Ministries matter: technocrats and regime loyalty under autocracy

How do technocrat ministers affect governance under autocracy? Autocrats frequently appoint non-par-tisan actors with technical competencies to bureaucratic leadership roles. Though their competencies might predict positive performance in office, these ministers are also dependent on the regime for their position and should thus demonstrate loyalty to its interests. I test this in the context of horizontal accountability to the legislature, using data on more than 27,000 legislative requests submitted to minis-tries in Morocco. I use both exact matching and difference-in-differences analyses to show that technocrat ministers are more than 25 percentage points less likely to respond to legislative queries than partisan cabinet members. The results imply that outside (partisan) participation in government strengthens weak institutions of executive oversight. They also cast doubt on the presumption that technocrat participation in government is universally beneficial to governance.

Morocco offers a useful example of within-case variation on the variable of interest, as both partisan appointees and non-partisan technocrats are represented in the cabinet during a single term. I adopt two strategies for dealing with lack of exogeneity in assignment of technocrats to ministry posts and in the data generating process. First, I use exact matching to pair legislative queries submitted to technocrats with those submitted to partisan ministers on a wide range of observable attributes. Next, I exploit a midterm cabinet reshuffle in which the regime gained control of two influential ministry portfolios, replacing partisan actors with technocrat appointees, to conduct a difference-in-differences (DiD) analysis using queries submitted before and after the change.
I estimate effects on minister responsiveness and find a large disparity: legislative queries submitted to regime-appointed technocrats were an estimated 26-32 percentage points less likely to receive a response than comparable requests submitted to partisan ministers. This suggests that technocrat ministers are less cooperative with institutions of horizontal accountability between branches of government and act as an impediment to elected MPs and their ability to address constituency needs through institutional authorities.
These findings extend the study of technocrat behavior into autocracy and offer distinct contributions to both literatures. First, they reveal a limitation of institutional function under autocracy. Autocrats have increasingly adopted democratic institutions; I show here that these institutions function quite differently in practice, allowing the regime to impede the activity of competitively elected bodies. Discretionary ministerial appointment power combined with a lack of legislative enforcement authorities weakens horizontal accountability and threatens legislative performance in a setting where this already faces many challenges.
Second, the results contradict the notion that technocrat ministers are universally beneficial to governance, despite their technical qualifications for office. Where technocrats in democracies are expected to enhance government function by providing expertise absent partisan rancor, I find evidence of reduced institutional cooperation in this autocratic setting, likely as a consequence of regime loyalty. The emphasis in this paper is on only one aspect of executive performance -accountability to the legislature. However, the evidence offers a useful first look at behavior in office and suggests a need to more closely examine how regime loyalty might manifest in other outcomes related to managing bureaucratic portfolios, including distributive metrics and citizen responsiveness.

Technocrat ministers under autocracy
Though much of the literature on non-partisan ministers is from democratic contexts, descriptive data suggest such appointees are actually more prevalent in autocratic cabinets, comprising around 30 percent of all autocratic ministers today ( Figure 1). 2 This disparity is not simply a function of the differing partisan structure across regime type: even single-party autocracies regularly appoint non-partisan ministers. 3 Research on non-partisan ministers in democratic settings largely focuses on the technical expertise of such actors, their level of autonomy, and the normative impact on governance, particularly in rescuing the state in times of economic crisis. 4 Camerlo and Pérez-Liñán (2015) add an important dimension to the study of non-partisan ministers, noting that such appointees may be chosen for their loyalty in addition to their technical competency.
2 Cabinet makeup estimated using data from Nyrup and Bramwell (2020), which categorizes non-partisans as "independents." Regime type identified using the Varieties of Democracy (V-Dem) "Regimes in the World" measure. I exclude regimes that have banned political parties altogether, though results are largely unchanged with their inclusion. 3 On average, they comprise 20-50 percent of the cabinet in such regimes. A parallel literature on autocracies further disambiguates the concept of competence: this includes technical expertise, which scholars argue contributes to long-term performance and stability of the regime (Bell, 2016), as well as political skill, which helps the regime maintain dominance even with competitive elections (Reuter and Robertson, 2012). Lee and Schuler (2020) provide evidence that ministers are promoted when they display technical skill but do not engage in political activity.
