dummy variablesMilitary, Civilian, and Monarchy code whether the au-thoritarian government prior to its transitionto democ-racy was headed by a professional military, civilians, ora hereditary monarch, respectively. When a country

was not independent prior to transition, the dummyvariable Not independentassumes a value of one. Thesedata are based on Beck et al. (2001), Cheibub andGandhi (2005), Correlates of War Project (2005), Van-hanen (2003), Svolik and Akcinaroglu (2006), and onmy own data collection. After accounting for missingobservations, my data contain 153 democratic spellsfrom 103 countries, which corresponds to a total of3,402 democracy-year observations between 1848 and2001.

I now extend the split-population model employedso far and let covariates affect the hazard of authoritar-ian reversals in transitional democracies as well as theprobability that a democracy is consolidated. Covari-

ates affect the timing of reversals in transitional democ-racies via the scale parameter of the Weibull andthe log-logistic parameterizations. In order to trans-form the range (,) of the linear combination ofcovariates and parameters to the natural domain of, which is (0,), I use the exponential link function(McCullagh and Nelder 1983) according to which =eX

and = (1, . . . , k) is a parameter vector asso-ciated with the covariates X. A positive coefficient jimplies a smaller , which in turn implies an increasein the expected duration of democracy for positivechanges in the values of the covariates. Alternatively, anegative coefficient implies that a covariate accelerates

the onset of an authoritarian reversal.The covariates X may vary over time, which is thecase for both of the economic covariates as well asfor the type of the democratic executive. In order toassure their exogeneity with respect to the survival ofdemocracy, I lag each covariate by one year. For theeconomic covariates, this controls for the possibilitythat the survival of democracy would affect economicperformance in the same year, rather than the otherway around. In order to facilitate exposition, I suppresstime subscripts for all covariates.

I also let covariates affect the probability that ademocracy is consolidated rather than transitional. Iuse the logistic link function to model the effect ofcovariates on , the probability that a democracy isconsolidated. Thus we have

=eX

1+ eX,

where = (1, . . . , k) is a parameter vector associatedwith the covariates X. A positive coefficient j impliesthat a democracy is more likely to be consolidated forlarger values of covariate j.

However, unlike when I estimate the timing of re-versals in transitional democracies, I cannot use time-varying covariates when estimating the probability that

a democracy is consolidated. An existing democracyis either transitional or consolidated, although thiscannot be observed. Therefore, when estimating theprobability that a democracy is consolidated, I onlyemploy covariates that are constant with respect totime throughout the entire democratic spell. In orderto estimate the effect of GDP per capita and GDPgrowth on the probability that a democracy is consol-

idated, I use a 10-year average of each indicator overthe 5 years preceding and the 5 years following tran-sition to democracy.12 In order to examine the effectof executive-legislative relations on the likelihood ofconsolidation, I use the type of executive adopted aftertransition to democracy. This approachallows me to ex-amine the impact of political institutions and economicconditions at the time of transition to democracy on thelikelihood of democratic consolidation.

The parameter estimates for the log-logistic andthe Weibull parameterizations of the split-populationmodel with covariates are presented in the first andthird column of Table 2, respectively. Under the re-versal timing model, I list parameter estimates for the

effect of covariates on the timing of reversals in tran-sitional democracies. Under the consolidation statusmodel, I list parameter estimates for the effect of co-variates on the probability that a democracy is con-solidated rather than transitional. I now turn to theinterpretation of these estimates.

The Impact of Economic Covariates:Economic Development and Growth

I find that the level of economic development deter-mines the extent to which a democracy issusceptible tothe risk of a reversal, but the eventual timing of a re-

versal is only associated with economic recessions. Thiscan be seen by comparing the statistical significance ofthe coefficients on GDP per capita and GDP growthin the first and third column of Table 2. Under bothparameterizations, the coefficient on GDP per capita isstatistically significant at the 1% level in the consolida-tion status model, but it is not significant in the reversaltimingmodel. The converse is true forthe coefficient onGDP growth. This coefficient is statistically significantat the 1% level in the reversal timing model, but it isnot significant in the consolidation status model.13

The level of economic development strongly af-fects whether a democracy is transitional or consoli-dated. The magnitude of this effect is remarkable: a

$1,200 increase in GDP per capita raises the proba-bility that a democracy is consolidated from 20% to80%. As I demonstrate later, whether a democracy istransitional or consolidated is also strongly associatedwith its type of executive and its past authoritarian

12 When I use averages for different time periods (10 and 20 yearsafter or prior to transition), estimation results are almost identical tothose obtained here.13 Under the Weibull parameterization, GDP per capita is statisti-cally significant at the 10% level in the reversal timing model. Thissignificance, however, disappears when I control for unobservableheterogeneity via a frailty term.

