The bureaucratic politics of authoritarian repression: intra-agency reform and surveillance capacity in communist Poland

Abstract

Coercive institutions' internal structures remain poorly understood. Bureaucratic reorganizations within security institutions cause significant variation in their behavior, however. Intra-agency reforms interact with officers' careerist incentives to cause changes in coercive capacity or repression. In this paper, I test the effects of intra-agency reforms on surveillance capacity. I exploit a rare source of exogenous variation in the structure of the secret police in communist Poland. Difference-in-differences models find that when security headquarters were duplicated through an administrative reform, the proliferation of higher-level posts within the service caused a large and statistically significant increase in the number of informants it employed. Intra-agency reform substantially altered the agency's coercive capacity. Previously overlooked dynamics within coercive institutions have important effects on authoritarian repression.

The threat and use of violence are essential aspects of authoritarian rule. Dictators empower security institutions to detect and repress challengers. However, we know relatively little about the internal structures of these institutions.¹ In the growing body of research focused on dictators’ secret police agencies, they have predominantly been characterized by their number, size, and social composition at the national level. Variation among coercive institutions has been explained by the nature and magnitude of threats to incumbent autocrats. Dictators facing an acute threat of mass opposition are likely to create large, socially inclusive security forces. Those threatened by intra-elite coups are likely to divide the security apparatus into multiple, smaller, socially exclusive agencies (Svolik, 2012; Talmadge, 2015; Chestnut Greitens, 2016; Hassan, 2020; Thomson, forthcoming).

It is unsurprising that we know little about the inner workings of autocrats’ security forces. Their organization and activities are shrouded in secrecy. Nonetheless, it is important that scholars understand the internal bureaucratic politics of authoritarian coercive institutions. Authoritarian regimes use frequent, incremental changes within coercive institutions to respond to threats and challenges; address inefficiencies or shortcomings in the operation of the secret police; and optimize the capacity, loyalty, and efficacy of their security apparatus. These shifts are as frequent within the authoritarian security apparatus as in any bureaucracy (Moe, 1987: 234). Far from unimportant or uninteresting, these intra-agency reforms are key to understanding how repression works under autocracy. They do not change the number of coercive agencies at the national level, or their social composition. They are not necessarily efficient responses to changes in threats facing a regime. Intra-agency reforms operate through the “terrifyingly normal” (Arendt, 1963: 253) bureaucratic and career incentives of staff working within the agencies.

Bureaucratic factors determine how security forces follow the directives of their masters, for example via standard operating procedures or criteria for officer promotion. They are a previously overlooked factor affecting the principal–agent relationship between the dictator and security agents or “guardianship dilemma” (Egorov and Sonin, 2011; McMahon and Slantchev, 2015; Tyson, 2018; Dragu and Przeworski, 2019; Paine, 2020). Intra-agency reforms can create institutional autonomy for agents within the secret police and exacerbate dictators’ fears of insubordination when they issue an order for repression, for example (Svolik, 2012). Similar dynamics also affect a regime's ability to repress popular threats (Hassan et al., 2022). A bureaucratic reform might create unclear or overlapping lines of communication and reporting that hamstring repressive operations, for example. Moreover, the secret police are a prominent source of coercive power and information within authoritarian regimes. Reforms to internal structures causing them to become more efficient in surveillance, for example, could destabilize existing power-sharing arrangements among elites by changing the distribution of information among them (Boix and Svolik, 2013; Casper and Tyson, 2014; Little, 2017; Meng and Paine, 2022; Meng et al., 2022; Luo and Rozenas, 2023).

In this paper, I do not directly examine the occurrence of dissent or repression under authoritarian regimes. Instead, I analyze the capacity of agencies that carry out repression. I examine how intra-agency reforms affect coercive institutions’ capacity at the local level. I argue that individuals working within the security apparatus are self-interested careerists seeking to maximize their prestige, influence, and resources (Wintrobe, 1998). These coercive agents respond to incentives and constraints provided by the bureaucratic structures surrounding them, and changes in those structures. Specifically, increases in the number of units tasked with a discrete task provide incentives for yardstick competition among self-interested individuals seeking promotion or more favorable assignments. When intra-agency reorganizations interact with these career concerns to cause yardstick competition, we may see changes in repressive capacity or outcomes. These would not be predicted by the current literature, because they do not reflect the thereat environment or popular grievances.

I test this argument empirically using an exogenous shock to the organization of the secret police in communist Poland. In 1975, the United Workers’ Party (PZPR) regime increased the number of administrative districts (voivodeships) in the country from 17 to 49. This reform caused a corresponding increase in the number of district state security offices—the “basic units of the Ministry of Internal Affairs at the voivodeship level” (Piotrowski, 2003: 53). State security offices organized repression, not least by administering the network of civilian secret collaborators who provided vital information and assistance to the security service. The reform caused short-run disorganization in the agency and a decline in its collaborator numbers that alarmed its leadership. Post-reform, local state security commandants faced pressure to grow their collaborator networks. Commandants in districts with more newly created administrations faced greater competition from proximate colleagues as they tried to distinguish themselves for promotion to desirable positions in the agency's hierarchy. Moreover, compared to before the reform there were greater opportunities for promotion as senior officials were rotated and barriers to entering the upper ranks of the agency declined.

