Politicians, bureaucrats and the public–private choice in public service delivery: anybody there pushing for remunicipalization?

ABSTRACT Empirical evidence on remunicipalization remains scarce, and even more so as regards potential differences in the roles played by politicians and bureaucrats in service delivery reform. We use information obtained from a survey of Spanish municipalities to investigate differences in the service delivery preferences of politicians and technical staff, as well as differences in their respective propensities to reform. The results we obtain suggest that bureaucrats have both a stronger preference for private participation in service delivery and for reforming services than do politicians.


Introduction
The remunicipalization of local public services has been a heavily debated issue in Europe in the last decade. Empirical analyses of the factors driving remunicipalization have accordingly increased in recent years. However, the existing literature has paid little attention to whether politicians and managers share similar views on remunicipalization, and whether, in practice, this might eventually influence service provision decisions. To the best of our knowledge, this issue has been addressed solely for the US case (Hefetz and Warner (2007), Warner and Hefetz (2012), and Warner and Aldag (2019)).
However, no evidence has been obtained about potential differences in the views of elected politicians and bureaucrats (i.e. technical staff) on the question of privatization and remunicipalization in Europe. While the "professional" form of government is not as common in Europe as it is in the US, it can be assumed that technical staff working in departments responsible for local public services management and supervision exert CONTACT Germà Bel gbel@ub.edu some influence on the decisions taken regarding delivery choices and, in particular, about privatization or remunicipalization. This paper seeks to fill this gap in the literature. To do so, we use data obtained from a survey conducted for this specific purpose in Spanish municipalities, and for a wide set of public services. We find that technical staff express a stronger preference for private participation in service delivery than is shown by politicians, regardless of whether the latter are conservative or progressive in ideology. Similarly, we find that technical staff show a greater propensity for reform (either to privatize or remunicipalize) than do politicians.

Empirical evidence of drivers of remunicipalization
Interest in the empirical analysis of the factors driving remunicipalization has grown in recent years, mirroring the attention devoted to it in political debates and case study reports. The first multivariate analyses would appear to be those conducted for the US by Warner and Hebdon (2001) and, later, by Hefetz and Warner (2004. These studies conclude that, in general, reversals of privatization tend to be more pragmatic than ideologically driven decisions. Only recently have multivariate empirical studies been carried out for European countries. Chong, Saussier, and Silverman (2015) analyze the factors accounting for the renewal of water contracts in France and find that ideology has little influenceif anyin decisions not to renew, while pragmatic considerations are more persuasive. Similar results are reported by Campos-Alba et al. (2017) in a study for Spain, where non-renewal decisions are driven by pragmatic considerations. Gradus and Budding (Forthcoming) analyze privatization and remunicipalization in the Netherlands and find that conservative parties are more strongly opposed to remunicipalization while progressive parties are more opposed to privatization. In another study conducted in the Netherlands, Gradus, Schoute, and Budding (2019) find that remunicipalization negatively relates to measurement difficulty and positively relates to average household income, but any potential association with ideology is not considered. Therefore, although robust evidence remains scant, the findings available suggest that the decision to remunicipalize is, as Voorn, van Genugten, and van Thiel. (2019) conclude, pragmatic rather than ideological.

Politicians, managers and service delivery choices
While conventional public choice theory recognizes the individual objectives of politicians and bureaucratsbasically, the preference for public delivery as a tool that allows them to promote their own goals, such as budget maximization, in recent years the theoretical literature in economics has begun to address differences in the policymaking objectives of politicians and bureaucrats. For example Tabellini (2007, 2008) claim that bureaucrats have a comparative advantage in the case of technical policymaking.
That bureaucrats should have an advantage in policy implementation is consistent with insights from public administration that managing contracts (and managing reforms, in general) requires managerial skills (Andrews and Boyne 2010), and these are more prevalent in managerial governments than in politician-based governments. Hence, the former are expected to be more effective in dealing with problems of contract management and market competition .
