Data source and sample
Financial inclusion is the dependent variable of this study and measured through a composite index based on ATMs, bank branches, deposit accounts with commercial banks, and borrowers from commercial banks per 100,000 adults. The data of financial inclusion proxies are retrieved from Financial Access Survey—IMF. On the other hand, economic freedom is measured through an index compiled by the Fraser Institute. Although an index for economic freedom is also compiled by Heritage Foundation, Fraser Institute’s index is widely used in previous studies. There are five items in the index, i.e., regulation, freedom to trade internationally, sound money supply, legal system and property rights, and size of government. These components have covered most aspects that may motivate the general population to move their savings to formal financial institutions. A wide range of studies have utilized this index to measure economic freedom (M. R. [29, 55, 61]
The data related to governance quality are collected from the World Governance Indicators (WGIs). The WGI has provided six individual and aggregate dimensions of governance including voice and accountability, political stability and absence of violence, government effectiveness, regulatory quality, rule of law, and control of corruption. Since some studies found mixed evidence for the individual dimensions, we used an aggregate measure of governance quality developed through principal component analysis (PCA) to eliminate the ambiguity. This process is used and justified by previous studies [6, 63]
Financial literacy is measured through the data provided by a survey of Gallup in collaboration with the Global Financial Literacy Center and the World Bank. The survey was conducted in 143 countries in the year 2014. The five survey items were related to interest compounding, interest rate, inflation, and risk diversification. The same data and measurement were used by Grohmann et al. [22] in their study. Since the time-series data for financial literacy are not available, we relied on the cross-sectional analysis to estimate the financial literacy model with 111 countries. On the other hand, we employed panel data composed of 98 countries from the year 2007 to 2018 (after eliminating missing data) for the economic freedom model. We control for gross domestic product (GDP) growth and inflation. Studies revealed that economic growth is positively associated with financial inclusion [7, 22]. However, inflation leads to income inequality and negatively influences saving behavior, which hinders financial inclusion [44]. Studies reveal that countries with high inflation experience low financial inclusion [5, 47]. The data related to control variables are retrieved from the World Bank database.
Principal component analysis (PCA)
Principal component analysis (PCA) is a dimensionality-reduction method used to transform large datasets into smaller ones without losing important information. Accordingly, we used PCA to develop the composite index of financial inclusion (FINDEX) and government of quality (GINDEX). By using PCA method, the jth factor index can be written as:
$$\begin{aligned} {\text{FINDEX}}_{j} & = {\text{InW}}_{J1} X_{1} + {\text{ InW}}_{J2} X_{2}\\ &\quad + {\text{ InW}}_{J3} X_{3} + \cdots + {\text{ InW}}_{JP} X_{P} \end{aligned}$$
$$\begin{aligned} {\text{GINDEX}}_{j} & = {\text{InW}}_{J1} X_{1} + {\text{ InW}}_{J2} X_{2} \\ &\quad+ {\text{ InW}}_{J3} X_{3} + \cdots + {\text{ InW}}_{JP} X_{P} \end{aligned}$$
whereas FINDEXj and GINDEXj are financial inclusion index and governance index, respectively; Wj is the factor score weight of the parameter; X is the initial value. The FINDEX is developed from the four factors, while GINDEX is developed from six factors of governance indicators. For measuring the sample adequacy, Kaiser–Meyer–Olkin (KMO) and the Bartlett test of sphericity are used.Footnote 2 The value of KMO for FINDEX and GINDEX was 0.71 and 0.74, respectively, along with significant Bartlett test values, which are considered in an acceptable (middling) range (H. F. [30].
Empirical model
Since the data on financial literacy are available for the only year (i.e., 2014), we used simple OLS regression (with robust standard errors) to estimate the effect of financial literacy on financial inclusion along with the moderating role of government quality. Owing to the fact that we are also estimating the mediating role of financial literacy on the relationship between government quality and financial inclusion, the procedure of Baron and Kenny [9] is employed:
$${\text{FINDEX }} = \, \beta_{1} {\text{GINDEX }} + \, \beta_{2} X \, + \, u$$
(1)
$${\text{FL }} = \, \beta_{1} {\text{GINDEX }} + \, \beta_{2} X \, + \, u$$
(2)
$${\text{FINDEX }} = \, \beta_{1} {\text{GINDEX }} + \, \beta_{2} {\text{FL }} + \, \beta_{3} X \, + \, u$$
(3)
The OLS regression for the moderating role of government quality takes the following form:
$$\begin{aligned} {\text{FINDEX }} & = \, \beta_{1} {\text{FL }} + \beta_{2} {\text{GINDEX }} \\ &\quad + \, \beta_{3} \left( {{\text{GINDEX }} \times {\text{ FL}}} \right) \, + \beta_{4} X \, + \, u \end{aligned}$$
(4)
where FINDEX is the financial inclusion, GINDEX is the government quality, FL is the level of financial literacy, X is a matrix of control variables. However, to estimate the effect of economic freedom on financial inclusion along with the moderating role of government quality, the following OLS model is developed:
$${\text{FINDEX }} = \, \beta_{1} {\text{EF }} + \beta_{2} {\text{GINDEX }} + \, \beta_{3} \left( {{\text{GINDEX }} \times {\text{ EF}}} \right) \, + \beta_{4} X \, + \, u$$
(5)
where EF is economic freedom. Although we begin our analysis with OLS estimations, we posit that the association between financial inclusion, economic freedom, and government quality is not unidirectional and could include endogeneity bias (omitted variable bias and reverse causality). Financial inclusion implies that the general public especially adult members of society should have access to financial products and services at affordable costs. For the reason that financial inclusion is strongly correlated with saving behavior and accessing credit from the formal financial institution, financially inclusive individuals demand a higher level of economic liberty, better regulatory quality, less information asymmetry, and strong investor protection policies [15, 43]. Therefore, financial inclusion may also lead to economic freedom and a higher degree of government quality.
Although fixed-effect estimations in panel data can eliminate the first source of endogeneity, i.e., omitted variable bias, however, it is not an efficient tool for resolving the reverse causality issue. Accordingly, we have utilized the generalized method of moment (GMM) with lagged values of independent variables as instruments to deal with both types of endogeneity. Since we are using a short panel (N > T), the system GMM model is an appropriate instrument to mitigate the correlation between the error term and the endogenous variable. It also eliminates the autocorrelation of the error term [58]. Our dynamic panel data model has the following form:
$${\text{FINDEX}}_{it} = \, \alpha_{it} + \, \beta_{1} {\text{FINDEX}}_{i,t - 1} + \, \beta_{2} {\text{EF}}_{i,t} + \, \beta_{3} {\text{GINDEX}}_{it} + \, \beta_{4} \left( {{\text{GINDEX }} \times \, EF} \right) \, + \beta_{5} X + \varepsilon_{it}$$
(6)
where i is the cross-sectional unit (country), t stands for time-period, X refers to control variables, country, and time effects,α is the unobserved fixed effect; and ε are error terms being independently and identically distributed over the whole sample with constant variance. System GMM also assumes the stationarity of all variables due to which we estimated the panel unit root testFootnote 3of Im, Pesaran, and Shin [27]. The results indicate the stationarity of all variables in level form.