I hypothesize that increases in the number of secret police administrations within a geographic area after 1975 exacerbated yardstick competition among officers. This competition caused them to register more secret collaborators in that area. Smaller post-reform voivodeships were nested within larger pre-reform districts. I can therefore estimate the effects of an increase in regional administrations on the growth of the agency's secret collaborator network while holding the population and area under surveillance constant. In a series of difference-in-differences models, I find that districts where the number of local secret police administrations increased saw significantly faster growth in collaborator numbers after 1975. I show that these effects were increasing with the magnitude of the administrative change and through time. In an extension of my core research design, I compare the effects of the administrative reform on the size of the collaborator network to those of a nationwide wave of unrest that occurred in June 1976. I find that the effects of reform and unrest on the collaborator network were similarly sized, demonstrating the salience of intra-agency reforms for coercive capacity, even compared to mass threats highlighted in previous research.

This paper makes a novel contribution to the literature on coercive institutions under authoritarian regimes. I uncover new sources of variation in coercive capacity and shed new light on determinants of regime stability (Hager and Krakowski, 2021; Nalepa and Pop-Eleches, 2022). I also make a novel contribution to our understanding of the subnational determinants of coercive capacity. Reforms inside coercive institutions have significant effects on their capacity at the local level that must be considered alongside arguments that repression will be targeted at threatening (Hassan, 2017; Thomson, 2017; Blaydes, 2018; Piotrowska, 2020) or concentrated groups (Tyson, 2018). I also make a contribution to the literature on authoritarian bureaucracies. Recent research has investigated promotion patterns and competition among civil servants under authoritarian regimes (Hassan, 2020; Pierskalla et al., 2020) and similar dynamics within coercive institutions (Wang, 2014). However, no previous studies demonstrate a link between bureaucratic reforms within coercive institutions and surveillance or repression.

1. Intra-agency reforms, administrative district proliferation, and coercive capacity

Authoritarian governments face a well-known dilemma when designing coercive institutions. They endow these agencies with capabilities to detect, deter, and eliminate threats. However, by establishing coercive institutions authoritarian regimes also create powerful actors with the means to challenge or undermine their masters (Egorov and Sonin, 2011; Svolik, 2012: 56; McMahon and Slantchev, 2015; Chestnut Greitens, 2016: 22–23; Tyson, 2018; Dragu and Przeworski, 2019; Thomson, forthcoming). The dilemma of enabling and controlling coercive institutions, a particularly pernicious form of principal–agent problem, determines important features of agency design. On the one hand, autocrats facing a more acute threat of mass opposition are more likely to create larger security agencies whose social composition closely mirrors that of the citizenry at large (Svolik, 2012; Chestnut Greitens, 2016; Blaydes, 2018). This allows for more effective cooperation in repression, and facilitates collection of intelligence on opponents. On the other, threats to the regime arising from within the ruling elite—including the security forces—are likely to create incentives to divide capabilities across multiple agencies and restrict enlistment to trusted social groups (Talmadge, 2015; Chestnut Greitens, 2016). These fragmented and socially exclusive institutions are expected to be more loyal to the dictator, and more cohesive in applying repression.

Changes in the size, number, and social composition of coercive institutions are important. However, they are rare compared to what I call intra-agency reforms: those affecting the number, function, or location of units; lines of reporting and oversight procedures; personnel policies such as those around appointments, promotion, and compensation; and standard operating procedures and performance indicators for staff (Moe, 1987: 234). Clearly, changes in any of these areas can be consequential enough to affect the institutional macro-structures discussed above, and warrant the attention of the regime leadership. However, intra-agency reforms are distinct because they occur within a single coercive institution and are primarily directed by the chief executive of the agency or their subordinates, not the ruling elite (even if they are endorsed or approved by elites). As Browder (1990: 3) notes in the case of Nazi Germany, “[a] great number of developments in the evolution of Sipo [Security Police] and SD [Security Service] … occurred without any evidence of Hitler's involvement or concern.”

Intra-agency reforms are the measures by which authoritarian regimes continually, incrementally optimize the design of their security apparatus, even while macro-structures of institution size, number, and social composition remain relatively stable. Intra-agency reforms do not have direct effects on entire agencies or their leadership. They cause variation in the incentives and constraints facing individual bureaucrats within institutions, and therefore variation in these individuals’ behavior. They have the potential to cause indirect, but significant, changes in the characteristics and behavior of entire agencies. For these reasons, they are also important determinants of patterns of repression and opposition (Hassan et al., 2022; Thomson, forthcoming), and intra-elite power-sharing (Boix and Svolik, 2013; Casper and Tyson, 2014; Little, 2017; Meng and Paine, 2022; Meng et al., 2022; Luo and Rozenas, 2023) that are of central interest to scholars of authoritarian politics.

