Data
Research on decentralisation has long suffered from a shortage of adequate country-level data. This problem is exacerbated when the analysis is conducted at subnational level (
Martínez-Vázquez et al., 2017). Testing our hypotheses, however, requires regional data for all independent, dependent, and control variables over a reasonably large time period. Hence, one of the contributions of this research is putting together a large dataset containing information on the degree of decentralisation and the dimension of unfunded mandates for 518 regions in 30 OECD countries. The dataset covers, in an unbalanced way, the period between 1997 and 2018. The data are extracted, depending on availability, mainly from national and regional statistical offices (see
Tables A1 and
A2 in the
Online Appendix) and complemented with data from international organisations. The extensive scope of this study in terms of regions and years, coupled with a lack of a single database compiling the data of interest, explains the slightly unbalanced nature of the panel.
The term ‘regions’ refers to the subnational tiers of government with sufficient data availability to measure the variables of interest (see
Table A3 in the
Online Appendix for the regions included in the analysis). The decision to study regions in the OECD responds to practical reasons of data availability. It also simplifies the task of controlling for unobserved heterogeneity and avoiding omitted variable bias, thanks to the relatively similar characteristics of most OECD countries (
Ezcurra and Rodríguez-Pose, 2013).
Measuring fiscal and political decentralisation remains a highly contentious (
Martínez-Vázquez and McNab, 2003). While the share of total public expenditure spent by subnational governments remains the most commonly used proxy for fiscal decentralisation in cross-country studies, it is only available at the national level. Therefore, we use the per capita expenditure capacity of each of the 518 regions as a proxy for their degree of fiscal decentralisation. This indicator is not available in international databases, meaning that the data collection involved checking the budgets for each of the 518 regions individually. Total regional public expenditure comprises all expenses undertaken by a particular subnational authority, irrespective of how they are funded (be them through own-source revenues, shared ones, or transfers).
2 Values have later been divided by population and rendered comparable, converting them to constant 2015 USD, adjusted by purchasing power parity (see
Table A1 in the
Online Appendix).
Political decentralisation is too elusive a concept for it to be captured with a single measure. Scholars have created different indexes that aggregate various measures into an overall score that denotes the overall level of political decentralisation. Due to the regional-level data requirement of this study, we use the regional authority index (RAI) calculated by
Hooghe et al. (2016,
2021), as it is the only index capturing within-country regional differences and the best at including a large variety of factors (
Ezcurra and Rodríguez-Pose, 2013;
Filippetti and Sacchi, 2016;
Lessmann, 2012). The RAI overall score results from the aggregation of the values in eight sub-categories that are grouped under two main pillars: self-rule and shared rule.
3 The former estimates the degree of authority exerted by the region over its territory, while the latter calculates a region’s influence over central government decisions. To avoid collinearity, we recalculate the index excluding the indicators related to fiscal decentralisation (fiscal autonomy and fiscal control) from the RAI overall score and use the resulting values as a proxy for political decentralisation at the regional level.
We use the above data to measure unfunded mandates. As the imbalance between power and resources, we make the variables for fiscal and political decentralisation comparable by standardising both with a mean value of 0 and a standard deviation of 1. We then subtract the values of fiscal decentralisation from the values of political decentralisation obtaining a relative index of unfunded mandates.
4 With this conversion, we measure which regions have a larger or smaller unfunded mandate depending on whether their value is above or below the mean of 0, respectively. This index does not provide an absolute value of unfunded mandates for each region. Rather, it offers an estimated degree of unfunded mandates for each region relative to the gap between political and fiscal decentralisation in all other regions in the OECD. This has the advantage of comparing regions with one another and establishing which regions have wider or narrower unfunded mandates relative to the rest of the sample.
Finally, following previous literature on the link of decentralisation and economic growth, several control variables are incorporated into the model to avoid inconsistent parameters (
Canavire-Bacarreza et al., 2017). We include regional population and region size, as larger regions in terms of population and/or land may have further resources to exploit to deliver a better economic performance (
Arzaghi and Henderson, 2005;
Congleton, 2006). Similarly, the level of development of a region may affect its growth potential. We therefore add regional gross domestic product (GDP) per capita in the model. We also control for human capital, a fundamental driver of growth. We measure human capital using the share of individuals in a region between the ages of 25 and 34 with a completed secondary education degree (
Canavire-Bacarreza et al., 2020). Government size, often correlated with declines in economic growth (e.g.
