INTRODUCTION
Bureaucrats play a central role in government as key providers of public services, from healthcare and education to transportation and law enforcement. This position gives them both substantial influence on who gets what from government (Slough Reference Slough2022; Xu Reference Xu2023) and grants them significant political power. By affecting service provision, bureaucrats can shape citizens’ perception of government performance and how they hold politicians accountable. Because citizens find it difficult to accurately attribute responsibility for poor service delivery—whether it stems from bureaucratic actions or politicians’ policies—bureaucrats can influence public perceptions and electoral outcomes. This article examines the conditions under which bureaucrats can leverage their control over service delivery as a political tool to resist reforms they oppose, undermine the reelection prospects of unaligned politicians, and influence policymaking to align with their preferences.
While often overlooked in social science research, anecdotal evidence increasingly points to politically motivated service provision by local bureaucrats. In 2021, protests erupted among municipal employees in several cities over vaccine mandates for their employees. Consequently, garbage accumulated noticeably in various neighborhoods across the country. In New York City, for example, sanitation workers in Staten Island and South Brooklyn left trash uncollected for more than a week around the implementation of the city’s COVID-19 vaccine mandate (Burkett Reference Burkett2021). City Sanitation Commissioner Edward Grayson attributed this lapse in service to the vaccine mandate, acknowledging that municipal garbage trucks were completing their routes with half-empty loads (Gross Reference Gross2021).
Similarly, recent research suggests that local police adjust their services to oppose reforms and influence city politics (Kyriazis, Schechter, and Yogev Reference Kyriazis, Schechter and Yogev2023; Wirsching Reference Wirsching2025). For example, officers of the San Francisco police department strongly opposed the progressive policies of District Attorney (DA) Chesa Boudin. During his recall campaign, San Francisco residents repeatedly raised concerns to city officials and the media that police were not responding to crime and justified their lack of engagement as due to the DA’s reluctance to press charges (Knight Reference Knight2021; Swan Reference Swan2021). In an interview, Chesa Boudin complained that “we’ve seen, on body-worn camera footage, police officers telling victims there’s nothing they can do and, ‘Don’t forget to vote in the upcoming recall election’” (Pearson Reference Pearson2023). This blame-shifting by police might have resonated with voters in a high-crime environment, who recalled the progressive DA by a significant margin. Immediately after the “unfriendly” attorney was successfully removed, police notably intensified their effort in making stops and arrests again (Kyriazis, Schechter, and Yogev Reference Kyriazis, Schechter and Yogev2023).
Yet, the logic, conditions, and consequences of such politically motivated bureaucratic resistance remain puzzling and largely unexplored. Why would bureaucrats engage in actions that disrupt public services for political reasons, knowing that voters will consider this possibility? And if this resistance affects how voters view reform policies, why would politicians ever push for reforms that bureaucrats oppose? In this article, we study how and when politicians’ electoral vulnerability motivates bureaucrats to undermine service provision, and how the potential for bureaucratic resistance influences voter behavior and an incumbent’s willingness to pursue reforms.
We integrate bureaucratic resistance into a model of electoral politics and policymaking, where politicians and bureaucrats co-produce public services. An incumbent chooses between a reform and the status quo after observing the true value of the reform. It is commonly known that the incumbent has a pro-reform bias, and the opponent is biased against it.Footnote 1 The voter observes the incumbent’s policy choice together with government service quality as a noisy signal of the true value of the incumbent’s policy choice. The voter uses the observed service quality to glean the reform’s merit and decides whether to retain the incumbent for a second period or to elect the opponent. Importantly, the reform’s inherent value and the bureaucrats’ performance are complementarities in the coproduction of government service quality. Bureaucrats who have an unknown degree of distaste for the reformFootnote 2 can privately choose to disrupt public service provision at a personal cost (e.g., by refusing to work diligently).Footnote 3 This complexity obscures the voter’s evaluation of the policy since he is unable to assign clear responsibility for poor service provision. For example, when a community experiences a decline in safety after police reform (e.g., a budget cut), it is challenging for residents to determine whether the drop in security is due to the reform itself or because police officers are resisting the changes. Even if the reform could potentially improve services, voters could still see a decline in quality due to bureaucratic pushback. We show that, in equilibrium, incumbents implement reform if they are sufficiently biased in its favor, bureaucrats resist if they are sufficiently anti-reform, and voters reelect their representative if government performance is sufficiently high.
The assumption of coproduction in our model primarily applies to street-level bureaucracies, where bureaucrats can directly influence service provision and, consequently, voters’ perceptions. For example, voters may reassess the merit of a budget cut or the restructuring of an agency based on their waiting times for emergency responders, delays in mail services, or their ability to obtain building permits in a timely fashion.
Our model produces several key insights. First, we demonstrate why and when bureaucrats undermine public service provision for political leverage. Since voters cannot perfectly identify who is responsible for poor service quality and can only probabilistically determine whether bureaucratic resistance has occurred, it becomes optimal for bureaucrats to engage in resistance despite voters’ awareness of this possibility (provided the costs of resistance are low enough). After incumbents introduce the reform, bureaucrats can exploit their intermediary role in government to affect voters’ inference about the reform and undermine the incumbent’s reelection chances in favor of the anti-reform opponent.
We also find that bureaucrats’ incentive to resist is non-monotonic with respect to the voter’s prior belief about the reform’s value. The incentive to resist depends on whether the voter is susceptible to information that bureaucrats mediate. When voters strongly favor the reform, bureaucrats have little incentive to resist, as they cannot significantly influence voter support for the reforming incumbent. Conversely, when voters are already pessimistic about the reform, bureaucrats have little incentive to resist, as the voter is already likely to perceive the reform as a failure. As a result, bureaucrats are most incentivized to resist when voters are torn between the reform and the status quo and, therefore, more receptive to interpreting poor service as a signal about the reform’s effectiveness.
The implications of bureaucratic resistance for voter learning and policymaking are not immediately clear. A naive conclusion could be that bureaucratic resistance makes incumbents more cautious about reform by directly jeopardizing service quality. Simultaneously, resistance should make a rational voter more forgiving of poor service provision, which would incentivize politicians to introduce reforms. We show that while these opposing mechanisms are at play, they are more complex than this simple logic suggests and depend on the reform’s true merit and popularity.
One reason for this complexity lies in how resistance shapes voter learning. By reducing the informational value of service provision, resistance forces Bayesian voters to rely more heavily on their prior beliefs about the reform’s value when making their election decision. This has asymmetric effects on voter behavior: it makes an initially lenient, pro-reform voter even more lenient, as it makes policy failure more excusable by introducing a plausible alternative explanation. Counterintuitively, however, resistance also makes an a priori strict, anti-reform voter more strict, even though bureaucratic resistance can only worsen service provision. This is because, by introducing noise, resistance prevents the voter from confidently attributing policy success to the reform itself.
Consequently, we find that the possibility of resistance can either incentivize or deter incumbents from implementing reform, depending on the voter’s prior beliefs. When reform is initially unpopular with the voter, resistance discourages reform efforts as incumbents fear the costs of resistance for service provision and voter backlash. Conversely, when reform is popular, bureaucratic resistance provides a convenient scapegoat for politicians and increases the incumbent’s electoral incentive to introduce reform. For intermediary levels of reform popularity, these tendencies lead to policy inefficiency, that is, the systematic deviation from socially optimal policy choices. Incumbents avoid implementing beneficial reforms (under-reform) and pursue ineffective ones too frequently (over-reform). Notably, the ability to resist can sometimes harm bureaucrats themselves. Particularly, in cases where bureaucrats are used as scapegoats for incumbents, bureaucrats would benefit from being able to commit to noninterference ex ante. However, once reform is implemented, resistance remains beneficial to undermine the reelection chances of reforming incumbents.
RELATED LITERATURE AND CONTRIBUTIONS
Our article contributes to existing scholarship on bureaucratic politics, interest group influence, and political economy.
Bureaucratic Politics and Interest Groups
First, our theory addresses a fundamental debate in bureaucratic politics between the public choice school of thought (Downs Reference Downs1967; Niskanen Reference Niskanen1971; Tullock Reference Tullock1965) and theories of bureaucratic control and delegation (Banks and Weingast Reference Banks and Weingast1992; Brehm and Gates Reference Brehm and Gates1997; McCubbins Reference McCubbins1985; McCubbins, Noll, and Weingast Reference McCubbins, Noll and Weingast1987; Miller and Moe Reference Miller and Moe1983). Niskanen positioned bureaucrats as primary strategic actors and famously argued that self-interested bureaucrats use their private information to extract rents by making take-it-or-leave-it offers to incumbents. In contrast, theorists of legislative control criticized Niskanen’s framework for ascribing outsized power to bureaucrats. They framed the politician–bureaucrat relationship as a top-down principal–agent model and focused on incumbents’ strategies to minimize agency loss and leverage bureaucratic expertise. We reconcile these two long-standing ideas on bureaucratic politics by synthesizing a principal–agent perspective on strategic politicians with the notion of politically powerful bureaucrats who can sway the incumbent’s policy decisions by leveraging their private information, exploiting the incumbent’s electoral vulnerability, and adjusting their work effort.
