What drives variation in the implementation gap in African Anti-Corruption Efforts?


What role do formal institutions in African countries play in attenuating and controlling the rampant levels of corruption that have hampered development over the past 70 years? Scholarship and policy have not always seen eye-to-eye in determining paths toward reduced corruption and better governance. While policymakers have generally (and unsurprisingly) focused on the impact that laws and policies can have, scholars have cast a much more skeptical eye toward the role that codified anti-corruption efforts play.

Scholars have long underscored the importance that informal institutions have played in the development and maintenance of African states[i]. Bratton and Van de Walle’s seminal work on African neopatrimonialism[ii] launched a fertile field of study[iii] of the role that informal patronal linkages play in driving state development and encouraging corrupt behavior. Indeed, the failure of the African Union to make meaningful inroads in reducing corruption in its constituent states calls into doubt the effectiveness of anti-corruption laws.

The misalignment between policymakers and scholars reveals analytical territory that can be further explored. This blog serves as an entry-point into that endeavor. Here at Global Integrity, we are particularly interested in the distance between de jure laws and de facto reality. As such, we have structured our dataset to allow for meaningful 1:1 comparison between these two. The recent release of the fifth round of Global Integrity’s Africa Integrity Indicators provides enough cases (216) to allow for an entry-point into truly analyzing the distance between de jure and de facto, which we refer to as the ‘implementation gap’.

This blog post presents initial findings about whether formal laws and policies are meaningful in controlling corruption in African governments. Drawing from the perspective of policymakers, I present the alternative hypothesis that formal anti-corruption laws are correlated with lower levels of corruption in practice. In order to test this, I employ a mixed-effects ordinal response model with random intercepts at the country level. I also run a standard mixed-effects generalized linear model as a simple robustness check to ensure that the ordinal response model is working properly.

The Data

In order to answer this question, I rely on the recently-updated Africa Integrity Indicators dataset, which was recently appended with a fifth round of data. The dataset covers each African country[iv] over the course of a five-year period, from 2012 to 2016. The total number of observations is 270, and there are over 100 governance-oriented variables, built on both de jure indicators about the presence of laws, as well as de facto indicators that measure on-the-ground governmental performance and behavior. In addition to these Global-Integrity based variables, the dataset includes data from the World Bank[v] concerning country-level economic factors, such as GDP, export rents on resources, population, and access to electricity. Drawing from the UCDP data[vi] project, each observation was coded to account for the present of conflict within the past five years.

There are several challenges inherent with the data that might complicate the analysis. First, the dataset is relatively small, with only 270 observations total. Due to some missing data from the World Bank, most analyses are able to analyze between 205 and 252 observations at any given time. This has a deleterious effect on the size of standard errors, and may possibly drive Type II errors. Second, Global Integrity shifted its methodology slightly between Round 1 and Round 2, which means that the first set of data is not completely comparable to Rounds 2 through 5. In order to address this, I run regressions with both the full set as well as a limited sub-set that does not include Round 1.


The first model analyzed is a mixed-effects cumulative linked ordinal response model. Because the Africa Integrity Indicators data are recorded in ordinal categories, this model is both statistically and theoretically appropriate. Three models are presented. The first model regresses the covariates on the implementation of anti-corruption investigations by the government. The second model regresses the covariates on the general efficacy of the anti-corruption agency. The third model regresses the covariates on the measured independence of appointments to the anti-corruption body. Each model is run with random intercepts at the country-level. Random slopes are not included due to the small number of observations.

The results, presented in Table 1, demonstrate tentative support for the hypothesis that de jure anti-corruption legislation is correlated with improved anti-corruption efforts by the government. The first model indicates a positive, but statistically insignificant correlation. The second model indicates that the present of in-law anti-corruption legislation is correlated, on average and holding all else constant, with a moderate improvement in the efficacy of anti-corruption efforts. The third model indicates that, on average and holding all else constant, the presence of in-law anti-corruption legislation is strongly correlated with the appointment of independent members to the anti-corruption authorities.