Afonso and Furceri, 2010), is measured as the total general government spending as a percentage of GDP and added in the full regressions (see
Tables 2,
3, and
A7 in the
Online Appendix).
Table A1 in the
Online Appendix offers the description and sources of all variables above.
Model Specification
To test our hypotheses with our original panel dataset, a static panel-data region and time fixed-effects model is estimated with heteroscedasticity-robust standard errors (
Newey and West, 1994). In this study, the baseline model adopts the following form:
where
represents annual GDP growth in region i for year t;
stands for unfunded mandates and denotes the difference between political decentralisation and fiscal decentralisation;
depicts the degree of fiscal decentralisation;
captures the level of political decentralisation;
encapsulates the relevant control variables (population size, level of development, educational attainment of young adults, region size, and national government size); µi and
are the region and time fixed-effects, respectively; while
denotes the error term.
Fixed-effects models have been normally used to estimate the economic impact of decentralisation in analyses of long-term decentralisation processes. However, one of the main problems of fixed-effects specifications when dealing with decentralisation is linked to the limited change over time of some decentralisation variables and, in particular, of political decentralisation. Fixed-effects models can also not consider time-invariant factors, such as region size or the presence of a particular region in a given country or continent. Random-effects estimators allows for both time- and time-invariant regressors, but have the drawback that region-individual effects can be correlated with some independent variables, leading to inconsistent coefficients (
Hausman, 1978).
We, therefore, resort to Hausman–Taylor (HT) estimators (
Hausman and Taylor, 1981) as our econometric approach. The use of HT, on the one hand, allows to calculate consistent coefficients for time-variant variables using their within-transformation as in the fixed-effects model; on the other hand, HT can also estimate coefficients for the time-invariant regressors. HT classifies variables as exogenous (i.e. correlated with the disturbance term only) or endogenous (i.e. correlated with the region-specific individual effects only), thereby partially controlling for endogeneity, a common concern in the literature (
Baltagi et al., 2003;
Canavire-Bacarreza et al., 2020). It also uses the between variation of time-variant exogenous regressors to derive internal instruments and hence does not require an additional external instrumental variable (
Baltagi and Liu, 2012;
Mitze, 2009). This is a key advantage because few strong instrumental variables exist for national-level studies, and this scarcity is aggravated at the regional scale.
The main model is specified as follows:
where,
captures annual GDP growth per region and acts as the dependent variable;
includes the time-variant exogenous control variables on regional population, level of development, education, and national government size. It also contains year dummy variables;
comprises three time-variant endogenous independent variables estimating the degree of unfunded mandates, the level of fiscal decentralisation, and that of political decentralisation, respectively;
represents the time-invariant exogenous variables and includes a series of supraregional dummy variables as well as regional size; µi denotes the fixed-effects term, while
stands for the disturbance term.
To test hypothesis 2, we also seek to determine the potentially mediating effect of a region’s level of development on the relationship between unfunded mandates and regional economic growth. Hence, following
Lessmann (2012) and
Filippetti and Sacchi (2016), we estimate an extended version of
equation (2), where
includes three interaction terms between the variable for unfunded mandates and fiscal decentralisation, political decentralisation, and level of development, respectively.
Finally, we are not oblivious to discussions about reverse causality. It could be the case that lower economic growth spurs unfunded mandates, instead of the other way around. We therefore run a series of robustness checks. Due to the lack of appropriate external instrumental variables, model 2 is transformed from a static into a dynamic panel-data system-generalised method of moments (GMMs) (
Arellano and Bover, 1995;
Blundell and Bond, 1998;
Roodman, 2009). This model is further discussed in the robustness checks section and its equation reads as follows:
where
captures the influence of past regional GDP growth on the next year’s growth rate, and the rest of variables are as stated above. The variables for unfunded mandates, and fiscal and political decentralisation are classified as endogenous in all system-GMM regressions.