In the specific case of the relationship between managerial governments and remunicipalization, we are aware of evidence only from the US. Thus, while Hefetz and Warner (2007) and Warner and Hefetz (2012) find that professional managers do not remunicipalize more, Warner and Aldag (2019) find that a professional manager form of government is related to more remunicipalization. Hence, to date, the evidence is scarce and far from conclusive.
In most European continental countries, city governments are non-managerial or political in style. In Spain, city government comprises elected city counselors, with responsibility for the city's different services and departments, in which technical staff oversee all technical matters. Typically, members of this technical staff are public servants, and retain their positions even when there are changes in city government. For this reason, technical staff may not be as adverse to undertaking reforms as politicians, who will tend to avoid actions for which they can be blamed (Hood 2010). However, we should not ignore other evidence that suggests that when politicians are under pressure from electoral competition, they may be more prone to initiate reforms (see Berliner and Erlich 2015,;Bernecker 2016). As such, the evidence is inconclusive, and the matter requires additional research.
The data obtained from our survey allow us to distinguish between the delivery choice preferences of politicians and technical staff. Hart, Shleifer, and Vishny (1997) point out that politicians can use government delivery of services to engage in patronage and so obtain political rents. Enikolopov (2014) finds positive evidence for this hypothesis, which is also consistent with views on bureaucrats' preferences for externalization as presented by Dunleavy (1991) and James (2003). Hence, in line with theoretical arguments and available empirical evidence for city governments, we believe that differences in delivery preferences may exist because of discrepancies in technical abilities to manage contracts. Furthermore, these differences may be attributable to the fact that public production is associated with the more active role played by elected officials, who have the political control of departments and agencies, whereas external production may be associated with a more active role being played by technical staff, who have a greater capacity for supervising and monitoring externalized services.
Our first proposition (P 1 ), therefore, is that politicians have a stronger preference than technical staff for public production. Our second proposition (P 2 ) is that politicians have a weaker preference than technical staff for reforming the organization of service delivery.

Institutional context
The municipality is the core jurisdiction of local government in Spain. According to data from the Spanish National Statistics Institute (INE), in 2018, there are 8,124 municipalities. Many of these are extremely small: 5,872 municipalities (72% of the total) have fewer than 2,000 inhabitants. This means the average population per municipality is low, standing at 5,751 in 2018. Municipal councils in Spain are elected for a four-year term. Voting is based on party lists and municipal council members are elected on a proportional basis (but a party list must obtain at least 5% of the vote to be allocated council seats). Thereafter, council members elect the mayor at the first meeting of the newly elected city council (Warner and Bel, 2008).
The members of the municipal government are appointed by the mayor, all of whom must be elected members of the city council. This means all government departments are headed by elected officials appointed by the mayor and staffed by technical workers, with the administrative status of public servants, to implement policies. Occasionally, and then only in Spain's largest cities, professional managers are appointed to undertake the technical management of the city departments, but always under the direction of the local counselor who heads the department.
This brief overview of the basic characteristics of Spain's municipal government shows that the mayor-council form of government is prevalent. The mayor is responsible for appointing the heads of departments from among the elected counselors sitting on the city council. Then, each department has technical staff (often made up of public servants) responsible for implementing the policies decided by the council.

Data on privatization and remunicipalization
Our sample comprises the 97 Spanish municipalities that answered our survey, sent out to all municipalities with more than 1,000 inhabitants in 2018. The size of the sample municipalities ranges from 1,150 inhabitants (Torrecampo) to 1.6 million (Barcelona), the largest city in our sample and the second largest city in Spain. The survey included questions about the current model of production of the 17 services described in Table A1 (see Appendix) and asked the respondents to express their opinion about the optimal production model.
The survey was sent to the municipalities' main council address without specifying the name or role of the addressee and asked the respondent to specify name, contact details and position on the city council. Thus, we received answers from quite a mix of respondents holding different positions on their respective city councils. Politicians are identified by their either being mayors, elected councilors, or holding positions of political confidence having been appointed directly by elected politicians. Technical staff are identified by their being public servants. Among the technical staff we find municipal and service managers, economic area managers, municipal technical engineers, administrative staff, technical service coordinators, public accounts controllers, service quality and evaluation managers, among others. Thus, we obtained one completed survey per municipality, but we had no influence over who the exact respondent was completing and submitting the information.
