The influence of organisational and staff characteristics on local governments’ efficiency: the case of Mexican municipalities

ABSTRACT

Administrative capacity depends on organisational and human factors that can enhance or diminish the performance of local governments. To analyse the effect of these factors, efficiency scores of municipal governments were calculated and regressed against organisational and staff characteristics, using double bootstrap data envelopment analysis, on a sample of 1470 Mexican municipalities. Of the variables related to staff characteristics, the proportion of employees with college education, and remunerations, have a positive and statistically significant effect on efficiency. Of the organisational characteristics of local governments, the number of computers per employee was the main variable explaining the efficiency of municipal governments.

KEYWORDS:

• Double bootstrap data envelopment analysis
• local governments performance
• administrative capacity
• municipal services
• human resources

Introduction

During the decades of 1980 and 1990, reforms that led to decentralisation of the public sector were carried out in many countries (Rondinelli 2006), under the premise that local governments were more accountable to their citizens and had a better understanding of their needs, so they were more likely to provide services such as street lighting, drinking water, and public safety, more efficiently than other levels of government (Morán-Figueroa and Ayvar-Campos 2020).

Such reforms were carried out in Mexico as a result of a process of decentralisation from the federal to local governments. Article 115 of the Mexican constitution was reformed in 1983 to make municipalities responsible for a series of services such as waste collection, street lighting, public safety, and drinking water, among others. The authority to collect property taxes was also transferred to municipal governments (Cabrero-Mendoza 2013). These reforms were supposed to increase the control of municipalities on their own affairs, thus empowering them to provide better services to their citizens. However, the performance of many Mexican municipalities has been disappointing, despite the increasing amount of resources they have received from the federal government (Cabrero-Mendoza 2013). According to Cabrero-Mendoza (2004), the performance of Mexican municipal governments is constrained by their low levels of administrative capacity, which can be observed in the poor professionalisation of their staff, the lack of staff development plans and the low property tax revenue collection, which are in turn a result of administrative lags, regulatory neglect, obsolete systems, and a generalised deficiency of financial, human and technical resources. Díaz-Aldret (2015) points out that the poor performance of municipalities is also the result of deficient planning capacities. Thus, these authors opine that the poor performance of municipal governments can be explained by their low administrative capacity both at the organisational and the human resources level.

The objective of this study is to measure the efficiency of Mexican municipal governments and to evaluate the effect that some factors of administrative capacity have on their efficiency. To analyse the effect of these variables, the efficiency scores of municipal governments were calculated and regressed against a set of indicators of the organisational and staff characteristics of municipal governments, by applying double bootstrap data envelopment analysis. This study aims at contributing to the literature on the efficiency of municipal governments in two ways: first, to help filling the gap that exists between the number of studies based in Latin American and developed countries; and second, to address the effect of administrative capacity, especially the human factor component, on the efficiency of local governments.

Although municipal governments efficiency has been studied for municipalities of some Mexican states (Morán-Figueroa and Ayvar-Campos 2020; Cázares, Ortiz Medina, and Cernas Ortiz 2015), to the best knowledge of the author, there is no study which covers all Mexican states. Most studies on local government efficiency have focused in developed countries, while Latin American countries have been less studied, and although there is a vast literature on the study of factors affecting local governments efficiency (Narbón-Perpiñá and De Witte 2018b; Milán-García, Rueda-López, and De Pablo-Valenciano 2022), human factor variables have been mostly neglected. In Mexico, decentralisation led to different degrees of institutional capacity development, which provides a valuable oportunity to investigate the causes of such divergence. Decentralisation processes yielded similar results in other countries (Grindle 2009) so studying the Mexican case can shed light on the causes of the performance divergence observed in other nations.

This study analyses a sample of 1470 municipalities, using data from the Mexican Municipal Government Census municipalities, which contains information of staff and organisational characteristics, to assess the effect of these variables on efficiency. This paper is organised as follows: in the next section, a theoretical framework for administrative capacity is presented, followed by a literature review of the state of the art of studies on local governments efficiency and the factors of administrative capacity that affect their performance; following, the methodology and the sources of data, and variables used are explained; then, results from the double bootstrap analysis are presented, followed by a discussion on the variables that affect efficiency; finally, the conclusions of the study are presented.

Administrative capacity of local governments

Grindle and Hilderbrand (1995) define capacity as ‘the ability to perform appropriate tasks effectively, efficiently and sustainably’, which depends on a series of internal and external factors that can enhance or diminish the impact or success of the local government’s administration. Internal factors can be classified into organisational and human factors. The organisational factors of local governments are the structures, resources, processes, and management methodologies that influence how these organisations set their goals, design work procedures, define hierarchies and relations, and implement the incentives framework. Organisational factors, which include resources such as financial, capital and technological (Donahue, Selden, and Ingraham 2000) enhance or diminish performance because they influence the functioning of the organisation and the behaviour of its human resources (Grindle and Hilderbrand 1995), which is its most important resource, since governments are labour-intensive (Agustin and Totaro 2020) and much of their performance relies on the capacity of their employees.