Intra-agency reforms carried out by repressive agents can, but do not always, follow their elite principals’ goals and the priorities handed down to agencies as a whole. It is beyond the scope of this paper to explain intra-agency reforms or speculate on their intended purposes. Here, I am concerned with their effects. I assume that bureaucrats within coercive institutions are self-interested actors who aim to maximize their prestige, policy influence, and the resources at their disposal (Wintrobe, 1998). These motivations exist at the agency level, even for a force under the strict hierarchical control of an authoritarian regime, because it is difficult for the incumbent dictator to monitor agency behavior and sanction insubordination. The performance and power of a coercive institution have real effects on the influence, prestige, and living standards of bureaucrats within it. Individual staff, therefore, face collective incentives to promote the agency's interests as a whole.

Individual bureaucrats also have narrower, selfish interests in career advancement and prestige, however. In order to achieve promotion, greater compensation, and privileges, or simply to retain their positions and avoid sanctions from superiors, employees in coercive institutions respond to the incentives and constraints provided by their position in the agency. Individual coercive agents’ interests are thus not determined only by those of the institution as a whole, but by features of the bureaucracy affected by intra-agency reform.

Intra-agency reforms are diverse and—given the nature of authoritarian regimes and their security services—very opaque. To link this broad class of institutional changes to readily observable reforms and the hypotheses I test in this paper, I focus on changes in the number of units within the coercive apparatus caused by reform of the number of administrative units within states (Grossman et al., 2017). The geographic reorganization of governance implied by administrative district reform is often a function of broader political conflict, particularly under authoritarian regimes. One implication of this type of reform for coercive institutions is a change in the number of units within the bureaucracy that fulfill the same task. When regional administrative centers are created or abolished in provincial cities, corresponding offices of the security services responsible for organizing and overseeing their local operations are established or dissolved alongside those of other arms of the state bureaucracy such as the courts, tax administration, or primary and secondary education institutions.² Such reforms occurred not only in communist Poland, but also in the neighboring German Democratic Republic, where five states were replaced with 14 smaller counties in 1952. In communist Bulgaria, the number of administrative districts was reduced from 28 to 9 in 1987. In a similar more recent reform, the Chinese military consolidated its seven former military regions into five new theater commands in 2015.

Changes in the regional organization of the coercive apparatus have effects on agencies’ activities. The number of agency chiefs accountable to central authorities and responsible for coercion within a given geographic area influences the degree of competition among these individuals. Increases in the number of regional secret police administrations create new senior posts aspired to by officers and reduce barriers to entry to these positions. They therefore increase competition among similarly qualified administrators within the security forces who wish to acquire these posts. Because the number of senior roles and the amount of compensation, privileges, and resources available within agencies is finite, increasing the number of individuals performing a task increases their incentives to out-perform each other, even if their authoritarian principal's goals remain the same. Especially under authoritarian bureaucracies, where power and resources are highly concentrated at the top of the hierarchy, competition among bureaucrats resembles a winner-take-all tournament. The creation of more senior positions induces greater effort among subordinates as they compete more intensely to attain them (Li and Wang, 2023: 4–5). This has particular relevance for yardstick competition, or production of tangible, quantifiable outputs that are used for individuals’ performance evaluation against their peers by superiors. Where competition is among individuals responsible for functionally identical units, promotion-seeking security chiefs have incentives to signal their competence and loyalty to superiors by out-performing others on these sorts of tangible performance indicators (Maskin et al., 2000: 360; Grossman et al., 2017: 827–29).

2. Administrative district proliferation and secret police surveillance capacity in communist Poland

The secret police were a core pillar of the authoritarian Polish United Workers’ Party regime from 1945 to 1989. Known as the Bezpieka, the Polish security apparatus was constructed under Soviet tutelage to follow the model of the NKVD, the predecessor of the KGB in the USSR.³ Like its counterparts in the other state socialist regimes of central and eastern Europe, it was a large, unified, socially, and ideologically homogeneous institution.⁴ It penetrated deep into Polish society, using a large network of collaborators and informants, surveillance and postal censorship to detect opposition to the PZPR. It interrogated, tortured, imprisoned, and killed these opponents at the behest of the communist regime.

Reforms of the Bezpieka's macro-structures were rare. The position of the coercive agency within the broader state bureaucracy changed twice between 1944 and 1956, but its mission, social composition, organizational structure, and status vis-a-vis the military remained relatively stable. The greatest change in the agency's history was in its size, which declined significantly after the death of Soviet dictator Josef Stalin in 1953, stagnated from 1956 to 1980, and grew again during the period of martial law and the final crisis of the regime through 1989 (Dudek and Paczkowski, 2009; Thomson, forthcoming).