Additionally, we contribute to the literature on the “deep state” (i.e., the perceived capacity of unelected bureaucrats to resist, undermine, or influence elected officials’ policy decisions through their control over implementation and service delivery) in two key ways. First, we examine “deep state” dynamics among street-level bureaucrats at the subnational level and develop a novel micro-founded explanation for why bureaucratic agencies might systematically undermine the programs and services they provide. Focusing on the federal bureaucracy, some scholars argue that agencies undermine their own work because U.S. presidents pursue retrenchment and direct the administrative state to sabotage itself when they cannot secure new legislation from Congress (Noll Reference Noll2022). Others examine the expressive benefits of “guerrilla” governance (O’Leary Reference O’Leary2020) and find that bureaucratic resistance stems from civil servants who must navigate tensions between professional norms and personal policy convictions (Kucinskas and Zylan Reference Kucinskas and Zylan2023). Notably, voters are absent from these accounts. In contrast, we focus on how voters’ dependence on bureaucrats to learn about policy outcomes can result in bureaucratic resistance as a strategic choice. Second, while previous work focuses on how bureaucratic autonomy constrains politicians who benefit from direct control and appointments to secure reelection (Huber and Ting Reference Huber and Ting2021; Lewis Reference Lewis2008), we depart by examining when bureaucratic independence becomes electorally advantageous for politicians. Specifically, we show the conditions under which politicians strategically exploit “deep state” rhetoric when they attribute policy failures to bureaucratic resistance rather than their own choices.
Last, we contribute to the growing literature on bureaucrats as interest groups within government. We build on Moe’s (Reference Moe2006) argument that bureaucrats leverage politicians’ electoral vulnerability to influence who their principals are and what policies they choose in office. An extensive literature highlights bureaucrats’ various means of direct political influence through their public sector unions, including collective bargaining (Anzia and Moe Reference Anzia and Moe2015; Moe Reference Moe2009; Reference Moe2011; Paglayan Reference Paglayan2019; Zoorob Reference Zoorob2019), union endorsements (Hartney Reference Hartney2022; Hartney and Flavin Reference Hartney and Flavin2011; Moe Reference Moe2006), electoral mobilization of their members (Anzia Reference Anzia2014; Flavin and Hartney Reference Flavin and Hartney2015; Leighley and Nagler Reference Leighley and Nagler2007), political contributions (Carreri, Teso, and Yu Reference Carreri, Teso and YuForthcoming; DiSalvo Reference DiSalvo2015; Moe Reference Moe2011), or direct lobbying (Anzia Reference Anzia2022). In contrast, we focus on a more fundamental source of bureaucratic power and explain how and when bureaucrats can exert policy influence through their roles in government, that is, merely by virtue of being bureaucrats.
Formal Political Economy
Our model is closely connected to several strands of literature in formal political economy. First, it is related to the political accountability literature (Bils Reference Bils2023; Canes-Wrone, Herron, and Shotts Reference Canes-Wrone, Herron and Shotts2001; Fox Reference Fox2007; Gersen and Stephenson Reference Gersen and Stephenson2014), which explores how voters’ imperfect observations of policy outcomes create electoral incentives for incumbents to prioritize popular policies, irrespective of their intrinsic value. Joining Calvert (Reference Calvert1985), Patty (Reference Patty2009), Ashworth, Bueno de Mesquita, and Friedenberg (Reference Ashworth, de Mesquita and Friedenberg2017; Reference Ashworth, de Mesquita and Friedenberg2018), Gailmard and Patty (Reference Gailmard and Patty2019), Prato and Wolton (Reference Prato and Wolton2017), and Schnakenberg, Schumock, and Turner (Reference Schnakenberg, Schumock and Turner2024), we study how the information environment influences voter learning within the accountability framework. Our contribution lies in studying an accountability game where both policymaking and changes in the information environment are endogenously determined in equilibrium by strategic actors. Specifically, as in Prato and Wolton (Reference Prato and Wolton2017) and Gailmard and Patty (Reference Gailmard and Patty2019), the voter’s information environment endogenously arises from the incumbent’s strategic choice in our model. The key difference is that the voter’s information can be affected by private actions of other strategic players, namely, bureaucrats’ resistance.Footnote 4
This article also contributes to formal models of bureaucracy in which government outcomes are jointly produced by politicians and bureaucrats (see Gailmard and Patty Reference Gailmard and Patty2012a for a review of this large literature). Seminal work in this literature examines dyadic relationships in administrative policymaking between politicians and expert bureaucrats who actively shape policies within delegated discretion (e.g., Forand Reference Forand2025; Gailmard and Patty Reference Gailmard and Patty2007; Reference Gailmard and Patty2012b; Huber and Shipan Reference Huber and Shipan2002) or who provide policy advice to politicians (e.g., Denisenko, Hafer, and Landa Reference Denisenko, Hafer and Landa2024; Gailmard and Patty Reference Gailmard and Patty2012b). In contrast, we offer a theoretical framework that investigates the triadic strategic interaction among an incumbent politician, a representative voter, and a group of street-level bureaucrats who act collectively as service providers to execute policy chosen by the politician. The key difference is twofold: bureaucrats execute policies rather than make them, and they strategically anticipate the voter’s response when doing so.
Several existing models study how government coproduction makes it difficult for the voter to attribute responsibility between bureaucrats and politicians (Awad, Karekurve-Ramachandra, and Rothenberg Reference Awad, Karekurve-Ramachandra and Rothenberg2023; Foarta Reference Foarta2023; Forand and Ujhelyi Reference Forand and Ujhelyi2021; Fox and Jordan Reference Fox and Jordan2011; Li, Sasso, and Turner Reference Li, Sasso and Turner2024; Martin and Raffler Reference Martin and Raffler2021; Slough Reference Slough2024; Ujhelyi Reference Ujhelyi2014; Yazaki Reference Yazaki2018). Yet, most of these models do not provide an explanation for why and when bureaucrats are willing to engage in costly resistance.Footnote 5
One exception is Ujhelyi (Reference Ujhelyi2014), who also examines bureaucrats’ strategic resistance and its implications for policymaking. However, while Ujhelyi (Reference Ujhelyi2014) assumes that politicians’ and bureaucrats’ choices are substitutes in government production and focuses on the learning between politicians and bureaucrats about each other’s type, we assume complementarity in government coproduction and extensively discuss how resistance affects voters’ inference.Footnote 6
Last, this article is closely related to models of policy obstruction and sabotage (Fong and Krehbiel Reference Fong and Krehbiel2018; Gibbs Reference Gibbs2023; Gieczewski and Li Reference Gieczewski and Li2024; Hirsch and Kastellec Reference Hirsch and Kastellec2022; Patty Reference Patty2016). The key difference between our argument and existing work is the observability of sabotage (i.e., resistance in our model). Unlike sabotage by the political opposition, which is overt and observable by the voter, the bureaucrats in our model resist covertly. In turn, the voter in our model must guess whether the observed government outcome is or is not affected by bureaucratic resistance.
MODEL
Consider a two-period (
$ t=1,2 $
) electoral competition model with an incumbent (she), an opponent, a median voter (he), and the bureaucrats (they).Footnote 7 There is an election after
$ t=1 $
, where the voter chooses between the incumbent and the opponent as a new officeholder for
$ t=2 $
.
Policymaking
An incumbent facing reelection decides whether to introduce a reform policy in
$ t=1 $
. Its value to voter welfare
$ \omega \in \{0,1\} $
is unknown to the public. The common prior for the reform’s value is
$ \Pr [\omega =1]=1/2 $
. Alternatively, the incumbent can keep the status quo with known value
$ q\in (0,1) $
.Footnote 8
$ t=1 $
is the window for reform. That is, a reform policy rejected in
$ t=1 $
cannot be reintroduced in
$ t=2 $
after the election. In
$ t=2 $
, the reform can only be repealed or maintained.Footnote 9 For simplicity, we assume that players do not discount their payoffs.
At
$ t=1 $
, the incumbent privately observes
$ \omega $
and chooses whether to introduce reform (
$ {a}_1=1 $
) or not
$ ({a}_1=0) $
, that is,
$ {a}_1\in \{0,1\} $
.
Partisan Policy Preference
Politicians are both office- and policy-motivated. They obtain
$ 1 $
from winning the election and
$ 0 $
otherwise. In addition, they obtain intrinsic policy payoff by choosing the policy their party prefers while they are in office, independently drawn from a uniform distribution. Each politician knows this partisan policy payoff, but the voter only knows that each politician’s payoff is drawn from a uniform distribution.
