Introduction

The scourge of corruptionFootnote 1 has been noted across civilisations, under varying political and economic systems and at all stages of development, for millennia (Ceschel et al., 2022; Klitgaard, 1988; Rose-Ackerman, 1999; Uberti, 2022). A non-exhaustive list of factors that have been theorised to be associated with corruption include poverty, inequality, economic underdevelopment, cultural and demographic factors, different forms of governance structures, historical determinism, a lack of competition, and ethical frameworks associated with individual decision-making and behaviour (see Fisman & Golden, 2017). The role of institutions, as the rules of the game, have featured prominently in research on corruption because they are posited to affect incentive structures and constraints on individual behaviours (for example, Acemoglu & Robinson, 2012; Beesley & Hawkins, 2022; Dang et al., 2023; North et al., 2013; Shleifer & Vishny, 1993; Treisman, 2000; Wu et al., 2023). In this paper we examine the relationship between institutions and levels of corruption, but take it a step further and investigate whether it is the presence of inclusive institutions (rules that promote equal access to opportunities and resources) and/or the credible and consistent implementation of the rules of the game that matter. These two dimensions are generally conflated in corruption research, and we contribute to the literature by showing that theoretically conceptualising and empirically operationalising these as analytically distinct dimensions—isolating the de jure (what is formally documented and recognised) and the de facto (what happens in reality over time)—allows us to capture different facets of how institutions relate to corruption. We argue that the extent of the de facto realisation of institutional ideals that economic agents observe when making choices of actions and strategies matter. But the extent to which the de facto realisation will be observed, will itself depend on the quality of the de jure ideal that has been defined for de facto implementation. Both factors are relevant, and their relationship may not be straightforward. For example, less-than-ideal institutions that are consistently and predictably implemented might be preferable to ambitious standards that are practically unattainable. We use a novel approach to theoretically conceptualise and empirically operationalise institutions along these two dimensions using a panel dataset for 148 countries covering 2012 to 2018.

We make various empirical and theoretical contributions. We have already highlighted our analytical conceptualisation of institutions into two distinct elements: the nature of the institutions (de jure), and the extent to which they are credibly and consistently implemented (de facto). We show that it is not just improvements in the formal rules that are related to decreased corruption, but that their consistent and credible implementation is also related to a country’s ability to effectively manage corruption. We find that when both are present, this relates to lower corruption levels than if either inclusive institutions or credible and consistent implementation are present on their own. We delve deeper into the nature of this relationship and show that these relationships differ across national income levels. The nature of the formal institutions is important across all income levels, and is more so at upper middle income levels, whilst the credible and consistent implementation of the formal rules is primarily important at upper middle income levels. This has significant implications for countries trying to transition from lower to higher income levels. We unpack these relationships and their implications.

Literature Review

Institutions and Corruption

Research examining the role of institutions in accounting for corruption provides alternative mechanisms through which institutions are argued to influence corruption. One school of thought focuses on political institutions (including, but not limited to, written constitutions, the type of government in place [for example, democracy or autocracy], and the power and capacity of the state to regulate and govern society). It argues that these impact how much control and influence citizens have over politicians, and the extent to which politicians can be held accountable. This perspective contends that the nature of the political institutions (whether they are extractive or inclusive)Footnote 2 enables economic institutions that are similarly structured, and that where these are extractive, those in positions of political power are able to structure the economic institutions such that rents can be extracted for the benefit of the elite (Acemoglu & Robinson, 2012). Corruption is argued to be more likely in societies that are characterised as having extractive institutions, where corrupt behaviours are incentivised and rewarded.

Whereas the aforementioned emphasises the primacy of political institutions, a variation on this theme highlights the institutional complex (political, economic and social institutions) and positions these as providing the ‘rules of the game’ (North, 1995) for social, political and economic interactions and transactions. Institutions are theorised to provide the choice sets, opportunities and incentives for actors. Such institutions may be classified on a continuum, with extractive institutions at one end, and those that promote equality, inclusion, impartiality and fairly enforced property rights at the other. Societies characterised as the former may be more likely to have greater amounts of corruption, with the converse more likely at the other end of the continuum. Such arguments are widely used in the corruption literature (for examples in the corruption literature, see Aidt et al., 2008; Chadee et al., 2021; Chowdhury, 2004; Shleifer & Vishny, 1993; Wu et al., 2023).

Institutions that are argued to control corruption better are described as having three key characteristics. First, property rights that are equally and impartially enforced across the society. Second, the elites, politicians and other powerful individuals or groups are constrained from expropriating the property of others or manipulating political and economic markets. Last, the institutions are structured such that there is equal opportunity to participate in economic and political markets, and these markets can accommodate the many varied interests of those in the society (Acemoglu, 2003; Acemoglu & Robinson, 2012; North et al., 2013). In this paper, we follow Acemoglu and Robinson (2012), and refer to institutions structured in this manner as inclusive institutions (with the opposite being extractive institutions). Based on these arguments we therefore hypothesise that:

H1

Countries with inclusive institutions have lower levels of corruption.