160

8/7/2019 Svolik2008p2979

9/16


TABLE 2. Estimation Results for Models with CovariatesSplit Simple Split Split & Frailty

Log-logistic Log-logistic Weibull Log-logistic

Reversal timing modela

GDP per capita 0.093 0.345 0.117 0.093

(0.078) (0.061) (0.070) (0.078)

GDP growth 0.045 0.033 0.041 0.045

(0.015) (0.014) (0.013) (0.015)Military (vs. Not independent) 0.287 0.880 0.307 0.287

(0.316) (0.325) (0.274) (0.316)

Civilian(vs. Not independent) 0.136 0.085 0.077 0.136

(0.345) (0.332) (0.314) (0.344)

Monarchy(vs. Not independent) 0.930 0.100 0.979 0.930

(0.533) (0.559) (0.454) (0.532)

Parliamentary(vs. Mixed) 0.295 0.018 0.246 0.295

(0.310) (0.330) (0.318) (0.310)

Presidential(vs. Mixed) 0.390 0.010 0.370 0.389

(0.290) (0.326) (0.308) (0.290)

Intercept 2.298 2.562 2.589 2.298

(0.402) (0.401) (0.379) (0.401)

Shape parameter 1.944 1.603 1.457 1.944

(0.211) (0.168) (0.151) (0.211)

Frailty variancec 0.000

(0.005)

Consolidation status modelb

GDP per capita 2.121 2.045 2.121

(0.586) (0.555) (0.586)

GDP growth 0.014 0.048 0.014

(0.227) (0.246) (0.227)

Military (vs. Not independent) 4.061 3.985 4.061

(1.895) (1.857) (1.895)

Civilian(vs. Not independent) 0.421 0.549 0.421

(1.097) (1.067) (1.097)Monarchy(vs. Not independent) 20.158 15.844 13.965

(2888.609) (680.185) (891.870)

Parliamentary(vs. Mixed) 2.231 2.290 2.231

(2.230) (2.326) (2.230)

Presidential(vs. Mixed) 8.310 8.186 8.310

(3.958) (4.035) (3.958)

Intercept 6.144 5.920 6.145

(2.646) (2.644) (2.647)Note:Standard errors in parentheses. Significance levels 10%, 5%, 1%.aModel estimates the timing of reversals among transitional democracies via an exponential link function for the scaleparameter .bModel estimates , the probability that a democracy is consolidated, via a logistic link function.cGamma frailty; significance levels are based on the 1

2

2

0

+ 1

2

2

1

likelihood ratio test statistic.

institutions. These institutions will therefore determineat which level of GDP per capita the relevant $1,200interval begins. For a democracy with median levels ofall covariates, that critical interval starts at a GDP percapita of $3,900 and ends at $5,100.

How does economic growth affect the timing of au-thoritarian reversals? Greater growth lowers the haz-ard of reversals in transitional democracies. Althoughthis effect is nonlinear and exhibits increasing returns,

it is almost constant within the range from 7.5% to9.1% of annual economic growth, which covers 90%of the data: a 1% increase in growth corresponds to aroughly 8-month increase in the median survival timeof a transitional democracy. Alternatively, a countrywith a 0% GDP growth faces the largest reversal haz-ard in its eighth year. A growth of 10% in the previousyear will lower the reversals hazard in the current yearby 42%, whereas a 10% recession in the previous year

161

8/7/2019 Svolik2008p2979

10/16


raises the hazard by 43% in the current year.14 Finally,recall that consolidated democracies are not at risk ofreverting to dictatorship and are thus immune to thenegative impact of economic recessions on their oddsof survival.

Compare these findings to those based on a modelthat does notallow for the possibility that some democ-racies are consolidated. Parameter estimates for a sim-

ple log-logistic survival model are displayed in thesecond column of Table 2. The coefficients of bothGDP per capita and GDP growth are statistically sig-nificant, yet we are unable to distinguish their effecton the timing of reversals in transitional democraciesfrom their effect on the probability that a democracy isconsolidated rather than transitional. In other words,the simple survival model cannot uncover the specificmechanism by which economic conditions promotedemocratic survival.

The Impact of Past Authoritarian and NewDemocratic Political Institutions

How do political institutions affect the survival ofdemocracy? I find that a military past and presiden-tial executive both have a large, negative and indepen-dent effect on the probability of democratic consol-idation, whereas neither institution raises the hazardof authoritarian reversals in transitional democracies.As discussed previously, the latter effect is associatedprimarily with economic recessions.

First, consider the effect of past authoritarian institu-tions on the survival of democracy. I find that democ-racies that were preceded by a military dictatorshipface significantly lower chances of becoming consoli-dated than democracies that were preceded by a civil-

ian dictatorship or a monarchy, or those democraciesthat were not independent countries prior to transition.(See the coefficient estimates for Military, Civilian,andMonarchy in the consolidation status model in Table 2;I use Not independent as the reference category.) Thelikelihood ratio test indicates that merging Civilianand Not independent into a single category does notsignificantly worsen the fit of the model. Therefore,setting aside the four democracies that were precededby monarchies, the difference in the odds of consolida-tion is primarily that between those democracies thatwere preceded by a military dictatorship and those thatwere not. My estimates suggest that at median levels ofthe other covariates, only about 1 in 8 democracies thatwere preceded by a military dictatorship consolidate,while the odds of consolidation are 9 out of 10 fordemocracies that were preceded by any other type ofdictatorship.