Intra-agency reforms within the Bezpieka, on the other hand, were frequent. The security service's multi-divisional geographic organization remained stable through time. Functionally identical self-contained administrations existed in regional centers and essentially replicated the organization of the agency in Warsaw. The internal organization of administrations in the regions and at the center, on the other hand, followed a unitary governance form. As I illustrate for the national level in Figure 1, administrations contained functionally distinct, specialized directorates responsible for operational tasks such as observation or encryption; targeting specific opposition groups or sectors of the economy; and other tasks such as maintaining archives or running hospitals and cafeterias.

Figure 1. Selected Units, Security Service (SB), Polish Ministry of Internal Affairs (MSW), 1975.

Units containing significant numbers of secret collaborators are shown in white.

Source: Ruzikowski (2003), Dudek and Paczkowski (2009), Piotrowski (2008: 19, 35–42).

These specialized directorates were relatively frequently renamed, merged with others, or abolished. New units were created to address novel threats, such as in June 1962, when a Fourth Directorate was created to target the Catholic church. There were some large-scale purges of Bezpieka personnel, however they predominantly occurred before 1956. These sorts of staff fluctuations undoubtedly had significant consequences for specific individuals’ chances of promotion within the agency, and in the mid-1950s were associated with institutional reform and a dramatic reduction in the overall capacity of the Bezpieka, but they did not change the core mission of the coercive apparatus.

In 1975, the Polish security service was subject to a major reorganization. This change was caused by a reform of the geographic administrative divisions of the country. Poland's three-tier regional governance structure of 17 districts or voivodeships (województwa), over 300 powiaty, and over 4000 communes (gromady) was abolished.⁵ It was replaced with a two-tier administrative structure of 49 smaller post-reform voivodeships and around 2500 communes (gmina). This reform affected the entire state apparatus (Letowski, 1976: 65–67; Gorzelak, 1992: 481). Importantly, there is no evidence that the reform was directed at the secret police or intended to strengthen its apparatus at the local level. Because the reform was so wide-ranging, affecting the judicial system, schools, and other core administrative functions, it was decided on and organized by the highest national regime leadership. It is unlikely that it was targeted at the security service, let alone at specific functionaries responsible for the secret informant network. Instead, it was part of a broader power struggle within the PZPR between the central party leadership under Edward Gierek, who had succeeded Władysław Gomułka after a wave of mass unrest in 1970, and regional bosses. The administrative reform weakened powerful regional political elites within the ruling party by dividing the relatively large provincial administrations under their control into much smaller units and placing them in the hands of loyal functionaries. The reform came during a period of relative stability in the size and institutional makeup of the secret police. It had the unintended short-run effect of creating administrative chaos, including within the Bezpieka, and caused the agency's informant networks to temporarily contract (Surazska et al., 1996: 442–44; Ruzikowski, 2003: 115; Pleskot, 2021: 8, 11–12). Indeed, the reform aimed to reduce the power of local institutions vis-a-vis the center (Lewis, 1989: 19–28; Gorzelak, 1992: 481; Yoder, 2007: 436).

The proliferation of voivodeships in 1975 caused a simultaneous reorganization of the state security service within the Ministry of the Interior. The Bezpieka continued to follow a multi-divisional organizational form, with functionally equivalent self-contained units in regional centers. However, there was a proliferation of these Bezpieka administrations as they were established in each post-reform voivodeship. The 1975 administrative reform was therefore wide-ranging, affecting many state institutions aside from the Bezpieka, but it led to an intra-agency reform of the security apparatus. It did not change the number of security agencies at the national level, their oversight structures, or the social composition of their membership. The Bezpieka remained unfragmented and inclusive. The proliferation of state security administrations across the country led to a significant increase in the number of high-ranking positions within the agency. There were more than three times as many new regional leadership positions in the Bezpieka as there had been under the previous Polish administrative divisions, and these positions were prestigious: compared to pre-reform regional chiefs and their deputies, who had regularly held the rank of Lieutenant Colonel or even Major, leadership positions in the post-reform Bezpieka administrations were predominantly held by Colonels, who were sometimes promoted to Brigadier Generals (Piotrowski, 2006, 2008).

The hierarchy of the Polish secret police was strict, with officers organized by military rank. Promotion was usually methodical and slow, and determined predominantly by seniority and time served in position. Following the nomenklatura system, the leadership of both the party and security service were involved in the construction of lists of suitable candidates for Bezpieka managerial posts (Tymiński, 2017). Bezpieka officers were predominantly careerists, seeking promotion from lower to higher rank, and to move from secret police administrations in smaller provincial towns to larger cities and eventually the Bezpieka's central offices in Warsaw. The 1975 reform did not change this general career track or the aspirations of Bezpieka officers (Pleskot, 2021: 9). The sudden creation of dozens of coveted, high-ranking positions within the security apparatus provided new opportunities for officers who may have otherwise languished on the lists of candidates for promotion without advancing, however. As a consequence of the administrative reform, there was a major rotation of senior Bezpieka officials and promotion of outsiders to the new regional leadership positions, suggesting that barriers to entry were newly reduced (Piotrowski, 2006, 2008). In addition, the Ministry of the Interior's local offices—within which the secret police were housed—came to be increasingly dominated by the Bezpieka, as the share of provincial Ministry chiefs who were secret policemen increased from only 8 percent in 1975 to 37 percent by 1981 (Pleskot, 2021: 5–6). This was another avenue for promotion newly open to ambitious and successful secret policemen.