Therefore, local governments must create incentives to attract talented staff by means of paying competitive salaries, which are comparable to those of the private sector (Grindle 2009). Donahue, Selden, and Ingraham (2000) argue that having flexible schemes for hiring, retaining, and firing employees is more likely to have a positive impact on performance compared to rigid bureaucratic systems, so local governments must be able to hire good workers, but also terminate contracts of those who are not fit for the position. The performance of public servants can be enhanced by means of training, the adoption of technological resources to increase productivity, and the institutionalisation of rules and procedures to establish the goals of the organisation and the means necessary to achieve these objectives (Grindle and Hilderbrand 1995).

Organisational and human factors do not operate in a vacuum but are influenced by the action environment, i.e., the economic, political, and social conditions where local governments perform their tasks. The performance of local governments can be affected by the economic development and structure of the municipalities, since in more developed localities there are usually more advanced technological and human resources available to the administration of the municipalities, which allows them to obtain better outcomes. The political legitimacy of the local government and the institutional context also affects the performance, and it includes the rules and procedures that constrain local government operations as well as the access to financial resources (Grindle and Hilderbrand 1995).

Literature review

There are studies testing the influence of variables related to administrative capacity on local governments efficiency. The ability to collect taxes is usually considered as an indicator of capacity (Besley and Persson 2009) and most of the papers testing its influence have found a positive correlation (Leif and Sørensen 2015; Doumpos and Cohen 2014; El Mehdi and Hafner 2014; Sørensen 2014). The adoption of information technology can increase the performance of municipal governments, which has been corroborated by recent studies (De Sousa, Da Conceição, and Stošić 2005; Sung 2007; Seol et al. 2008).

Some studies of local government efficiency have included the effect of human factors in their analysis, focusing on the salaries of majors and council members. According to Benito et al. (2014) paying higher salaries to politicians does not improve the financial management of municipalities, however, Benito, Guillamón, and Martínez-Córdoba (2020) found that the salary levels of majors have a positive correlation to efficiency, while city councillors’ salaries are negatively correlated. Loikkanen, Susiluoto, and Funk (2008) analysed the effect of the characteristics of city managers, and found that the years of education of these managers had a positive impact on efficiency, while other characteristics such as age and sex did not have any significant effect.

Thus, studies focusing on human factors have been mainly based on the study of politicians and city managers, but the characteristics of the administrative staff have been less studied. To the best knowledge of the author, only one study (Loikkanen and Susiluoto 2005) has investigated the effect of the characteristics of the staff on the efficiency of local governments. The study found that the age structure of municipal employees and the educational level of citizens in the municipality, as a proxy for the education of municipality employees, affect efficiency. However, they did not use a direct measurement of the municipality workers education level, thus, the characteristics of the municipal governments’ workforce have been barely studied, leaving an important component of administrative capacity out of the research on factors affecting efficiency.

Although the organisational and staff characteristics have been seldom studied, there is an important number of studies that analyse the effect of the action environment on the efficiency of local governments. Some studies have analysed the effect of population size, and have found that municipalities with a larger population are more efficient, which suggest the existence of economies of scale (Balaguer-Coll, Prior, and Tortosa-Ausina 2007; Nakazawa 2013, 2014; Bernardino, Bastida, and Garcia 2010; Bruns and Himmler 2011; Boetti, Piacenza, and Turati 2012; Asatryan and De Witte 2015; Perez-Lopez, Prior, and Zafra-Gómez 2015; Giménez and Prior 2007; De Sousa, Da Conceição, and Stošić 2005; Fellows, Dollery, and Tran 2022), while other studies find that smaller municipalities are comparatively more efficient (Loikkanen and Susiluoto 2005; Ashworth et al. 2014; Sung 2007; Sørensen 2014; Šťastná and Gregor 2015).

Population structure has also been found to be correlated to local governments’ efficiency, specifically, the age structure and population density (Cruz and Marques 2014; Nakazawa 2014, 2013). A variable of special interest in the literature is economic development and some studies have found a positive correlation of income to efficiency (Boetti, Piacenza, and Turati 2012; Asatryan and De Witte 2015; Afonso and Fernandes 2008; Afonso, Schuknecht, and Tanzi 2010), while other studies have found a negative correlation (De Sousa, Da Conceição, and Stošić 2005; Loikkanen and Susiluoto 2005; Giménez and Prior 2007; Bruns and Himmler 2011; Ashworth et al. 2014; Benito, Solana, and Moreno 2014), so the observed effect of this variable is ambiguous.