However, note that transitional democracies thatwere preceded by a military dictatorship are not ex-pected to revert to dictatorship any sooner than eitherdemocracies that were preceded by a civilian dictator-

14 Thus I do not find any support for the conjecture that growth maydestabilize democracy (Huntington 1968; Olson 1982), even after Iseparate theeffectof growthon the timingof reversalsin transitionaldemocracies from its effect on the probability that a democracy is

consolidated rather than transitional.

ship or those democracies that were not independentprior to their transition. As the estimates of the rever-sal timing model indicate, the only institutional factorthat affects the risk of authoritarian reversals in tran-sitional democracies is a monarchical pastthis typeof authoritarian legacy actually lowers the hazard of areversal. In fact, when they finally reverted to dictator-ship, democracies that were preceded by a monarchy

lasted 34 years on average, while those democraciesthat were not preceded by a monarchy lasted less than10.5 years on average.15

Now I consider the effect of the type of executiveon democratic survival. As discussed earlier, existingresearch is not unanimous regardingthe negative effectof presidentialism on democratic survival. The presentmodelallows me to take thefollowing, newapproach tothis issue: is the effect of presidentialism (1) to raise thehazard of authoritarian reversals in transitional democ-racies, (2) to make democracies more susceptible toother negative factors that will eventually lead to ademocratic breakdown, or (3) is it both? The resultsof my analysis support only the second, indirect re-

lationship between presidentialism and authoritarianreversals: presidential democracies are less likely tobecome consolidated and are thus more susceptible toother factors that eventually determine the timing ofdemocratic breakdowns.

The magnitude of presidentialisms negative effecton consolidation is remarkable: my results imply thatonly about 1 in 6,800 presidential democracies willconsolidate at median levels of other covariates. Thusthe odds that a presidential democracy with medianlevels of all other covariates will become immune toan authoritarian reversal are practically zero. Com-pare this to the odds of consolidation for democracies

with mixed executives (3 in 8) and for parliamentarydemocracies (6 in 7). Furthermore, the likelihood ratiotest indicates that merging parliamentary and mixedsystems into a single category does not significantlyworsen the fit of the model. Thus when it comes tothe effect of different types of executive on democraticconsolidation, the present model suggests a large andimportant difference between presidential and non-presidential systems.

As mentioned earlier, however, this large, negativeeffect of presidentialism on consolidation assumes ademocracy with median levels of all remaining covari-ates. An increase in GDP per capita at the time of tran-sition to democracy from the median $2,858 to $7,831

will compensate for the dismal odds of survival thata presidential executive implies. Nonetheless, severallong-lived presidential democracies, and particularlythe United States, are outliers in terms of the presentmodel. Furthermore, although this analysis suggeststhat the type of democratic executive is an importantdeterminant of consolidation, it also suggests that it

15 This finding should be interpreted with caution since onlyfour democracies in the data were preceded by a monarchy;these observations are France (18701940), Portugal (19111926),Germany (19191933), and Nepal (19912001). Of these, only Nepal(19912001) is currently existing, which explains the large standard

error on the coefficient on Monarchy in the consolidation model.

162

8/7/2019 Svolik2008p2979

11/16


has no effect on the hazard of authoritarian reversalsin transitional democracies. According to the estimatesof the reversal model, the median time until a reversalis 14 years for a democracy at median levels of all othercovariates, regardless of the type of executive.

Why cannot this strong, negative, and independenteffect of presidentialism on democratic consolidationbe seen in a simple survival model? Estimates of

the log-logistic parameterization of such a modelare displayed in the second column of Table 2. Asin Cheibubs (2007) analysis, a model that ignoresthe existence of consolidated democracies suggeststhat a military past has a statistically significant andnegative effect on the survival of democracy, butfinds no such effect for a presidential executive. Infact, among those democracies that did revert todictatorship, presidential democracies survive almostequally long as non-presidential ones: the meansurvival time is 11.89 and 11.31 years for presidentialand nonpresidential democracies, respectively. On theother hand, when only currently existing democraciesare considered, nonpresidential democracies have, on

average, survived twice as long as presidential ones:the mean survival time is 15.97 and 36.95 years forexisting presidentialand non-presidential democracies,respectively (see also Przeworski et al. 2000, 129). Thepresent model is able to infer whether we observe thisdifference in survival between existing presidentialand non-presidential democracies because (1) non-presidential democracies, despite ultimately revertingto dictatorship, tend to survive longer, or because (2)nonpresidential democracies indeed consolidate athigher rates than presidential democracies. In contrast,the simple survival model confounds the two processesand, consequently, does not reveal the important fact

that non-presidential democracies indeed consolidateat higher rates than do presidential democracies.

THE DYNAMICS OF DEMOCRATICCONSOLIDATION

I now use the split-population model to address a ques-tion that is at the heart of many studies of democratic

FIGURE 3. The Age of an Existing Democracy and the Probability That It Is Consolidated

.3

.4

.5

.6

.7

.8

.9

1

Probability

0 5 10 15 20 25 30 35 40 45 50Years democratic

Probability that an existing democracy is consolidated, given that it has survived for T years

95% confidence intervals

Currently existing democracies as a fraction of all democracies

consolidation: Supposea countryhas beendemocraticfor T years. Can we conclude that this country is notat risk of reverting to dictatorship? In order to an-swer this question, I jointly consider the findings aboutdemocratic consolidation so far as well as the additionalinformation of the age of an existing democracy. Asin the previous section, I will proceed by first clarify-ing how the present model addresses the question just

raised using age of democracy alone, without economicand institutional covariates. I will then include thesecovariates in my examination of the effect of age ofdemocracy on consolidation.