Taken collectively, these findings imply that the autocrat prefers a bureaucrat with the technical competence required to oversee her administrative portfolio 5 but who exhibits loyalty to the regime in dealing with potential political threats. This loyalty can be secured through institutional structure: in most autocracies, the regime has appointment power over ministries and can dismiss ministers if they contravene its interests. 6 I posit that ministers' behavioral incentives depend on this accountability to the regime as well as on their partisan affiliations. Partisan appointees were put forward by their party leadership and remain affiliated with that party while in office. On the other hand, because technocrat ministers have no partisan affiliation, there are no non-regime political actors to whom they answer. Their appointment was the result of regime support and requires it to continue.
I therefore argue that we should expect differing behavior from technocrat and partisan appointees, particularly in the area of legislative responsiveness. In autocracies with competitive legislatures, the regime seeks to limit the authority of the legislature in order to prevent political challengers from becoming too popular. Technocrat ministers with greater dependence on the regime will behave in accordance with its preferences and more frequently ignore legislative requests. I expect that they will be less responsive to legislative queries than partisan ministers, who maintain linkages to their nominating party and its representatives in the legislature. That  Autocrats seem to favor engineers over economists, but technocrat ministers in this setting meet the criteria established by McDonnell and Valbruzzi (2014); see Appendix A1 for more discussion. 6 Roughly 70 percent of autocratic leaders have full discretion over cabinet appointments, per V-Dem data. is to say, I do not expect technically competent ministers to behave as wholly independent, expertise-driven actors in autocratic settings, but rather to serve as regime agents in the cabinet.

Data and approach
To test this hypothesis, I use data on legislative requests and executive response from Morocco, a hybrid regime with parliamentary institutions. The king is head of state and possesses the bulk of political power, including influence over executive appointments. Following legislative elections, the king appoints a prime minister from the winning party and tasks him with forming a governing coalition and cabinet. Yet because ministers are appointed by the king on the recommendation of the prime minister, the palace is often able to secure a number of cabinet portfolios for its preferred candidates.
Historically, most Moroccan cabinets were almost entirely non-partisan, comprised of technocrats and regime insiders (Willis, 2002). In recent years, however, the cabinet has represented a mix of partisan appointees nominated by the parties in the governing coalition as well as nonpartisans selected by the king. The regime has, however, largely maintained control over the most important or sensitive ministries-a common practice in autocratic settings.
This within-case variation makes Morocco a useful setting in which to examine the behavior of technocrats in comparison to partisan ministers. The 2011-2016 term is of particular interest: the Islamist Party of Justice and Development (PJD) won a plurality of votes and partnered with three other parties to form a governing majority. The resulting coalition secured control of a majority of ministries, including several prestige portfolios normally headed by regime appointees. The list of newly partisan ministries included the Ministry of the Interior-a notoriously sensitive position and one never before held by a partisan actor.
But this government was relatively short-lived: in mid-2013, the Istiqlal party announced it would withdraw from the coalition over unresolved policy tensions between its leadership and that of the PJD (Fakir, 2018). 7 The split led to new cabinet negotiations, and the departing party was replaced in the governing coalition. In the ensuing reshuffle, regime-backed technocrats replaced partisans to lead the ministries of education and the interior-evidence of considerably increased palace influence within the cabinet. 8 Both the new appointees met technocrat criteria: they possessed advanced degrees, had a history of relevant experience in public service, and lacked pre-existing partisan affiliation. Each had been tapped by the palace for administrative positions in the past, an indication of longstanding cooperation with the regime. Rachid Belmokhtar, the new Minister of Education, had held the same post for a period in the 1990s; he had also been selected by the king to lead the National Observatory of Human Development, a research institute established by the palace, upon its founding in 2006. Mohamed Hassad, the new Minister of the Interior, had a prior ministerial appointment as well as an 11-year stint as regional governor (wali), an administrative role then appointed directly by the king. Further evidence of his relationship with the palace came during the term; Hassad (along with other regime-appointed ministers) accompanied the king personally on an international tour related to the rejoining of the African Union (de Larramendi and Tomé-Alonso, 2017). Finally, there is evidence that, once in office, the two technocrats pursued a palace-friendly policy agenda that ran contrary to the wishes of the PJD leadership. Hassad refused to make 2015 electoral data available for nearly a year, despite repeated requests by the government, and Belmokhtar reversed efforts at Arabization of the national school 7 The split was spurred by a disagreement over consumer subsidies: see "Ministers to quit Moroccan coalition," Al Jazeera, July 9, 2013 (https://www.aljazeera.com/news/africa/2013/07/20137991633558497.html). I do not find that the changes to the cabinet affected the data generating process (Appendix A3). 8 The partisan ministers being replaced were from the Istiqlal and Popular Movement parties. A full list of ministry appointments during the 2011-2016 parliamentary term is in Appendix Table A1. curriculum despite this serving as a core aspect of the PJD platform (López García and Hernando de Larramendi, 2017). In other words, these appointees possessed technocrat qualifications while exhibiting signs of loyalty to the palace.

