4.1 Econometric specification and estimation strategy
Empirical approaches to study differences in income levels have been proposed in economics using growth dynamics, which brings to a debate between neoclassical economics and the endogenous growth theory. Neoclassical economics assumes that economic growth can be explained by external forces related to the combination of the factors of production, mainly capital and labor (Solow
1956). On the other hand, endogenous growth theory proposes an alternative explanation of economic growth through endogenous drivers, such as human capital, knowledge, and innovation (Romer
1994). Both theories have emphasized the combination of physical capital, human capital, and technology, together with a residual that captures other factors that cannot be explained. Institutional quality may be helpful to explain this residual (Rodríguez-Pose and Ketterer
2020; Rodríguez-Pose
2020).
For this study, we use a generalized production function approach (e.g., Zellner and Revankar
1969), which allows us to measure the impact that technology, geography, and institutions exert on regional development. We present our baseline model in Eq. (
1). This equation is theoretically and empirically rigorously founded since we follow the common generalized production functions together with the augmented Solow model from Mankiw et al. (
1992). Both cases propose a specification where the logarithm of GDP per capita in levels is the dependent variable. Besides, we also follow the empirical specification from Rodríguez-Pose and Ketterer (
2020) to include other geographical, institutional, and economic determinants. In contrast to previous studies, we extend the empirical model incorporating ICT variables.
$$\begin{aligned} \ln {\text{GDPpc}}_{it} = & \beta_{0} + \beta_{1} \ln {\text{HHINT}}_{it} + \beta_{2} \ln {\text{HHBR}}_{it} + \beta_{3} \ln {\text{EMPLICT}}_{it} + \beta_{4} {\text{QoG}}_{it} \\ & \quad + \beta_{5} {\text{HDD}}_{it} + \beta_{6} \ln {\text{ACCESS}}_{i} + \beta_{7} \ln {\text{TRI}}_{i} + \beta_{8} \ln {\text{INV}}_{i,t - 1} \\ & \quad + \beta_{9} \ln {\text{POP}}_{it} + \beta_{10} \ln {\text{EDUC}}_{it} + \varepsilon_{it} \\ \end{aligned}$$
(1)
where ln denotes the natural logarithm and subscripts
i and
t refer to region and time, respectively.
\({\mathrm{GDPpc}}_{it}\) is the regional income per capita in region
i at time
t.
\({\mathrm{HHINT}}_{it}\) denotes the percentage of households having access to the Internet,
\({\mathrm{HHBR}}_{it}\) is the percentage of households with broadband access,
\({\mathrm{EMPLICT}}_{it}\) is the share of employment in the Information and Communications Technology (ICT) sector,
\({\mathrm{QoG}}_{it}\) is the regional quality of government,
\({\mathrm{HDD}}_{it}\) is the number of heating degree days,
\({\mathrm{ACCESS}}_{i}\) is the accessibility index, and
\({\mathrm{TRI}}_{i}\) is the average terrain ruggedness index in the region.
\({\mathrm{INV}}_{it-1}\) denotes the 1-year lag of investment, and
\({\mathrm{POP}}_{it}\) is population in the region
\({\mathrm{EDUC}}_{it}\) is human capital. Finally,
\({\varepsilon }_{it}\) denotes the error term.
Growth and economic development theories have incorporated technology as an explanatory variable by assuming that technology increases efficiency and productivity (Basu and Weil
1998). Despite the relevance of ICT as a regional growth driver in the European Union, existing literature on the effects of ICT is scarce and has only focused on a single variable. For this reason, we use three variables to measure ICT. First, the share of the ICT sector in the regional gross value added to capture the size of the ICT sector in the economy, as in Martínez and Rodríguez (
2009). Although the share of this sector in the economy depends on the region production structure itself, descriptive statistics in Table A.2 show that lagging regions have a small share. A traditionally agricultural region would have a smaller share of the ICT sector. However, it might have higher levels of economic development and less need for ICT convergence in the production share of the economy. Therefore, it is necessary to complement this variable with other indicators of ICT penetration in the region. Second, we follow other studies (Vicente and López
2011; Billón et al.
2017) by using ICT variables related to households: the percentage of households with access to the Internet—to measure the quantity—and the percentage of households with broadband access—as a measure of the quality (speed).