According to our survey, only 17.5% (17) of respondents reported at least one reform in the municipal production model over the preceding 5 years (See Table 1). Among these, 8.2% (8) reported one reform, 3.1% (3) reported two and the remaining 6.2% (6) reported more than two reforms, with four being the maximum. Among the municipalities reporting reforms, the vast majority corresponded strictly to either a privatization or remunicipalization, with the exception of two municipalities where both kinds of reform were reported to have been implemented in the same period. According to our survey results, only 8.2% of municipalities declared having privatized at least one of their services by contracting it out or signing a public-private partnership (PPP) in the preceding 5 years. Half of these 8.2% municipalities carried out just one privatization (Table 1); the other half more than one, with three being the maximum number reported by one respondent. Of the services affected by these privatizations, one third were water utilities and a quarter sewerage collection (See Figure 1). Privatizations were also recorded in solid waste collection, civil protection, parks and social action.
In contrast, 11.3% of municipalities reported having implemented at least one remunicipalization in the preceding 5 years. Thus, only a few municipalities carried out some form of reversed privatization and about a half of these did so as an isolated instance, having no effect on the rest of their services. Only 5.1% of all municipalities reported implementing more than one remunicipalization and just 1% reported more than two remunicipalizations, with four being the maximum number in the sample.
Interestingly, the services most frequently remunicipalized were very similar to those that were privatized during the period (see Figure 1). Thus, reverse privatizations were most frequent in water distribution and sewerage and waste collection. To a lesser degree, we also find remunicipalizations in road pavement, parks, public lights and civil protection.
Following these reforms, the current distribution of production models in our sample can be summarized as in Table 2, ordered by the percentage of public production. Technical-type services related to solid waste management, urban transportation, and water distribution are those for which private involvement is highest. In contrast, personal services and those related to environmental action have the highest degree of public production. In terms of the dynamics of these service delivery choices, a comparison of the current distribution with Source: Authors data available for 2003 for solid waste collection and water distribution (Bel 2006: 113 & 125) shows that the share of public production has remained extremely stable.

Service delivery preferences of politicians and bureaucrats
To evaluate differences in perception of the relative optimality of production models as expressed by politicians and technical staffi.e. stated preferences for public or private involvementwe exploit our survey responses. Respondents were also asked to record the official post they held on the city council which enabled us to identify and assign the respective roles of either politician or member of the technical staff. From our sample, 45% of respondents were classified as politicians and 55% as technical staff. Note that we assume the distribution of respondents between the two classesi.e. politicians vs. technical staffto be random and unrelated to any potential determinant of differences of opinions that might emerge from the position held on the council. It might be claimed that the likelihood of receiving a survey response from a politician correlates with the size of the municipality and that this might influence the respondent's opinion regarding the optimal model of production. However, the pair correlation between the binary variable denoting with 1 a politicianand 0 a technical staff memberand the municipality's population is −0.0087, that is, less than one percent. Furthermore, a probabilistic regression on the probability of the respondent being a politician shows that the coefficient associated with the population variable is not statistically significant (p-value: 0.953) [results available upon request].
We employ three empirical approaches to compare the opinions expressed by the respondents about the three production models considered in the survey for each of the 17 services included: Public production, contracting out and PPPs.
1.-Our first approach is to estimate a two-sample t-test (with equal variances) by the position held by the respondent. To do so, we create a categorical variable on the degree of privatization deemed as being optimal, which takes a value of 0 if the respondent considers the optimal model to be public production, 1 if it is considered to be a PPP (both public and private participation) and 2 if it is considered to be contracting out (private production). We use this variable to estimate mean differences between politicians and technical staff. 2.-In the second approach we drop PPPs, which are complex contracting modes with heterogeneous private involvement, and compare differences in the production model preferences (public production vs. contracting out) of politicians and technical staff by means of logistic regression. 3.-Third, we apply a multinomial logistic model to predict differences in the choice of production model (here, reintroducing PPPs) made by politicians and technical staff, without assuming any strict increase in the degree of private participation depending on the production model considered.