Data
To test the prediction that technocrat ministers are less responsive to the competitively elected legislature, I construct a dataset of all 27,196 unique written queries submitted by MPs to cabinet members during the 2011-2016 parliamentary session, gathered from the official parliamentary website in March 2017. The parliamentary database reports the subject of each query, asker(s), ministry being queried, date submitted, and date of response, if one was received.
I supplement this information with data on minister affiliation, MP characteristics, and content analysis of the queries themselves. Minister affiliations were obtained through media research and verified with the assistance of a local NGO focused on citizen-parliamentary relations. MP characteristics were gathered from the parliamentary website. Finally, I used dictionary-based text analysis to code the subject of each query, including geographic references, casework on behalf of individual constituents, and critical or sensitive queries.
Technocrats held six out of 32 ministries (19 percent) during the term; they received a proportionate number of queries (5929, or 22 percent). Summary statistics at the query level on these and other attributes are reported in Appendix Table A2.

Approach
The data used in this study are observational, which poses challenges for inference. Ministers are not randomly assigned to high-and low-prestige ministries, which could confound the relationship between minister type and responsiveness. Query submission could also be subject to selection bias; MPs may self-censor if they anticipate non-response, or be more confrontational with ministers from the regime in the hopes of scoring political points.
To address these selection issues, I adopt two distinct analytic approaches: matching on observable characteristics and difference-in-differences (DiD) estimation using the midterm change in ministry leadership. For the former, I employ exact matching with estimation using stratification weights. The goal is to generate a counterfactual group for treated observations that is as similar as possible with respect to other attributes, such that the only observable difference is whether queries were submitted to a technocrat versus partisan minister. Exact matching accomplishes this by sorting treated and control observations into subclasses defined by precise covariate values, then weighting observations based on the proportion of treated observations within blocks and the overall proportion of treated observations. Exact matching has the advantage of generating complete balance on variables included in the matching algorithm, such that there are no differences between treatment and control when weights are taken into account.
I match treated and control observations on attributes related to query timing (year, cabinet period, and proximity to the end of a cabinet period); content (whether it includes a georeference, constituent casework, or critical or sensitive language); and deputy characteristics (party, leadership role, and whether the submitting deputy was elected as part of the national list). Additionally, I match on two metrics of ministry importance. First, I code an indicator for whether the ministry received a greater-than-median number of queries; this is intended to proxy for the size and capacity of the bureaucratic profile involved. 9 Second, I code an indicator 9 These include ministries with broad bureaucratic portfolios, like the Ministry of the Interior, as well as service ministries, like the Ministry of Health. The full list of large ministries includes craft, economy, employment, energy, health, housing, industry, interior, islamic affairs, justice, national education, prime minister, transportation, women and family, and youth and sports.
for what scholars have termed the "ministries of sovereignty:" sensitive portfolios that the regime has sought to control historically (Sater, 2003;Darif, 2012). These include the ministries of the interior, foreign affairs, justice, and Islamic affairs. This indicator is intended to serve as a measure of the ministry's strategic relevance. 10 Exact matching keeps all observations for which there is a covariate analog in the alternative category (Stuart, 2010). In this case, the matched dataset preserves 93 percent of treated observations (5542 queries) and 71 percent of untreated observations (13835 queries). 11 Because nearly all queries submitted to technocrats have covariate analogs in the control group, the estimand can be interpreted to approximate the average treatment effect among the treated (ATT). Summary statistics for the matched data are shown in Table A3.
For the DiD estimation, I define treatment as a switch in ministers from partisan to technocrat, and I therefore subset the dataset to include only queries submitted to ministries that were partisan during the first cabinet. 12 This approach has the advantage of holding ministry constant and allowing us to differentiate between response rates for questions that would have been assigned to a non-technocrat minister had the cabinet change not taken place. Questions asked before the change in ministers serve as a reference category for subsequent queries, and the design is thus able to account for other non-treatment-related aspects of ministry responsiveness, including time-based variation.