The prevalence of some economic activities, specially tourism, has been found to affect the efficiency of municipal public services provision (Benito, Guillamón, and Ríos 2021; Benito, Martínez-Córdoba, and Guillamón 2021; Benito et al. 2021; Benito, Martínez-Córdoba, and Guillamón 2020; Cruz and Marques 2014; Pérez-López, Prior, and Zafra-Gómez 2021). Political factors (Asatryan and De Witte 2015), including the ideology of the party governing the municipality (Benito, Martínez-Córdoba, and Guillamón 2021; Benito et al. 2021, 2021; Benito, Martínez-Córdoba, and Guillamón 2020) and organisational characteristics of municipalities (Narbón-Perpiñá and De Witte 2018b) have also been incorporated in the analysis as part of the action environment, and have been found to have a significant influence on efficiency.

The literature then has investigated the effect of certain features of administrative capacity on efficiency, using some organisational and contextual characteristics as explanatory variables, but has practically left the human factor component out of the analysis. The present study aims at filling this gap by analysing the effect of these factors by using the methodology and data described in the next section.

Methodology and data

The methods used for the evaluation of efficiency can be classified into two types: parametric and non-parametric. Non-parametric methods have been preferred for the evaluation of local governments performance because these approaches do not require the specification of an a priori model (Simões, De Witte, and Marques 2010) and allow for more flexibility in the choice of inputs and outputs, which is adequate for the analysis of municipal governments, which produce multiple non-market outputs (Worthington and Dollery 2000), such as waste collection, drinking water, public safety, etc. (Narbón-Perpiñá and De Witte 2018a). In this study, efficiency is analysed by applying data envelopment analysis (DEA) which is the most widely applied non-parametric approach.

The basic DEA model, or Charnes Cooper Rhoades (CCR) model of DEA (Charnes, Cooper, and Rhodes 1978) assumes constant returns to scale (CRS) to calculate the efficiency of decision-making units (DMU). In this study DMUs are municipal governments, which use a set of inputs to provide services (outputs) to their citizens. It is assumed that these governments look forward to minimising the inputs used to produce services, therefore an input-oriented focus is applied. The CCR model assumes that there are N inputs and M outputs, for each I DMUs. For the ith DMU, these inputs and outputs are represented by y and x. The N×I X matrix represents the inputs data for all DMUs and the M×I Y matrix represents the corresponding output data (Coelli et al. 2005). The CRS input-oriented efficiency can be calculated by solving the linear programme:

(1) $min{ }_{\theta }{,}_{\lambda }\mathit{\theta }$(1)

subject to

$-{\mathit{y}}_i+\mathit{Y}\mathit{\lambda }\ge 0$

${\theta }_{X_i}-\mathit{X\lambda }\ge 0$

$\mathit{\lambda }\ge 0$

where λ is a I × 1 vector of constants, and the value of $\mathit{\theta }$ is the DEA score for each DMU, which satisfies 0 $\le \mathit{\theta }\le 1$, where a value of 1 indicates that the DMU is located on the efficient frontier and is technically efficient. In contrast, DMUs with $\mathit{\theta }<1,$ are comparatively inefficient. To obtain the efficiency that is the result only of management operations, without taking into account the inefficiencies that result from scale effects, it is necessary to calculate DEA under the assumption of variable returns to scale (VRS), using the Banker Charnes Cooper (BCC) model (Banker, Charnes, and Cooper 1984), which is obtained by adding one constraint to the CCR model 1:

(2) $\mathit{I}1\prime \mathit{\lambda }=1$(2)

As previously explained, studies on the efficiency of local governments aim at identifying the factors that explain these efficiency levels. To identify which factors affect efficiency, DEA scores are usually regressed against a set of explanatory variables in a second stage. This regression model can be represented as:

(3) ${\theta }_i={\mathbf{z}}_i\mathit{\beta }+{\epsilon }_i$(3)

where $\mathbf{z}$ is the set of explanatory variables, $\mathit{\beta }$ is the set of regression parameters, and $\mathit{\epsilon }$ is the corresponding error term. Regression parameters can be obtained by applying econometric techniques such as ordinary least squares (OLS) or Tobit regression (see for example Afonso and Fernandes (2008, 2005)). However, the use of these techniques in the second stage provides unreliable estimators because DEA scores are biased and serially correlated (Simar and Wilson 2007; Bǎdin, Daraio, and Simar 2014; Simar and Wilson 2011). To avoid this problem, Simar and Wilson (2007) developed an algorithm that produces unbiased estimators by applying bootstrap techniques. In the first stage, a pseudo frontier is generated from bootstrap estimates to estimate the DEA scores’ bias, which is used in turn to obtain the bias corrected DEA scores. In the second stage, the bias corrected scores obtained from the first stage are regressed against the explanatory variables, using bootstrap to obtain unbiased confidence intervals for the regression parameters. The algorithm, which consists of seven steps, is described next:

