2. Theoretical considerations

Institutions consist of formal rules and informal norms that lend stability and predictability to human interaction and behaviour (North, Reference North1989; Reference North1990). They do so by providing incentives and constraints that jointly determine the set of permissible choices, the relationships among individuals, and ‘who may do what to whom’ (Bromley, Reference Bromley1989). Like other human constructs, institutions can and do change, and the factors underpinning this change have been the subject of a large body of social science literature.

North (Reference North1989, Reference North1990) further emphasises the role of ideas and tastes in shaping institutions. Changes in either can alter societal preferences about existing institutions as well as beliefs about how the world works or should work. North brings up the example of slavery, whose abolition in several parts of the world was eventually brought about by voting, which more easily channelled shifting social preferences regarding the institution (North, Reference North1989, p. 241). In North's framework, institutional change is achieved by actions of political and economic entrepreneurs holding these beliefs, who work to change existing institutions to their benefit.

Development economists often gauge the impact of ideas through the lens of ideology and culture.Footnote ² The premise is culture determines the kinds of ideas and beliefs that can take root. Williamson (Reference Williamson2000) notably identifies informal norms, customs, traditions, and beliefs as the most socially embedded of all institutions. These ‘Level 1’ institutions then constrain the development of ‘Level 2’ institutions where formal rules of the game lie. In a similar vein, Roland (Reference Roland2004) proposes the distinction between ‘slow-moving’ and ‘fast-moving’ institutions, arguing that changes in the latter likely come about through the process of slow but continuous evolution of culture, beliefs and norms. Cultural persistence can thus explain the existence of multiple institutional equilibria, not all of which are considered ‘efficient.’

What is the mechanism by which culture affects institutions? The evolutionary view is that new institutions arise in a decentralised fashion, and those that minimise transaction costs will emerge and survive (Kingston and Caballero, Reference Kingston and Caballero2009). Culture can thus inspire the process of random or deliberate change in formal rules, and the most efficient rules for a given cultural context tend to be the ones that succeed. Thus, the prediction of the evolutionary approach is that the most efficient institutions arise deterministically. Yet economic policies encompassed within the EFW index come about as a process of collective choice directed by some political body such as the parliament.

Under the collective choice framework, institutional equilibria arise from the interaction between supply (government) and demand (constituents) of institutional change (Alston, Reference Alston, Alston, Eggertsson and North1996). The direction and the intensity of institutional change depend on the relative bargaining powers of the two sides. Democracy is one such gauge of bargaining power: when governments are responsive to citizens, institutional change moves in the direction of, in part, the cultural ‘consensus.’ In the total absence of democracy, institutional supply are the unelected elites. In either situation, culture shapes the cost of establishing particular institutional arrangements (Ruttan, Reference Ruttan2006). This suggests a crucial role for political and legal processes in shaping institutions. We thus model institutional (EFW) change as being conditioned culture, democracy levels, and the interaction between democracy and culture, among other covariates. This approach implies that the impact of culture on institutions depends on democracy levels (which capture the bargaining power of constituents (the median voter—institutional demand)), and that the impact of democracy likewise depends on culture.

We are not the first to propose an interaction between culture and (formal) institutions as a promising explanatory variable for institutional change. Cline and Williamson (Reference Cline and Williamson2017) interact democracy with individualism, arguing that culture determines preferences over the level of contract regulation, and democracy serves to aggregate those preferences and channel them into policy on contract enforcement. Davis and Williamson (Reference Davis and Williamson2016) find that individualism has a stronger impact on decreasing the regulatory burden of start-up entry in common law jurisdictions. Davis and Williamson (Reference Davis and Williamson2018) find a similar result when explaining the regulation of entry by foreign firms. Ang and Fredriksson (Reference Ang and Fredriksson2018) further show individualism decreases the rigidity of labour market regulations more so in common law countries.Footnote ³

Notably, democracy itself may depend on the levels of individualism within a country. Gorodnichenko and Roland (Reference Gorodnichenko and Roland2020) argue that individualistic cultures exhibit lower aversion toward institutional innovation. Once these societies overcome their collective action problems, transition to democracy is more likely to ensue. This stands in contrast to collectivist cultures that value conformity and usually transition from predatory to non-predatory autocracies. The implication is that a culture of individualism is thus causally linked to democracy.