I define the following additional variables: post is a binary indicator for whether or not a query was raised after the cabinet reshuffle. Treat is a binary indicator for whether or not the question was assigned to a ministry that became technocrat in 2013. 13 I then estimate the equation where Y is the relevant metric of responsiveness, i denotes individual queries, and X i indicates the battery of additional control variables pertaining to ministry importance, timing, content, and the submitting MP (identical to those used in exact matching). The DiD design relies on the assumption of parallel trends; I find no evidence that this assumption is violated (Appendix A3). I additionally find evidence that the data generating process is not altered by the introduction of treatment. I consider two outcome metrics of ministry responsiveness. The first is Received Response, a binary indicator for whether a query received an official response before the government it was directed to was dissolved. The second is Time to Response (TTR), measured as the number of days between the initial question and its response, if one was issued. Queries receive an answer only once, so each appears as a single observation in the dataset.
I use OLS regression in specifications using the Received Response outcome. For the TTR outcomes, I estimate a semi-parametric Cox proportional hazards model. All specifications using the matched data employ stratification weights; I estimate treatment effects with and without controls (the same variables included in the matching algorithm) and ministry fixed effects and cluster robust standard errors at the ministry-cabinet level. I likewise estimate the DiD equation with and without controls and construct robust standard errors clustered at the ministry level. 10 In Appendix A5 I construct an alternate metric using historial regime appointments; results are substantively very similar. 11 Exact matching is among the most restrictive forms of matching and imposes no functional form assumptions on the assignment algorithm. In this case, the large sample size and binary covariates help to ensure comparability for a large portion of the total sample. 12 This excludes 2290 queries (8%) submitted to ministries held by technocrat or non-partisan ministers throughout the parliamentary term (Appendix Table A1). Because no ministries changed from technocrat to partisan hands in 2013, I am unable to test the reverse treatment. 13 The two ministries that switched from partisan to technocrat control fielded 6843 queries throughout the 2011 term (25 percent of the total dataset).

Results
I begin with results from exact matching, reported in Table 1. 14 In all specifications using the binary response outcome, queries submitted to technocrat ministries were significantly less likely to be answered (models 1-3). The difference in response was substantively large: the estimated effect of Technocrat Minister was a roughly 32 percentage point decrease in the probability of receiving a response. Models 4-6 report results from estimating a Cox model on TTR outcomes. Consistent with expectations, technocrat ministers responded more slowly than their partisan counterparts. The treatment coefficient is negative and significant in all specifications, meaning the hazard rate is reduced for technocrat ministers.
In this case, matching ensures that differences in the type and timing of queries submitted to different minister types are not driving the differences in response rate; yet it cannot account for confounding by factors not included in the set of matching covariates. In particular, we might be concerned that different ministries have different capacities for response or receive questions of more or less complexity to address in ways that are not captured by the ministry indicators (size and regime importance) included in the matching algorithm. To this end, the DiD approach has the added benefit of holding ministry constant. We can thus compare outcomes for queries submitted to the same ministry under differing leadership conditions. Table 2 reports output from the DiD specifications. Models 1 and 2 present estimates of the average treatment effect among the treated using the binary response outcome. 15 As with the matching analysis, technocrat ministers are associated with a considerably lower response rate: for ministries taken over by technocrats, the probability of response declined by 26 percentage points relative to those that remained partisan. The coefficient is significant in both specifications with standard errors clustered at the ministry level. Before the technocrat ministers took over, treated and untreated queries experienced similar response rates (59 and 61 percent respectively). This changed dramatically after the technocrat "treatment" was administered: while the response rate rose for untreated queries (to 70 percent), it fell to 42 percent among treated queries.