Credible and Consistent Implementation of Institutions

The institutional logics view of corruption (see Misangyi et al., 2008) positions both corruption and anti-corruption oriented institutional logics as existing in a particular context—and this context matters. The extent to which either of these logics is lived and experienced by actors depends on the extent to which resources are allocated in support of these logics and the extent to which actors believe that the institutions are credible and their rules consistently applied (Sewell, 1992). The resources being referred to include physical capital, human capital, symbolic influence, and social status (Misangyi et al., 2008; Sewell, 1992), which are hereafter referred to as institutional resources. Where institutions are both credible and the rules consistently applied then those institutions are more likely to be propagated within the context, and to be lived, believed and experienced by actors, and be more effective in constraining corrupt behaviour.

Similar arguments around policy credibility and time-consistency have been made in research focused on monetary, fiscal, trade and environmental policy (Nemet et al., 2017). Such policy work focuses on the potential conflict between policy credibility and policymakers’ discretion to change their decisions over time, as well as the implications of this for policy outcomes (Shepsle, 2019). Common definitions of policy credibility focus on the extent to which ‘beliefs about the current and future course of economic policy are consistent with the program originally announced by policymakers’ (Blackburn & Christensen, 1989, p. 2). Or, more simply put, whether or not deeds match words (Blinder, 2000)—essentially, the extent to which policy commitments are enacted. Such credibility has both a motivational and imperative quality. The former when the policymakers have the motivation and commitment to honour the dictate and there is an alignment of incentives, and it becomes self-reinforcing. They are seen as credible in the imperative sense if policymakers are unable to act otherwise and their discretion to deviate from the original commitment is curtailed. Discretion is seen as generally being in conflict with credibility because arbitrarily changing the policy after the initial announcement may result in actions and outcomes that are different from those originally announced (Shepsle, 2019).

Maintaining motivational and imperative credibility is not a simple exercise because policymakers and stakeholders have numerous incentives and opportunities to attempt to deviate from initial policy commitments. Because actors in a region are heterogeneous and have heterogeneous preferences, policy decisions create winners and losers (Carraro et al., 2012; Jenkins-Smith et al., 2014; Meckling et al., 2015). This creates incentives for actors to manipulate the policymaking process for their own purposes. Additionally, because many policies result in short-run costs and yield potential benefits at a later stage, policymakers are frequently tempted to shift costs associated with such policies to the future. Weitzman (1980) also highlights that any future policy that is contingent on the present state will be anticipated, and this is argued to distort incentives away from those that are aligned to the initially intended optimal outcome. When policymakers act on these incentives, they thereby undermine prior policy decisions and commitments. This may to lead to suboptimal outcomes (Gerlagh & Michielsen, 2015).

This connects to the literature on rent seeking and corruption, which highlights the dangers associated with elites being able to manipulate formal institutions at will for their own benefit and the importance of this being constrained. For example, North and Weingast (1989) note that a government that is powerful enough to protect its borders from internal and external threats, is also powerful enough to compromise property rights and their enforcement to benefit the elite. A political system is therefore not seen as being credibly constrained from rent seeking (and the time-inconsistency problem) until its public agents are constrained from acting as agents of the political elite (Knott & Miller, 2006; North et al., 2013).

The argument is therefore that the policy credibility and time-inconsistency problem is important to acknowledge when considering inclusive institutions and their ability to curtail corruption (as well as for anti-corruption initiatives in general). This positions the credible and consistent implementation (the de facto dimension) of inclusive institutions (the de jure dimension) as a central issue in the relationship between institutions and corruption. Corruption inquiry has generally conflated the de jure and de facto dimensions, focusing more specifically on certain types of institutions and their relationship to corruption. It is therefore hypothesised that:

H2

Countries with inclusive institutions that are credibly and consistently implemented have lower levels of corruption.

It is not just defining, documenting and signing into effect inclusive institutions that matter; their time-consistency and credibility are also important. Furthermore, while either inclusive institutions or credible and consistent implementation may lower corruption levels, it logically follows that both being present simultaneously is likely to lead to lower levels of corruption than if either of the two are present independently. It is therefore hypothesised that:

H3

Countries with inclusive institutions that are credibly and consistently implemented will have lower levels of corruption than if only one of these aspects is present.

The Influence of Economic Development and Income Levels

The modernisation theory hypothesises that economic development underpins inclusive institutions, and so whilst institutions are important, they themselves may be outcomes of processes of economic modernisation. Economic development leads to changes in the values and preferences of those in society, and these changes increase the probability that a society becomes a liberal democracy and develops inclusive institutions (Fukuyama, 2014; Inglehart, 1997; Przeworski & Limongi, 1997; Welzel et al., 2018). As such, both inclusive institutions and lower corruption levels are argued to be related to higher levels of economic development. Earlier versions of the modernisation hypothesis posited a linear relationship, but later works argued that the relationship may be non-linear (Lipset et al., 1993). The non-linear argument maintains that as economic development increases, a society may initially see a deterioration in institutions or no positive institutional progress, but once a certain level of development is reached the probability that the society develops inclusive institutions increases. This is said to continue up until a threshold level, after which an equilibrium point is reached and the relationship flattens out (Lipset et al., 1993).

More recently, corruption scholars have offered similar arguments, namely that the effectiveness of institutions in reducing corruption depends on whether or not a country has surpassed particular levels of development. Jetter et al. (2015), for example, maintain that where a country has not achieved a GDP per capita of greater than approximately USD2000 (in 2005 USD), then opening up political markets to be more inclusive may serve to increase corruption, rather than decrease it. They argue that, in poorer countries, work in the productive sector may not be that lucrative, and therefore opening up political markets may lead to more people engaging in political corruption. They continue that once this level of development has been surpassed, and as the country gets richer, more inclusive political and economic markets are associated with lower levels of corruption. Given our explicit inclusion of the de jure dimension, we contextualise these arguments with ours and hypothesise that:

H4

The strength of the negative relationship between inclusive institutions, their credible and consistent implementation and corruption increases as a country becomes more economically developed.