The results from running our first approach are presented in Table 3. For each of the services considered we compute the real mean value for the existing production model (Column 1) and the mean for the declared optimal model according to respondents (Column 2). This allows us to compare the average real situation with the average preferred situation. Respondents assigned a higher mean value (i.e. a preference for greater private participation) to most services, except for water distribution, sewerage collection and street cleaning.
To verify differences in preferences between politicians and technical staff, Figure 2 shows the average score awarded to the existing models in the sample for all services considered and the average scores for the optimal model as reported by technical staff and politicians, respectively.
Next, we estimate the mean difference and its statistical significance between politicians and technical staff. Column 1 shows the mean value of the overall sample, columns 3 and 4 record the mean values reported by both politicians and technical staff and, Table 3. Two sample t-test on the optimality of production models.

Service
Real Mean (1)  finally, column 5 shows the mean difference and its statistical significance. Inspection shows that politicians tend to favor production models with less private involvement than is preferred by technical staff. This holds for all services (except for social action, which is not statistically significant), with politicians assigning lower values than the mean assigned by technical staff. This means that our first hypothesis is not rejected: Politicians have a stronger preference than technical staff for public production. The differences that are statistically significantat least at the 10% leveland, therefore, the opinions that can be assumed to be significantly different between the two groups of respondents, are those associated with water, sewerage, street cleaning, library facilities, passenger transport, and environmental action. Here, it is worth recalling evidence in the literature (Bel and Fageda 2017;Petersen, Houlberg, and Christensen 2015) that suggests privatization in the case of social and personal services tends to have a stronger ideological influence than privatization in technical services. Our results in Table 3 (also those below, in Tables 4 and 5) indicate that technical staff preferences for privatization are particularly significant in the case of technical services (those that tend to appear in the lower part of the tables, for which private delivery is more frequent), which is consistent with existing evidence.
In the above analysis, however, we assume an ordered interpretation of production models based on the amount of private involvement, considering PPPs an intermediate model between pure public delivery and contracting out. PPPs can be considered complex contractual forms that differ from contracting-out models in terms not only of their public entity participation, but also of the allocation of tasks, risks and rewards among parties. Indeed, some PPPs may involve much more actual private participation (financing, maintenance, operation, etc.) than contracting out and this private management may operate in a more flexible, unconstrained regulatory framework than exists in contracted-out firms. For this reason, we verify our results by reversing the order of PPPs and contracting-out models, but we obtain very similar results. To avoid problems from including PPP contracts, we also test for differences between public production and contracting out without PPPs; thus, we create a binary variable that takes a value of 0 for public production and of 1 for contracting out. We then recompute the mean values for both politicians and technicians in order to estimate mean differences for each of the services. In this case, without PPPs, we find statistical differences between politicians and technical staff in their preferences for water  Note: Standard errors in parentheses are robust to heteroskedasticity and clustered by region (NUTS 2). *,**,*** statistical significance at 10%, 5% and 1%, respectively. distribution, sewerage, passenger transportation and environmental actions (results available upon request). In all these services, politicians reveal a stronger preference than technical staff for public production. However, these descriptive statistics fail to take into consideration the size of the municipality. To address this, we apply a multivariate logistic regression to predict how being a politician affects the relative opinion expressed on production model optimality with respect to that of technical staff. Here we include two covariates that account for size effects. We include population (in thousands) and its square, to account for the inverse U-shape of the relationship between population and contracting out in Spain (Bel and Miralles 2003;Bel 2006). Size effects are relevant because they are proxies of both economies of scale and the managing capacity of the city council. Thus, our logistic regression takes the following form: where the probability functions of choosing contracting out and public production as optimal models are regressed against the three covariates of interest. Politician refers to a binary variable that takes a value of 1 if the respondent is a politician and 0 if the respondent forms part of the technical staff. The rest of the regressors are the number of inhabitants in each municipality (Population) and its square (Population^2). Table 4 displays our main results, with preferences for the delivery of water distribution, sewerage and urban passenger transport presenting statistical differences between politicians and technical staff. In the case of library facilities and environmental actions, we are unable to derive a probability because the characteristic of being a politician determines perfectly the outcome of public production, a finding that is consistent with our previous results. This is a clear indication of politicians' strong preference for public production. In all these services, politicians are less likely than technical staff to choose contracting out as the optimal model. Recall that we present our results in terms of odds ratios to facilitate interpretation. Politicians and technical staff have the same probability of choosing contracting out as the optimal model for a specific service if the odds ratio is equal to 1. When lower than 1, it means politicians are less likely than technical staff to choose contracting out; when higher than 1, politicians are more likely to choose contracting out as the optimal model. According to our results, technical staff are more than twice as likely to choose contracting out than politicians for water distribution and passenger transport, and almost three times more likely to do so in the case of sewerage.