(1) The original DEA scores ${\theta }_i$for every DMU are calculated using the basic DEA model

(2) Maximum likelihood is applied to obtain estimates $\hat{\mathit{\beta }}$ of parameters $\mathit{\beta }$, as well as ${\hat{\sigma }}_{\mathit{\epsilon }}$ of ${\sigma }_{\epsilon }$ for ε of truncated regression given by Equation (3)(3) ${\theta }_i={\mathbf{z}}_i\mathit{\beta }+{\epsilon }_i$(3) .

(3) Obtain bootstrap estimates ${\hat{\theta }}_{\mathit{i}}^{\ast }$ by repeating sub steps 3.1 to 3.4 L₁ times:

(3.1) Produce ${\epsilon }_i$ from the $\mathit{N}\left(0,{\hat{\sigma }}_{\mathit{\epsilon }}\right)$ distribution, truncated by the left at $\left(1-z_i\hat{\beta }\right)$

(3.2) Compute ${\theta }_i^{\ast }={\mathbf{z}}_i\hat{\mathit{\beta }}+{\epsilon }_i$

(3.3) Produce pseudo data $x_i^{\ast }=x_i,$and $y_i^{\ast }=y_i\left(\frac{{\theta }_i}{{\theta }_i^{\ast }}\right)$

(3.4) Apply DEA to $x^{\ast }$ and $y^{\ast }$, replacing the original x and y data, to obtain efficiency scores ${\hat{\theta }}_{\mathit{i}}^{\ast }$

(4) The bias corrected estimator ${\widehat{\stackrel{ˆ}{\theta }}}_{\mathit{i}}$ for each DMU is obtained using the bootstrap estimates ${\hat{\theta }}_{\mathit{i}}^{\ast }$ and the original estimates ${\theta }_i$

(5) Apply maximum likelihood estimation to regress unbiased scores ${\widehat{\stackrel{ˆ}{\theta }}}_{\mathit{i}}$ on ${\mathbf{z}}_{\mathbf{i}}$ to obtain estimates $\widehat{\stackrel{ˆ}{\mathit{\beta }}}$ and $\widehat{\stackrel{ˆ}{\mathit{\sigma }}}$

(6) Repeat sub steps 6.1 to 6.3 L₂ times to produce a set bootstrap estimates ${\widehat{\stackrel{ˆ}{\mathit{\beta }}}}^{\mathit{\ast }}$ and ${\widehat{\stackrel{ˆ}{\sigma }}}^{\ast }$(6.1) Produce ${\epsilon }_i$ from the $\mathit{N}\left(0,\widehat{\stackrel{ˆ}{\sigma }}\right)$ distribution, with left truncation at $\left(1-z_i\widehat{\stackrel{ˆ}{\beta }}\right)$(6.2) Compute ${\theta }_i^{\ast \ast }={\mathbf{z}}_i\widehat{\stackrel{ˆ}{\mathit{\beta }}}+{\epsilon }_i$(6.3) Use maximum likelihood estimation ${\theta }_i^{\ast \ast }$ on ${\mathbf{z}}_{\mathbf{j}}$ to obtain estimates of ${\widehat{\stackrel{ˆ}{\mathit{\beta }}}}^{\mathit{\ast }}$ and ${\widehat{\stackrel{ˆ}{\sigma }}}^{\ast }$

(7) Calculate confidence intervals for $\mathit{\beta }$ using bootstrap estimates ${\widehat{\stackrel{ˆ}{\mathit{\beta }}}}^{\mathit{\ast }}$ and ${\widehat{\stackrel{ˆ}{\sigma }}}^{\ast }$ and original estimators $\widehat{\stackrel{ˆ}{\mathit{\beta }}}$ and $\widehat{\stackrel{ˆ}{\sigma }}$

To analyse the effect of organisational and staff characteristics of municipal governments on their efficiency, double bootstrap DEA was applied on a sample of Mexican municipal governments. Calculation of unbiased scores and second-stage regression estimation was carried out using the R statistical software, by applying the rDEA package (Simm and Besstremyannaya 2020), using L₁ = 100 for the first stage and L₂ = 2000 for the second, which are the number of replications suggested by Simar and Wilson (2007). The data and variables used are described in the next subsection.