The incumbent is in the pro-reform party and obtains
$ \rho \sim U[0,1] $
if the reform is in place in each period. In contrast, the opponent is commonly known to be anti-reform and prefers the status quo.Footnote 10
Bureaucratic Resistance
Bureaucrats intrinsically dislike the reform and obtain disutility of unknown value
$ -\kappa $
with common prior
$ \kappa \sim U[0,1] $
. After observing
$ {a}_1 $
and
$ \omega $
if
$ {a}_1=1,\hskip0.20em $
the bureaucrats privately choose whether to undermine the policy,
$ {b}_1\in \{0,1\} $
, where
$ {b}_1=1 $
is to undermine the policy and
$ {b}_1=0 $
not to undermine. Such resistance to policy can comprise a variety of measures, including slowing the delivery of services, overlooking service infractions, misusing their authority, or mismanaging funds. Resistance is costly for bureaucrats (i.e., they incur a known cost of
$ c\in [0,1] $
if they resist). c captures material/reputational punishments for noncompliance (Ujhelyi Reference Ujhelyi2014), bureaucrats’ public service motivations and utility from high-quality service provision (Forand, Ujhelyi, and Ting Reference Forand, Ujhelyi and Ting2022; Yazaki Reference Yazaki2018), or coordination efforts of bureaucrats necessary to engage in resistance.
Government Outcome
The government outcome
$ {g}_1\in \mathbb{R} $
is produced by
$$ \begin{array}{rll}{g}_1=\left\{\begin{array}{l}(1-{b}_1)\omega +{\eta}_1\hskip0.3em \mathrm{if}\hskip0.3em {a}_1=1\hskip1em \qquad \\ {}\hskip1em \qquad \\ {}(1-{b}_1)q+{\eta}_1\hskip0.3em \mathrm{if}\hskip0.3em {a}_1=0,\hskip1em \qquad \end{array}\right.& & \end{array} $$
where
$ {\eta}_t $
is an i.i.d. shock drawn from a log-concave density
$ h(\cdot ) $
that has full support on
$ \mathbb{R} $
and is symmetric around
$ 0 $
. Let
$ H(\cdot ) $
denote the associated CDF of
$ h(\cdot ) $
.
The density of
$ {g}_t $
is
$ h({g}_t-1) $
if the reform works (
$ \omega =1 $
) and bureaucrats do not resist (
$ {b}_t=0 $
), and
$ h({g}_t) $
if the reform does not work (
$ \omega =0\hskip-0.15em $
) or bureaucrats resist (
$ {b}_t=1 $
).
Election
After observing the chosen policy
$ {a}_1 $
and the realized government outcome
$ {g}_1 $
, the voter chooses between the incumbent and the opponent. If the voter is indifferent between the two candidates, he flips a fair coin and reelects the incumbent with probability
$ 1/2 $
.
Second Period
The incumbent’s policy decision in
$ t=1 $
affects the set of policies from which the election winner can choose in
$ t=2 $
.
$ {a}_2=0 $
indicates the election winner’s choice of the status quo and
$ {a}_2=1 $
her choice of the reform. If the incumbent chooses the status quo in period one, the second-period policy is fixed as the status quo:
$ {a}_1=0\Rightarrow {a}_2=0 $
. If the incumbent chooses the reform, the election winner can choose between maintaining or repealing it:
$ {a}_1=1\Rightarrow {a}_2\in \{0,1\} $
.
The government outcome in
$ t=2 $
,
$ {g}_2 $
, is given by
$$ \begin{array}{rll}{g}_2=\left\{\begin{array}{l}(1-{b}_2)\omega +{\eta}_2\hskip0.3em \mathrm{if}\hskip0.3em {a}_1={a}_2=1\hskip1em \qquad \\ {}\hskip1em \qquad \\ {}(1-{b}_2)q+{\eta}_2\hskip0.3em \mathrm{if}\ \mathrm{otherwise},\hskip1em \qquad \end{array}\right.& & \end{array} $$
where
$ {b}_2\in \{0,1\} $
is the bureaucrats’ decision to undermine the policy, and
$ {\eta}_2 $
is a shock drawn from
$ h(\cdot ) $
.
Payoffs
The voter obtains the government outcome in each period:
$$ \begin{array}{rll}{g}_1+{g}_2.& & \end{array} $$
The incumbent obtains policy payoff
$ \rho $
in each period if she chooses the reform. Also, she obtains
$ 1 $
if she wins the election:
$$ \begin{array}{rll}{a}_1\rho +1\{\mathrm{reelection}\}(1+{a}_1{a}_2\rho ).& & \end{array} $$
The opponent obtains
$ 1 $
if she chooses the status quo while she is in office:
$$ \begin{array}{rll}1\{\mathrm{election}\}(1-{a}_2).& & \end{array} $$
The bureaucrats obtain
$ -\kappa $
in each period if the reform is in place. Also, they obtain
$ -c $
if they engage in resistance:
$$ \begin{array}{rll}-{\displaystyle \underset{\mathrm{disutility}\ \mathrm{from}\ \mathrm{the}\ \mathrm{reform}}{\underbrace{a_1(\kappa +{a}_2\kappa )}}}-{\displaystyle \underset{\mathrm{cost}\ \mathrm{of}\ \mathrm{resistance}}{\underbrace{c({b}_1+{b}_2)}}}.& & \end{array} $$
Timing
To recap,
-
0. Nature draws the reform’s value
$ \omega $
, partisan policy payoff for
$ \rho $
, the bureaucrats’ disutility from the reform,
$ \kappa $
. -
1. The incumbent privately observes the personal value
$ \rho $
and
$ \omega $
, and publicly chooses whether to introduce the reform (
$ {a}_1=1 $
) or not (
$ {a}_1=0 $
). -
2. The bureaucrats privately observe the reform’s value
$ \omega $
and their disutility from the reform
$ \kappa $
. Then they privately choose whether to undermine the chosen policy (
$ {b}_1=1 $
) or not (
$ {b}_1=0 $
). -
3. The government outcome
$ {g}_1 $
is produced, and the voter observes it. -
4. The voter chooses between the incumbent and the opponent as the new officeholder in the election.
-
5. The election winner chooses the policy
$ {a}_2 $
, and the bureaucrats chose
$ {b}_2 $
. -
6. Payoffs are realized, and the game ends.
The solution concept is weak perfect Bayesian equilibrium with pure strategies (henceforth, equilibrium). Every player plays a pure-strategy best response given their beliefs about other players, and beliefs regarding every on-path event are formed following Bayes’ Rule. Specifically, we focus on the equilibrium where the voter conditions his reelection decision on the expected value of the policy to be enacted in the next period, inferred from the information available to him before the election.
Modeling Choices
Before solving the model, we discuss some crucial modeling choices and their relevance to our results.
Window for Reform and Policy Continuity
$ t=1 $
in our model is a critical “watershed” point, where the reform is either implemented or abandoned (Keeler Reference Keeler1993). We assume that if the incumbent decides not to introduce the reform at
$ t=1 $
, she is committed not to revisit it at
$ t=2\hskip-0.15em $
. As we show in detail in Appendix B of the Supplementary Material, this assumption reflects that incumbents benefit from commitment in a two-period game and policy continuity also arises in equilibrium in a dynamic game extended beyond two periods, where the voter conditions reelection decisions on the reform’s expected value in every period.
To summarize the logic, suppose that the game has more than two periods. As our analysis shows, the pro-reform incumbent has more incentives to implement the reform when it works (
$ \omega =1 $
) than when it does not (
$ \omega =0 $
) as long as the voter cares about the reform’s expected value, and
$ {g}_t $
is an informative signal in every stage game. Consequently, the pro-reform incumbent’s decision not to implement the reform in t will always negatively affect the voter’s posterior belief about the reform’s value. She cannot delay the reform without hurting its reputation:
$ \Pr [{a}_t=1|\omega =1]\ge \Pr [{a}_t=1|\omega =0]\hskip2.77695pt \iff \hskip2.77695pt \unicode{x1D53C}[\omega |{a}_t=0]\le \unicode{x1D53C}[\omega ]. $
That is, the pro-reform incumbent who did not adopt the reform but nonetheless won reelection in t faces an even less favorable environment for the reform going forward.Footnote 11
Politicians’ Policy Preferences
We assume that the incumbent is strictly pro-reform, and the opponent is strictly anti-reform. This assumption can be relaxed. The key is that the incumbent must be more pro-reform than the opponent at the point when the reform is deployed.Footnote 12
We also assume that the incumbent’s policy payoff remains independent of bureaucratic resistance. Instead, resistance imposes costs on the incumbent solely through reduced reelection prospects when she introduces reform. Specifically, we model the incumbent’s policy payoff as a “partisan” or “ideological” payoff
$ \rho $
, which does not depend on the reform’s implications for voter welfare,
$ {g}_t\hskip-0.15em $
. Nevertheless, because the reelection probability strictly increases with
$ {g}_t $
and resistance can only harm
$ {g}_t $
, adding a term for the incumbent’s intrinsic valuation of policy success as an increasing function of
$ {g}_t $
does not qualitatively alter our results.