The common argument of the literature surveyed above is that certain cultures and political and legal systems co-determine policy outcomes. Davis and Williamson (Reference Davis and Williamson2016) term this the Interdependent Institutions Hypothesis, whereby culture determines social policy preferences while institutions such as democracy or legal origins determine the extent to which culture can influence economic policy. Once social preferences are sufficiently aggregated within given political systems, culture's impact is magnified and reform is more likely to succeed if the prevailing culture and the direction of institutional reform are compatible. This notion echoes that of Pejovich's (Reference Pejovich1999) Interaction Thesis according to which transaction costs of maintaining, enforcing, and protecting new formal rules will be lower if formal and informal institutions are in harmony. Along a similar vein, Bisin and Verdier (Reference Bisin and Verdier2017) argue that depending on whether culture and institutions are complements or substitutes, their co-evolution may either mitigate or magnify their effects on socio-economic outcomes.

Why is individualism found to be robustly associated with ‘good-for-development’ outcomes? Greif (Reference Greif1994) uses a game-theoretic approach to find individualistic cultural beliefs are more likely to foster institutions that support anonymous exchange and innovation, thus enabling faster growth. Gorodnichenko and Roland (Reference Gorodnichenko and Roland2017) also find individualistic countries tend to innovate more because of the social status and recognition conferred to innovators.

3. Data

EFW index

The EFW index measures the quality of countries' economic policies and institutions on a 0–10 scale. The composite score is an aggregate of five broad dimensions of economic freedom: the size of government, legal system and protection of property rights, access to sound money, freedom to trade internationally and regulation of business, labour and credit. A growing cross-country empirical literature shows that EFW correlates positively with higher income levels and growth, reported happiness and life expectancy (Hall and Lawson, Reference Hall and Lawson2014), while it also correlates with higher income inequality (Bergh and Nilsson, Reference Bergh and Nilsson2010).

Although the EFW index measures institutions in line with the ideals of small government, large governments are not necessarily ‘bad’ for economic development. Indeed, countries with large public sectors and high overall levels of economic freedom (e.g. Sweden) have nearly identical levels of development compared to economically free countries with smaller governments. To explain this phenomenon, Bergh (Reference Bergh2020) introduces the concept of Hayekian welfare state which combines a large fiscal apparatus, rule of law, and the ability to both implement policies (state capacity) and alter them based on past experiences.

According to Bergh, the development effect of government size is moderated by the amount of knowledge available to the state when engaging in planning and fiscal interventions. If knowledge needed is relatively low, even large public expenditures (e.g. pension schemes) do not generally inhibit economic growth. If knowledge needed is considerable, interventions tend to be less successful and wasteful. The knowledge problem itself may be partly overcome with sufficient state capacity and the ability to learn from past policy failures. This implies that the relationship between government size and economic outcomes is not monotonic, and that it warrants a closer look into the types of interventions governments conduct.

The EFW scores we employ (version 2018) are assigned to countries every 5 years between 1970 and 2000 and annually over 2000–2017. Murphy and Lawson (Reference Murphy and Lawson2018) extend the composite index back to 1950 for 54 countries and to the 1960s for up to 95 countries. They do so by reconstructing two separate economic freedom ratings for 1950–1980: one based on a set of eight available variables, and the other obtained by backcasting baseline data using autoregressive modelling. The two new measures are then combined in a regression model used to fit the final EFW values over 1950–1970. This paper merges the 1950–1970 extension with the original EFW data to obtain economic freedom ratings over 1950–2015.Footnote ⁴

Figure 1 shows the coefficient of variation for the EFW index from 1950 through 2015. The figure suggests countries were becoming more dissimilar with regard to economic freedom until the 1980s, after which a large contraction in institutional disparities occurred. The general trends over the last 25 years indicate a decrease in cross-country dispersion in economic freedom has occurred. Figure 2 shows the same shrinking dispersion of the density by decade. However, the mass of the EFW distribution also concentrates at higher levels of EFW over time, indicating not only are economic institutions becoming more similar, but also increasingly liberalised.

Figure 1. EFW coefficient of variation (std. dev. over mean) over time.

Figure 2. EFW kernel density over time.