Suppose that a democracy has survived for T yearsand denote by t the time at which this democracy re-verts. We can use Bayes rule to compute the proba-bility that an existing democracy is consolidated, giventhat it has survived for Tyears,

Pr(C= 1|T< t) =Pr(T< t|C= 1)Pr(C= 1)

Pr(T< t).

Using the fact that Pr(T< t) = Pr(T< t|C= 1)

Pr(C= 1) + Pr(T< t

|C= 0)Pr(C= 0), and the factthat Pr(T< t|C= 1) = 1 and Pr(T< t|C= 0) = 1 F(T|C= 0) = S(T|C= 0), we obtain

Pr(C= 1|T< t) =

S(T|C= 0) + F(T|C= 0), (1)

where is the unconditional probability that a democ-racy is consolidated. Large sample properties of thisestimator are discussed in greater detail in Maller andZhou (1996, Chapter 4 and Section 9.3).

Figure 3 plots this estimator of democratic consoli-dation based on the parameters of the split-populationlog-logistic model for the period of the first 50 years

after the transition to democracy. The dotted lines plotthe associated 95% confidence interval, which mea-sures our statistical confidence in the estimated prob-ability that an existing democracy is consolidated. In-tuitively, this probability is increasing in the length oftime T that this country remains democratic. We cansee in Figure 3 that at time zero, this probability corre-sponds to the estimate of the fraction of consolidated

163

8/7/2019 Svolik2008p2979

12/16


democracies . As time progresses, our belief that asurviving democracy will never revert to dictatorship isrevised upwards.

Using only the age of an existing democracy, anycountry that has been democratic for 52 or more yearsas of 2001 is estimated to be consolidated with atleast 90% probability. Thus the youngest consolidateddemocracy would be India, which became a democracy

in 1950. We also learn that despite the unprecedentedsurge in the number of democracies after 1950, nodemocracy that emerged during that period had by2001existed longenoughto be considered consolidatedwith a sufficient degree of confidence.

However, note that with the exception of the firstcouple of years of a democracys life, the increase in theprobability that an existing democracy is consolidatedis greater during the first 20 years after its transitionthan during any later period. Thus our belief that ademocracy is consolidated depends most crucially onits survival during these initial two decades. The sub-stantive relevance of this point is underscored by thefact that 71 out of the 116 currently existing democ-

racies have been democratic for 20 years or less. Thesurvival of democracy in these countries in the next fewyears will therefore be essential to our belief that thesedemocracies are consolidated.

For expositional purposes, the discussion so far hasconsidered the relationship between democratic con-solidation and the age of an existing democracy only,ignoring any covariate effects. Building on the modelwith covariates, I can now extend this analysis and ad-dress the following question: knowing the economicand institutional history of a country and given thatit has been democratic for T years, how strong is ourbelief that this country is not at risk of reverting to

dictatorship?Extending the result in equation (1) to a settingwith covariates, I compute the conditional probabilitythat an existing democracy is consolidated, given itsage T and its economic and institutional history Xtfor t= 1, . . . , T. The results are displayed in Table 3.

TABLE 3. Level of Economic Development, Duration of Democracy, Type of Executive, andDemocratic Consolidation

Probability that an existing democracy is consolidated

Presidentiala Nonpresidentiala

GDP p.c.b

0 10 20 30 40 50 0 10 20 30 40 50$ 1,000 0.00 0.00 0.00 0.00 0.00 0.00 0.08 0.09 0.12 0.16 0.21 0.27$ 2,000 0.00 0.00 0.00 0.00 0.00 0.00 0.42 0.45 0.52 0.61 0.69 0.75$ 3,000 0.00 0.00 0.00 0.00 0.00 0.00 0.86 0.87 0.90 0.93 0.95 0.96$ 4,000 0.00 0.00 0.00 0.00 0.00 0.01 0.98 0.98 0.99 0.99 0.99 1.00$ 5,000 0.01 0.01 0.01 0.02 0.03 0.04 1.00 1.00 1.00 1.00 1.00 1.00$ 6,000 0.10 0.10 0.12 0.16 0.21 0.27 1.00 1.00 1.00 1.00 1.00 1.00$ 7,000 0.42 0.45 0.52 0.61 0.69 0.75 1.00 1.00 1.00 1.00 1.00 1.00$ 8,000 0.86 0.87 0.90 0.93 0.95 0.96 1.00 1.00 1.00 1.00 1.00 1.00$ 9,000 0.98 0.98 0.99 0.99 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Note:Based on a model that distinguishes only between presidential and nonpresidential executive.aFirst democratic executive/Age of an existing democracy.bAverage GDP per capita over the 5 years preceding and the 5 years following transition to democracy; the remaining covariatesare held at their median/modal values.

