Transparency and its drivers. A study of the Colombian experience (2012-2016)

Transparência e seus drivers. Um estudo baseado na experiência colombiana (2012-2016)

La transparencia y sus impulsores. Un estudio desde la experiencia colombiana (2012-2016)

Víctor Mauricio Castañeda-Rodríguez vmcastanedar@unal.edu.co

Universidad Nacional de Colombia, Colombia

Jeimi Maribel Leon-Silva jmleons@eafit.edu.co

Universidad EAFIT, Colombia

In recent decades, there has been a growing interest in studying transparency due to its importance for public administration and international initiatives aimed at improving the access and availability of public information (Gandía et al., 2016). Transparency is considered a good governance tool that allows citizens to participate in public decision-making (Tejedo-Romero & Araujo, 2020), as well as monitoring and evaluating the performance of mayors and other public servants (Albalate, 2013). Transparency also contributes to accountability (Lyrio et al., 2018; Piotrowski & Van Ryzin, 2007) and confidence in governments (Bearfield & Bowman, 2017; Saez-Martin et al., 2017).

Despite the above, studies on the matter are focused principally on Southern European countries (Portugal and Spain) and the United States (e.g., Bearfield & Bowman, 2017; Garrido-Rodríguez et al., 2019; Piotrowski & Van Ryzin, 2007; Tavares & Cruz, 2020). It shows very little research interest in the case of developing regions (e.g., Adiputra et al., 2018; Krah & Mertens, 2020). Therefore, we identified a research gap regarding transparency in developing countries that need to be covered since the results of available papers cannot be extrapolated.

It then becomes necessary to analyze what variables affect the level of transparency of governments in developing countries (Colombia, for this paper) and establish whether the findings already known on the matter are still valid. In this regard, it is essential to point out that transparency practices differ from country to country, depending on the contexts and jurisdictions associated with laws or regulations related to Freedom of Information (FOI) (Tavares & Cruz, 2020). Added to the above, the existing literature suggests conducting studies in countries with different administrative cultures (Grimmelikhuijsen & Welch, 2012; Tejedo-Romero & Araujo, 2020).

Furthermore, transparency has become a relevant issue in the multilateral agendas of several countries (e.g., the 2030 Agenda for Sustainable Development). In this scenario, Gilli (2017) highlights the importance of transparency to achieve Sustainable Development Goal 16, which establishes (in its target 16.6) the need to develop effective, accountable, and transparent institutions at all levels for building effective, accountable and inclusive institutions (United Nations, 2015).

Considering the above, and using the Colombian case, we model the relations between transparency (captured by a transparency index built by the Office of the Attorney General of the State - Procuraduría General de la Nación, in Spanish) and a set of political, socioeconomic, demographic, and fiscal variables. It is carried out through static and dynamic approaches and using a dataset with information from 1,101 Colombian municipalities during five years (2012-2016).

According to the agency and legitimacy theories, transparency drivers could influence citizens’ or public administrators’ behaviors or involve special conditions to enact and enforce transparency initiatives. We find that fiscal autonomy, some population age ranges (i.e., 35-44, 55-64, and 65-74), education, local gross domestic product (GDP) per capita, population, and political participation contributes to the study of the differences in the progress made regarding public information disclosure and accessibility among Colombian territorial entities.

Therefore, this research contributes to the transparency literature in three ways. First, we incorporate a variable called “fiscal autonomy” (ratio of local tax revenues to total municipal revenues), typically overlooked in the literature. Second, we find empirical evidence that transparency in Colombia follows a path-dependence process, which means that the current level of public transparency depends on its past. Finally, we provide empirical evidence on the drivers of subnational transparency in a developing country (i.e., Colombia) from the perspective of citizens (agency theory) and governments (legitimacy theory).

Besides this introduction, this paper is organized into the following five sections. First, we present the conceptual background and the transparency initiatives for the Colombia case. Second, we describe the theoretical approaches and variables of interest. Then, we describe the data and methods used. Afterward, we summarize and comment on the results obtained. Finally, we present our conclusions.

To capture the public transparency level in the Colombian territorial entities, we used the OGI provided by the Office of the Attorney General of the State for 1,101 municipalities (out of a total of 1,103) and 32 provinces (the total number of Colombian provinces) during the period 2012-2016. This composite index measures to what extent the information on the activities conducted by subnational governments, in terms of their management and results, is disclosed through the country’s information systems (Procuraduría General de la Nación, 2016).

The OGI latest comparable version (i.e., for 2016) collects and evaluates information using three weighted dimensions (i.e., organization, exhibition, and dialogue regarding public information), eight weighted categories (i.e., internal control, document management, procurement, basic territorial reporting, administrative and financial management systems, E-government, transparency and accountability, and attention to the citizen), and 24 weighted indicators (Box 2). Each indicator score is between 0 and 100, where 100 is the maximum score concerning reporting and compliance in the respective matter.

