Kokkonen, A., & Sundell, A. (2014). Delivering
Stability—Primogeniture and Autocratic Survival in European Monarchies
1000–1800. American Political Science Review, 108(2), 438-453.
Question 1: The article tests one hypothesis: that
autocratic states that use primogeniture succession (that is, succession by the
ruler’s oldest son alone) have more stability than autocratic states that use
either electoral succession (the elites elect the ruler’s successor), agnatic
succession (succession by the ruler’s oldest brother prior to any children), or
designated succession. The authors justify their hypothesis theoretically both
positively and negatively. Beginning by delineating the succession dilemma,
they note that elites have incentives to support the regime so long as a) the
regime appears stable and b) the ruler shares rents. However, succession
implies a lack of stability where no successor is designated, which creates
perverse incentives to “jump the gun” on deposing the ruler. However,
designating an older heir (through either agnatic or designated succession)
simply concentrates the incentives to trigger a succession crisis into the
designated heir, who cannot expect to outlive the ruler naturally. Designating
a young heir, such as the ruler’s son, properly aligns elite and heir
incentives toward stability, as the heir can afford to be patient, and the
elite observe the longer time horizon provided by the prospect of the son’s
inheritance. They note that other succession orders presented short-term
benefits not necessarily available under primogeniture, which explains
variance. Examining similar literature from a non-monarchic perspective, they
note that personalist regimes and one-party systems demonstrate similar, albeit
not congruent, succession orders. They then examine prior historical studies,
which either fail to directly test the hypothesis, or are, in one example, a
case study with generalizability issues. They then note that they need to
control for the rise of parliaments and state capacity, both of which have a
positive, confounding influence on regime stability.
Question 2: The dependent variable is the likelihood of
deposition over time. Succession failure is thus the threshold measurement, and
is coded as one for monarchs who were murdered, forced to abdicate, or died in
a civil war. All other observations were right-censored. Then, the authors use
a Cox proportional hazard model to measure the relative probability of deposition
with each succession order, and every control variable. This measure is
internally valid because of the analysis outlined in Question 1, and is
externally valid because the question tested – namely, the role of succession
order on autocratic stability – can be generalized to other situations. Indeed,
the authors note in their literature review two studies of modern autocracies
that suggest similar findings, although they are quick to note that neither one
exactly replicates their approach. They also use their same test on the few
remaining modern autocratic monarchies. This measure is reliable because, while
the authors are not paragons of clarity in explaining their methods, once their
approach is parsed, it is easy to replicate; they even promise that the
historical sources supporting their coding choices are offered in their
Appendix (I couldn’t find the Appendix, so I can’t evaluate that claim
directly; presumably the reviewers would not have let something that obvious
pass); thus, other scholars can evaluate, not only the models and methods used,
but also the coding choices for cases, and can determine if the authors made
coding mistakes.
Question 3: The primary independent variable is succession
order: election, agnatic succession, and primogeniture. This is measured first
with elective monarchies as the reference (i.e., IV=0), and agnatic or
primogeniture succession as variation (IV=1). Then, the authors separate
agnatic and primogeniture succession by testing whether monarchs who were
succeeded by sons survived longer than monarchs who were succeeded by brothers
or other relations. They acknowledge that this may introduce endogeneity, but
provide that it’s the best of a set of all-bad choices. These measures lack
internal validity, as acknowledged by the authors. The first, dichotomous
measure is internally valid, as the authors argue that agnatic and
primogeniture successions may be usefully distinguished from elective
monarchies because of the coordination problem. They acknowledge the lack of internal
validity on their second measure, but argue that nothing better presents itself
due to the paucity of data. They similarly assert that their modern data set
presents a pattern so strong as to “inhibit statistical analysis.” I will argue
in Question 6 that a modern data set which encompasses both monarchies and
personalist regimes (the modern version of monarchy) could successfully test
the crown-prince problem, as a modern study will be able to get the data the
authors are missing. The measure is reliable for the same reasons explained in
Question 2.
Question 4 and 5: The authors explain why they chose the two
alternative hypotheses that they control for. Parliaments are controlled for
because a prior study supported the notion that parliamentarianism drove
political stability regardless of regime type. Since parliamentarianism varies
even within autocracies, it is necessary to control for it. They similarly
control for state capacity because it varied over time; thus, to control for
the ability of the regime to suppress dissent and thus raise the costs of a
succession crisis regardless of the succession order, state capacity must be
controlled for. The authors then cite two historical studies for the
proposition that other than these two variables, that medieval monarchies did
not vary in any relevant way. Methodologically, they control for coding issues
in a variety of ways. First, because they reconstructed a major dataset
rejecting certain coding choices, they run their models using the original
dataset in order to check for robustness. Second, they run their models using
the original dataset less certain cases involving foreign depositions. Thus,
they argue, this test captures only “clear-cut cases of [domestic] elite
depositions.” They control for sovereignty unions by counting depositions that
lead to divisions, and censoring depositions that lead to death outside of the
state where the deposition occurred. They control for the stabilizing effect of
designation by testing whether sons or brothers who succeeded lasted longer
than other relatives or non-relatives. They also attempt to test the
“crown-prince problem” by evaluating a monarch’s survival based on whether they
were succeeded by a son or some other relation, acknowledging that this
presents methodological problems of its own. Finally, they code de jure
succession orders, but re-run the model including de facto primogeniture as a
check, and find that it confirms their findings. The authors also include four
control variables intended to eliminate non-hypothesis confounders: branch of
Christianity, foreign depositions and enemies, and whether the ruler’s
predecessor was domestically deposed (to control for civil wars or lengthy
instability). I am concerned about the use of religious sectarianism as a
control variable, as the authors provide no justification for it – an argument
that sectarian depositions are somehow qualitatively different than “pure”
political depositions would have been nice, but I don’t think it could have
been legitimately made. However, that variable could, arguably, serve as a proxy for the control variables I would
have liked to have seen, which is rulers and elites in a state who traveled to
the Holy Land as Crusaders. That variable, which may be uncollectable in any
reliable way, would capture a) the instability of regimes governed by absentee
rulers and b) the increased stability achieved by sending the most violent and
ambitious elites to the Holy Land. I don’t like religious sect as the authors
code it for that purpose, because the Protestant Reformation was post-Crusades,
and thus coding for Protestant rulers accomplishes nothing for my concern; the
First Crusade, at least, was called at the request of the Byzantine Emperor,
who was Orthodox, which means that distinguishing between Orthodox and
Catholic, at least early in the period, does not capture the value of the
Crusader variable (by the time of the Fourth Crusade, of course, such a
distinction becomes very important);
and for the purpose of Crusading, I’m pretty sure religion doesn’t vary.
Question 6: I would extend this study by creating a modern
(post-twentieth century) autocratic dataset that would encompass both the
monarchies that the authors briefly examined, and personalist regimes such as
Cuba, North Korea, and Syria. This would allow me to test the coordination
problem and the crown-prince problem. However, it would introduce ambiguity in
the succession order coding, as personalist regimes rarely codify a succession
order. However, by using the same methods of analysis as the authors, with the
exception of right-censoring peaceful transitions and unclear transition, as
that would eliminate most cases, given modern censorship by personalist
regimes, then the study can be usefully replicated.