The aforementioned therefore provides the basis of our theoretical contribution. Whereas the literature has generally focused on the relationship between institutions and corruption, we tease apart the nature of institutions (de jure) from their consistent and credible implementation (de facto), and position these as analytically distinct. We additionally contextualise our arguments with those emphasising the influence of economic development, in order to both account for the influence of economic development, and to investigate how the relationships between the two dimensions and corruption might vary across income levels.

Econometric Model

As the study uses panel data, we begin with the following Pooled Ordinary Least Squares (OLS) econometric specification as a baseline:

$${CPI}_{it}= \alpha + {\beta }_{1}{FORM\_INST}_{it}+ {\beta }_{2}{INST\_DUR}_{it}+ {\beta }_{3}{LNGDP}_{it }+ {\varepsilon }_{it}$$
(1)

where CPI is the level of corruption, and \(\alpha\) is the pooled intercept term. FORM_INST is the extent to which the inclusive institutions are established. INST_DUR is a durability score of the institutions present in the region, and focuses on the extent to which the formal institutions are credibly and consistently implemented, such that the primary nature of the institutions is stable. LNGDP is the natural log of GDP per capita. GDP per capita is included to take account of theoretical arguments that posit that economic development driven by modernisation may be triggering the changes in institutions that influence the corruption levels in a country. Lastly, ɛ is the error term that captures any residual variation. The subscript it indicates that each variable represents the observation for country i at time t.

One concern immediately evident from the above specification is the omitted variable bias. This results in concerns that cross-sectionally heterogeneous but time invariant (unobserved effects) and time variant but cross-sectionally invariant (time effects) related sources of variation are not explicitly catered for in the specification. We deal with these concerns by including robustness tests in the form of additional estimations that explicitly control for these aspects in several ways. First, unobserved effects are controlled for by carrying out a within-group estimation. Secondly, estimations are undertaken where time effects are explicitly controlled for. Lastly, further estimations are conducted as additional robustness tests that include several other factors that are hypothesised to explain corruption.

H3 is tested by including an interaction term of the formal institutions on the institutional durability variable. This is to ascertain if these two aspects together lower corruption more than if each is present in isolation.

To test H4 and to establish whether the relationship is non-linear, where inclusive institutions and their credible and consistent implementation may work differently at different country income levels, this specification is extended to:

$${CPI}_{it}= {\alpha }_{i}+ {\beta }_{1}{FORM\_INST}_{it}+ {\beta }_{2}{INST\_DUR}_{it}+\Sigma {\beta }_{4j}{INC\_CAT}_{jit}+\Sigma ({\beta }_{5j}{INC\_CAT}_{jit}. {FORM\_INST}_{it})+\Sigma ({\beta }_{6j}{INC\_CAT}_{itj}. {INST\_DUR}_{it}) + {\varepsilon }_{it}$$
(2)

Where \(INC\_CAT\) is a categorical variable that indicates the income category, j, of country i at time t. The j categories include low, lower middle, upper middle, and high income. Each category is included as a dummy variable in the specification to control for each income level. In doing so, the influence of formal institutions and institutional durability can be assessed in each income level category.

Estimation Methods

As our specification is in pooled form, it assumes homogeneity of cross-sectional units, which therefore leads to homogeneous intercepts and slope coefficients for each country. The countries in the panel may contain country-specific factors that lead to cross-sectional heterogeneity, and therefore the above specification is extended to include country-specific unobserved effects. Time effects (both with zero and non-zero restrictions on time-effects) are also included. To do so, we use the fixed effects within estimator, which takes the form of:

$${Y}_{it}= {\beta }_{0} + {\alpha }_{i}+ {\beta }_{1}{X}_{it} + \lambda t + {\varepsilon }_{it}$$
(3)

where \({\beta }_{0}\) is the intercept term, and \({\alpha }_{i}\) represents country-specific unobserved effects which are not directly estimated, but are instead removed by within-group demeaning of the above econometric model. X is a vector of explanatory variables (those detailed in the “Econometric Model" Section), and \({\beta }_{1}\) is a vector of coefficients of X. Lastly, \(\lambda t\) represents the explicit inclusion of time effects, where \(\lambda\) is the coefficient of each time period, \(t\).

The econometric specification has potential for endogeneity. Theoretical arguments have posited that economic development may affect corruption, that corruption may affect economic development, and that corruption may affect the institutions themselves (for example, see Serra, 2006; Treisman, 2000). Additionally, arguments in the corruption literature posit that corruption being commonplace in a society can lead to more corruption (Barr & Serra, 2010; Ivanyna et al., 2016; Robert & Arnab, 2013; Schulze & Frank, 2003) resulting in potential autocorrelation.