Data

The main source of data was the 2018 Municipal Governments Census (INEGI 2019). This dataset provides information about the services provided by every municipal government in Mexico during that year. It also contains information on the organisational and staff characteristics, and the resources of each municipal government. From this dataset, a set of inputs and outputs was selected to carry out double bootstrap DEA modelling. Inputs are the total number of employees in the municipal government (STAFF), and the total current expenditure of the municipal government minus staff expenditures (EXP). Outputs consist of three of the most important services provided by Mexican municipalities: public lighting, waste collection, and police services. The indicator used to measure public lighting output was the total number of lighting points in use (LIGHT), as it has been used in recent studies (Perez-Lopez, Prior, and Zafra-Gómez 2015; Doumpos and Cohen 2014; Narbón-Perpiñá et al. 2020). Waste collection services were measured as the quantity of collected waste (WASTE), as in Giménez and Prior (2007) and lo Storto (2016). Finally, the number of police interventions (POLICE) was used to measure the output of policing services, as used in Bernardino, Bastida, and Garcia (2010) and Morán-Figueroa and Ayvar-Campos (2020).

It must be noticed that not all the quantities of services provided by local governments are available in official datasets, thus limiting the scope of the modelling of the total output of local governments. According to (Perez-Lopez, Prior, and Zafra-Gómez 2015), population can be used as a proxy of services which are not listed in the data, although the inclusion of this proxy introduces the implicit assumption that all municipalities provide the same per capita quantity of these unlisted services. To account for these unobserved services, a second model was estimated including population (POP) as an output, which also provides an opportunity to test the robustness of the effect of explanatory variables z.

To evaluate the influence of administrative capacity, a group of organisational and staff characteristics of municipal governments was selected. As previously discussed, the capacity to collect taxes is one of the most used indicators of administrative capacity. In this study, the proportion of the total expenditure that comes from taxes collected by municipal governments (TAX) is used as an indicator of this variable. Another variable used as an indicator of administrative capacity, is the informatization or intensity of computer use in the workplace. In this study, the number of computers per worker (COMP) was used as an indicator of this variable.

One variable that is in the dataset but has not been analysed in the revised literature is the impact of employees’ training on efficiency. In this study, TRAIN is a dummy variable that indicates that the municipality provided training to their property registry employees during the year 2018. Another variable indicating capacity is PROG, which indicates the number of sub-programmes that are included in the Municipal Development Programme, which is the instrument used by Mexican municipalities to plan their activities as a means to comply with their objectives (Mejía-Lira 1992).

To analyse the human factor component of administrative capacity, the education level of employees was included among the explanatory variables, which is measured as the proportion of employees with college level education (COLLEGE) as an indicator of the capacity of the municipal government to attract qualified workers. It would be expected that higher education levels are associated with higher efficiency. Average remuneration of employees (REMUN) was also included. It has been argued that low salaries can reduce the capacity of governments (Grindle and Hilderbrand 1995); therefore, it would be expected that higher remunerations incentivise workers and thus induce higher efficiency. TEMP indicates the proportion of temporary workers in the municipality which can be interpreted as an indicator of hiring flexibility, a factor which is thought to affect local government performance (Donahue, Selden, and Ingraham 2000). Finally, two characteristics which have been included in previous analyses of human factors affecting local government performance, the age and sex structure of employees were also included: AGE_34_49 and AGE_50 indicate the percentage of employees aged 34-to 49 and employees aged over 50, respectively. Variable WOMEN indicates the percentage of employees of the municipal government who are women.

The contextual variables can be subdivided into three groups: the demographic, the socio-economic, and the political variables. In the first category, the variables considered were population (POP), the proportion of the population aged over 60 (P60) and population density (DENS), since these variables have been found to influence efficiency in previous studies (Cruz and Marques 2014). The expected influence of these variables is undefined because published results suggest both positive and negative correlation of these variables to efficiency. To address the effect of socio-economic variables, the Marginalisation Index of 2015 (MARG) was included, which was constructed by the National Population Council from several indicators of education income, and housing conditions. A larger positive index score indicates a high degree of marginalisation, while negative scores indicate lower levels. As part of the economic environment, an indicator of touristic activity (TOUR) was included.

To address the influence of the political context, the party leading the ruling coalition in each municipality was indicated by dummy variables PRI, PAN, GREEN, and LEFT, the latter formed by grouping together the three major left leaning parties. The sample contains 153 municipalities from the state of Oaxaca governed under the Indigenous Normative System (dummy variable INS) where authorities are elected by means of traditional methods. All other smaller and local political parties, as well as independent candidates were grouped together and were designated as the reference category, so no dummy variable was assigned to this group. Dummy variable ALIGN indicates partisan alignment between the municipality and the state, since it would be expected that these municipal governments would find a more favourable environment. State dummies were included in the model to control the effect of state political and institutional characteristics.