In the European Union, Internet access has become relatively cheap due to the high competition in the market of Internet service providers as a result of liberalizing market access. Therefore, one might think that the percentage of households with access to the Internet might not be a good proxy for digital technology in explaining growth, being speed a more relevant proxy. However, Fernandez-Portillo et al. (
2020), using data for the OECD European countries, find that the number of Internet users is the ICT indicator with the highest performance on explaining GDP growth. We consider relevant to include both the share of households with access to the Internet and households with broadband access as a proxy for the speed.
Intrinsically related to technology, human capital has also been demonstrated to be important for increasing economic growth since differences in education can help to explain differences in income per capita (Mankiw et al.
1992) and this pattern is also observed at the regional level (Gennaioli et al.
2014).
Geography has also been recognized as a growth driver (Sachs et al.
2001), but finding a suitable variable to measure the impact of geography has been subject to debate by academic scholars. Albalate et al. (
2021) find that temperature, measured through the heating degree days, is the most important factor explaining population density in European regions. Using a model that predicts the regional population distribution from geographical factors, they find that misadjustment between observed and predicted population distribution harms economic growth. A rugged terrain may make it difficult to develop certain living conditions (Nunn and Puga
2012) and be negatively associated with growth and development. Lower accessibility and higher ruggedness increase the cost of transport and trade (Ketterer and Rodríguez-Pose
2018). Regions with lower accessibility can find in ICT a solution to overcome the curse of geography. However, greater ruggedness increases the cost of deploying the necessary technological infrastructure.
Growth theory has demonstrated that a combination of the aforementioned factors is not sufficient to explain growth patterns. Although growth theory has been continuously improving, regional growth patterns show an increasing residual factor that suggests the existence of missing elements (Rodríguez-Pose and Ketterer
2020). Among all the potential elements that could trigger regional growth, the role of institutions is found to be a major regional growth driver, especially after the progress made in measuring the quality of government at the subnational level in the European Union (Charron et al.
2019). A positive effect is expected since those regions with better quality institutions register higher levels of regional growth (Ketterer and Rodríguez-Pose
2018; among others).
One of the most critical aspects of this article is how to measure inequality, given the importance of this variable to explain research objectives and the heterogeneity of potential indicators.
5 We begin with the concept of social exclusion, coined in Sociology, as this is a broader concept that may comprise inequality. Lenoir (
1974) defined social exclusion as a new source of inequality that prevents selected individuals or social groups from their full participation in society. As poverty has been considered as the main indicator to measure social exclusion by institutions and national bureaus of statistics, we hypothesize that risk of poverty may accurately capture differences in regional inequality accurately, in contrast to other variables like the Gini Index, which is the most widely used indicator to measure inequality but has been subject to criticism.
6
The lack of social variables could also be mentioned as another factor that has increased the growth residual. To overcome this issue, following Perugini and Martino (
2008), we define a second Eq. (
2) to study whether the previous determinants that explain economic development can also explain the risk of poverty and social exclusion in European regions
7:
$$\begin{aligned} \ln {\text{RISKP}}_{it} = & \beta_{0} + \beta_{1} \ln {\text{HHINT}}_{it} + \beta_{2} \ln {\text{HHBR}}_{it} + \beta_{3} \ln {\text{EMPLICT}}_{it} + \beta_{4} \ln {\text{QoG}}_{it} \\ & \quad + \beta_{5} {\text{HDD}}_{it} + \beta_{6} \ln {\text{ACCESS}}_{i} + \beta_{7} \ln {\text{TRI}} + \beta_{8} \ln {\text{INV}}_{i,t - 1} + \beta_{9} \ln {\text{POP}}_{it} \\ & \quad + \beta_{10} \ln {\text{EDUC}}_{it} + \varepsilon_{it} \\ \end{aligned}$$
(2)
where
\(\mathrm{ln}{RISKP}_{it}\) is the logarithm of the risk of poverty and social exclusion in a specific region. The remaining variables are as in Eq. (
1).
Whether the results are driven by endogeneity due to the possible correlation of the percentage of households with access to Internet and broadband, the share of the ICT sector in the regional economy, and the institutional quality variables with the unobserved individual random effect is an important fact to address. To this end, the existence of potential problems of endogeneity between economic development and technology (Grossman and Helpman
1991), geography (Sachs et al.
2001), and institutions (Glaeser et al.
2004) has been identified in previous studies. As investment is included in the measure of GDP, we include the first lag of the investment variable to overcome the possible problem of endogeneity.