To control for the above issues, we carried out a two-stage systems Generalised Method of Moments (GMM) dynamic panel estimation (Arellano & Bover, 1995; Blundell & Bond, 1998; Roodman, 2009). The systems GMM estimation controls for autocorrelation through the explicit inclusion of lags of the dependent variable as an explanatory variable. There is an assumption of exogeneity as lags of dependent and independent variables are used as instruments of potentially endogenous ones. The systems GMM estimator is noted to control for omitted variable bias, unobserved panel heterogeneity, and potential measurement errors through the lags and differences that are utilised as instruments (Arellano & Bover, 1995; Blundell & Bond, 1998; Roodman, 2009).Footnote 3

We recognise that, in addition to time-invariant country-specific unobserved effects, there may also be random effects with between country variation. We therefore employ a mixed model to account for both fixed and random effects. This mixed effects model takes the form of the following maximum likelihood estimation:

$${Y}_{it}= {\beta }_{0} + {\alpha }_{i}+ {\beta }_{1}{X}_{jit} + \lambda t+ {\alpha }_{i}{X}_{jit}\theta + \lambda t{X}_{jit}\phi + {\varepsilon }_{it}$$
(4)

where \(\theta\) allows for a random slope on \({X}_{jit}\), which varies depending on the country, i, while \(\phi\) allows for a random slope on \({X}_{jit}\), which varies depending on the time period, t.

Lastly, in order to investigate H3, Eq. 3 (fixed effects within specification) is extended in order to ascertain whether the formal institutions interact with institutional durability to reduce corruption, or whether these variables have independent influences:

$${CPI}_{it}= {\beta }_{0} + {\alpha }_{i}+ {\beta }_{1}{FORM\_INST}_{it}+ {\beta }_{2}{INST\_DUR}_{it}+ {\beta }_{3}({FORM\_INST}_{it}.{INST\_DUR}_{it}) + {\beta }_{4}{LNGDP}_{it }+ \lambda t + {\varepsilon }_{it}$$
(5)

Data

Dependent Variable

By its covert nature, corruption is hard to measure because the phenomenon is concealed by the perpetrators. Despite various measures of corruption being available, each has limitations because we cannot directly observe and objectively measure all instances of corruption in a country. In this study, Transparency International’s (2021) Corruption Perception Index (CPI) is used. It is widely used in the literature as it is considered to be one of the better measures available in approximating actual corruption (see Lambsdorff, 2007; Wilhelm, 2002), and provides the benefits of good cross-country and time-series coverage. The CPI scores countries on a scale from 0 to 100, where a higher score indicates less corruption. The measure is formulated as a poll-of-polls, where multiple surveys that are carried out on a sample of citizens, risk analysts, and experts in countries around the world are aggregated into a composite measure (Serra, 2006).

Independent Variables

In order to model the formal institutions present in a region, data from the Freedom House (Freedom House, 2021) and Polity V (Centre for Systemic Peace, 2021) datasets were used. From the Freedom House dataset, we used variables underlying the civil and political rights measures. These variables indicate the real-world rights and freedoms that individuals enjoy and whether the country’s institutions can be classified as inclusive. From the Polity V dataset, three variables were used: polity2, which indicates the extent to which a country can be considered a democracy or autocracy; durable, which provides a count of the number of years that a regime type (and its institutions) has consistently been in place; and constraint on the executive, which indicates the extent to which the decision-making powers of a country’s executive are constrained to prevent abuses of power. Table 1 provides summary statistics of the variables,Footnote 4 and more detailed descriptions of the data and its sources are provided in Table 2. Table 3 presents the results of a Pearson correlation analysis of the aforementioned variables. With the exception of constraint on the executive (XCON) and polity (POL), the CPI has a correlation of greater than 0.5 with all of the institutional variables. One point of concern is the correlations of the institutional variables with each other (that are greater than 0.75), which may lead to issues of multicollinearity which may bias the standard errors of the estimations. We address this through a principal components analysis.

Table 1 Descriptive statistics
Full size table
Table 2 Freedom house and polity V variables
Full size table
Table 3 Pearson correlation matrix
Full size table

As the hypotheses are focused on the strength of inclusive institutions as a single analytical construct and not on individual institutional aspects, a factor analysis was undertaken on the institutional variables in order to create principal components for use in the estimations. We included electoral process (EP), political pluralism and participation (POLP), functioning of government (FUNC), freedom of expression and belief (FE), associational and organisational rights (AOR), rule of law (RULE), personal autonomy and individual rights (PAIR), polity (POL), durable (DUR), and constraint on the executive (XCON) in the factor analysis. Durable, while not highly correlated with the other institutional variables, was included because it is related to the level of stability of the institutional variables. The results of the factor analysis are shown in Table 4, and summary statistics of the principal components are shown in Table 5.

Table 4 Principal component factor analysis (orthogonal varimax rotation)
Full size table
Table 5 Descriptive statistics
Full size table

This resulted in two factors with eigenvalues over the recommended threshold of 1. Factor 1 is correlated significantly with the institutional variables that relate to the type of formal institutions in place. Factor 2 is most strongly correlated with durable, and less so with functioning of government, rule of law, and personal autonomy and individual rights. This second factor can therefore be argued to capture whether or not the policies in place are implemented consistently for longer time periods (durable), whether the public are able to see the policy decisions and actions of government officials (functioning of government), and whether there are channels through which discrepancies can be raised and addressed (rule of law). The way in which the variables load into Factor 2 therefore represents the consistent and credible implementation of the formal institutions (named as institutional durability, represented by INST_DUR), while Factor 1 can be seen as a composite measure of the nature of the formal institutions themselves (named as formal institutions, represented by FORM_INST).