Politicians’ Ability to Reform and Collective Bargaining
One concern may be that, due to successful bargaining of public sector unions in many U.S. jurisdictions, elected officials have effectively lost their ability to control and reform the bureaucracy. While collective bargaining does constrain the scope of potential reforms, particularly around work rules and employment terms, significant categories of reform remain within elected officials’ authority. Examples range from cuts in agency funding to civilian complaint review boards, ending qualified immunity, body camera mandates, or book bans and legislative gag orders in schools. Moreover, where collective bargaining provides bureaucrats with stronger employment protections and discretion (Anzia Reference Anzia2022; DiSalvo Reference DiSalvo2015; Hartney Reference Hartney2022; Moe Reference Moe2006), it may actually enhance their capacity for the subtle service manipulation, we model by reducing the likelihood of retaliation (i.e., decreasing the costs of shirking c).Footnote 13 Hence, collective bargaining operates as both a constraint on and an enabler of the strategic dynamics we examine. It limits which policies can be proposed while strengthening bureaucrats’ ability to resist those policies that do advance through the political process.
Nature of Bureaucratic Resistance
Our concept of bureaucratic resistance differs from canonical shirking in principal–agent models. In seminal dyadic models of administrative policymaking, delegation creates opportunities for administrative drift driven by policy disagreement or effort minimization (Epstein and O’Halloran Reference Epstein and O’Halloran1999; Fox and Jordan Reference Fox and Jordan2011; Gailmard and Patty Reference Gailmard and Patty2007). Notably, these frameworks treat voters as absent or only implicitly present. Our framework departs by conceptualizing bureaucrats as public servants who maintain direct relationships with citizens and can shape public opinion. In this triadic relationship, bureaucratic resistance becomes a strategic tool for political influence rather than merely a byproduct of preference divergence or effort avoidance. It is worth noting that our model allows for a baseline level of resistance due to standard delegation and moral hazard problems. Our theoretical contribution focuses on the additional layer of strategic resistance that arises when uncertain voters are incorporated into the game.
Readers familiar with canonical principal–agent models may also question the idea that shirking (rather than working) is costly for bureaucrats. However, we are not the first to assume this mirror image in which policy-motivated bureaucrats face a trade-off between the benefits of sabotaging an unwanted policy and the material, reputational, or psychological costs of doing so (Brehm and Gates Reference Brehm and Gates1997; Ujhelyi Reference Ujhelyi2014; Yazaki Reference Yazaki2018). Instead of minimizing the costs of positive effort for government output while accounting for its benefits (e.g., higher wages and avoiding political oversight), bureaucrats in this setting maximize the benefit from negative government output while taking the costs into account.
Another important aspect of bureaucratic resistance here is that it can only damage the quality of government services when the reform is effective. This assumption can be relaxed (see Appendix C of the Supplementary Material). The qualitative results still hold if bureaucrats can resist the government service under an ineffective reform, but the damage to an ineffective reform is smaller than the damage to an effective reform.Footnote 14
Finally, we assume that bureaucrats know the reform’s value when they choose to resist, which reflects the idea that bureaucrats tend to be better informed about policies that affect their operations (Lipsky Reference Lipsky1980). Again, relaxing this assumption does not change our results qualitatively (see Footnote Footnote 18).
(Dis-)Continuity of Bureaucratic Resistance
Another important feature of bureaucratic resistance in our game is that its effect travels through voter inference. Bureaucrats do not resist in
$ t=2 $
because the game ends without any further elections, eliminating their incentive to resist. Consequently, when evaluating the reform, the voter disregards resistance implications, knowing it will cease after the election. We therefore assume that once the voter supports the pro-reform incumbent, she both continues the reform and can prevent future bureaucratic resistance. This often reflects reality: reform packages typically constrain bureaucratic behavior or reelected incumbents consolidate reforms through policy expansion. From this perspective,
$ t=1 $
is truly a watershed moment: if the pro-reform incumbent wins with the reform, it not only continues, but bureaucrats’ incentives and abilities to resist are also curtailed.
Yet, we relax this assumption in the “Dynamics: Bureaucrats’ Continued Resistance” section to study cases where voters might expect the reform to face continued bureaucratic resistance after the election, at least in the short run. We investigate this scenario by examining a steady state in the dynamic game, where reelection decisions depend on both the reform’s value and the implications of continued resistance.
Voter’s Uncertainty about Resistance
In our model, voters cannot observe whether bureaucrats resisted the policy. If resistance were observable, voters would rationally adjust their interpretation of
$ {g}_t $
, thus eliminating bureaucrats’ incentives to resist.
One could argue that bureaucratic resistance is sometimes well-documented and often covered by local media outlets. However, while voters (as well as journalists and scholars) can form a rational conjecture about whether resistance occurred for a specific incident or its frequency, observing resistance would imply full awareness of bureaucrats’ intentions. For example, increasing police response times (low
$ {g}_t $
) can indicate a work slowdown, as in the case of police response to Chesa Boudin’s policies. However, conclusively attributing these delays to intentional resistance requires evidence that delays are indeed the result of a slowdown (
$ {b}_t=1 $
) rather than ineffective policy (
$ \omega =0 $
).
One may also be concerned that incumbents and bureaucrats aim to persuade the voter about the reform’s value in a preceding communication game. However, neither party can provide credible information about the reform’s effectiveness. As our analysis demonstrates, once the reform is introduced, the incumbent has an incentive to claim the reform is effective regardless of its actual value, while the bureaucrats are always incentivized to argue that it is ineffective. This yields a babbling equilibrium, where messages from both parties are uninformative.
Finally, our assumption that resistance is unobserved, and thus affects the information voters receive about government services, distinguishes our argument from an alternative noninformation story of resistance. In the latter case, resistance would be a visible tool for bureaucrats to pressure politicians and demonstrate their political strength and indispensability. In this case, voters would punish politicians for provoking resistance and poor services. However, since voters condition their election on bureaucrats’ behavior, bureaucrats’ dominant strategy would be to fully reveal their actions and claim responsibility for resistance to ensure that voters punish the incumbent. This is inconsistent with multiple examples, such as the Chesa Boudin case, where bureaucrats refuse to take responsibility for poor services rather than claim it.
EQUILIBRIUM ANALYSIS
All proofs are relegated to Appendix A of the Supplementary Material.
Second-Period Behavior
Regardless of what happened in the first period, the incumbent who wins the election with the reform keeps it,
$ 1\{{a}_1=1\}\times 1\{\mathrm{reelection}\}\Rightarrow {a}_2=1 $
, since she gets policy payoff
$ \rho \ge 0 $
by doing so and
$ 0 $
otherwise. If she does not introduce the reform or the opponent wins the election, the status quo is chosen,
$ {a}_2=0 $
. Regardless of the election winner or the policy she chooses, bureaucrats have no incentive to resist with cost
$ -c<0 $
, that is,
$ {b}_2=0 $
.
First-Period Strategies
From here forward, we focus on the first period and drop the subscript for
$ t=1 $
as discussion focuses on the first period.
If the incumbent introduces the reform, the voter’s strategy is a function of his Bayesian conditional expectation of the reform’s value given g,
$ \unicode{x1D53C}[\omega |g] $
,
$ 1\{\mathrm{election}\}(g):\mathbb{R}\to \{0,1\} $
. He reelects the reforming incumbent if and only if the expected value of the reform given g is larger than that of the status quo:
$$ \begin{array}{rl}\unicode{x1D53C}[\omega |g]\ge q.& \end{array} $$
If the incumbent does not introduce the reform, he flips a fair coin and reelects the incumbent with probability
$ 1/2 $
.
The incumbent’s strategy in the first period is a function of the reform’s merit and partisan payoffs from it,
$ (\omega, \rho ) $
,
$ a(\omega, \rho ):\{0,1\}\times [0,1]\to \{0,1\}. $
She introduces the reform if and only if the combination of the reelection gain and partisan benefits from the reform is larger than the reelection gain from the status quo:
$$ \begin{array}{rll}\rho +(1+\rho )\Pr [reelection|a(\omega )=1]\ge & & \\ {}\Pr [reelection|status\hskip0.3em quo]=1/2.& \end{array} $$
If the incumbent introduces the reform, the bureaucrats’ strategy is a function of their disutility from reform and the value of the reform
$ (\kappa, \omega ), $
$ b(\kappa, \omega ):[0,1]\times \{0,1\}\to \{0,1\}. $
Bureaucrats engage in resistance only if the product of their gain from reducing the reforming incumbent’s reelection probability and their disutility from the reform outweighs the cost of resistance:
$$ \begin{array}{rl}\begin{array}{rll}& -\kappa \Pr [reelection|a=1,\omega, b=0]\le & \\ {}& -\kappa \Pr [reelection|a=1,\omega, b=1]-c\hskip2.77695pt \iff \hskip2.77695pt \\ {}& \kappa \cdot \left(\right.\Pr [reelection|a=1,\omega, b=1]-\\ {}& \Pr [reelection|a=1,\omega, b=0]\left)\right.\ge c.\end{array}& \end{array} $$
If the incumbent does not introduce the reform, the bureaucrats do not resist because the reform will not be in place in the second period with or without their resistance.
Equilibrium without Bureaucratic Resistance
As a benchmark, we solve the game without any resistance. One can think of this equilibrium as if the costs of resistance are very high.