Finally, our third approach allows us to include PPPs and avoids the arbitrary ordering of models. We predict the differences in opinion held by politicians and technical staff independently, without considering any scale of degree of private delivery for the three categories of production model. The most appropriate method for undertaking an analysis of this kind in which the different values of a categorical variable (i.e. the production model) cannot be meaningfully ordered is the multinomial logistic regression. To run this model, we use the categorical variable with the three values directly reported in the survey, where 0 denotes public production, 1 denotes PPPs, and 2 denotes contracting out. Our model is analogous to a logistic regression where the response variable's probability distribution is multinomial as opposed to binomial. The J-1 multinomial logistic equations compare categories 1, 2 . . . J-1 to category J (Public Production in our case). The covariates used in the analysis are the same as those employed in the logistic regression (i.e. Politician, Population and Population^2). Table 5 shows our main results regarding differences in opinion between politicians and technical staff as to what they consider to be the optimal production model. The multinomial logistic regression analysis, however, has to be interpreted by taking the omitted base cases into consideration. First, we interpret the coefficients associated with contracting out and PPPs with respect to public production (base case omitted). Second, because we include the politicians' opinions and omit those of the technical staff (to avoid perfect collinearity), the coefficients have to be interpreted as a mean difference between both groups, that is, a comparison of their views on production models with private participation and public production. Thus, a statistically significant coefficient means that politicians have a different opinion regarding a specific service to that of technical staff concerning the relative optimality of private models vs. public production. Here, our results are presented as relative risk ratios as opposed to coefficients to facilitate interpretation. A relative risk ratio of 1 means there is no difference in the probability of choosing a particular model, a ratio lower than 1 means the probability is less, while a ratio higher than 1 means the probability is greater.
Our results show that for water, sewerage, urban passenger transport, fire-fighting and prevention and environmental actions (as well as libraries) there are significant differences in the opinions expressed by politicians and technical staff. For these specific services, politicians seem to consider public production more optimal than other forms of production involving private participation than do technicians. For most of the services, with the exception of water distribution, statistically significant differences are found in relation to opinions concerning contracting out with respect to public production. For instance, the probability of politicians choosing contracting out as the optimal model for sewerage is one third that of technical staff making the same choice. As regards PPPs, only water distribution appears to offer a significantly different prediction for this form of delivery and, so, we opted to only show the results for contracting out. Note, however, that for several services there are no statistical differences in the opinions of politicians and technical staff.
Our analysis can also usefully consider the ideology of the politicians that responded to the survey. 1 Using a scale between 1 (extreme left) and 7 (extreme right)with 4 being the "center"survey respondents indicated the ideological position of the council. We identified as leftist councils those declaring values below 4, while values from 4 to 7 are considered to be center or rightist. The average ideology of municipalities is 3.2, with a standard deviation of 1.0 and minimum and maximum values of 1 and 5, respectively. Interestingly, politicians report slightly lower values on the ideology scale (2.8), indicating more leftist municipalities than the ideology reported in surveys completed by technical staff (3.6).