Based on the analysis with covariates, we know thatboth a higher GDP per capita and a non-presidentialexecutive increase the likelihood that a democracy isconsolidated. Table 3 illustrates how the age of an ex-isting democracy leads to an upward revision of thatbelief. We see that the effect of the age of a democracycan be large, but is restricted to a GDP per capita under$5,000 for nonpresidential democracies, and to a GDP

per capita above $5,000 for presidential democracies.Thus depending on its executive type, a democracymay be too poor or too rich for its age to help us inferwhether it is consolidated.

Nonetheless, these estimates imply that the survivalof democracy for 52 years in India raises our estimateof the odds that India is consolidated from 7% in 1950to 70%in 2001, despite its lowGDP percapita through-out that period. In contrast, models typically employedin the study of democratic survival predict that Indiashould have reverted to dictatorship with a high proba-bility, both in 1950 and in 2001, because of its low GDPper capita.

By jointly considering the age of a democracy and

its covariate history within the present model, we alsoarrive at another notable finding: if a democracy sur-vives an economic recession, our belief that it is con-solidated is revised upwards. This is intuitive; giventhat recessions raise the hazard of reversals when ademocracy is transitional, a democracy that survives arecession is more likely to be consolidated than tran-sitional. Formally, this can be seen be conditioningPr(C= 1|T< t) in equation (1) on covariate valuesXt and considering the effect of a change in Xt onPr(C= 1|T< t,Xt). A recession in year T leads toan increase in F(T|C= 0,Xt) and the associated de-crease in S(T|C= 0,Xt), which results in an increase

in Pr(C= 1|T< t

,X

t). Importantly, however, modelsthat fail to account for the existence of consolidateddemocracies imply only that recessions raise the haz-ard of reversals. Those models tell us nothing aboutwhat to infer when a democracy has in fact survived arecession.

164

8/7/2019 Svolik2008p2979

13/16


To summarize, the split-population model can beextended to consider jointly the age of an existingdemocracy andits economic andinstitutional history inorder to further examine the dynamics of democraticconsolidation. Intuitively, the longer a democracy lives,the greater our confidence that it is consolidated. How-ever, I show that the age of a democracy leads to asignificant increase in our belief that it is consolidated

only for moderate values of economic developmentand depending on its type of executive. The presentmodel also implies that if a democracy survives an eco-nomic recession, it is more likely to be consolidatedthan transitional, a finding that cannot be obtainedwith models typically employed in the literature ondemocratic survival. This analysis therefore provides astatistical counterpart to qualitative research that at-tempts to summarize our belief about the stability ofnew democracies.

MODEL ROBUSTNESS AND GOODNESS

OF FITIn this section, I demonstrate that the results of myempirical analysis are robust even after accounting forthe limited availability of covariate data for the pe-riod under study. This lack of data on various potentialcovariates of interest does not allow me to addressseveral findings in the literature on democratic transi-tions. For instance, I cannot address the relationship be-tween the survival of democracy and income inequality(Acemoglu and Robinson 2005; Boix 2003), the legaland colonial origin of democracies (La Porta et al.1999), ethnolinguistic and religious fractionalization(Przeworski et al. 2000; Fish 2002), oil exports (Ross

2001), trade openness (Milner and Kubota 2005), ormembership in international organizations(Pevehouse2002). Thus even after employing the available eco-nomic and institutional covariates, a significant amountof heterogeneity may remain unaccounted for and biasparameter estimates.

In this section, I account for such spell-specific, un-observed heterogeneity with a frailty extension of thesplit-population model and confirm the robustness ofmy findings. The inclusion of a frailty term in the split-population model accounts for the possibility that twotransitional democracies with the same covariate val-ues may be subject to different risks because of un-observable, spell-specific risk factors. A frailty term

is an unobservable, multiplicative, random effect withunit mean = 1 and variance that affects the speedof reversals among transitional democracies. Obser-vations with > 1 are more frail for reasons unex-plained by the covariates and have an increased riskof failure; the converse holds for observations with < 1.

I present parameter estimates of a log-logistic split-population model with one commonly used frailtydistributionthe Gamma frailtyin column 4 ofTable 2. All parameter estimates are now conditionalon the frailty variance . However, as Table 2 indi-cates, these estimates are very close to those of the

original split-population model. Because research onfrailty models indicates that estimates may be sensitiveto the specific distribution posited for frailty (see e.g.,Heckman and Singer 1984), I used an alternative,inverse Gaussian frailty distribution to re-estimatethe split-population model. Yet the results of thatanalysis are almost identical to those obtained us-ing the Gamma frailty distribution.16 We can there-

fore be confident that the results in the previoussections are not skewed by spell-specific, unobservedheterogeneity.

So far, my results do not do not indicate whether thelog-logistic or the Weibull parameterization providesa better fit to the data. I therefore consider an addi-tional, information-based criterion for goodness-of-fit,Akaikes information criterion (AIC) (Akaike 1974).This criterion allows me to compare the fit of bothnon-nested models (the log-logistic versus the Weibullparameterization of the split-population model) andnested models (the Gamma versus the inverseGaussianfrailty within the split-population model) andis a useful indicator of model overfitting. AICis defined

asAIC= 2 ln L+ 2k, where L is the model likelihoodand k is the number of parameters in the model. Themodel with the smaller AIC is considered the better-fitting model.