Control variables

We account for the wide variety of countries' socio-economic characteristics that plausibly affect the institutional quality: real per capita income (output side, Penn World Tables 9.1), population size (Penn World Tables 9.1), incidence and intensity of armed conflict (Marshall, Reference Marshall2017), human capital (average years of schooling for working-age population, 15–64; Barro and Lee, Reference Barro and Lee2010), legal origins (British and socialist tradition; La Porta et al., Reference La Porta, Lopez-de Silanes, Shleifer and Vishny1999), ethno-linguistic fractionalisation (Esteban et al., Reference Esteban, Mayoral and Ray2012), geography (latitude and continent fixed effects), dominant religion and year fixed effects.

Another major independent variable is a democracy, measured using the Varieties of Democracy (V-Dem) index (Coppedge et al., Reference Coppedge, Lindberg, Skaaning and Teorell2016). Our aim is to capture democracy purely as the rule by the people. Though by no means novel, this conceptualisation satisfies the condition of citizens exercising power through freely elected officials while simultaneously discarding any normative implications regarding economic freedom reform. The idea is to ‘let culture speak’ and use democracy only as a vehicle for cultural preferences. While Polity IV and Freedom House measures have received criticism on grounds of being too narrow, not sensitive enough to changes in democracy, or reflecting concepts other than democracy (Munck and Verkuilen, Reference Munck and Verkuilen2002), the multidimensionality of the V-Dem data allows for moving beyond the minimalistic view of democracy as the mechanism for removing the executive, as well as those varieties of democracy with a potentially normative or local flavor.Footnote ⁷ Table 1 reports summary statistics for these variables.

Table 1. Summary statistics, largest regression subsample

4. Methodology

We begin by running the following monadic regression:

(1)$$\Delta EFW_{i, t} = \alpha + \beta Democ._{i, t} + \gamma Cult._i + \phi Democ._{i, t} \times Cult._i + \rho EFW_{i, t-1} + \Omega {\boldsymbol X}_{{\boldsymbol i}{\bf , }{\boldsymbol t}} + \delta _t + \varepsilon _{i, t}, \;$$

where ΔEFW _i,t is the 5-year difference in the EFW index, Democ._i,t is the simple average between electoral and participatory democracy indices for country i in year t, Cult._i is any of six Hofstede cultural dimensions, X is a vector of country-specific controls and δ _t are year fixed effects. Equation (1) has 782 observations across 80 countries when all controls are present. We estimate separate regressions for each cultural dimension.

Because Hofstede indices are cross-sectional, we interact them with time-varying democracy levels. In doing so, we implicitly assume that culture's impact on economic freedom depends on whether a country is democratic, and that democracy's impact depends on the underlying culture. Due to this interaction, the coefficient on culture, γ, and the coefficient on democracy, β, should not be interpreted as unconditional effects. Rather, the impact of culture is now conditional on the country's level of democracy:

(2)$$\displaystyle{{\partial \Delta EFW_{i, t}} \over {\partial Cult._i}} = \gamma + \phi Democ._{i, t}, \;$$

and the impact of democracy is conditioned on the prevalence of a particular cultural orientation:

(3)$$\displaystyle{{\partial \Delta EFW_{i, t}} \over {\partial Democ._{i, t}}} = \beta + \phi Cult._i.$$

Thus, if a cultural dimension promotes economic freedom, we expect $\hat{\phi } > 0$; i.e. the positive effect on reform should be stronger in more democratic countries (equation (2)), while democratisation in the said cultural environment should stimulate reform (equation (3)). We choose to observe the effect of culture primarily through the marginal effect of democracy in equation (3). We do so to exploit the longitudinal variation in the marginal effect of democracy that Hofstede culture indices lack.

We report our results using pooled OLS rather than within estimates. We do so for two reasons: first, country-fixed effects would subsume the time-invariant Hofstede indices, potentially biasing ϕ. Second, we model the variation in EFW more richly by drawing from information on between-country differences in culture. To deal with omitted variables, we control for a large number of time-invariant factors, which plausibly correlate with some of the variation captured by country fixed effects.