An increase in senior, prestigious roles within the Bezpieka and reduced barriers to entry to these positions created incentives for promotion-seeking local chiefs to distinguish themselves by recruiting greater numbers of secret collaborators to their regional administrations. Secret collaborators were private citizens who had made a formal commitment—usually in writing to a full-time officer—to provide information and assistance to the Bezpieka. As in the other state socialist dictatorships of central and eastern Europe, a large network of secret collaborators was an essential component of the Polish coercive agency (Ruzikowski, 2003; Kamiński et al., 2009). Collaborators contribute to authoritarian strategies of political control by helping coercive agencies infiltrate society, collect information on opponents, mobilize or demobilize groups, and induce compliance with the regime's directives (Hassan et al., 2022: 10). The number of individuals employed in such strategies of social penetration is a key indicator of coercive institutions’ intelligence capacity and likely associated with specific strategies of repression and social control (Chestnut Greitens, 2016: 42–49). Polish secret collaborators’ ubiquitous presence was “a tool of terror ... producing an aura of fear” among the population that deterred dissent (Dudek and Paczkowski, 2009: 304).

Bezpieka collaborators covertly gathered information on potential threats to the ruling communist parties from their professional and social networks, and passed it on to their handlers within the coercive agency. They also provided practical assistance to the security services, for example by allowing them to use their homes for meetings. Recruitment was often the result of blackmail or intimidation, but individuals’ personal vendettas, political convictions, and desire to help the state security apparatus played a non-trivial role. Collaborators were sometimes paid and were regularly given non-monetary compensation such as permission to apply for a passport to travel abroad.

By early 1976, the central Bezpieka administration had issued orders to “immediately counteract” the downward trend in secret collaborator numbers caused by the 1975 administrative reform (Pleskot, 2021: 12). This was an era of growing mass threats to the PZPR regime, and surveillance units at the local level “began to dominate numerically over the other structures of voivodeship headquarters” (Piotrowski, 2003: 67). Regional secret police chiefs were thus expected to increase the size of their informant networks. The number of secret collaborators was a key element of superiors’ assessments of officers’ work, making these officers grow their informant networks seeking promotion. Collaborators were formally registered at the local level and carefully tracked, either annually or quarterly, by the central office of the institution in Warsaw. As I depict for selected units of the Polish apparatus in Figure 1, at the time of the administrative reform in 1975, secret collaborators were predominantly recruited and enumerated in Directorates II–IV, those units tasked with counter-espionage, the protection of the state and the economy, and repressing opposition within the Catholic church.⁶ Depending on time period, secret collaborators were also distinguished and documented in different categories. For example, in the 1950s, informants only provided information to the Bezpieka, while residents ran their own networks of informants and agents actively infiltrated and subverted opposition groups. It is likely that a similar categorization of collaborators persisted into the 1970s. Readily quantifiable and even able to be distinguished by target and function, secret collaborators were a tangible and objective indicator of unit performance within the repressive bureaucracy—and therefore well-suited for yardstick competition among officers.

The Polish security service was aware that secret collaborators could be recruited not to provide information but to boost administrators’ apparent performance. High-ranking officers therefore sometimes expressed frustration at over-recruitment by their subordinates. The agency adopted strict formal criteria to discourage the employment of collaborators who did little to help the agency. It also carried out audits of collaborators and their output. In the mid-1950s, large numbers of unproductive secret collaborators were dismissed, for example (Ruzikowski, 2003: 113–15) and a similar “cleaning out” of the informant network occurred after the 1975 administrative reform (Pleskot, 2021: 12). However, the short slump in collaborator numbers that was caused by the reform obviously alarmed the Bezpieka elite, causing them to issue orders to reverse the trend and grow the informant network quickly. Simultaneously, the logic of yardstick competition among a suddenly expanded group of senior regional Bezpieka administrators affected the incentives of individual regional chiefs to follow this directive, with those facing greater local competition facing greater incentives to recruit collaborators. This bureaucratic logic of informant recruitment leads to the following testable hypothesis relating intra-agency reform to the surveillance capacity of the Polish coercive agency:

Hypothesis: Increases in the number of secret police offices within a geographic area after 1975 caused increases in the number of secret collaborators registered by the coercive agency within that area.