Table 4 displays AIC scores for the alternativeparameterizations of the simple survival model andthe split-population models with and without frailty.The AIC scores indicate that the log-logistic split-population model without frailty provides the best fitto the data.17 In Table 4, I also display the resultsof a likelihood ratio test of the hypothesis that thefrailty variance does not improve the fit of the split-population model (H0 : = 0) as well as the hypothe-

sis that there are no consolidated democracies in thesample (H0 : = 0). The test of the former hypoth-esis indicates that the inverse Gaussian frailty termimproves the fit of the Weibull parameterization at the5% significance level, but neither frailty parameteriza-tionsignificantly improvesthe fit of the split-populationlog-logistic model. Importantly, the latter test clearlyindicates that accounting for consolidated democraciesimproves the fit of the present model regardless of theparticular parametric assumptions about the hazardfunction or the frailty distribution, as I have arguedthroughout this paper.

16 Estimation results based on the inverse Gaussian frailty distribu-tion as well as other robustness checks are available as an appendixon the authors website.17 The qualitative implications of my empirical analysis are identicalnot only under the log-logistic and Weibull parameterizations that Idiscuss in this paper butalso under thealternative log-normal or gen-eralized gamma parameterizations of the reversal model. The sameis true when I use the complementary log-log link function insteadof the logistic link function in the consolidation status model. Fur-thermore, the qualitative implications of my empirical analysis arealso preserved when I use the non-mixture split-population model ofTsodikov et al. (2003) instead of the mixture split-population modelemployed in this paper.

165

8/7/2019 Svolik2008p2979

14/16


TABLE 4. Comparison of Goodness of Fit of Alternative Simple and Split-Population Modelswith CovariatesModel ka lnLb AICc H0 : = 0

d H0 : = 0e

Weibull parameterizationSimple 9 270.909 559.819 Split-population 17 253.843 541.686 34.133 Split-population with Gamma frailty 18 252.607 541.214 36.605 2.472Split-population with inverse Gaussian frailty 18 251.841 539.681 38.137 4.004

Log-logistic parameterizationSimple 9 269.461 556.923 Split-population 17 252.614 539.228 33.695 Split-population with Gamma frailty 18 252.614 541.228 33.695 0.000Split-population with inverse Gaussian frailty 18 252.614 541.228 33.695 0.000

Note:3402 observations (democracy-year), 153 democratic spells, 103 countries, 63 reversals.aNumber of parameters, including auxiliary parameters such as the shape parameter and frailty variance .bModel log-likelihood.cLower AIC indicates better fit. The model with the lowest AIC is denoted by .dThe critical value of the 12

20 +

12

28 likelihood ratio test statistic is 18.17 at 1% significance level,

10%, 5%, 1%.eThe critical value of the 12

20 +

12

21 likelihood ratio test statistic is 5.41 at 1% significance level,

10%, 5%, 1%.

CONCLUSION

In this paper, I establish a new approach to the empir-ical study of democratic survival. A key feature of thisapproach is the intuitive assumption that some democ-racies may be consolidated and thus immune to the riskof an authoritarian reversal, while otherstransitionaldemocraciesface that risk. Importantly, the differ-ence between thetwo types of democracy is not directlyobservable.

I show that a substantial number of existing democ-racies are indeed consolidated, and that our confi-dence that an existing democracy is consolidated in-

creases with its age. This is in contrast to the influentialquantitative empirical literature on democratic transi-tions, which finds no statistical evidence that the ageof a democracy is associated with greater chances ofits survival (see e.g., Epstein et al. 2006; Przeworskiet al. 2000). The present approach therefore bridgesthe divide between these existing quantitative findingsand the qualitative research on democratic survival, inwhich the concept of consolidation receives consider-able theoretical attention and bears out empirically inqualitative evidence.

I also investigate the effect of prominent economicand institutional factors on democratic survival. Cru-cially, I identify the factors that explain whether a

democracy survives because it is consolidated fromthose that separately lower the hazard of authoritar-ian reversals in transitional democracies and thus alsopromote democratic survival. I find that democracieswith low levels of economic development, a presiden-tial executive, and a military authoritarian past are lesslikely to consolidate. However, these three factors haveno effect on the hazard of authoritarian reversals intransitional democracies; that risk is only associatedwith economic recessions.

At the theoretical level,these results stronglysuggestthat two separate theories may be needed in order toexplain democratic survival: one that accounts for the

onset of authoritarian reversals in transitional democ-racies, and a second one that explains why only somedemocracies consolidate. In terms of public policiesthat aim to improve the survival of new democracies,these processes pertain to two related but distinct pol-icy ends: the first is relevant when we want to reducethe immediate risk of authoritarian reversals in transi-tional democracies, the second when we want to devisepolicies that will transform a transitional democracyinto a consolidated one.

