When Purges Fail: Organisational Structure and Mass Purges

• In Event: The Structure of Armed Groups

Abstract

Mass purges are ubiquitous, both in transfers of power, either by revolution or electoral upset, as in democratic spoils systems e.g., the 19th century US, and cyclically in authoritarianism. Previous work on purges has focused on strategic aspects, such as purging to dissuade future coups d’etat (Easton and Siverson 2018) or principals violently motivating agents (Montagnes and Wolton 2019). Nevertheless, many purges are based on observed traits, with scant regard for one’s behaviour. Holding office in the old regime, regardless of rank or importance, is an especially salient trait because rulers often gain power by promising to throw out the bums and give newly-vacated positions to supporters. A purge may therefore be considered a failure if too many old faces remain in important posts. Unfortunately, previous research has overlooked both purge failures and the role of ascriptive traits.
This paper addresses such shortcomings by modelling how a purge of incumbents (those who held office before a certain date) impacts hierarchical organisations, such as churches or state bureaucracies. This is accomplished via a computational, agent-based model (ABM; de Marchi and Page 2014) of mobility, e.g. promotion, recruitment. This incorporates a key element of hierarchical organisations: mobility is only possible when a job becomes vacant, usually because someone leaves the system; and purges are precisely synchronised mass dismissals. Such models allow us to examine how organisational structure can dampen purges, sometimes to the point of failure.
I examine five dimensions, each attested historically, that condition a purge’s success in removing old regime cadres from positions of authority. 1) speed: a Night of Long Knives vs. the decade-long effort to replace “bourgeois specialists” with “red experts” in the USSR; 2) breadth: 15,000 bureaucrats replaced after Romanian elections vs. modest purges in Portuguese ministries in 1974; 3) expansion/contraction of organisational size: Bolsheviks growing the Academy of Sciences vs. shrinking the Orthodox Church; 4) the addition/removal of hierarchical levels: Nicolae Ceauşescu’s reorganisation of the Romanian territorial administration vs. Romanian communists eliminating appellate courts; 5) the strictness of seniority rules for promotion: Romanian communists loosening seniority rules in the late 1940s to colonise the judiciary vs. tightening said rules ten years later to cement their control of the law.
Specifically, I introduce an ABM featuring hundreds of agents of two types (vacancies and actors), interacting over fifty steps (representing years), across a hierarchy of positions. Vacancies swap positions with actors (so a vacancy moving down one level implies an actor being promoted) subject to organisational rules, e.g. an actor can only be moved once per step. Actors can retire each step and must retire after thirty steps. Actor retirement introduces vacancies, as do mass dismissals and the creation of new, vacant positions. While agent movement probabilities are derived from empirical observations, both are constrained by organisational rules. Actors are aimless and purely stochastic, while vacancies try to exit the system as quickly as possible.
A first set of ceteris paribus results is derived from computational experiments simulating stylised purges. Afterwards, I parametrise ABMs using real mobility data on three US churches, one US police force, and the Romanian national judiciary and procuracy. Finally, I subject these empirically-parametrised models to distinct historical purges: Soviet church purges from 1917 to 1940, Saakashvili’s post-2003 purge of the Georgian police force, and the 1945-1952 purge of the Romanian judicial system. These realistic purges combine dimensions, thereby enabling us to explore the net effects of complex purges.
I find that while all purge types reduce the number of incumbents, certain purge tactics actually inflate the old cadres’ percentage in the upper levels, paradoxically creating organisations in which old timers are more prominent than they would have been without the purge, indicating a failed purge. The speed of a purge, for example, affects its direction. A one-year purge leads to a slower rate of retirement in middle and high levels than would’ve been the case otherwise, while spreading out the purge over five years accelerates retirement across the board, as most purges actually intend. I derive a rich set of results for all five purge dimensions.
To conclude, this paper addresses important gaps in research on mass purges by modelling how hierarchical organisations condition the effects of purges of incumbents. The proposed approach is not limited, however, to political purges, but can also apply to other situations featuring exogenous, personnel-oriented shocks, such as plagues, wartime conscription, or economic recessions. Ultimately, this approach helps us answer an overlooked question: when do purges fail?

Author

• Radu Andrei Parvulescu, Cornell University