3. Research design

I use a difference-in-differences research design to model the effect of the division of a district on secret collaborator numbers. Large pre-reform districts were affected unequally by the 1975 administrative reform. Some were not altered, while others were split into up to four smaller post-reform voivodeships. Because the post-reform voivodeships were imperfectly nested within the larger pre-reform districts, I am able to model changes in collaborator numbers in the same geographic area—the pre-reform districts—in response to a proliferation of regional Bezpieka administrations.

The causal effect of secret police offices on collaborator numbers can be estimated if the trends in collaborators across districts that were split by the reform and those that were not are parallel before the administrative reform (Angrist and Pischke, 2009: 230–31). In the left-hand panel of Figure 2, I show the trends in the average number of collaborators from 1970 to 1980 for the districts that were split and those which were not. In both groups, average collaborator numbers were increasing at a very similar rate before the 1975 reform. There is a striking divergence in the trends of average collaborator numbers after 1975. In the right-hand panel of Figure 2, I present an identical graph, distinguishing by the magnitude of the voivodeship split. Here, the parallel trends assumption also holds.

Figure 2. Trends in secret collaborator numbers by district split, 1970–1980.

Data: Ruzikowski (2003: 128–129).

I estimate the effect of secret police offices on collaborator numbers using the following difference-in-differences linear panel regression:

(1)$$\matrix{Collaborators_{i, t} & = \beta_{1} + \beta_{2}District\ Split_{i} + \beta_{3}Reform_{t} \cr & \hskip-1pc + \; \beta_{4}District\ Split_{i}\times Reform_{t} \cr & \hskip-5.4pc + \, \gamma_{t} + \upsilon_{i} + \epsilon_{i},\; }$$

where Collaborators _i,t is the number of secret collaborators in district i in year t, District Split indicates whether the pre-reform district was divided through the reform, Reform is a binary variable which is coded one for all districts after 1974, γ _t is a year fixed effect, υ _i is a district fixed effect, and $\epsilon _{i}$ is an error term clustered by district.⁷

In these models, the estimate of the effect of an increase in the number of regional secret police commandants on the number of informants in a district is the coefficient β ₄ on the interaction between District Split and the Reform indicator. Because the exact nature of the reform varied across voivodeships, in some models District Split is a binary indicator of any division, while in others it captures the Split Magnitude, or the increase in the number of local secret police administrations compared to the pre-reform status quo. Another set of models uses event study estimation to allow effects to vary through time by interacting District Split with a series of annual dummy variables. In further analyses, I include an indicator of mass unrest in 1976 as an independent variable alongside District Split.

I restrict most of my analyses to the period between the conclusion of Stalinist repression in 1956 and the declaration of martial law in 1981. This was a period of relative stability within the Polish security apparatus. As I show in Figure A3 in the online Appendix the collaborator network in the mid-1970s grew steadily at a rate of around ten percent per year. As a robustness check, I estimate some models including only the years 1969–80, the period after the Soviet invasion of Czechoslovakia in 1968 and including the wave of popular opposition to the PZPR regime that began to escalate in 1970.

It is important to consider whether the District Split variable represents an endogenous treatment, whereby the state was responding to opposition or other political objectives in carrying out the administrative reform. This does not appear to be the case. In Figure A4 in the online Appendix, I present the results of linear regressions modeling differences in six pre-reform (1965) characteristics across treated and untreated districts. These balance tests indicate no significant differences in these characteristics across the two groups of districts, except population.

4. Data

To construct my dataset and the District Split variable, I match post-reform voivodeships to the territory of pre-reform administrative districts. I do so using maps by Martí-Henneberg (2005) shown in Figure A1 in the online Appendix and generating the coding presented in Table A1. For more details on the data collection process and descriptive statistics, see the online Appendix.

4.1. Dependent variable

Collaborators, is the annual number of secret collaborators registered with the Security Service of the Polish Ministry of Internal Affairs in each voivodeship Ruzikowski, 2003. I aggregated annual data on the number of collaborators per voivodeship following the coding scheme laid out in Table A1. All collaborators registered with a post-reform voivodeship state security office are therefore assigned to the pre-reform voivodeship within which that office was located. In this way, I create a balanced voivodeship–year panel dataset from 1950 to 1984, where the units of analysis are pre-reform voivodeships.

4.2. Independent variables

The binary District Split variable measures whether the administrative reform affected a pre-1975 voivodeship. It takes a value of one after 1975 for all areas except Olsztyn on the Baltic coast, Opole in the south, and Szczecin on the north-western border. The Split Magnitude variable captures the increase in the number of regional state security organizations created through the administrative district reform. Unrest 1976 is coded as one if a pre-reform voivodeship experienced a strike or more violent form of social unrest during the wave of mass anti-regime contention in June 1976 (Bernhard, 1987). Readers should note that the inclusion of this post-treatment control does raise endogeneity concerns, specifically if there was a direct or indirect effect of the 1975 reform on the likelihood of unrest in 1976. I base my strongest conclusions on the causal effect of intra-agency reforms on models that do not include the Unrest 1976 variable.