The current approachfurthermore allows me to eval-uate empirical propositions about democratic consoli-dationwhile being realistic about theavailable data and

even without identifying a theoretical consensus aboutthe proper measures of consolidation. Considerabledebate persists about the factors that contribute to theconsolidation of democracy (see e.g., ODonnell 1996).Still, most research agrees that consolidation greatlyimproves a democracys chances of survival. I translatethat point of agreement into the statistical assump-tion that consolidated democracies are markedly moreresilient in the face of adverse political or economicconditions than transitional ones. Mine is therefore aprobabilistic rather than a substantive concept ofconsolidation. The current approachallows me to studyconsolidation without presuming that it actually occursand without imposing an arbitrary age threshold that

implies that all democracies consolidate by a certainage. In this way, we can advance empirical researchon consolidation despite the lack of consensus on itsparticular determinants or measures.

The empirical approach established here may bea fruitful framework for the study of other politicalsettings with similar unobservable heterogeneity. Forinstance, Fortna (2004) studies the effect of cease-fireagreements on the durability of peace, while Diehl andGoertz (2000) study the determinants of enduring in-ternational rivalries. In the context of these studies,whether peace or rivalry are truly enduring or merelytemporary can only be observed indirectly, via the

166

8/7/2019 Svolik2008p2979

15/16


absence of violence. The approach that I establishhigh-lights the distinction between those factors that lowerthe hazard of the resumption of violence and those thatlead to a permanent peace settlement, and provides amethodology that can identify the potentially differentdeterminants of each trend.

REFERENCES

Acemoglu, Daron, and James A. Robinson. 2005. Economic Originsof Dictatorship and Democracy. New York: Cambridge UniversityPress.

Akaike, Hirotugu. 1974. A newlook at the statistical model identifi-cation. IEEE Transaction on Automatic Control 19 (December):71623.

Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer, andPatrick Walsh. 2001. New toolsin comparative political economy:The Database of Political Institutions. World Bank EconomicReview 15 (September): 16576.

Bernhard, Michael, Timothy Nordstrom, and Christopher Reenock.2001. Economic Performance, Institutional Intermediation, andDemocratic Survival. Journal of Politics 63 (August): 775803.

Boix, Carles. 2003. Democracy and Redistribution. Cambridge:

Cambridge University Press.Boix, Carles, and Sebastian Rosato. 2001. A complete dataset of

political regimes, 18001999.Boix, Carles, and Susan C. Stokes. 2003. Endogenous Democrati-

zation. World Politics 55 (July): 51749.Box-Steffensmeier, Janet M., and Bradford S. Jones. 2004. Event

History Modeling. New York: Cambridge University Press.Box-Steffensmeier, Janet M., Peter M. Radcliffe, and Brandon L.

Bartels. 2005. The Incidence and Timing of PAC Contributionsto Incumbent U.S. House Members, 199394. Legislative StudiesQuarterly 30 (November): 54979.

Cheibub, Jos e Antonio. 2007. Presidentialism, parliamentarism, anddemocracy. New York: Cambridge University Press.

Cheibub, Jos e Antonio, and Jennifer Gandhi. 2005. Classifying Po-litical Regimes: A Six-Fold Measure of Democracies and Dicta-torships.

Cohn, Jr., Samuel K. 2002. The Black Death: End of a Paradigm.The American Historical Review 107 (3): 70338.Correlates of War Project. 2005. State System Membership List.Diamond, Larry. 1999. Developing Democracy: Toward Consolida-

tion. Baltimore: Johns Hopkins University Press.Diehl,Paul F., andGary Goertz. 2000. War and Peace in International

Rivalry. Ann Arbor: University of Michigan Press.Douglas, Stratford. 1998. The duration of the smoking habit. Eco-

nomic Inquiry 36 (January): 4964.Elkins, Zachary. 2000. Gradations of Democracy? Empirical Tests

of Alternative Conceptualizations. American Journal of PoliticalScience 44 (April): 293300.

Epstein, David L., Robert Bates, Jack Goldstone, Ida Kristensen,and Sharyn OHalloran. 2006. Democratic Transitions. Ameri-can Journal of Political Science 50 (July): 55169.

Farewell, Vernon T. 1977. A Model fora BinaryVariable with Time-Censored Observations. Biometrika 64 (l): 4346.

Findley, Michael G., and Tze Kwang Teo. 2006. Rethinking Third-Party Interventions into Civil Wars: An Actor-Centric Approach.

Journal of Politics 68 (November): 82837.Fish, M. Steven. 2002. Islam and Authoritarianism. World Politics

55 (October): 437.Forster, Martin, and Andrew M. Jones. 2001. The Role of To-

bacco Taxes in Starting and Quitting Smoking: Duration Anal-ysis of British Data. Journal of the Royal Statistical Society 164:51747.

Fortna, Virginia Page. 2004. Peace Time: Cease-Fire Agreements andthe Durability of Peace. Princeton, NJ: Princeton University Press.

Gasiorowski, Mark J. 1995. Economic Crisis and Political RegimeChange: An Event History Analysis. American Political ScienceReview 89 (December): 88297.

Gasiorowski, Mark J., and Timothy J. Power. 1998. The struc-tural determinants of democratic consolidation: Evidence from

the Third World. Comparative Political Studies 31 (December):74071.

Haggard, Stephan, and Robert R. Kaufman. 1995. The PoliticalEconomy of Democratic Transitions. Princeton: Princeton Uni-versity Press.