EXP, TAX and REMUN were obtained from official municipal finances data (INEGI 2021). Inputs, outputs, and staff and organisational characteristics were obtained from the 2018 Municipal Governments Census (INEGI 2019). POP, DENS, and P60 were obtained from interpolation of the 2010 and 2020 population census data (INEGI 2020; 2010). TOUR was obtained from the economic census (INEGI 2014). MARG and information on the political parties governing each municipality were obtained from the National System of Municipal Information (INAFED 2022). A total of 1470 municipalities were included in the sample, which were the municipalities that had available data for all variables studied. This sample represents about 60% of total municipalities, which were home to 90.7 million people, or 71.8% of the Mexican population in 2018. Table 1 shows the descriptive statistics and definitions of the variables analysed in the study.

Table 1. Descriptive statistics of inputs, outputs and explanatory variables.

Variable	Description	Min	Median	Mean	Max
Inputs
STAFF	Total number of employees	4	198.0	518.3	12,041
EXP	Total expenditure minus staff expenditures (millions of pesos)	1.703	57.20	174.0	5115
Outputs
LIGHT	Number of lighting points in use	22	1374	4062	108895
WASTE	Quantity of waste collected (tons/day)	0.023	9	52.55	1769
POLICE	Number of police interventions	0	159.5	2727	509190
POP	Population	254	16568	61737	1849955
Explanatory variables
TAX	Tax revenue as a percentage of total budget (own revenue)	0.000006	0.0214	0.0446	0.56
PROG	Number of programs in the municipal development plan	0	5	6.381	20
TRAIN	Dummy variable that indicates that property registry employees received training	0	0	0.1714	1
COMP	Number of computers per staff	0	0.2	0.22	2.38
TEMP	Proportion of temporary employees	0	0	0.1401	1
COLLEGE	Proportion of employees with a college degree	0	0.15385	0.16431	0.99340
REMUN	Average expenditure per municipal employee	1619	123714	133885	1769521
AGE_35_49	Percentage of employees aged 35 to 49	0	0.3998	0.4073	1
AGE_50	Percentage of employees aged 50 or more	0	0.2240	0.2172	0.7143
WOMEN	Proportion of employees who are women	0	0.3314	0.3292	0.8750
MARG	Marginalisation Index	−2.2190	−0.3560	−0.2752	4.8450
P60	Proportion of the population aged 60 or more	0.01855	0.13014	0.13758	0.33740
DENS	Population density (inhabitants per km^2)	0.0011	0.5456	2.9629	172.318
TOUR	Tourism (hotels per 10 thousand inhabitants)	0	0	2.034	45.386
ALIGN	Dummy variable that indicates that the party governing the municipality was aligned with the party governing the state	0	0	0.3612	1
PRI	Dummy variable that indicates that the municipality was governed by PRI	0	0	0.3653	1
PAN	Dummy variable that indicates that the municipality was governed by PAN	0	0	0.2238	1
LEFT	Dummy variable that indicates that the municipality was governed by a major national left leaning party	0	0	0.2068	1
GREEN	Dummy variable that indicates that the municipality was governed by the Green party	0	0	0.0408	1
INS	Dummy variable that indicates that the municipality was under the Indigenous Normative System	0	0	0.1040	1

Based on these data, 2 double bootstrap DEA models were estimated. Model 1 includes three outputs (LIGHT, WASTE, and POLICE), and Model 2 includes four outputs (LIGHT, WASTE, POLICE, and POP). These models were used to test the following hypotheses:

• The efficiency of municipalities is positively correlated to the education of employees (COLLEGE) and the remunerations (REMUN) they are paid.

• Flexibility in the hiring of workers (TEMP) correlates positively to efficiency.

• Municipalities that train their staff (TRAIN), have higher levels of planning (PROG), have a higher level of informatization (COMP) and have higher levels of tax collection (TAX) are more efficient.

• Higher levels of marginalisation (MARG) correlate negatively to efficiency.

Results

Results from the first-stage bootstrap DEA indicate that the average VRS efficiency was 0.34 in Model 1, and 0.43 in Model 2, which implies that municipalities could reduce their inputs somewhere between 57% and 66% on average, to obtain the same quantity of output. Results from the second-stage bootstrap regression are provided in Table 2.

Table 2. VRS regression results.