Thus, our logistic regression takes the following form: where Left_Politician is a binary variable that takes a value of 1 if the respondent belongs to a leftist political party and 0 otherwise. Right_Politician takes a value of 1 if the respondent belongs to a rightist or center political party. Results should be interpreted in comparison to the omitted category, which in our case is technical staff. The rest of the regressors are the number of inhabitants of the municipality (Population) and its square (Population^2). Table 6 shows the results of the logistic regression when PPPs are excluded from the sample. We observe that our previous logistic regression results concerning the lower probability of politicians choosing contracting out as the optimal model for water distribution, sewerage and fire services are determined by leftist politicians. The results present interesting differences according to ideology. Indeed, some of the differences in opinion found in the previous analysis (Table 5) seem, in most services, to be led by respondents with a left-wing ideology. Center-and-right politicians present statistically significant differences in opinion with respect to those technical staff (i.e. between public production and contracting out) for just fire-fighting and prevention, libraries, environmental actions and public transportation. For all other services, their views are the same as those of technical staff.
All the analyses performed are consistent with our first proposition, namely, that politicians have a stronger preference for public production than technical staff. For many services we find this difference to be statistically significant, and we do not find the opposite result (politicians expressing a preference for private participation) in any case when comparing public production and contracting out. Interestingly, we find that the statistically significant preference for public production (with respect to technical staff) is more frequent in the case of leftist politicians than in that of their center and rightist counterparts. Notes: Standard errors in parentheses are robust to heteroskedasticity and clustered by region (NUTS 2). Info on population in the last columns belongs to estimations for services in the central column "Services". *,**,*** statistical significance at 10%, 5% and 1%, respectively.

Reforms in service delivery choices
By cross referencing responses regarding the status quo of the production model and opinions about the optimal delivery model, we can obtain the reform preferences of both politicians and technical staff. We understand a reform preference as being any difference between the current model and the optimal model, regardless of whether more or less private sector involvement is preferred. Thus, we can test how inclined politicians are to introduce reform with respect to technical staff for all 17 services. To do so, we run a logistic regression that takes the following form: where now the probability of declaring a preference for reform is regressed against three covariates. The first of these is a binary variable, denoted by Politician, that takes a value of 1 if the respondent is a politician and 0 otherwise. The rest of the regressors are the number of inhabitants of the municipality (Population) and its square (Population^2) The second column in Table 7 shows our results. For four of the services, we find statistically significant differences between politicians and technicians. Thus, politicians seem less inclined than technical staff to undertake reforms in water distribution, sewerage, cemeteries and environmental actions. There is no single service for which politicians are more inclined to initiate reform.
Next, we obtain more detailed outcomes by accounting for the ideology of the politicians. Columns 5 and 7 in Table 7 indicate whether politicians are more or less inclined to undertake reforms with respect to technical staff. We find that leftist politicians are relatively less inclined to undertake reforms in the same four services listed above, but to these we can now add road pavement and signaling and access roads to population enclaves. As for center and rightist politicians, we find their stance to be closely aligned with that of technical staff in relation to all services, with the exception of libraries, civil protection and waste treatment. For these services, politicians are also more reluctant to initiate reform. All in all, the results obtained in this section are consistent with our second proposition: Politicians have a weaker preference than technical staff for reforming the organization of service delivery. This outcome could indicate, as is to be expected, that politicians' preferences do not differ greatly from the policies they actually apply, given that elections allow governments to be changed frequently if the citizens are unhappy with their service delivery models. In this regard, our results may reflect that politicians are more successful to realize their preferences than technical staff. In contrast, technical staff cannot be replaced so easily and, therefore, there is a greater mismatch between the current model and the preferred model of delivery among this latter group.

Going private vs. going public
So far, we have found that respondents seem to prefer a production model reform that best fits with the model they consider to be optimal. However, we have not yet examined the direction that this reform should take, i.e. whether it should seek more or less private participation. Figures A1 and A2 (see Appendix) show the percentage number of respondents that declared a preference for a production model reform that would imply, respectively, a lower and a higher presence of private participation for the 17 services.
The preference for remunicipalizations seems to be highly concentrated in services such as urban passenger transport, waste collection and treatment, and water and sewerage, which are the main services that would be remunicipalized if respondents could implement the production model they believe to be optimal. In contrast, libraries and environmental actions are the services mentioned least by respondents in this regard.