This novel method of empirically operationalising the nature of institutions as being analytically distinct from their credible and consistent implementation allows us to investigate the hypotheses developed in the “Literature Review" Section. The formal institutions factor effectively provides a 3.866 unit scale with inclusive institutions at the top of the scale, whilst the credible and consistent implementation factor has a 5.993 unit scale, where greater values represent higher levels of credible and consisted implementation.

Figure 1 illustrates the relationships between FORM_INST and INST_DUR and the CPI and GDP per capita. In all figures, the relationships appear to be positive and potentially non-linear, with FORM_INST showing a concave shape, while INST_DUR shows a convex shape with both the CPI and GDP per capita.Footnote 5 This may indicate that the two factors function differently at different levels of income (as we have hypothesised).

Fig. 1
figure 1

Principal components’ relationships with CPI and GDP per capita

Full size image

Unit Roots and Panel Cointegration Testing

As panel data are being used, Im et al. (2003) (IPS) unit roots were tested to ascertain the stationarity and order of integration of the variables. The test requires a balanced panel to run and given that our panel is unbalanced, we ran this test on an interpolated and extrapolated version of the dataset.Footnote 6 Additionally, for further robustness, we also carried out the IPS tests on only those countries with no missing observations.Footnote 7 We additionally used the Fisher augmented Dickey–Fuller (ADF) unit root test (Fisher 1992), which can be run on unbalanced panels (Maddala & Wu, 1999).

The results of the unit root tests for all variables are presented in Table 6. CPI is stationary at level and first difference across the IPS and Fisher ADF tests. GDP per capita is stationary at level according to the Fisher ADF, and at first difference across all tests. While the interpolated and extrapolated version of the IPS test was unable to run on the two principal components at level, it showed that both are stationary at first difference. The full panel version of the IPS test shows that the institutional durability factor is stationary at level and first difference, while the formal institutions factor is stationary only at first difference. The Fisher ADF test indicates that the formal institutions and institutional durability factors are stationary at level and first difference. We explain how we address this below.

Table 6 Unit root tests
Full size table

In addition to testing for unit roots, we tested for panel cointegration which provides insight into the long-run relationship between variables. When variables are cointegrated, they share a long run relation (Granger, 1981). In such cases, results would not be spurious, despite unit roots being present. We used the panel cointegration tests of Pedroni (2004) and Kao (1999), and the results are shown in Table 7. The null hypothesis of both tests is that the dependent and independent variables are not cointegrated. In both cases, the null hypothesis can be rejected. It can therefore be concluded that the variables are cointegrated, and econometric specifications at the levels form would not be spurious because of issues relating to cointegration and/or stationarity.

Table 7 Panel cointegration tests
Full size table

As regards the issues concerning stationarity, Table 5 shows that the specification may be unbalanced, where a stationary dependent variable is being (at least partly) explained by one (or more) non-stationary one(s). Scholars argue that the violation of classical assumptions and problems arising from issues of stationarity are associated more with large T and small N panels (e.g., Wooldridge, 2015, p. 440). As the panel has large N and small T dimensions, issues relating to stationarity should therefore be of less concern. However, we take a more cautious approach, and add two additional robustness checks to ensure that the results at levels are not spurious. First, as all the four variables are stationary at first difference, we undertook an additional first differenced fixed effects within regression. Secondly, additional Fisher ADF unit root tests were carried out to assess stationarity when a time trend is included. The results are presented in Table 8, and show that once time effects are included, all variables are stationary at level. We therefore also confirm the robustness of results by controlling for time effects in the fixed effects within, systems GMM, and mixed effects maximum likelihood estimations.

Table 8 Fisher ADF unit root tests with time trend
Full size table

Our approach therefore provides an effective and novel empirical operationalisation strategy that separates the nature of institutions from their credible and consistent implementation and allows each dimension’s individual and combined influence to be tested. It additionally controls for multiple issues that are known to bias results, and therefore provides a robust manner of testing the hypotheses developed.

Results

Investigating H1 and H2: The Influence of Inclusive Institutions and Their Credible and Consistent Implementation

The results of the pooled OLS and fixed effects within estimations are presented in Table 9. Both the Pooled OLS (regression 1 in Table 9) and fixed effects within (regression 2) estimation results support H1 and H2 because the coefficient values of formal institutions (FORM_INST) and institutional durability (INST_DUR) are > 0 and are statistically significant at the 1% level. Comparing the coefficient sizes of both regressions shows that when country-specific unobserved effects are controlled for, the influence of formal institutions and institutional durability reduces, whilst that of GDP per capita (LNGDP) increases. This shift in coefficient sizes illustrates that country-specific unobserved effects do need to be controlled for. The results of the fixed effects within regression show that, on average, for a one-unit increase in the formal institutions, the CPI would increase by 5.183 units, and for a one-unit increase in institutional durability, the CPI would increase by 2.32 units. The coefficient of GDP per capita has the greatest magnitude, where a 1% increase leads to a 5.535 unit increase in the CPI. To put this into perspective, a 6-unit move on the CPI is roughly half the difference in CPI scores between the Czech Republic (a country in which corruption is considered more widespread) to the United States of America (ranked as the 27th least corrupt country in the world in 2021). It is also half of the decline in Hungary’s CPI between 2012 and 2022—a period anticorruption agencies note to involve the ‘erosion of the rule of law, state capture, and the development of systemic corruption’ that saw Hungary shift from the 19th least corrupt country of the then EU-27 to the most corrupt (Transparency International, 2023, p. 5). To contextualise a 1 unit move on the formal institutions (FORM_INST) scale, the noted decline in Turkey’s formal institutions between 2014 and 2018 is approximately 1 unit. Political rights advocates have noted that this period, following an attempted military coup, involved a ‘deeply flawed constitutional referendum’ that centralised presidential power, forcing elected mayors to resign and replacing them with government appointees, arbitrary prosecutions of rights activists, and harassment, intimidation, attacks and arbitrary prosecutions aimed at opposition candidates (Freedom House, 2019).