	Model 1			Model 2
Variable	Regression Coefficient	90% CI LB	90% CI UB	Regression Coefficient	90% CI LB	90% CI UB
TAX	−19.793**	−28.700	−11.817	0.424	−1.421	2.386
PROG	0.011	−0.039	0.060	0.005	−0.007	0.019
TRAIN	−1.103**	−1.918	−0.349	0.086	−0.117	0.322
COMP	−4.358**	−6.901	−2.032	−0.724*	−1.361	−0.076
TEMP	−1.173*	−2.144	−0.175	−0.130	−0.369	0.191
COLLEGE	−2.892*	−5.823	−0.169	−1.032**	−1.897	−0.173
log(REMUN)	−1.579**	−2.319	−0.970	−0.743**	−0.959	−0.563
AGE_35_49	−3.196**	−5.719	−1.083	−0.051	−0.716	0.609
AGE_50	−2.982**	−5.607	−0.604	−0.956**	−1.736	−0.232
WOMEN	0.834	−1.893	3.712	−0.140	−1.307	0.439
log(POP)	−0.162	−0.530	0.174	−0.831**	−0.95	−0.73
P60	1.664	−6.282	10.266	5.291**	3.04	7.59
DENS	0.007	−0.025	0.052	−0.006	−0.014	0.013
TOUR	−0.019	−0.078	0.049	0.027**	0.013	0.044
MARG	3.535**	3.178	4.081	0.536**	0.408	0.67
ALIGN	0.256	−0.379	0.921	−0.141	−0.273	0.095
PRI	−1.428**	−2.687	−0.309	−0.640**	−1.010	−0.302
PAN	−1.420**	−2.713	−0.262	−0.464**	−0.835	−0.114
LEFT	−1.114*	−2.338	−0.021	−0.566**	−0.916	−0.223
GREEN	−0.446	−1.985	1.177	−0.739**	−1.221	−0.274
INS	−7.591**	−9.795	−5.986	−2.478**	−3.095	−1.952

*and ** indicate statistical significance at the 0.1 and 0.05 level, respectively. State dummies are not shown.

The rDEA package calculates the second stage truncated regression model using the reciprocal of the DEA score as the dependent variable. Therefore, a positive sign of the regression coefficient for each variable should be interpreted as a negative correlation between the DEA efficiency score and the explanatory variable. Conversely, a negative sign should be interpreted as a positive correlation. Table 2 provides the 90% confidence interval for the regression coefficient. When both the lower bound (LB) and upper bound (UB) of the confidence interval have the same sign, it means that the variable is significant in the model at the 0.1 level. The same procedure is applied to evaluate significance at the 0.05 level, but the 95% confidence interval was not included for space reasons.

Of the variables associated to the organisational characteristics of the municipal governments, TAX, COMP and TRAIN are statistically significant at the 5% level in Model 1. TAX has a positive correlation to efficiency, which indicates that the capacity to collect taxes is positively correlated to efficiency. COMP has also a positive correlation, which indicates that the level of informatization might have a positive effect on the efficiency of municipal governments. TRAIN has a positive correlation to efficiency, which indicates that training the staff has a favourable effect on efficiency. The other variable that describes the organisational characteristics in the model, PROG has no statistical significance in the model. In Model 2, only COMP is significant at the 5% level and has a positive correlation as in Model 1.

Of the variables related to human factors, REMUN is statistically significant at the 5% level in Model 1 and has a positive correlation to efficiency, which implies that a higher remuneration of the staff incentivises efficiency. COLLEGE and TEMP are statistically significant at the 10% level. COLLEGE has a positive correlation to efficiency, which means that a larger proportion of employees having a college degree is associated to a greater efficiency in municipal governments. TEMP has a positive correlation to efficiency, indicating that a larger proportion of temporary workers, a signal of hiring flexibility, is associated with higher efficiency levels. The age structure of employees is also significant in the model, AGE_35_49 and AGE_5 have a positive correlation to efficiency, which indicates that employees older than 35 are more productive than those of younger cohorts, which suggests that more experienced workers contribute to increasing the efficiency of local governments. Model 2 provides similar results, except for AGE_35_49 and TEMP, which are not statistically significant.

Of the contextual demographic and economic variables, only MARG is significant at the 5% level in Model 1, and has a negative correlation to efficiency, which indicates that more marginalised municipalities tend to have less efficient local governments. This coincides with the statement that economic development in general correlates positively to government performance. The other contextual variables, POP, P60, DENS, and TOUR are not statistically significant in Model 1. However, in Model 2, in addition to MARG, some of these variables are also statistically significant at the 5% level: POP is positively correlated to efficiency; and P60 and TOUR are negatively correlated.

Of the political variables, the political party dummy variables PRI and PAN are significant at the 5% level, and have a positive correlation to efficiency, while LEFT is statistically significant at the 10% level and is also positively correlated to efficiency. These results indicate that municipalities governed by the major national parties are more efficient compared to smaller and local parties. INS is statistically significant at the 5% level, suggesting that municipalities under the Indigenous Normative System are more efficient than similar municipalities governed by small and local parties. ALIGN is not statistically significant, so the party alignment of the municipal to state government does not translate into efficiency gains of municipal governments. Model 2 provides similar results, with GREEN being also statistically significant at the 5% level.