BOX 2
OGI STRUCTURE

Note 1: The numbers in brackets are the corresponding weights. Note 2: SICEP is the Spanish acronym for “Sistema de información para la Captura de Ejecución Presupuestal” (Information System for Capturing Budgetary Execution).

This index considers through the exhibition dimension the extent to which financial and management information (e.g., procurement, financial reporting, and government spending by sector) is available. Likewise, the OGI considers the channels citizens can use to access public information and make their requests so they can be met (i.e., the dialogue dimension). This index also establishes whether there are guarantees on the reliability of the information, so it measures - through the organization dimension - the level of implementation of the standard internal control and accounting control model in a territorial entity, as its document management capabilities.

Table 1 presents basic statistics about our data panel considering only the observations for 1,101 Colombian municipalities during 2012-2016. Data is missing for some variables and years, which explains why the total number of observations per variable varies slightly. Despite this, we have a highly balanced panel. This data structure allows us to model common and individual behaviors among groups and increase observations. The analysis period is 2012-2016, as the OGI (our transparency proxy) is comparable and available for those years.

TABLE 1
DESCRIPTIVE STATISTICS

Source: Elaborated by the authors.

It is important to highlight that variables measured in monetary units or related to the number of inhabitants (i.e., GDP per capita, and population) are transformed through their natural logarithms. We transformed them to avoid common statistical problems that are likely to occur with differences in scales; for example, non-stationarity in the models.

Concerning econometric modeling, a static approach can first be considered. In general, it is required to choose between a Pooled Ordinary Least Square Estimator (POLSE), which neglects the data panel structure, fixed effects (FE), and random effects (RE). Therefore, we run the Breusch and Pagan Lagrangian multiplier and F tests, the last one under the null hypothesis that all municipality dummies are zero. The results suggest that both RE and FE estimates are better than POLSE since the P-values for both tests are zero. Similarly, and through the Hausman test, it is determined that the FE is preferable over the other estimates.

It is also relevant to contrast basic assumptions such as homoscedasticity and not cross-municipality or serial correlations using three well-known and standard tests: the modified Wald test for GroupWise heteroskedasticity, the Breusch-Pagan LM test of independence, and the Wooldridge test for autocorrelation. These show that basic econometric assumptions are not met in our case, so we resort to a Panel Corrected Standard Errors (PCSE) model, following Beck and Katz (1995).

The general structure of the static models mentioned above is defined in Equation 1. Here, the transparency level for a pair it (municipality-year) depends on a set of K factors ( with ), presented in the previous section, so is the k-th variable for the pair it. Equation 1also includes municipality- specific characteristics represented by , so it is a dummy variable that equals 1 for the municipality i (i.e., municipality dummies).

However, institutional changes can also depend on the results of past initiatives, as suggested by serial correlation, which motivates including the lag of the dependent variable on the right side of Equation 1. Besides, other variables may be endogenous for several reasons, mainly because causality can run in both directions. For example, as fiscal autonomy would favor transparency extension (H11, Box 1), better and greater disclosure of public information, in turn, could improve citizens’ trust in their government and foster tax compliance (Castañeda-Rodríguez, 2021). Consequently, dynamic models also deserve attention.

In this regard, the difference GMM proposed by Arellano and Bond (1991) does not provide reliable coefficients under heteroskedasticity and excludes constant-time regressors (e.g., the dummy variable for capital cities). Therefore, we applied a system GMM estimator that starts from a level equation and includes a lag of the dependent variable, as observed in Equation 2. Moreover, according to Benedek et al. (2014, p. 75), “the resulting system-GMM estimator has much better finite sample properties in terms of bias and root mean squared error than that of the difference-GMM estimator.”

Table 2 displays our static and dynamic estimates applying the aforementioned techniques (i.e., PCSE and system GMM). Dynamic models (last two columns) add three rows showing Arellano-Bond tests for the first differences and the Hansen test, the last one to verify the instruments’ validity. Including the lag of the dependent variable as a regressor in Equation 2 generates serial correlation of order 1 in differences. However, there is no serial correlation of order 2 at the 99% significance level. Furthermore, the instruments (i.e., suitable lags of the first differences of the variables in levels) are statistically valid, as suggested by the Hansen test.

TABLE 2
STATIC AND DYNAMIC ESTIMATES

Note 1: Significance levels: * 0.05 ≤ p-value < 0.10; ** 0.01 ≤ p-value < 0.05; *** p-value < 0.01. Note 2: We used STATA 15 for estimating the respective models. Source: Elaborated by the authors.