Table 9 Fixed effects, systems GMM, and mixed effects regression results
Full size table

Robustness Tests Relating to the Testing of H1 and H2

The results of the systems GMM regression (regression 3 in Table 9), which we run primarily to address endogeneity concerns, show consistent coefficient signs and significance levels to those in the pooled OLS and fixed effects within estimations. However, the coefficient size of both the formal institutions and institutional durability decreases. The coefficient size of 0.913 on the first lag of the CPI is statistically significant. This indicates that the existing level of corruption is influential, and highlights the persistence of corruption as a phenomenon, where the level of corruption at time t-1 is, on average, almost able to explain the corruption at time t on a unit-for-unit basis, ceteris paribus. These results suggest that the impact of endogeneity bias on standard estimators is to overstate the influence of institutions, as correcting for this, through the use of the systems GMM estimator, shows a more moderate influence. However, the statistical significance of the variables is robust to controlling for endogeneity bias, thereby suggesting that any estimated impact of formal institutions and their credible and consistent implementation is not purely an artificial or spurious result of endogeneity bias. This increases our confidence in the robustness of the influence of both institutional dimensions that we control for.

When allowing for heterogeneity across intercepts, and slope coefficients, and controlling for time-effects in the mixed effects maximum likelihood regression (regression 4 in Table 9), the coefficient signs and significance levels of the variables related to H1 and H2 remain consistent with the three estimations explored thus far, providing further support for these hypotheses.

To ascertain whether the results may be biased by any potential ambiguities in the order of integration of the variables, we carried out a first differences regression as all of the dependent and independent variables are stationary at first difference. The results of the first differences regression (regression 5 in Table 9) also show similar coefficient signs and significance levels.

As the null hypothesis of the Fisher ADF was rejected for all variables when a time trend was controlled for, all variables are stationary at levels under this condition. In the regressions carried out thus far, time effects are controlled for in the dynamic Systems GMM regression (regression 3), and mixed effects maximum likelihood regression (regression 4). An additional fixed effects within estimation that includes time effects was also carried out (regression 6). These results are also consistent with the other estimations in terms of coefficient signs, significance levels and coefficient magnitudes, which supports the robustness of the results.

Lastly, to confirm that the results are robust to the inclusion of other variables that have been argued to explain corruption, we carried out a Random Effects Generalised Least Squares (GLS)Footnote 8 estimation. This, in addition to the primary explanatory variables, included the Human Capital Index (HCI), unemployment (UNEMP), a measure of ethno-linguistic fractionalisation (ELF), the percentage of the country that ascribes to the Muslim (PERC_MUS) and Protestant (PERC_PROT) religions, a categorical variable for the legal origins of the country, and a second categorical variable indicating the colonial origins of the country.Footnote 9 As can be seen in Table 10, the results remain robust when omitted variables are controlled for in this manner.

Table 10 Random effects GLS estimation
Full size table

Investigating H3: Whether the Influence of Inclusive Institutions and Their Credibility and Consistent Implementation is Greater When Combined or Independent

To investigate H3, we carried out a fixed effects within estimation that helps ascertain whether formal institutions and institutional durability interact in a manner that results in a non-linear relationship to corruption in a country. The results of this regression are shown in Table 11. The results support H3, where the interaction term is significant at the 10% level and the coefficient is > 0, which indicates that the results of each variable depend on the value of the other. As in prior regressions, the results indicate that higher levels of both institutional dimensions are independently associated with lower corruption. Additionally, the interaction term coefficient indicates that the higher the credible and consistent implementation of institutions, the stronger the impact of formal institutions. Symmetrically, the more formal institutions are inclusive in nature, the stronger the impact of their credible and consistent implementation. The influence of this interaction term was also explored across the range of the formal institutional variable to ascertain whether this relationship is more influential at certain levels of formal institutions. Dummy variable specifications were explored where the formal institutional variable was split into deciles, quarters and thirds, and the results of these specifications did not add substantive insights to the investigation. Therefore, this result appears to be true on average across the sample and does not work differently across different parts of the sample.

Table 11 Fixed effects (within) regression with interaction term
Full size table

Investigating H4: Whether Formal Institutions and Their Credible and Consistent Implementation Operate Differently at Different Levels of Development

Recall that Fig. 1 illustrating the relationships between formal institutions and institutional durability with the CPI and GDP per capita revealed non-linear associations. To explore whether the hypothesised relationships hold across the various levels of economic development, and whether the dynamics of the variables change as income levels rise, dummy variables were included based on the World Bank’s (2021) four-stage income level classification. This classification categorises countries in terms of whether the country falls into the low (variable name: inc_l), lower middle (variable name: inc_lm), upper middle (variable name: inc_um), or high (variable name: inc_h) income categories. The lower middle income category was used as the reference category. We then included interaction terms between the remaining income level dummy variables and the three variables included in the base specification. The results of the regression are shown in Table 12, and graphs of the changes in coefficients and significance across income levels are illustrated in Fig. 2.