Discussion

Three of the variables related to organisational characteristics are significant in Model 1. The capacity to collect taxes (TAX) is positively correlated to efficiency, which coincides with most previous studies (Doumpos and Cohen 2014; El Mehdi and Hafner 2014; Leif and Sørensen 2015), since it is not only an indicator of administrative capacity, but it is also more likely to make local governments accountable to taxpayers (Boetti, Piacenza, and Turati 2012). TRAIN is positively correlated to efficiency which indicates that training helps to increase the productivity of employees in local governments (Grindle 2009).

COMP is significant in both models, which coincides with previous findings by De Sousa, Da Conceição, and Stošić (2005), Sung (2007) and Seol et al. (2008), providing new evidence of the positive impact of informatization on productivity. Luna-Reyes et al. (2020) found that the use of information technology may be linked to increased innovation and knowledge processing, which would be in turn related to increased performance.

Of the variables related to human factors, COLLEGE is significant in both models, which implies that hiring more educated employees might help to increase efficiency in municipal governments. It had been previously found that the education level of city managers had a positive correlation to efficiency (Loikkanen, Susiluoto, and Funk 2008) as well as the education of citizens (Loikkanen and Susiluoto 2005), but the effect of the education of employees had not been previously tested at the local governments level. Afonso, Schuknecht, and Tanzi (2010) found that the competence of civil servants is positively correlated to efficiency at the national level, so the result of this study provides evidence that there might be a positive effect of this variable at the local government level too. REMUN was statistically significant in both models as well, as it was expected from the incentives provided by higher salaries (Grindle and Hilderbrand 1995). The proportion of temporary workers (TEMP) is statistically significant in Model 1, which suggests that flexibility in hiring could help improve the performance of local governments (Donahue, Selden, and Ingraham 2000).

Cabrero-Mendoza (2004) pointed out that larger Mexican municipalities tend to outperform smaller ones. However, results from Model 1 show that after controlling for other variables, there is no effect of the population size of municipalities on efficiency. On the other hand, results from both models suggest that more developed municipalities are more efficient than more marginalised ones in this study, which coincides with results from previous studies (Boetti, Piacenza, and Turati 2012; Asatryan and De Witte 2015; Afonso and Fernandes 2008; Afonso, Schuknecht, and Tanzi 2010) that support the conjecture that higher economic development creates an environment that is more favourable to better government performance (Grindle and Hilderbrand 1995). In Model 2, POP is statistically significant, however, since this model includes the municipal population in the output set, this variable is only controlling the effect that population has on the total output, which is expected to increase with a larger population. In Model 2, P60 and TOUR are statistically significant, suggesting that older citizens and tourists tend to put a higher pressure on the demand for services, which leads to lower efficiency scores as observed in (Cruz and Marques 2014).

The political variables in both models show that that municipalities governed by major national parties and those under the indigenous normative system (INS) tend to be more efficient than those governed by smaller and local parties. The INS regression coefficient is even greater than those of other national party dummy variables. This seems counterintuitive since these municipalities are usually less developed than other similar municipalities. A plausible explanation is that these communities have the institution of tequio, which is a form of taxation in kind, where people are required to provide their labour for public works and services (Cohen 1999). Thus, in these municipalities, the input provided by the variable STAFF is being underestimated in the calculation of efficiency, because the labour provided by all the inhabitants of the municipality is not being accounted for.

Conclusions

The efficiency of Mexican municipal governments was evaluated and measured against a set of explanatory variables related to administrative capacity to identify which factors affect their performance using two double bootstrap DEA models, with different specification of outputs (3 and 4 outputs). An organisational characteristic that explains efficiency is the level of informatization of the workplace (COMP), which is statistically significant in both models. Results also indicate that some characteristics of the municipal governments’ workers such as the educational level of employees (COLLEGE) and the salaries they are paid (REMUN), have a positive influence on efficiency. These variables are statistically significant in both models. These results suggest that increasing administrative capacity through hiring better educated employees, attracting them with competitive salaries, and providing them information technology tools, has the potential to improve the performance of municipal governments.

Hiring flexibility (TEMP) and training of the workforce (TRAIN), are human resources variables that are statistically significant and positively associated to efficiency, but only in the 3-output model. The capacity to collect taxes (TAX) was also found to have a positive correlation to efficiency in this model, as in most studies analysing this variable. This model does not consider unlisted outputs, but focuses on the main services provided by Mexican municipalities (waste collection, street lighting and public safety) so the effect of these variables can be considered to apply to this specific output set.

To the best knowledge of the author, this study is the first to provide evidence that the human resources variables included in this study influence efficiency. It is recommended that future studies include these and other variables related to the characteristics of employees of local governments, because despite the importance that human resources have on the functioning of these organisations, their effect on their performance has been barely studied.

Disclosure statement

No potential conflict of interest was reported by the author.