Models 1 and 3 in Table 2 correspond to static and dynamic estimates, respectively, including all the previously discussed variables (Box 1). On the other hand, models 2 and 4 (Table 2) are the product of an iterative process in which statistically non-significant variables were removed, starting with those with higher P-values. This process was performed in order to obtain parsimonious estimates that capture the most relevant factors associated with transparency.

We also ran two suitable panel unit tests for relatively short datasets in T (our case) to verify the series stationarity and avoid any doubt about spurious relationships. We used the Harris-Tzavalis test, as it is less restrictive than other ones (e.g. allows non-normality), and the Fisher test, which is asymptotically optimal and nonparametric (Maddala & Shaowen, 1999).

Considering the possible non-stationarity of some variables (i.e., those encoded as Log_Mun_GDPP, Urban_Pop, Pop_25-34, Pop_35-44, Pob_45-54, Pob_55-64, Pob_65-74, Pob_75+, Turnout, and Perc_Win) in our models, according to the tests mentioned above, we studied whether there were long-run relationships between them. Therefore, we used the Kao test for panel cointegration, which allows the inclusion of 10 regressors (our case) and displays five test statistics. The corresponding results, (Table A, Annex) provide evidence that variables in all panels are cointegrated and build confidence in our estimates (Table 2).

Before proceeding to discuss our findings, it is important to mention that the considered municipalities are distributed in provinces with particular features and that estimates (Table 2) give equal weight to observations, without considering differences between cross-sections (e.g., Bogotá has around eight million inhabitants, while Busbanzá, in Boyacá, less than one thousand). Therefore, Table B (Annex) shows the basic models previously presented, but this time applying the jackknife resampling technique.

We divided the dataset by municipalities into ten population-based deciles and ran as many regressions as clusters (i.e., 10) by iteratively leaving out each group to control possible sample bias. Hence, the corresponding coefficients are the averages of the ten coefficients previously estimated. Table B (Annex) indicates that the sample distribution by population is not a driver of our findings.

The most populated territorial entities can influence the approval of public policies, for example, the determination of tax powers or the application of public transparency norms, without considering the needs and interests of minorities. This behavior is explained by the predominance of the principle of popular representation in the conformation of the upper and lower chambers of the Colombian Congress (Rodríguez, 2001, 2017), one of the main shortcomings of the 1991 Political Constitution to implement the figure of a decentralized unitary state and achieve a vertical distribution of political power in the country.

With that in mind, we can also apply the jackknife resampling technique by dividing the dataset by provinces, considering that Colombian provinces present differences in aspects such as their economic situation and population (Rodríguez, 2017) and include a dissimilar number of municipalities (e.g., while Antioquia has 125 municipalities, Amazonas groups only two).

Public transparency results depend partly on the institutional strength to promote and enforce policies in this area (Schnell & Jo, 2019), so it is important to consider whether the differences between provinces could affect our results. Consequently, there are 33 clusters if we take into account that the Colombian Constitution of 1991establishes that the country is a unitary and decentralized republic divided administratively and politically into 33 units (32 provinces and a capital district); however, the regressions are similar to those in Table B (Annex) (hence omitted).

Likewise, we could re-estimate the models by considering a dataset that includes only observations for the 32 Colombian provinces during the period 2012-2016.⁶Nevertheless, the respective coefficients are consistent with those estimated in Table 2, which were excluded for space considerations.

After checking the robustness of our econometric models, it is time to discuss the findings. From here on, we will focus on Model 4 (Table 2) as it shows that transparency follows a path-dependence process (according to the significance of its lag) (Equation 2) and only includes statistically significant variables. We will first refer to the results that deviate from the respective hypotheses (Box 1) or are contrary to what was expected.

This model (i.e., Model 4, Table 2) shows that there is not a linear relationship between transparency and population age; only three age population ranges (i.e., 35-44, 55-64, and 65-74) correlate positively and significantly with our endogenous variable. This suggests the need for future research that investigates what features distinguish people in each age group and how this affects their interests and abilities to monitor governments.

Although our final model excludes the urban population as an independent variable, the other three models in Table 2 show a negative association between the former and transparency (contrary to H4, Box 1), which suggests that people living in cities do not always have the required time or interest to keep themselves informed about public issues. This result could be due to the multiple occupations of urban residents that would lead to lower demand for information. For instance, national surveys for the United States indicate that rural residents are more rooted in their communities (Pew Research Center, 2018) and therefore more interested in participating in discussions of public affairs (which requires information) than their counterparts in urban areas. This point requires further research in the context of developing countries.

Additionally, Colombia is a highly urbanized country (81.1% of its population lives in urban areas, according to the World Bank, 2020). However, some of its provinces stand out in this regard but with poor results in terms of public transparency. For example, Atlántico and Bolívar, where most of the population lives in urban areas (95.7% and 77.7%, respectively), at the same time belong to the Caribbean, a region that has not achieved outstanding results in public information disclosure (Figure 1). Hence, our estimates could be capturing this fact.