Next, we collapse all six cultural dimensions into a single metric for cultural distance to compare each country's ‘overall culture’ to that of a reference country. We choose the US as the benchmark, due to its long history of having liberal economic institutions. Thus, the hypothesis is that increased cultural distance from the US negatively correlates with ΔEFW ($\hat{\phi } < 0$). Our first measure of cultural distance is based on the standard Euclidean metric for the distance between two points in n-dimensional space:

(4)$$E.Cult.Dist._{i, j} = \sqrt {\sum\limits_{k = 1}^6 {\left[{\displaystyle{{{( {D_{k, i}-D_{k, j}} ) }^2} \over {V_k}}} \right]} } , \;$$

where D represents one of k = 6 Hofstede dimensions and V_k is the in-sample variance of the kth dimension (Kogut and Singh, Reference Kogut and Singh1988). To minimise bias, cultural distance is computed only for those country pairs that have observations across all six dimensions.

We also employ a generalised form of the Euclidean distance called Mahalanobis distance, which corrects for the covariances between different cultural dimensions and provides for more accurate estimates (Kandogan, Reference Kandogan2012). The general formula for Mahalanobis distance is given by the following equation:

(5)$$M.Cult.Dist._{i, j} = \sqrt {{( {{\boldsymbol D}_{\boldsymbol i} {-} {\boldsymbol D}_{\boldsymbol j}} ) }^TS^{{-}1}( {{\boldsymbol D}_{\boldsymbol i} {-} {\boldsymbol D}_{\boldsymbol j}} ) , \;} $$

where D is a vector of k cultural dimensions and S is the variance-covariance matrix. Note when S is a diagonal matrix, Mahalanobis collapses into Euclidean distance. In our sample, the correlation between Euclidean and Mahalanobis distance is r = +0.904, and the comparison of densities is shown in Figure 3.

Figure 3. Densities of different measures of cultural distance.

One typically examines the temporal dispersion of a longitudinally observed variable by regressing its cross-sectional measure of variation (such as standard deviation or coefficient of variation) against a time trend. However, such a specification would not allow us to condition institutional dispersion on a set of country-specific covariates. We therefore model institutional dispersion as a function of cultural distance between countries in a dyadic model:

(6)$$\vert {EFW_i-EFW_j} \vert = \alpha + \beta Cult.Dist._{i, j} + \Gamma {\boldsymbol X}_{{\boldsymbol i}{\bf , }{\boldsymbol j}} + \eta _{i, j}, \;$$

where Cult.Dist._i,j is a Euclidean (or Mahalanobis) cultural distance between countries i and j, and X_i,j is the vector of various measures of country-pair differences and similarities. The absolute bilateral difference in countries' EFW scores conforms remarkably well with the coefficient of variation of EFW for our sample (Figure 4), and serves as a reasonable proxy for cross-sectional variation in economic freedom.

Figure 4. Average EFW distance over time.

In bilateral specifications, we use a modified set of covariates to control for bilateral similarity (or distance) along institutional, economic, and demographic, lines. We transform time-varying independent variables and latitudes from equation (1) into absolute pairwise differences, and add dummy variables for common legal origins (French, British or socialist), continent, coloniser and for whether countries in the dyad were the same nation, in a colonial relationship or contiguous. To this, we add (log) geodesic distance (between most populated cities), and weighted religious, linguistic and genetic distances.Footnote ⁸

To further test the robustness of our results, in bilateral models we also employ replicated Hofstede dimensions developed by Beugelsdijk et al. (Reference Beugelsdijk, Maseland and Van Hoorn2015). The replicated dimensions are created from WVS data collected over 1981–2008. The authors assign a Hofstede dimension score each of the two non-overlapping age cohorts within countries: older cohort, born in 1941 on average (in 1902–1958) and younger, born in 1971 on average (after 1958). We exploit the generational variation in Hofstede indices to add a temporal dimension to our dyadic models.

We then re-estimate equation (6) as a two-period (unbalanced) panel with variables observed in 1980 and 2015. The idea is to take two sufficiently distant points in time to which we can assign separate generational Hofstede scores. For 1980, we use the older generations' scores to calculate bilateral cultural distance; for 2015, the younger. Though we recognise this approach will not perfectly capture prevailing cultural distances at different times, it may well provide a reasonable approximation: in 1980, the majority of the population in many countries were likely individuals from the older cohort; by 2015, the majority likely flipped.