5. Results

I begin my analysis by regressing Collaborators on District Split, and its interaction with an indicator for the period after the 1975 Reform. I present results of these difference-in-differences models as models 1.1–1.6 in Table 1 and in the left-hand panel of Figure 3. They strongly support the hypothesis that increases in the number of secret police administrations within a pre-reform voivodeship were associated with increases in the number of collaborators registered in that area.

Table 1. Difference-in-differences model results

Standard errors in parentheses.

*p < 0.10, **p < 0.05, ***p < 0.01.

Figure 3. Results of difference-in-differences models, Table 1.

In model 1.1, districts that were split have significantly more collaborators after the 1975 reform than those that were not. The effect of the 1975 reform on the size of the collaborator network in divided voivodeships is positive and statistically significant (p < 0.013). As I show in the left-hand panel of Figure 3, split voivodeships had an average of 1132 collaborators after 1975, holding all else constant, while undivided voivodeships had an average of 773, or 32 percent fewer. In model 1.2, I restrict the sample to include only the years 1969–80, the period following the Soviet Union's invasion of Czechoslovakia. Results remain very similar. The interaction of the Reform and District Split variables is positive and significant (p < 0.052). Districts that were split by the 1975 reform had 264, or 26 percent, more registered Bezpieka collaborators than districts that were not divided, holding all else constant. In model 1.3, I include the years 1957–80, as in model 1.1, but I also include all pre-reform voivodeships, including those where post-reform boundaries do not map neatly to pre-reform voivodeships. The results of this model are also very similar to those of model 1.1. Split districts have 435 more collaborators post-reform than those that were not split, and this difference is significant at the p < 0.01 level. In model 1.4, I exclude the district of Poznan that was split into four by the reform. Results are substantively similar to model 1.1. Districts that were split had 1048 informants compared to 764 in those that were not split, and this effect is significant at the p < 0.03 level.

In model 1.5, I add to previous specifications by controlling for total population. Population is not a significant predictor of collaborator numbers (p < 0.47). Its association with collaborator numbers is very imprecise. This is demonstrated by Figure A2 in the online Appendix and by the results of model 1.9 where the sign on Population is negative, versus model 1.5 where it is positive. The Polish security services did not respond mechanically to population growth with greater collaborator numbers, not least because of bureaucratic incentive structures. The inclusion of Population in model 1.5 only minimally changes the substantive size of reform's effect, which here is an increase in average collaborators from 788 to 1072. However, controlling for population reduces the statistical significance of reform somewhat to p < 0.11. The population data in my models are imperfect as they are interpolated over five-year periods and could miss key demographic developments. It is therefore likely that the Population variable is introducing some measurement error into the model, reducing the significance of the reform. In a final model 1.6, I control for the effects of Unrest in 1976. Although the inclusion of this post-treatment control raises endogeneity concerns, it does not change the main results. The substantive size and statistical significance of reform's effects on collaborators are almost identical to those in model 1.1. Interestingly, the effects of the 1975 reform are very similar in magnitude to that of unrest in 1976. Voivodeships that were split had 367 more collaborators, on average (p < 0.01), while those areas that saw unrest in 1976 had 324 more collaborators (p < 0.05). This model demonstrates that intra-agency reforms were a significant determinant of the size of the Polish Bezpieka's secret collaborator force, even when compared to the agency's response to a major episode of mass social unrest.

I now operationalize the administrative reform using the ordinal measure of Split Magnitude. This variable ranges from zero in those pre-reform districts that were not divided, to four in Poznan, which was divided into four post-reform districts. In models 1.7, 1.8, and 1.9 in Table 1, I interact indicators of each level of this variable with Reform, which is coded as one after 1975. Results indicate that the reform's effect on secret police collaborators was monotonically increasing in the number of new administrations created. I graph the results of model 1.7 in the right-hand panel of Figure 3. Pre-reform voivodeships that saw one new administration added through the reform had 209 more collaborators than those voivodeships that were not split, on average. The effect of a split of this magnitude is not statistically significant (p < 0.30). Voivodeships that saw two administrations added through the reform saw a greater increase in collaborators, 231 on average, though this effect also does not reach statistical significance (p < 0.14). The effect in areas where the reform added three secret police administrations is much larger—an increase of 554 collaborators, on average—and statistically significant at the p < 0.04 level. Although only one pre-reform voivodeship added four administrations in the reform, Poznan, the estimated effect of this change is an addition of 732 collaborators, holding all else constant (p < 0.01). In model 1.8, I restrict the sample to the years 1969–80, and results are similar to model 1.4 but weaker. The effects of each level of the Split Magnitude are smaller in magnitude, and their statistical significance is diminished. The effect of the addition of three secret police administrations in a voivodeship on collaborator numbers is around 28 percent smaller than in model 1.7, and only significant at the p < 0.14 level. Finally, in model 1.9 I include 1957–80 but control for district population. Here, the effect of Split Magnitude is very similar in size and statistical significance to model 1.7.