Again, there is evidence that technocrat ministers respond more slowly than their partisan counterparts. Models 3 and 4 report results from estimating a Cox model on TTR. As expected, the estimated effect is negative and significant in all specifications. Substantive interpretation of the semi-parametric Cox estimates is complex; to illustrate the real-world impact of this behavior, I plot Kaplan-Meier survival curves by treatment condition and cabinet period (Figure 2). During the 2011 cabinet (solid lines), the curves are similar for treated and untreated ministries. Though  14 For full regression output including controls, see Appendix Table A4. 15 For full regression output including controls and constituent terms from the DiD interaction, see Appendix Table A5.
the speed of response for both categories declined during the 2013 cabinet (dotted lines), the change was much more dramatic among treated ministries. During the 2011 cabinet, 25 percent of queries submitted to untreated ministries were answered within 91 days; this declined to 120 days during the 2013 cabinet. For treated ministries during the 2011 cabinet, 25 percent were answered within 66 days. But in the 2013 cabinet, this declined to 389 days-more than a year -for the newly technocrat ministries. Both the exact matching and DiD methods offer distinct benefits and drawbacks in enhancing inference. Matching only explicitly accounts for variation in the attributes on which observations were matched, which here include ministry, query, and MP-level attributes. DiD allows for comparison of partisan and technocrat ministers while holding ministry constant and accounting for possible time trends, yet in this case only two ministries switched to technocrat leadership. It is therefore encouraging that the estimates from both approaches are consistent in size, sign, and significance. Inference from the DiD analysis is contingent on the parallel trends assumption, Treatment effects from DiD estimation, with and without controls. Models 1-2 represent OLS estimation using a binary indicator for response received; models 3-4 estimate a Cox PH model with a TTR outcome. Standard errors are clustered at the ministry level. Note: *p < 0.05; **p < 0.01; ***p < 0.001. and therefore threatened if the technocrat treatment led to a shift in the type of queries submitted. In Appendix A3, I offer evidence that, based on submission rates and query content, the data generating process remained consistent across both cabinet periods. Another empirical challenge is the fact that treatment is administered at the ministry (or ministry-cabinet) level, and there are a comparatively small number of ministries overall. 16 This reduces the overall power of the design to detect effects. In this case, the estimated coefficient of interest is substantively large and significant when robust standard errors are clustered at the level of treatment. To bolster these results, I also conduct a set of robustness checks, including varying the approach adopted in matching (Appendix A5), using a more restrictive sample to reduce confounding in the DiD sample (Appendix A6), and looking at effects using the full observational dataset (Appendix A7). Results remain consistently negative and significant across the varying approaches.

Explanatory mechanisms
To what should we attribute this differential responsiveness to legislative requests? I have argued that it follows from the differential accountability of technocrat and partisan ministers to the regime versus political parties with representation in the legislature. The regime prefers not to enable the politically competitive legislature and seeks to obstruct legislative requests; institutional accountability ensures that technocrats serve as its agents in office.
Yet there may be other explanations for the reduced responsiveness of technocrats. Legislative competition in some Middle Eastern autocracies is centered around the use of personal connections (wasta) to obtain government services (Lust-Okar, 2006;Blaydes, 2010). Leaving aside a normative evaluation of this political strategy, 17 it may be the case that technocrats emphasize issues of broader policy in their administrative approach and are therefore less concerned about and responsive to constituent issues and localized service provision. Content analysis suggests that a majority of legislative queries address issues of constituency service (focused on district, sub-district, or individual constituent outcomes). Under this explanation, the effects found here would illustrate a qualitative difference in how technocrat and partisan actors approach administrative roles rather than revealing differential loyalties to the regime. If this explanation holds in Morocco, we would expect to see stronger treatment effects for queries that address localized or constituent issues and attenuated effects for those that do not (i.e. those queries that more plausibly address broader issues of policy). 18 To test this, I subset the dataset based on query content. Table 3 reports the technocrat response rate separately for queries that contain references to sub-district places or casework and those that do not, using the same matching procedure and weighted analysis as in the main results. 19 The subset analysis does not reveal a differential technocrat effect across the two subsets: the coefficient remains very similar. This suggests that the main effect is not driven by a technocrat bias against constituency-focused queries specifically.