Proposition 1. In the game without bureaucratic resistance, there exists a unique pure strategy equilibrium with a set of threshold values,
$ \{{g}_B^{*},{\rho}_{B0}^{*},{\rho}_{B1}^{*}\}\in \mathbb{R}\times {[0,1]}^2 $
such that:
-
• the voter reelects the reforming incumbent if and only if the observed government service quality is above the threshold,
$ g\ge {g}_B^{*} $
; -
• the incumbent introduces the reform with value
$ \omega $
if and only if the partisan payoff is above the threshold given
$ \omega $
,
$ \rho \ge {\rho}_{B\omega}^{*} $
such that
$ \frac{1}{2}>{\rho}_{B0}^{*}\ge {\rho}_{B1}^{*} $
; -
• the voter and the incumbent apply stringent thresholds when the reform is ex-ante unpopular;
$ {g}_B^{*}(q) $
is increasing in q and
$ {\rho}_{B\omega}^{*}(q) $
is weakly increasing in q.
Since g is an informative signal of the reform’s merit, a voter who observes higher service quality is more likely to believe that the reform is effective. Consequently, the conditional expectation of the reform’s value exceeds the value of the status quo if and only if the observed service quality exceeds a certain threshold,
$ {g}_B^{*} $
. Naturally, this threshold increases as the value of the status quo increases.
The incumbent introduces the reform only when the partisan payoff for the reform is large enough. This follows directly from Equation 2: as
$ \rho $
increases, the incumbent’s expected payoff from introducing the reform increases. Additionally, the incumbent is more likely to introduce an effective reform than an ineffective one because voters are more likely to observe a higher g when the reform is effective. Figure 1 illustrates this: the density of g when the reform is effective (
$ h(g|\omega =1)=h(g-1) $
) is shifted to the right relative to the density when the reform is ineffective (
$ h(g|\omega =0)=h(g) $
). Consequently, higher values of g are more likely to be drawn from the former distribution than from the latter.
Comparison of Conditional Densities of g
Note: The figure shows the densities for g, conditional on
$ \omega $
.
Apart from g, the voter also gains additional information by observing the incumbent’s action a. In equilibrium, the incumbent is more likely to introduce the reform when it is effective than when it is not, so the very act of introducing the reform serves as a positive signal of its value.
Last, the incumbent is less inclined to introduce the reform when the voter values the status quo more regardless of whether the reform is effective or not. This is because she is less likely to be reelected after introducing the reform when the voter highly values the status quo q.
Popularity of the Reform and Policy Distortion
Ideally for voter welfare, the incumbent should choose the reform only if it is effective (
$ {\rho}_{B0}^{*}=1 $
and
$ {\rho}_{B1}^{*}=0\hskip-0.25em $
). However, in reality, the incumbent can distort her choice, introducing the reform when it is not effective (
$ {\rho}_{B0}^{*}<1 $
, Over-Reform) or not introducing an effective reform (
$ {\rho}_{B1}^{*}>0 $
, Under-Reform) in equilibrium.
As in the canonical electoral accountability literature (Bils Reference Bils2023; Canes-Wrone, Herron, and Shotts Reference Canes-Wrone, Herron and Shotts2001; Fox Reference Fox2007; Gersen and Stephenson Reference Gersen and Stephenson2014), the voter can observe the incumbent’s policy choice but not fully observe its implication. In this situation, the incumbent’s reelection incentives biases the incumbent policy decision toward a more popular policy (i.e., pandering or posturing). For instance, it may encourage the incumbent to adopt the reform that does not work (
$ \omega =0 $
) only because it is popular or discourage her from adopting the reform that works (
$ \omega =1\hskip-0.25em $
) only because it is unpopular.
Remark 1. In the absence of bureaucratic resistance, higher popularity of the reform exacerbates over-reform while mitigating under-reform. Conversely, lower popularity of the reform reduces over-reform but worsens under-reform.
The incumbent’s incentive to over-reform arises from two components. The first is the partisan policy payoff
$ \rho $
. Even if the incumbent could never secure reelection with the reform, she would introduce it if
$ \rho \ge \Pr [\mathrm{reelection}|\mathrm{status}\ \mathrm{quo}]=1/2\hskip-0.20em $
. The second component arises from electoral incentives due to the voter’s inability to directly observe
$ \omega $
. Because g is a noisy signal, the voter can still observe g high enough to secure the incumbent’s reelection after introducing an ineffective reform. This incentivizes the incumbent to introduce an ineffective reform even when
$ \rho <1/2\hskip-0.25em $
. Specifically, for
$ \rho \in \left[{\rho}_{B0}^{*},\frac{1}{2}\right]\hskip-0.25em $
, policy distortion occurs only because of electoral incentives.
In contrast, because policy payoff
$ \rho $
always pushes the incumbent toward introducing the reform, under-reform is driven purely by electoral concerns. The incumbent avoids introducing an effective reform only when the probability of winning reelection is higher with the status quo than with the reform.Footnote 15
Equilibrium with Bureaucratic Resistance
We now present our main findings. We characterize the equilibrium when bureaucrats are able to resist, explore their incentives to do so, and compare how resistance affects the voter’s and the incumbent’s strategies compared to the equilibrium without resistance.
Proposition 2. In the game with bureaucratic resistance, there exists a unique pure strategy equilibrium with a set of threshold values,
$ \{{g}^{*},{\rho}_0^{*},{\rho}_1^{*},{\kappa}^{*}\}\in \mathbb{R}\times {[0,1]}^3 $
such that:
-
• the voter reelects the reforming incumbent if and only if
$ g\ge {g}^{*}(q,c) $
, where
$ {g}^{*}(q,c) $
is monotonically increasing in q; -
• the incumbent introduces the reform with value
$ \omega $
if and only if
$ \rho \ge {\rho}_{\omega}^{*} $
such that
$ \frac{1}{2}>{\rho}_0^{*}\ge {\rho}_1^{*} $
.
$ {\rho}_{\omega}^{*}(q) $
is weakly increasing in q; -
• bureaucrats resist if and only if the incumbent introduces an effective reform and disutility from the reform is above the threshold,
$ \kappa \ge {\kappa}^{*} $
. Bureaucrats are more likely to resist for an intermediately popular reform than extremely popular or unpopular reforms;
$ 1-{\kappa}^{*}(q,c) $
is weakly single-peaked in q.
Footnote 16
Bureaucrats’ Incentives to Resist
The bureaucrats’ incentives to resist in
$ t=1 $
stem from two sources: (i) their desire to prevent reform in
$ t=2 $
and (ii) their ability to influence the election by affecting the government outcome, g. Consequently, they do not resist if the status quo is maintained, as reform is already obviated. Similarly, they do not resist if the reform is adopted but ineffective (
$ \omega =0 $
), because in this scenario their actions have no impact on the distribution of g.
Footnote 17
If the reform works (
$ \omega =1\hskip-0.25em $
) and the incumbent implements it (
$ a=1 $
), g is drawn from
$ h(g-1) $
if bureaucrats do not resist (
$ b=0 $
) and from
$ h(g) $
if they do (
$ b=1 $
). If the voter reelects the reforming incumbent if and only if he observes
$ g\ge {g}^{\prime } $
, bureaucrats can lower the reelection probability of the reforming incumbent from
$ 1-H({g}^{\prime }-1) $
to
$ 1-H({g}^{\prime }) $
, and from Equation 3, resistance is incentive compatible for them if and only if
$$ \begin{array}{rl}\begin{array}{rll}& -\kappa \left(\right.1-H({g}^{\prime }-1)\left)\right.\le -\kappa \left(\right.1-H({g}^{\prime })\left)\right.-c& \\ {}& \hskip2.77695pt \iff \hskip2.77695pt \kappa \left(\right.1-H({g}^{\prime }-1)-1+H({g}^{\prime })\left)\right.\ge c\\ {}& =\kappa \left(\right.H({g}^{\prime })-H({g}^{\prime }-1)\left)\right.\ge c.\end{array}& \end{array} $$
Notice that the gain from resistance is nonnegative since
$ \kappa \ge 0 $
and
$ H({g}^{\prime })-H({g}^{\prime }-1)>0 $
.Footnote 18
Therefore, the bureaucrats undermine the reform if and only if it is effective (
$ \omega =1 $
) and, for
$$ \begin{array}{rl}\kappa \ge \widehat{\kappa}({g}^{\prime },c):=\frac{c}{H({g}^{\prime })-H({g}^{\prime }-1)}& \end{array} $$
if
$ \widehat{\kappa}({g}^{\prime },c)<1 $
. If
$ \widehat{\kappa}({g}^{\prime },c)\ge 1 $
, there is no
$ \kappa <1 $
such that resistance is profitable for bureaucrats.
As Figure 2 shows, the marginal negative impact on the reelection probability (i.e., the shift from the red to the blue line) is greatest when the voter’s prior preference for the reform (
$ 1-q $
) is in the intermediate range, and diminishes when prior preferences are at the extremes. Intuitively, if the voter strongly favors the reform a priori (a very lenient threshold
$ {g}^{\prime } $
), the probability of observing a sufficiently low g to overturn that preference is small even with resistance. Conversely, if the voter strongly favors the status quo (a very strict threshold), the voter is unlikely to be satisfied with a given level of service quality and unlikely to reelect a reforming incumbent, even without bureaucrats’ interference. The necessity for bureaucrats to resist and oust a reformer is therefore lower. In contrast, when the voter is ex-ante torn between the reform and the status quo, a small change in the distribution of g can meaningfully sway the voter’s decision, making resistance more rewarding for bureaucrats.