On the other hand, urban passenger transport, waste collection and water distribution are the main services for which respondents indicated a preference for a higher participation of the private sector, either in the form of a PPP or contracting out.
To understand better the preferred direction of reforms, we created two variables -Rem i and Priv i (specific to each of the i services)to estimate the differences in preferences of politicians and technical staff. Rem i takes a value of 1 if the respondent considers public production to be the optimal model although the current model employed in their municipality is either contracting out or a PPP, and 0 otherwise. This variable, therefore, captures the preference for full remunicipalization: A reform from private models towards in-house production. Priv i takes a value of 1 if the respondent considers private production (PPP or contracting out) to be the optimal model although the current model employed in their municipality is that of public production.
Our results for the two logistic regression models (Priv i and Rem i ) present differences between politicians and technical staff (see Table 8). Regardless of whether we examine the preference for privatization or for remunicipalization, we do not find any statistically significant differences for most services. However, where structural differences occur, they are always indicative of the weaker preference of politicians for reform. Given the size of our subsamples, politician status predicts failure perfectly in some services, which means they opt for "no reform" in all their responses.
The proposition that politicians have a weaker preference for reform finds some support in the case of both privatizations and remunicipalization. However, here, our results are less robust because of certain sample limitations, which result in excessively small subsamples as we fragmented our data. We obtain a weak indication that technical staff may exert some pressure for the remunicipalization of various serviceswith respect to politicians, but the same holds for reforms in favor of privatization. This is consistent with the view that the pressure for remunicipalization (as well as that for privatization) is more pragmatic than ideological.

Conclusion
Robust empirical evidence on the drivers of remunicipalization remains scarce. Preliminary findings suggest that the role played by ideology is minor. In most services we observe that more pragmatic reasons account for decisions to remunicipalize. However, very little is known about the potential differences between politicians and technical staff regarding choices of delivery form. The same is true of their respective propensities to engage in reforms, where remunicipalization is just one option.
All the analyses performed suggest that politicians as opposed to technical staff have a greater preference for public production, though this preference (compared to that of technical staff) is more frequent in the case of leftist than it is of center/rightist politicians. This is consistent with views that technical staff are more capable of managing contracts, which is a likely determinant of their preference for contractual forms. Similarly, it is consistent with the view that public production places more control in the hands of politicians, while technical staff may have a more relevant role to play in supervising and monitoring contracts.
Our results also suggest that politicians are less likely than technical staff to want to reform the organization of service delivery. This preference holds both for leftist and center/rightist politicians. When we distinguish between pro-privatization and pro- remunicipalization reforms, we still observe a lower propensity among politicians for reform, in both cases. However, we need to treat these last results with caution, because the more fragmented the sample became, the less robust our results were. Indeed, the preceding analysis suffers from various limitations. The main shortcoming is the limited number of available observations. On the one hand, the small number of completed surveys places stricter requirements on statistical significance throughout our results. This in turn prevents us from breaking down the data to undertake further analyses. Clearly, expanding the sample would help us obtain more robust and more refined results. On the other hand, given our sample size and the information this affords, potential endogeneity issues exist, as there is no quasi-random sorting mechanism into the group of politicians and bureaucrats. However, it is impossible to check for any potential biases, such as selection bias or endogeneity, which advises to interpret our results with caution. In this regard, further analyses need to take this point into account when designing the empirical setting for future surveys and experiments.
Our results open up interesting avenues for further research. While incentives and views on public and private delivery choices are likely to play a role in the difference in preferences between politicians and technical staff, it might well be that other management related issues such as contract management difficulties, available competition or citizen interests also play a role in explaining these differences. Note 1. Our survey measures the ideology of the local council, according to the survey respondent opinion. Note we do not have information about the ideology of the technical staff. However, in the case of politicians, who are either elected themselves or appointed by elected politicians to take political action within the city council, we believe it reasonable to assume they share the political ideology of the city council.  Figure A1. Percentage of respondents declaring a preference for remunicipalization by service (optimal model vs. current model).  Figure A2. Percentage of respondents declaring a preference for privatization by service (optimal model vs. current model).