Another possible explanation for these findings is that rather than reflecting the regime loyalty of technocrats, they result from partisan loyalty exhibited by partisan ministers. Because they remain active party members, partisan ministers retain a vested interest in the performance of their party in office as well as their own status within the party ranks. In other words, it is not the case that technocrats are less responsive, but that partisan ministers are more responsive than we would otherwise expect. To some extent, this is a matter of perspective and how we define "baseline" responsiveness. Yet if the results here are a function of partisan loyalty, we 16 The intra-class correlation coefficient is only 0.19, meaning assignment to a ministry is not a deterministic predictor of response-there is quite a lot of variation in responsiveness within ministry clusters. 17 In interviews, MPs emphasize the importance of fieldwork and constituency service in their role as representatives. 18 Constituency-focused queries were submitted to technocrat and partisan ministers at similar rates (82 and 78 percent, respectively). would expect to find evidence of co-partisan bias: increased responsiveness to queries submitted by MPs from the same party as the minister in question. I explore this possibility in Appendix A8 and find that partisan ministers are somewhat less likely to answer requests by fellow party members, and that the estimated effect of technocrat ministers remains negative when taking this into account. This suggests that the difference in technocrat and partisan responsiveness is not driven by positive co-partisan bias amongst partisan ministers.
Finally, I consider minister behavior related to overtly critical queries. If technocrats serve as regime agents on the cabinet, we might expect them to be especially obstructionist when it comes to sensitive or confrontational topics. While many written queries address issues (especially of service provision), they are not typically phrased in a confrontational manner. 20 Yet a small number of queries (4 percent) use provocative language (e.g. referring to something as a violation or injustice) or address sensitive issues such as protests or human rights. 21 I again use subset analysis to look at the technocrat effect among queries that are overtly critical or address sensitive issues and those that do not (Table 4). 22 The results produce two interesting insights. First, it does appear that technocrat ministers are especially unreceptive to confrontational or sensitive queries: for these, the matched analysis suggests they are nearly 50 percentage points less responsive than partisan ministers (models 1 and 2). 23 The technocrat coefficient is statistically significant even with the small sample and clustered standard errors. 24 Second, it does not appear that this is driving the core results: technocrat ministers are also less responsive to non-critical queries, with the estimated coefficient remaining similar to that in Table 1. In other words, there is evidence that technocrat ministers also engage in more generalized obstruction of legislative requests. This subset exploration should be seen as suggestive, but it is consistent with the paper's argument that technocrats act as regime agents in office. We would expect the regime to be especially sensitive to (and therefore less likely to answer) criticism. Results from OLS estimation of query response on technocrat treatment, with and without controls, estimated on subsets including constituency-focused queries (models 1 and 2) or those that do not address constituency issues (models 3 and 4). Models are estimated on datasets constructed using exact matching with stratification weights. Robust standard errors are clustered at the ministry-cabinet level. Note: *p < 0.05; **p < 0.01; ***p < 0.001. 20 This is likely due to the fact that other legislative actions, such as televised oral queries, offer a more public mechanism for critiquing the regime (Malesky and Schuler, 2010;Rozenberg and Martin, 2011). 21 I adopt an inclusive approach that codes as critical queries that explicitly address a problem, complaint, or issue; that refer to a violation, grievance, or injustice; and that address a sensitive subject, including crime, corruption, protest, or human rights. 22 Note that I do not find evidence that critical queries are directed disproportionately at technocrat ministers: 4 percent of queries submitted to technocrats were critical, compared to 4.5 percent of queries to non-technocrats. In the DiD parallel trends analysis (Appendix A3), I find that the rate of critical query submission is comparable before and after the minister switch and between treated and untreated ministries. 23 I find nearly identical results when using the full, observational dataset. 24 An interaction model (not shown) also reveals that the difference in coefficients between models 1 and 3 is itself statistically significant. competent party members (Lee and Schuler, 2020) and engage in legislative obstruction. Electoral competition offers an alternative path to political advancement subject to citizen rather than leadership preferences. Reducing legislative performance overall may help prevent the rise of political rivals, whether inside or outside the party.
The disruption to institutions of horizontal accountability demonstrated here is distinct from patterns in democratic contexts. 30 Technical competence is valued under both autocracy and democracy. But when it is developed in conjunction with dependence on the regime, we should not expect such actors to behave as agnostic experts in office. Regime-appointed ministers are often career bureaucrats who have acquired the necessary technical credentials, including advanced degrees and academic work, but who have also had regime loyalty instilled by years of service to the state. The results here are suggestive that this loyalty is, at least in some contexts, deployed in opposition to politically competitive institutions. Technocrats may thus serve a different purpose in autocracies, obstructing systems of electoral representation rather than simply cutting through partisan rancor. These findings suggest a number of interesting topics for future study, including developing an understanding of the conditions under which these patterns of horizontal (in)accountability hold and examining whether and how regime loyalty affects other aspects of bureaucratic performance.