Table 12 Fixed effects within estimation controlling for income levels
Full size table
Fig. 2
figure 2

Changes in slope coefficients and significance across income categories

Full size image

Allowing for interactions between formal institutions and institutional durability and levels of development, confirms the statistical significance of formal institutions. Moreover, the impact of formal institutions is particularly important substantively and statistically for the upper middle-income level of development (positive and significant coefficient of 4.061 on formal institutions, and the positive and significant coefficient on the interaction term of 2.285, providing a net formal institutions impact of 6.346 for upper middle income countries). By contrast, for less developed and highly developed income groupings there is no differential in the impact of formal institutions among the development levels (no significant interaction terms), though formal institutions remain important (positive and statistically significant formal institutions coefficient of 4.061). The implication is that ‘escaping’ lower middle-income status may require particularly close attention being paid to the quality of formal institutions for purposes of controlling corruption.

The behaviour of the institutional durability variable is distinct from that of formal institutions. The implication of the institutional durability variable and its interaction with the levels of development controls is that credible and consistent implementation of institutions is important substantively and statistically for upper middle-income countries, while for less developed, lower middle-income and high income countries this does not hold—see the statistically insignificant institutional durability variable, and its interactions with the less developed and high income group controls, contrasted with the significance of the interaction of the variable with the upper middle-income (at the 1% level) control (coefficient value of 5.269).

Collectively, this set of results confirms our general point that formal institutions, and the credible and consistent implementation of formal institutions are not only conceptually distinct, but have a differential impact on corruption, with the two dimensions of institutions coming to exercise a distinct impact at different levels of development.

Prior research has generally conflated the de jure and de facto dimensions of institutions and has emphasised institutions as a single analytical construct (see Fedderke et al., 2001; Brites Pereira & Luiz, 2020 for exceptions). Our findings show that the de facto dimension appears influential on its own, and that the extent to which the formal rules relate to corruption depends on the extent to which they are credibly and consistently implemented.Footnote 10 Furthermore, the de jure and de facto dimensions function differently across income levels. These results provide valuable insights that highlight the value of explicitly including the credible and consistent implementation dimension in institutional analysis. We discuss these findings and implications further in the next section.

Discussion and Conclusion

Our findings indicate that both inclusive institutions and the credible and consistent implementation of institutions matter as regards controlling corruption. Furthermore, the effect is stronger when both are present simultaneously. The results therefore support the notion that credible and consistent implementation is an important factor to consider when formulating and implementing anti-corruption policies, as well as policies aimed at establishing and strengthening inclusive institutions. It would therefore be beneficial for issues of institutional credibility and time-inconsistency to be taken into account by institutional scholars, and by policymakers and policy researchers when formulating anti-corruption policies and institutions.

In addition to the insights generated from operationalising institutions as de jure and de facto dimensions, our results also show that these aspects operate differently at different country income levels. The nature of the formal institutions is important across income levels, but is particularly important at the upper middle income level, whilst the credible and consistent implementation of institutions is highly influential during the transition through the upper middle income level. This illustrates the distinct nature of these institutional dimensions, and emphasises that addressing issues of time-inconsistency is critically important, particularly during the process of transitioning from lower to higher income levels.

The systems GMM estimation results highlight the ‘stickiness’ of corruption once it is institutionalised in a region, where the corruption present at time t-1 is almost able to fully explain the corruption present at time t. This emphasises that, while formal institutions and their credible and consistent implementation may have a noteworthy effect, there will likely be cumulative effects based on the level of corruption already present in the society. This also emphasises the importance of credible and consistent implementation of institutions, as the results achieved in one period appear to influence the results achieved in the next. It demonstrates the importance of rooting out corruption as quickly as possible to break the cycle.

Theoretically, this paper contributes to the literature by providing a more nuanced understanding of how institutions matter as regards corruption, and we demonstrate that it is not just the nature of the institutions (inclusive or extractive) but their credible and consistent implementation also matters. Operationalising these two analytically distinct institutional dimensions and the de jure versus de facto distinctions leads to the insights we have shown in this paper, and appears to be a promising avenue for further research on institutions and corruption, and in institutional research more generally.

One response to the difficulty of separating the de jure from the de facto dimensions of institutions is that it may not matter. For economic agents the technical distinction between formal existence and practical implementation may be irrelevant, since all that matters is the ‘net’ outcome of the combination of the formal rules and their implementation in creating the environment in which they exercise actions and strategic choices. Insofar an analytical interest is in the proximate drivers of action and strategy, this may be true. But it does not satisfy an analytical interest in the functioning of institutions. While the extent of the de facto realisation of institutional ideals matter to economic agents when deciding on actions and strategies, the likelihood of these ideals being effectively realised will depend on the quality of the de jure ideals. Formal de jure institutions can be thought of as an ideal benchmark to which an appeal can be made by agents acting within their jurisdiction in the event that they perceive a shortfall of practice relative to the de jure ideal. Such appeals against ideal de jure principle themselves rely on institutions (the typical recourse might be the legal system, such as courts) which come to determine both the extent to which the ideal of the de jure standard will be realized de facto, and indeed the transactions cost intensity of any such realization. Both plausibly matter—and not necessarily in any straightforward one-to-one sense. For instance, less-than-ideal de jure institutions that are sure to be implemented rigorously and predictably, may be preferrable to grandly ambitious de jure standards that have no hope of practical de facto implementation.