Resistance’s Marginal Effect on Reelection
Note: In panel a, the X-axis is the voter’s cutoff
$ {g}^{\prime } $
, and the Y-axis is the reelection probability. The gray area between the two lines captures the marginal effect of resistance as a function of the voter’s cutoff
$ {g}^{\prime } $
. In panel b, the X-axis is the voter’s cutoff
$ {g}^{\prime } $
, and the Y-axis is resistance’s marginal effect on reelection probability. The line
$ H({g}^{\prime })-H({g}^{\prime }-1) $
is the resistance’s marginal effect as a function of the voter’s cutoff
$ {g}^{\prime } $
. (The size of the gray area in panel a.) The shaded area indicates the range of
$ {g}^{\prime } $
where resistance is incentive compatible.
Resistance Effects on Voter Learning
Proposition 3. Bureaucratic resistance makes the voter more lenient for a popular reform and more stringent for an unpopular reform; there exists
$ {q}^{\dagger } $
such that
$ {g}^{*}(q,c)\ge {g}_B^{*}(q) $
if and only if
$ q\ge {q}^{\dagger } $
.Footnote 19
For ex-ante popular reforms (low q), bureaucratic resistance makes a voter who already favors the policy even more forgiving toward the incumbent. Conversely, for initially unpopular reforms (high q), bureaucratic resistance has the opposite effect: it discourages the voter from reelecting the reforming incumbent, even when service quality remains relatively high. A voter who initially opposes the policy becomes even less forgiving toward the incumbent. Since resistance only reduces g in expectation, this latter result may seem puzzling.
The explanation lies in that bureaucratic interference introduces noise and diminishes the informational value of government performance as a signal of the policy’s quality. This leads a Bayesian voter to rely more heavily on prior beliefs when evaluating the incumbent’s policy choice. For example, when confronted with poor service provision and a high likelihood of police resistance, an already pro-reform voter might reason: “Crime rates haven’t dropped as quickly as I hoped, but I know the police are not fully committed to the policy change. That’s probably why it didn’t work as well. It’s unfair to blame the policy when it hasn’t been properly implemented.” In contrast, after observing good service quality despite the high possibility of police resistance, an already skeptical voter is concerned that this high g could result from noise rather than a genuinely effective reform. This voter might say: “Crime rates improved, but that’s likely because of other factors, like broader trends in policing or community initiatives, not this policy. The police would have sabotaged an otherwise successful reform, so the apparent success is likely to be a fluke.”Footnote 20
This mechanism offers a rational basis for what might appear as “motivated reasoning” (Ditto and Lopez Reference Ditto and Lopez1992; Little Reference Little2019; Little, Schnakenberg, and Turner Reference Little, Schnakenberg and Turner2019): People do not simply dismiss signals that contradict their priors out of bias; rather, they do so rationally, because they recognize that interference from a third actor undermines the reliability of those signals.
For a formal intuition, consider the voter’s posterior beliefs
$ \mathrm{E}[\omega |g] $
.Footnote 21 Without resistance, the posterior belief is
$$ \begin{array}{ccc}\mathrm{E}[\omega |g]=\frac{\Pr [g|\omega =1]}{\Pr [g|\omega =1]+\Pr [g|\omega =0]}& & \\ {}=\frac{h(g-1)}{h(g-1)+h(g)}.& & \end{array} $$
Now suppose that bureaucrats resist with probability
$ 1-{\kappa}^{\prime}\in [0,1] $
. Then,
$$ \begin{array}{rll}\mathrm{E}[\omega |g,{\kappa}^{\prime }]=\frac{\kappa^{\prime }h(g-1)+(1-{\kappa}^{\prime })h(g)}{\kappa^{\prime }h(g-1)+(1-{\kappa}^{\prime })h(g)+h(g)}.& & \end{array} $$
It becomes clear that
$ \mathrm{E}[\omega |g,{\kappa}^{\prime }] $
is a perturbation of
$ \mathrm{E}[\omega |g] $
with extra noise from resistance, so the value of g decreases in general, inducing the voter to rely more on his prior. Formally,
$$ \begin{array}{ccc}\mathrm{E}[\omega |g]\ge \mathrm{E}[\omega |g,{\kappa}^{\prime }]& & \\ {}\hskip2.77695pt \iff \hskip2.77695pt h(g-1)\ge {\kappa}^{\prime }h(g-1)+(1-{\kappa}^{\prime })h(g)& & \\ {}\hskip2.77695pt \iff \hskip2.77695pt h(g-1)\ge h(g)& & \\ {}\hskip2.77695pt \iff \hskip2.77695pt g\ge 1/2,& & \end{array} $$
so
$ \mathrm{E}[\omega |g]\ge \mathrm{E}[\omega |g,{\kappa}^{\prime }] $
if and only if
$ g\ge 1/2 $
. As Figure 3 illustrates, this implies that the voter’s posterior belief with bureaucratic resistance has a more dispersed distribution and, in turn, is less sensitive to g than his posterior without resistance.Footnote 22
Resistance’s Countervailing Effects on Voter Learning
Note: The figure shows the voter’s conditional expectation of the reform’s value, with and without resistance.
Another way in which bureaucratic resistance influences the voter’s inference is through its effect on the incumbent’s decision. Because the incumbent knows the reform’s true value, her choice conveys information to the voter. By altering how the incumbent decides, resistance indirectly affects the voter’s inference. We elaborate on the incumbents’ reaction to resistance next.
Resistance Effects on Policymaking
Interestingly, bureaucratic resistance can either increase or decrease the incumbent’s willingness to reform, depending on the reform’s ex-ante popularity and its inherent value. The logic is that incumbents consider not only the direct threat of resistance to service quality but also its implications for how voters interpret the reform’s merit.
Proposition 4.
-
1. Bureaucratic resistance induces over-reform if the reform is popular: for
$ q<{q}^{\dagger } $
,
$ {\rho}_0^{*}(q,c)<{\rho}_{B0}^{*}(q) $
. -
2. Bureaucratic resistance causes under-reform, but alleviates over-reform when the reform is unpopular enough: there exists
$ {q}^{\dagger \dagger } $
such that
$ {\rho}_1^{*}(q,c)>{\rho}_{B1}^{*}(q) $
and
$ {\rho}_0^{*}(q,c)>{\rho}_{B0}^{*}(q) $
if
$ q>{q}^{\dagger \dagger } $
. -
3. If the cost of resistance is low enough, bureaucratic resistance causes both under-reform and over-reform for an intermediately popular reform: There exists
$ {c}^{\dagger } $
such that
$ {q}^{\dagger \dagger }<{q}^{\dagger } $
,
$ {\rho}_1^{*}(q,c)>{\rho}_{B1}^{*}(q) $
and
$ {\rho}_0^{*}(q,c)<{\rho}_{B0}^{*}(q) $
if
$ q\in ({q}^{\dagger \dagger },{q}^{\dagger }) $
when
$ c<{c}^{\dagger } $
.
If the reform is initially popular (
$ q<\min \{{q}^{\dagger },{q}^{\dagger \dagger}\} $
), the possibility of resistance can incentivize the incumbent to introduce it, resulting in more over-reform. Two factors explain this result. First, when the reform is popular, resistance poses little threat—or none at all when
$ \omega =0 $
—to the incumbent’s reelection chances, since even poor service provision can be sufficient to secure voter support.Footnote 23 Second, as per Proposition 3, the incumbent even benefits from resistance by making the voter more lenient toward her. Consequently, if a reform is popular yet ineffective, the incumbent can essentially use bureaucrats as scapegoats for any shortfall in service quality—a likely outcome given the reform’s ineffectiveness.Footnote 24 Hence, resistance creates policy distortions by increasing the number of reforms that are doomed to fail (see panel a in Figure 4).
Effect of Bureaucratic Resistance on Policy Distortion under Different q
Note: The figure illustrates how bureaucratic resistance affects policy distortions under three scenarios. Panel a: when
$ q<\min \{{q}^{\dagger },{q}^{\dagger \dagger}\} $
, bureaucratic resistance increases over-reform. Panel b: when
$ c<{c}^{\dagger } $
and
$ q\in [{q}^{\dagger \dagger },{q}^{\dagger }] $
, bureaucratic resistance increases both over- and under-reform. Panel c: when
$ q>\max \{{q}^{\dagger },{q}^{\dagger \dagger}\} $
, bureaucratic resistance reduces over-reform but increases under-reform. Hatched boxes indicate increases in policy distortions with resistance; solid boxes indicate reductions in policy distortions with resistance.