Table 1 – Cumulative Linked Ordinal Response Model

In Law Corruption 0.479 0.838* 2.421***
Petroleum Rents -0.019 -0.031 -0.066*
Nat. Gas Rents -0.125 -0.390 0.375
Mineral Rents -0.011 -0.015 -0.089**
Conflict Five Years -0.704* -0.981 -1.942*
Log of GDP 0.360 0.374 0.604
Log of Population -0.188 0.078 -0.187
Observations 252 252 205
0 | 25 4.361** 8.685*** 10.499*
25 | 50 5.649** 10.339*** 12.310**
50 | 75 7.167** 12.775*** 14.644***
75 | 100 8.216** 14.753*** 15.807***


The complexity of this correlation is made more clear once the coefficients from the ordinal model are translated into predicted probabilities. The plots below indicate relative probabilities that, given either the absence or presence of a law, we will observe a certain outcome. Across the board, the absence of anti-corruption laws predicts increased probabilities that a country’s in-practice anti-corruption efforts will achieve a score of 0 or 25. This relationship becomes more ambiguous as we examine predicted probabilities for higher rankings. On average, and holding all else constant, countries without anti-corruption laws are more likely to achieve scores of 50. This reverses for scores of 75, in which the presence of anti-corruption laws is correlated with a higher likelihood of attaining a ranking of 75 than countries with no such laws. Interestingly, there is a general convergence in predicted probabilities when examining the highest score, which is 100. In the first two models, the predictions seem to converge. In the third model, pertaining to appointments, countries without anti-corruption laws have a higher likelihood of scoring 100 than countries with anti-corruption laws.

Figure 1 – Anti Corruption Investigations

Investigation Probabilities

Figure 2 – Anti Corruption Effectiveness

Effectiveness Probabilities

Figure 3 – Anti Corruption Appointments

Appointments Probabilities

This relationship emphasizes the importance of understanding the multicausal and likely interactive relationship between inputs and outcomes. To ensure the robustness of these results, I run the same data through a generalized linear model with random slopes and intercepts at the country level. The results, which are found in Table 2, provide confirmatory support to the original model. Across the board, the presence of codified anti-corruption laws is correlated with higher overall anti-corruption performance.

Table 2 – Generalized Linear Model

In Law Corruption 6.864* 9.211** 26.919***
Petroleum Rents -0.288 -0.291* -0.478**
Nat. Gas Rents -2.080 -4.161 3.166
Mineral Rents -0.129 -0.152 -0.672***
Conflict Five Years -10.167** -9.355 -19.661**
Log of GDP 5.720* 4.040* 5.363*
Log of Population -2.981 0.577 -0.702
Constant -51.217 -71.882** -89.159*
Observations 252 252 205
Note: *p<0.1; **p<0.05; ***p<0.01

Figure 4 – Predicted Scores

Predicted Scores LMER

The results indicate that the implementation gap is largest with regard to the appointment of officials to the government anti-corruption agency. These predictions hold constant a number of covariates, including GDP, population, the presence of conflict, and levels of natural resource rents.


What can we take from this initial examination of the implementation gap? Perhaps most excitingly, we find that the presence of formal, codified anti-corruption laws are positively and statistically significantly correlated with higher anti-corruption performance. This is a promising finding, as it indicates that the type of government that is likely to enact such laws is also likely to step up its efforts in enforcing anti-corruption programs. There are reasons to be hesitant about this finding. First, it is important to note that this is a correlational relationship rather than a causal one. These results show us clusters of governance rather than delineating a clear, causal link. Likely, factor analysis would find that the presence of some laws informs us about the presence (or absence) of other laws, as well as government performance in many governance-related areas. Second, the models presented do not test for endogeneity. While the single-equation model treats the presence of anti-corruption laws as exogenous, the astute reader will understand that this is an unrealistic assumption. There is an entire ‘sausage-making’ process that leads countries to codify anti-corruption policies, and it is highly likely that there is either an active endogenous relationship or a selection issue at hand. Running a simultaneous equation model or a selection model would likely provide further insight into the relationship.

[i] (Hyden 2012; Jackson and Rosberg 1982a, 1982b)

[ii] (Bratton and Walle 1994, 1997)

[iii] (Bach 2011; Cammack 2007; Erdmann and Engel 2007; Pitcher, Moran, and Johnston 2013; Von Soest, Bechle, and Korte 2011)

[iv] Excepting South Sudan and the Western Sahara.

[v] The World Bank, 2017. World Development Indicators. Dataset. Available online: http://data.worldbank.org/products/wdi. Accessed July 24, 2017.

[vi] Kreutz, Joakim. 2010. “How and When Armed Conflicts End: Introducing the UCDP Conflict Termination Dataset,” Journal of Peace Research 47(2): 243-250. Accessible online at: http://ucdp.uu.se/downloads/. Accessed June 24, 2017.