Heckman, James J., and Burton Singer. 1984. A Method for Mini-mizing the Impact of Distributional Assumptions in EconometricModels for Duration Data. Econometrica 52 (March): 61523.

Huntington, Samuel P. 1968. Political Order in Changing Societies.New Haven: Yale University Press.

Huntington, Samuel P. 1991. The Third Wave: Demoratization in theLate Twentieth Century. Norman: University of Oklahoma Press.

Ibrahim, Joseph G., Ming-Hui Chen, and Debajyoti Sinha. 2001.Bayesian Survival Analysis. New York: Springer.

La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, andRobert Vishny. 1999. The Quality of Government. Journal ofLaw, Economics and Organization 15 (April): 22279.

Linz, Juan J. 1994. Presidential of parliamentary democracy: Doesit make a difference? In ed. Juan J. Linz and Arturo Valenzuela.The Failure of Presidential Democracy, Baltimore: Johns HopkinsUniversity Press, 387.

Linz, Juan J., andAlfred Stepan. 1996. Problemsof democratic transi-tion and consolidation: Southern Europe, SouthAmerica, and post-communist Europe. Baltimore: Johns Hopkins University Press.

Londregan, John B., and Keith T. Poole. 1996. Does high incomepromote democracy? World Politics 49 (October): l30.

Maddison, Angus. 2003. The World Economy: Historical Statistics.Paris: OECD.

Mainwaring, Scott, and Matthew Soberg Shugart. 1997. Presiden-tialism and Democracy in Latin America. New York: CambridgeUniversity Press.

Maller, Ross A., and Xian Zhou. 1995. Testing for the Presence ofImmune or Cured Individuals in Censored Survival Data. Bio-metrics 51 (December): 11971205.

Maller, Ross A., and Xian Zhou. 1996. Survival analysis with long-term survivors. New York: Wiley.

Marshall, Monty G., and Keith Jaggers. 2003. Polity IV: PoliticalRegime Characteristics and Transitions, 18002003.

McCullagh,Peter, and John A. Nelder. 1983. Generalized linear mod-els. London: Chapman & Hall.

McLachlan, Geoffrey, and David Peel. 2000. Finite Mixture Models.New York: Wiley.

Milner, Helen, and Keiko Kubota. 2005. Why the Move to Free

Trade?Democracy and Trade Policy in the Developing Countries.International Organization 59 (Winter): 15793.ODonnell, Guillermo. 1996. Illusionsabout consolidation.Journal

of Democracy 7 (April): 3451.ODonnell, Guillermo, and Philippe C. Schmitter. 1986. Transitions

from Authoritarian Rule: Comparative Perspective, vol. 4. Balti-more: Johns Hopkins University Press.

Olson, Mancur. 1982. The Rise and Decline of Nations: EconomicGrowth, Stagflation, and Social Rigidities. New Haven: Yale Uni-versity Press.

Pevehouse, Jon C. 2002. Democracy from the Outside-In? Interna-tional Organizations and Democratization.International Organi-zation 56 (Summer): 51549.

Przeworski, Adam. 1991. Democracy and the Market: Political andEconomic Reforms in Eastern Europe and Latin America. Cam-bridge: Cambridge University Press.

Przeworski, Adam, Michael E. Alvarez, Jose Antonio Cheibub, and

Fernando Limongi. 2000. Democracy and Development: Political Institutions and Well-Being in the World, 19501990. New York:Cambridge University Press.

Przeworski, Adam, and Fernando Limongi. 1997. Modernization:Theories and facts. World Politics 49 (January): 15583.

Ross, Michael. 2001. Does Oil Hinder Democracy? World Politics53 (April): 32561.

Rustow, Dankwart A. 1970. Transitions to Democracy: Toward aDynamic Model. Comparative Politics 2 (April): 33763.

Schedler,Andreas. 1998. What is DemocraticConsolidation?Jour-nal of Democracy 9 (April): 91107.

Schmidt, Peter, and Ann Dryden Witte. 1989. Predicting CriminalRecidivism Using Split Population Survival Time Models. Jour-nal of Econometrics 40 (January): 14159.

Svolik, Milan, and Seden Akcinaroglu. 2006. Government Changein Authoritarian Regimes.

167

8/7/2019 Svolik2008p2979

16/16


Tsodikov, A. D., J. G. Ibrahim, and A. Y. Yakovlev. 2003. Es-timating Cure Rates From Survival Data: An Alternative toTwo-Component Mixture Models. Journal of the American Sta-tistical Association 98 (464): 106378.

Vanhanen, Tatu. 2003. Democratization and Power Re- sources 18502000. Tampere: Finnish Social Science DataArchive.

Weingast, Barry R. 1997. The Political Foundations of Democracyandthe Rule of Law.AmericanPoliticalScience Review 91 (June):24563.

Yamaguchi, Kazuo. 1992. Accelerated Failure-time RegressionModels with a Regression Model of Surviving Fraction: An Appli-cation to the Analysis of Permanent Unemployment in Japan.

Journal of the American Statistical Association 87 (June): 28492.

Documents

Svolik2008p2979