Conversely, when the reform is unpopular (
$ q>\max \{{q}^{\dagger },{q}^{\dagger \dagger}\} $
), bureaucratic resistance leads the politician to be cautious with reforms, because both the direct and inference effects work to the incumbent’s disadvantage. For ineffective and unpopular reforms, bureaucratic resistance reinforces the skepticism of an already doubtful voter by lowering the informational value of g. When the reform is effective, the incumbent must further account for the direct effect of resistance, which lowers the distribution of service quality. By dampening enthusiasm for reform, resistance amplifies under-reform but can help curb over-reform in this case (see panel c in Figure 4).
What happens when the reform’s popularity falls within an intermediate range? If the cost of resistance is low and thus, resistance is likely, it prompts a reform when it is ineffective but deters it when it is effective, leading to both more over-reform and under-reform. Specifically, cheap and frequent resistance makes the voter sufficiently lenient to accept an ineffective reform while simultaneously providing a substantial direct threat to service quality that deters the incumbent from implementing an effective reform. As a result, there exists an intermediate range of reform popularity where resistance induces both types of policy distortion (
$ q\in ({q}^{\dagger \dagger },{q}^{\dagger }) $
, panel b in Figure 4).
Tying Bureaucrats’ Hands
Given these results, it is worth considering whether bureaucrats would benefit from committing not to resist. If the reform is ex-ante popular (low q), they are better off tying their hands. Moreover, the pro-reform incumbent and bureaucrats’ ex-ante interests over bureaucrats’ ability to privately resist tend to be misaligned; when one party clearly benefits from constraining resistance, the other is clearly harmed.
Proposition 5. In terms of ex-ante expected payoff, bureaucrats’ ability to privately resist the reform benefits the incumbent (hurts the bureaucrats) if
$ q\ge {q}^{\dagger } $
and harms the incumbent (benefits the bureaucrats) if
$ q\le {q}^{\dagger \dagger } $
.Footnote 25
This implies that (1) bureaucrats are better off without their ability to privately resist a policy they dislike if the policy is ex-ante popular and (2) the incumbent and bureaucrats cannot voluntarily agree on institutional factors that constrain bureaucrats’ resistance capacity without additional transfers between them.
DYNAMICS: BUREAUCRATS’ CONTINUED RESISTANCE
The main model implies that bureaucrats resist only in the first period to influence the incumbent’s reelection. We now extend this to examine how continued bureaucratic resistance across multiple periods affects incumbent and voter decision-making in a dynamic game with periods
$ t=0,1,2,\dots \hskip0.3em $
, the period-invariant value of the reform
$ \omega \in \{0,1\} $
, and overlapping generations of voters.
Assumption 1. In each period
$ t\ge 1 $
, a new generation of a representative voter is born and inherits the posterior belief about
$ \omega $
from the previous generation in
$ t-1 $
, makes a reelection decision for its own period after observing
$ {g}_t $
, and dies in the next period after receiving its policy
$ {g}_{t+1} $
.
This assumption allows us to maintain the voter inference as in the main model: each generation of voters born in period t cares only about its expected payoff in
$ t+1 $
when making reelection decisions.
The sequence of the extended dynamic game is as follows: At the beginning of the game (
$ t=0 $
), nature draws
$ \omega \in \{0,1\} $
. A pro-reform incumbent privately observes
$ \omega $
and publicly commits either to the reform or to the status quo for all following periods
$ t\ge 0 $
while she is in office.Footnote 26 If the incumbent commits to the reform at
$ t=0 $
, then the voter observes
$ {g}_t=(1-{b}_t)\omega +{\eta}_t $
in each period, where
$ {\eta}_t\sim H(\cdot ) $
is an i.i.d. shock in each period and
$ {b}_t\in \{0,1\} $
is the bureaucrats’ resistance decision in t.
After observing
$ {g}_t $
, the voter decides whether to reelect the incumbent, who will continue the reform, or the anti-reform challenger, who would revert to the status quo. If the pro-reform incumbent wins, then she faces another anti-reform challenger in the next election. If she loses, the game ends. If the incumbent commits to the status quo, the voter flips a fair coin to make a decision between the two candidates.Footnote 27 Again, if the pro-reform incumbent who was in office at
$ t=0 $
loses the election, the game ends.
The pro-reform incumbent who is in office at
$ t=0 $
receives
$ \rho $
in every period if the reform is implemented. In every period, she also receives
$ 1 $
by staying in office. Her payoff is multiplied by a discount factor
$ \delta \in (0,1) $
in every period. The voter born in t receives
$ {g}_{t+1} $
after his election decision if he chooses the pro-reform incumbent and q if he chooses the anti-reform challenger.
Unlike in the main model, the voter now takes into account resistance in
$ t+1 $
in his reelection decision after t if bureaucrats can resist. When the probability of resistance in
$ t+1 $
(after the election in t) is
$ 1-{\kappa}_{t+1}^{*} $
and
$ {\mu}_t({g}_t)=\unicode{x1D53C}[\omega |{g}_t] $
is the expected value of the reform given
$ {g}_t $
, the voter reelects the pro-reform incumbent in the election between t and
$ t+1 $
if and only if
$$ \begin{array}{rl}{\mu}_t({g}_t)\cdot {\kappa}_{t+1}^{*}\ge q.& \end{array} $$
In contrast, if bureaucrats cannot resist, then the voter reelects the pro-reform incumbent if and only if
$$ \begin{array}{rl}{\mu}_t({g}_t)\ge q& \end{array} $$
just as in the main model.
We focus on the steady-state solution where the voter applies a consistent cutoff rule that incorporates the effects of continued resistance in every period. Let
$ {g}_E^{*} $
denote the voter’s threshold in the extended game with bureaucratic resistance that solves Equation 6 with equality, and
$ {g}_{EB}^{*} $
denote the threshold in the game without resistance that solves Equation 7 with equality.
Since our goal is to understand the implications of bureaucratic resistance, we examine two cases: one where bureaucrats can and do resist from time to time, and another where they never do. To focus on the voter and the incumbent’s strategies, we assume that the probability of bureaucratic resistance in each period (i.e.,
$ \Pr [{b}_t=1] $
) is qualitatively the same as in the main model in the scenario where bureaucrats can resist.
Assumption 2. If the reforming incumbent commits to the reform at
$ t=0 $
and
$ \omega =1 $
, then bureaucrats resist with probability
$ 1-\widehat{\kappa}({g}_t^{*}) $
, where
$ {g}_t^{*} $
is the voter’s reelection cutoff in t, and
$ \widehat{\kappa}({g}_t^{*}) $
is
$ {\kappa}^{*} $
in the main model in the equilibrium where the voter’s cutoff is
$ {g}_t^{*} $
.
This assumption can be understood as bureaucrats playing a myopic strategy in every period that depends only on their payoffs in the next period as a function of the voter’s election strategy in that period. Under this assumption, there exists a unique cutoff
$ {g}_E^{*} $
that solves
$ {\mu}_t(g)\cdot \widehat{\kappa}(g)=q $
and a unique cutoff
$ {g}_{EB}^{*} $
that solves
$ {\mu}_t(g)=q $
. The key part of finding this solution is the fact that the posterior belief at
$ {g}_E^{*} $
must remain constant across periods,
$ {\mu}_{t-1}({g}^{*})={\mu}_t({g}^{*}) $
, for the voter to apply the same cutoff
$ {g}^{*} $
in every t in both situations.
As in the main model, our interest lies in understanding how the possibility of bureaucratic resistance affects the equilibrium reelection strategy of the voter and the incumbent’s initial policy decision. To do so, we compare
$ {g}_E^{*} $
to
$ {g}_{EB}^{*} $
.
Proposition 6. In the steady-state equilibrium of the extended game with periods
$ t=0,1,2,\dots \hskip0.3em $
, where the incumbent makes the policy decision only at
$ t=0 $
,
$ {g}_E^{*}>{g}_{EB}^{*} $
.
That is, in the steady state where the voter incorporates bureaucratic resistance into the reform’s expected welfare implications, the voter becomes more stringent toward the reform when bureaucrats can resist than when they cannot (
$ {g}_E^{*}>{g}_{EB}^{*} $
). As a result, the incumbent is less inclined to implement the reform when bureaucrats can resist than when they cannot.
This result implies that when bureaucrats can sustain resistance over time, voters rationally incorporate this persistent opposition into their evaluation of the reform. Unlike in the main game, the possibility of bureaucratic resistance thus unambiguously discourages reform: it makes voters more stringent and reduces the incumbent’s incentive to pursue reform.
EMPIRICAL EXAMPLES
To illustrate how our model helps explain various dynamics in bureaucratic politics, we provide examples of when resistance discourages or encourages reform. Importantly, we make no normative judgments about the social desirability of these reforms (
$ \omega \hskip-0.20em $
) or the realization of bureaucratic resistance (b) in these examples. Instead, we use these cases to demonstrate how variation in the key parameter q generates different reform efforts, either incentivizing more or less reform policies.