Finally, I use event study estimation to allow the effects of reform to vary through time. Models 2.1–2.4 in Table A3 in the online Appendix include interactions of District Split with year dummy variables. Results of these models indicate that there are no pre-treatment differences in collaborator numbers across split and unsplit districts. They also indicate that the reform's effects on collaborator numbers grew monotonically through time. This suggests that each area did not have a maximum carrying capacity or equilibrium level of informants, to which the network converged after the reform. Instead, competition among officers caused informant numbers to continue to grow for at least five years, showing no sign of reaching equilibrium.

I graph the results of model 2.1 in Figure 4. They show that before 1975, differences in collaborator numbers across split and unsplit districts are small and statistically insignificant, reaching a maximum of 107 (p < 0.23) in 1972. By the end of 1975, after the administrative reform was implemented on 1 June, voivodeships that had been split into two or more secret police administrations had 95 more collaborators than those that had not been split, on average (p < 0.31). By 1977, the positive effect of District Split is much larger, at 304 collaborators on average, and statistically significant (p < 0.04). Both the magnitude of the effect of District Split and its statistical significance increase monotonically through time. By 1980 districts that had been divided through the reform had 489 more collaborators than undivided districts, on average, and this difference is significant at the p < 0.01 level.

Figure 4. Effects of district split through time, model A3.1, Table A3.

Model 2.2 includes all districts, including those where post-reform districts nest only imperfectly. Here, the effects of the reform are similarly statistically significant and slightly greater in magnitude than in model 2.1, increasing from 141 collaborators in 1975 (p < 0.15) to 622 in 1980 (p < 0.01). Model 2.3 excludes both the problematic matches included in model 2.2 and Poznan, which was alone in being split into four post-reform districts. The statistical significance of the reform's effects is moderated here somewhat, reaching only p < 0.49 in 1975, p < 0.08 in 1977, and p < 0.03 in 1980. Model 2.4 includes the same sample of districts as model 2.1, but extends the time period under analysis back to 1957. Due to space constraints, I only show coefficients for 1970–80 in Table A3, but I graph the full range of years in Figure A5 in the online Appendix. This model shows slightly larger pre-treatment variation across split and unsplit districts, especially in 1972–73, but these differences remain statistically insignificant. Much larger differences emerge by one year after the 1975 reform (333, p < 0.03) and by 1980, split districts have 629 more collaborators on average (p < 0.01).

The results of these three sets of models all indicate that the 1975 administrative reform had significant effects on the development of the Polish Bezpieka's secret collaborator network. They are reconcilable with my argument that the reform created incentives for yardstick competition among promotion-seeking local secret police chiefs. Not only were divided voivodeships associated with significantly larger collaborator numbers, but greater numbers of new Bezpieka administrations within pre-reform voivodeships were associated with more collaborators. This suggests that local chiefs competed with close neighbors in their recruitment of collaborators, and where the potential for competition was heightened through greater increases in Bezpieka administrations, chiefs responded with collaborator recruitment. Furthermore, the effects of the reform on collaborator numbers were increasing through time. This suggests that local chiefs did not quickly reach a new equilibrium level of collaborator recruitment, but that they sought to continually outdo each other at least until the imposition of martial law in 1981.

6. Conclusion

In this paper, I draw attention to intra-agency reforms within coercive institutions. These reforms, which are relatively frequent but difficult to observe, have significant effects on the nature of repression under authoritarian regimes. This is a major contribution to our understanding of coercive institutions and authoritarian politics in a range of different settings. Autocrats’ design and use of coercive institutions should not be seen as an unproblematic response to political threats. Directives given to these agencies are filtered through internal bureaucratic politics that have significant effects on their activities and outputs. In some cases, such as Poland and its broader region in the wake of the World War II, authoritarian regimes create wholly new coercive institutions or have them imposed by foreign powers. In many others, for example in the post-colonial or post-Soviet worlds, regimes inherit coercive institutions from their predecessors. The internal structures of these institutions have long-lasting effects on regimes’ coercive capacity and patterns of repression and opposition.

Intra-agency reforms pose thorny problems for data collection and research design. It is rare for researchers to be able to observe and measure them or their consequences. Following the research design of this paper, I suggest that studying the effects of exogenous shocks to authoritarian bureaucracies is a promising strategy for analyzing intra-agency structures and reforms. Although it may be very difficult to directly observe the incentives and constraints facing individual bureaucrats within these agencies, the ways agency behavior changes in response to exogenous shocks can reveal much about how they operate. By directing our attention to the internal structures of coercive agencies and studying how their behavior changes under the influence of exogenous shocks, we can draw important new insights into how repression works under authoritarian regimes.