Regarding political parties (control variable), only two of them (i.e., MAIS and Opción Ciudadana) correlate negatively with transparency, according to Model 4 (Table 2). In the future, it is required to study whether some characteristics of a given political party (e.g., the method of selecting candidates) could affect the position of politicians regarding transparency in the Colombian case.

For its part, our findings on the correlation between education, GDP per capita, population, voter turnout, political cycle (i.e., elections), and fiscal autonomy, on the one hand, and transparency, on the other, are aligned with our hypotheses (i.e., H1, H2, H3, H7, H8, and H11, Box 1). In this regard, for instance, our results support the point of agency theory that the greater the interest of citizens in engaging in public affairs, the greater the demands for public information disclosure, as suggested by hypothesis 7 (Box 1).

We also establish that fiscal autonomy correlates positively, significantly, and consistently with transparency. The above is reasonable since citizens are more willing to allocate part of their time and efforts to monitor what is done with the public treasury and hold the administration accountable for its decisions and performance when paying high taxes (the numerator of our fiscal autonomy indicator). Likewise, and according to the legitimacy theory, officials are more likely to adopt and enforce transparency initiatives in order to legitimize the tax burden. Fiscal autonomy seems to be a core element in improving and deepening transparency.

When municipalities depend mainly on transfers (i.e., have low fiscal autonomy), as with most local administrations in Colombia (Hernández & Barreto, 2018), three effects derive from this situation. First, a reduction in the administration’s discretion in public spending since these resources are highly conditioned to two items: education and health care (about 80% of transfers through the General System of Participations). Second, citizen monitoring becomes less necessary due to fewer agency problems. Third, there is a diminished interest of officials to legitimize their policies through transparency initiatives. The above should receive special attention if we consider that, on average, 73.63% of the income of Colombian municipalities comes from national transfers (DNP, 2021).

In summary, our results support the hypotheses H1, H2, H3, H5 (partially), H7, H8, and H11. In contrast, we found no evidence for hypotheses H4, H6, H9, and H10 (see Box 1).

As discussed throughout this paper, public transparency depends on many variables related to citizens’ and officials’ behaviors from different perspectives, as explained by the agency and legitimacy theories. However, sometimes the theoretical effect of a variable can be ambiguous, which directs attention towards empirical work. For instance, based on the agency theory, high electoral turnout is expected to be positively related to transparency (Esteller-Moré & Polo, 2012; Mourao et al., 2020), which aligns with one of our hypotheses (H7, Box 1). On the other hand, low voter turnout may threaten the legitimacy of public organizations, so governments can foster transparency initiatives to restore public trust (Tejedo-Romero & Araujo, 2018). The above is valid for other potential transparency drivers.

In addition, studying the factors that motivate or restrain public transparency in developing countries, such as Colombia, is essential due to the difficulty of generalizing results obtained in different socioeconomic contexts (e.g., carried out in developed countries). Among other things, we find that fiscal autonomy is positively correlated with transparency, which suggests the benefit of making local governments more responsible for financing their expenditures when possible; it is especially applicable to territorial entities with tax capacity.⁷⁷Although other studies have discussed this association, such as that led by Tejedo-Romero and Araujo (2018) for the Spanish case, the evidence in this dimension was not conclusive.

This study contributes in several ways to the literature on public transparency. On the one hand, it provides empirical evidence on the drivers behind public transparency in a developing country such as Colombia and finds evidence about a path-dependence process (i.e., the level of public transparency today depends on the corresponding level reported yesterday). On the other hand, this research shows that transparency levels in Colombian municipalities also depend on different economic, political, and demographic factors, some of them related to the demand for information due to pressures from citizens and other stakeholders (agency theory), such as education, GDP per capita, population, and political participation. Other variables refer to officials’ motives for gaining or preserving legitimacy (legitimacy theory), such as the occurrence of executive elections.

For future research, we recommend studying the incidence of institutional factors on transparency levels and conducting comparative studies with other Latin American countries in order to show if similarities prevail or unobservable factors influence the outcomes. Furthermore, studying how local governments could raise tax revenues (or fiscal autonomy) and compliance when facing political, institutional, or economic constraints would be an interesting subject. Finally, we believe that it is necessary to delve into the progress and the effects of transparency in developing countries and its role in meeting the sustainable development goals of the 2030 Agenda.

We thank our universities for the time they allowed us to dedicate to this research and the anonymous referees for their enriching comments. The author, Leon-Silva, would like to thank Instituto Colombiano de Crédito Educativo y Estudios Técnicos en el Exterior (ICETEX), particularly the “Pasaporte a la Ciencia” program. This work contributes to the “Sociedad” focus area under the framework of the “Programa Colombia Científica”.