The dependence of the ‘translation’ of the de jure into the de facto dimension of institutions on the quality of the de jure institutions, its potential fragility, and the attendant uncertainties that result are readily illustrated. The South African constitution is crystal clear as to the unacceptability of corruption, but somewhat more ambiguous as to the precise institutions responsible for the implementation of anti-corruption measures. The result is that the pursuit of criminal charges against those allegedly involved in corruption and state capture has been slow, with the processes subject to apparent reversals and setbacks, and incomplete implementation of convictions that have been realised. Since the South African constitution is young, the translation of an impeccable formal de jure institutional structure is still in the process of being defined in de facto terms, with a final outcome that is yet to be determined through, in this instance, case law. Yet such uncertainty is not purely a function of the youth of de jure institutions. The USA has one of the oldest formal constitutions in the world. Nonetheless, the current legal processes surrounding former President Trump illustrate that many of the de facto implications of the de jure rules of the US constitution remain open to contestation, even close to a quarter of a millennium after the drafting of the constitution.

For these reasons exploring the importance of de jure as well as de facto institutions, and how they interact, is relevant and worthy of closer analytical examination.

As regards implications, our research points to the importance of credible and consistent implementation of anticorruption institutions, and by implication that planning for anticorruption initiatives should provide sufficient institutional resources in support of these initiatives over time. This appears particularly important as a country transitions from lower to higher income levels. This connects to prior research that has found that such transition paths are highly unstable, notably in the middle income range, and that socio-political and institutional factors are especially important in these stages (Przeworski & Limongi, 1997). Fukuyama (2014) argues that attempting to move from lower to higher income levels triggers pressures for social and institutional change and higher levels of volatility and uncertainty. Institutions may come under pressure as they attempt to adjust and accommodate new actors with the possibility of institutional decay and clientelism (Luiz, 2019). At these junctures when institutions are under pressure, the credibility and consistency of institutions are vital.

The case of Rwanda is illustrative. Following the genocide in the mid-1990s, Rwanda’s corruption was considered to be equivalent to its sub-Saharan African neighbours. But from the early 2000’s the country adopted a very consistent and strict anti-corruption institutional logic that was supported by institutional resources to the point of institutionalisation. Paul Kagame’s presidency focused on rebuilding the state, promoting economic development, and on the ideal of turning Rwanda into the Singapore of Africa. This was accompanied by an anti-corruption agenda (Rotberg, 2017). Kagame ran education campaigns to educate people to reject and report corrupt practices, fired members of his cabinet and other leaders for corruption and theft (Chêne & Hodess, 2008), enforced existing legislation against corruption, and enacted new laws that banned forms of corruption that were not covered by existing laws. Strict codes of conduct were enforced and the disclosure of assets by public officials was mandated. An ombudsman was appointed to examine these disclosures. Regulations were simplified and made clearer, which reduced the discretion that officials had—and thereby reduced the opportunities for bribery (Rotberg, 2017). Several institutions were established to help enhance transparency and accountability. These included the Internal Audit and Integrity Department of the Rwanda Revenue Authority, the Auditor General’s Office, and the Rwanda Public Procurement Authority (Chêne & Mann, 2011). Kagame mobilised significant physical capital, human capital, symbolic influence, and social status in support of the anti-corruption institutional logic and was able propagate this to the point of institutionalisation. Rwanda’s CPI rank saw a marked improvement during this period—from being ranked as poorly as the 121st (out of 180) least corrupt country to 50th by 2012—and has maintained a stable score since then. Such cases emphasise the importance of getting institutions right and in ensuring that they are credible and consistently implemented if anticorruption measures are to be convincing.

Theodore Roosevelt once said, ‘Rhetoric is a poor substitute for action… If we are really to be a great nation, we must not merely talk; we must act big.’ The same appears to be true for anticorruption-oriented institutions. To be effective, they not only need to be present but must be credible, resourced, and must be seen to be working without fear or favour.

Limitations and Recommendations for Future Research

Measuring corruption is challenging. We used Transparency International’s CPI, and despite the limitations of this measure it is considered one of the better and more widely used measures in the literature. Nevertheless, it has limitations. Likewise, measuring institutions has also been noted to be imperfect (Glaeser et al., 2004). This issue is not unique to this research but applies to the broader literature, and we have partially compensated for this by using various measures of institutions and employing extensive robustness checks. Given these data limitations, future empirical research could be complemented with more in-depth case study and ethnographic approaches. Our research utilised more aggregated measures, and future research could aim to understand the influence of the underlying institutional dimensions and how they can be deliberately targeted in strategic anticorruption initiatives. This might add additional insights as regards which dimensions may be more critical to support through institutional resource allocation. Further studies could expand on the aforementioned to identify which institutional actors’ (that are associated with the institutional dimensions) agency might matter more than others. Future research could also explore longitudinal country specific case studies, tracking the process of allocating institutional resources and the impact of this on institutional development and on corruption, to help further our understanding of the mechanisms through which the institutional influence takes place.