Examples of Dampened Reform Efforts
The deaths of Black Americans at the hands of police in recent years, including George Floyd, Daunte Wright, Breonna Taylor, and Tyre Nichols, have sparked a movement calling for sweeping police reform. While lawmakers across the aisle supported reform efforts in 2020, their impetus faltered in recent years and reform policies stalled (McCaskill Reference McCaskill2020; Pearson Reference Pearson2022). For example, despite efforts to cut police budgets in 2020, many jurisdictions increased funding again in 2021, often far beyond prepandemic levels (Fegley and Murtazashvili Reference Fegley and Murtazashvili2023; Manthey, Esposito, and Hernandez Reference Manthey, Esposito and Hernandez2022).Footnote 28 Similarly, while legislatures across U.S. states and cities initiated policy change through civilian oversight boards or restrictions on qualified immunity for police officers in 2020, most of the reform efforts soon died in the political process. Why?
Our model illustrates how resistance by powerful police organizations and their threats to sabotage reform policies might have contributed to politicians’ unwillingness to follow through with reforms aimed at police accountability and transparency. In particular, if voters are sufficiently wary about the effectiveness of reforms (q is high), our results predict that incumbents will shy away from reforms. This is because bureaucrats’ ability to resist reforms by undermining service quality can negatively affect both voters’ perceptions of the policy and incumbents’ reelection prospects.Footnote 29
A clear example of this are the difficulties of eliminating qualified immunity for police officers. In the aftermath of George Floyd’s killing, federal and state lawmakers nationwide attempted to reverse a legal principle that effectively shields police officers from being sued for violating individuals’ civil rights. Yet, the federal bill soon stalled in Congress, as bipartisan Senate negotiations failed, and by October 2021, at least 35 qualified immunity bills had been withdrawn or died in state legislatures (Kindy Reference Kindy2021).Footnote 30
The opposition to these reforms by police organizations played an important role in this development. Police unions bought ads in local newspapers warning that officers might hesitate to pursue criminals due to concerns about potential lawsuits, urging readers to call state legislators to oppose the reforms (Kindy Reference Kindy2021). Similarly, in opinion pieces, unions warned that crime would surge uncontrollably if the reforms passed (Kindy Reference Kindy2021). Against the backdrop of rising crime rates after 2020 (high q), this strategy effectively discouraged lawmakers from pursuing reforms that could make them appear soft on crime. In cases where police groups were unable to completely prevent immunity reforms (e.g., New Mexico), they often managed to shift the narrative and ensured that victims could only seek retribution from cities and counties, rather than individual officers (Kindy Reference Kindy2021). Hence, by underscoring their capacity to resist the policies (low c), leveraging citizens’ fear of crime (low g), and suggesting to voters that the reforms would be worse than the status quo and that incumbents would be responsible for any decline in service quality (
$ \omega =0 $
), police made reforms of qualified immunity electorally risky and unattractive for incumbents.
Examples of Increased Reform Efforts
Conversely, our model also explains how and when incumbents can leverage the possibility of resistance for their electoral gains. If reforms are fairly popular with voters (low q), incumbents are motivated to implement them and attribute negative outcomes to bureaucratic resistance rather than policy failure. Importantly, politicians can exploit the fact that voters expect bureaucrats to resist a policy they dislike, and bureaucrats cannot credibly deny their incentives to resist.
A prominent example of this is the strategy of populist incumbents to blame the “deep state” for policy failures (i.e., suggesting that bureaucrats are actively undercutting their political authority and thwarting the will of the people by sabotaging policies). Ron DeSantis’ efforts to blame teachers, librarians, and school administrators for implementation challenges with his education policies nicely illustrate this tactic. Since 2022, the Florida governor has implemented a series of laws that impose severe restrictions on classroom materials addressing topics such as gender identity, sexual orientation, racism, and slavery. These laws soon resulted in logistic chaos, as school districts were overwhelmed with requests from parents and conservative groups to remove a wide array of books from their curricula (Atterbury Reference Atterbury2024). Moreover, the policies led to empty bookshelves, as school districts started pulling even dictionaries and encyclopedias due to references to “sexual conduct” and a Miami school required parental consent for students to access a book by a Black author (Luscombe Reference Luscombe2024). Facing criticism over these outcomes, DeSantis asserted that school officials were strategically obstructing the policy. For instance, after a book about Puerto Rican baseball legend Roberto Clemente was removed for its discussion of racism, DeSantis argued that teacher unions were removing benign books to portray him as a racist, authoritarian zealot (Algar Reference Algar2023). Crucially, DeSantis’s narrative capitalizes on voter uncertainty about whether policy failures (low g) stem from strategic resistance by educators (
$ b=1 $
) or flaws in his policies (
$ \omega =0 $
). Given the vocal opposition from teachers and their unions, who protested and filed lawsuits against these policies, these bureaucrats cannot credibly deny their incentives to resist DeSantis’s policies. Hence, using the fact that the state bureaucracy might be working to undermine his administration, DeSantis leveraged expectations of bureaucratic resistance among his constituents to legitimize reform policies.
CONCLUSION AND DISCUSSION
Politicians inherently depend on bureaucrats to deliver policies to their voter base, and poor public service provision creates an electoral vulnerability for politicians. When and how can bureaucrats exploit this to affect policies they dislike? In this article, we argue that bureaucrats’ central position in government production, together with voters’ difficulty in attributing responsibility for service provision, vests bureaucrats with a unique source of political power. Our model illustrates how this leads to bureaucrats’ strategic resistance of public service provision, affects voters’ learning from policy outcomes, and can impact politicians’ policies and chances of reelection.
Using a three-player model with a politician, a bureaucrat, and a voter, we find that bureaucratic resistance leads to complex disruptions in electoral accountability relationships among voters and politicians. Depending on the voter’s beliefs about the reform’s merit, bureaucratic resistance (1) reduces the informativeness of public services for voters, making them either more or less favorable to the incumbent, (2) occurs more often if voters are more susceptible to the government outcome, and (3) can both promote and hinder reform efforts, sometimes resulting in too few beneficial reforms (under-reform) and too many ineffective reforms (over-reform) compared to the normative optimum.
Our model and analysis enrich our understanding of the degree of political motivation among bureaucrats and their consequences for voters’ learning and politicians’ behavior. In doing so, we highlight an underappreciated mechanism of political influence for bureaucrats as interest groups and micro-found a reason why bureaucrats act against the very programs and services they oversee. Additionally, we respond to recent calls to integrate interactions among politicians, bureaucrats, and voters within a single framework for studying political accountability (Grossman and Slough Reference Grossman and Slough2022). Compared to conventional models of electoral politics that examine the relationships between voters and politicians or between politicians and bureaucrats separately, this integration allows discovery of new mechanisms influencing voter learning, service quality, and government responsiveness.
This article opens several paths for future work. In our model, we focus on a simple two-period game and abstract away from potential dynamics. Particularly, we treat both the voter’s perceptions about the reform’s value relative to the status quo (q) and bureaucrats’ perceived costs of resistance (c) as exogenous. It appears fruitful for future theoretical research to explore how our results are affected by voters’ dynamic adjustment of their beliefs about the cost of resistance or the reform’s value over time. Additionally, while our focus on covert resistance reflects both empirical reality and our interest in informational manipulation, developing a unified framework that endogenizes bureaucrats’ choice between observable resistance in the form of strikes and covert resistance through subtle, deniable actions like police slowdowns would be valuable (Gibbs Reference Gibbs2023; Patty Reference Patty2016). Understanding when bureaucrats select each strategy, and how politicians and voters respond to this meta-game, would advance the theory of bureaucratic resistance.
Our model can also inform future empirical work on the drivers, conditions, and consequences of bureaucratic resistance in several ways. In particular, one could test the comparative statics described here (i.e., the effect of changes in voters’ beliefs about the reform’s value (q) and bureaucrats’ cost–benefit trade-off when resisting (c relative to
$ \kappa $
) on the probability of resistance (
$ 1-{\kappa}^{*} $
), the probability of reform (
$ 1-{\rho}^{*} $
), and the probability of reelection (
$ 1-{g}^{*} $
). Similarly, scholars could use surveys to empirically evaluate the impact of bureaucratic resistance (i.e., variation in c) on voters’ perceptions of reform merit (
$ \mathrm{E}[\omega |g,c] $
), conditional on the realized government quality (g).
SUPPLEMENTARY MATERIAL
The supplementary material for this article can be found at https://doi.org/10.1017/S0003055426101609.
ACKNOWLEDGEMENTS
The authors thank Benjamin Blumenthal, Roel Bos, Tak-Huen Chau, Eric Dickson, Justin Fox, Sanford Gordon, Catherine Hafer, Alexander Hirsch, Dimitri Landa, Kyuwon Lee, Justin Melnick, Anne Joseph O’Connell, Barbara Piotrowska, Ken Shotts, Tara Slough, Carolina Torreblanca, Stephane Wolton, Hye Young You, Antoine Zerbini, and Daniel Gibbs. The authors also thank participants at workshops and conferences at the University of Chicago, NYU, MPSA, and APSA for helpful comments and suggestions. We also appreciate insightful comments from two anonymous reviewers.
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors affirm that this research did not involve human participants.
Open access
Comments
No Comments have been published for this article.