Public sector reforms and their impact on the level of corruption: A systematic review

In spite of the large number of anti-corruption reforms implemented in different countries, there has been little research that empirically and systematically assesses the impact of these efforts.

The main objective of this review is to identify what works in curbing corruption in the public sector, by meta-analyzing the findings of published and unpublished evaluations of different types of anti-corruption interventions in different countries. The focus of this review is administrative corruption, namely corrupt acts involving civil servants in their dealings with their superiors, during the implementation of public policies, or while interacting with the public for service delivery. Political corruption (in the adoption of laws, regulations, and policies), and private-to-private corruption (involving only private actors) are excluded from this review.

The literature search was conducted by querying three widely recognized electronic databases: RePEc, SSRN, and Web of Science. These databases are considered the most comprehensive in the socio-economic field of research. The main grey literature repositories were also queried. Both published and unpublished studies were searched on the basis of specific combinations of keywords. The terms used to define queries were based on the "types of corruption", "types of interventions/policies/reforms" and "study design" search strings. Specific conventions were used to "explode" or "truncate" keywords as appropriate. Screening of the references (i.e., snowballing) of the identified studies was also performed, and a reverse snowballing approach on Google Scholar was used. In order to ensure replicability, all searches were stored into Covidence, an online software developed by the Cochrane community for screening studies and extracting data for systematic reviews.

Any study that included experimental evaluations (randomized controlled trials) of interventions developed for use in the public sector (e.g., public administration, education, health, etc.) to curb administrative corruption has been included in this review without any geographical or temporal limitations. Only studies written in or translated into English have been considered.

Two review authors read the titles and abstracts of identified studies in order to determine their eligibility against the inclusion/exclusion criteria. When a title or abstract could not be included or rejected with certainty, the full text of the article was reviewed. In case of disagreement about whether or not a study should be included, the lead author (Giulia Mugellini), together with Martin Killias acted as arbitrator. The relevant information from identified studies was extracted independently by two review authors, following the guidelines of the Campbell Collaboration. The studies were assessed using the Cochrane Risk of Bias checklist as a basis.The effect size selected for the analysis was the Fisher's z-score transformation of the partial correlation coefficient. For the meta-analysis, random effect(s) models were estimated. Meta-regression analysis models were then used to investigate the determinants behind the observed between- and within-study heterogeneity. Ten different covariates were included in the meta-regression models in order to control for the type of intervention, the type of corruption, the level of national income, the quality of the study and the type of participants involved in laboratory experiments.

The initial literature search led to the identification of 70 studies. Approximately one-third of the studies were excluded at the title/abstract stage because they either did not evaluate any anti-corruption intervention but simply assessed the relationship between corruption and other phenomena, or because the study design was not based on randomized controlled trials. Another 14 studies were excluded only after a full-text assessment. At this stage, the main reasons for exclusion were related to an unsuitable type of corruption (e.g., when the focus of the paper was political corruption, or private-to-private corruption instead of administrative corruption), the lack of regression output, or an unsuitable study design. At the end of the selection process, 29 studies resulted as eligible for inclusion.All the selected studies were written in English. The publication years ranged from 2007 to 2018. The majority of the selected studies (20) investigates the effect of anti-corruption interventions in high- and upper-middle income countries (Austria, Brazil, Canada, China, Germany, Italy, Mexico, the Netherlands, Thailand, the United Kingdom, and the United States). Nine studies focused on low- and low-middle income countries (Burkina Faso, Burundi, Ethiopia, India, Indonesia, Pakistan, Tanzania, and Uganda). All of them were randomized experiments. Twenty-five of these experiments were conducted in a laboratory, while four of them were field experiments.As to the type of outcome, the majority (18) of the selected studies addressed bribery (either active or passive), while 11 studies considered misappropriation of public resources (embezzlement). In terms of anti-corruption interventions, 19 studies tested the effect of deterrence interventions, while 10 studies focused on policies based on organizational and cultural change.Overall, the meta-analysis' findings indicate that the identified interventions decrease the level of corruption. Results are statistically significant (p < 0.01) and robust to different heterogeneity estimators-that is, (restricted) maximum likelihood and method of moment estimators. The observed high level of heterogeneity- $$ is equal to 92.36%, of which 43.78% is due to between-study heterogeneity and 48.57% to within-study heterogeneity-albeit in line with other meta-analyses in economics, suggests the need for meta-regression analyses.To investigate the determinants behind the between- and within-study heterogeneity of the observed effect, both a random effect model and a multilevel model were adopted.The results of the multilevel model show that: 1)Control and deterrence interventions are more effective than organizational and cultural reforms in reducing corruption in the public sector.2)Combining different interventions reduces corruption more than single interventions.3)Interventions are more effective in preventing misappropriation of public resources (embezzlement) than passive or active bribery. Finally, the Funnel Asymmetry Test (FAT), conducted with both additive and multiplicative dispersion terms, shows no evidence of a strong publication bias in the literature.

The results of this systematic review, based on a combination of laboratory and field experiments, demonstrate that increasing the expected monetary costs (e.g., sanctions) of corruption or the probability of detection (e.g., audit risk) is more effective than organizational, cultural and educational interventions in curbing administrative corruption, at least in the short term. However, this result might be due to the fact that the majority of selected studies are based on lab-experiments, where the assessment of the intervention is almost concurrent to its development. Short-term evaluations might fail to identify the effect of organizational and cultural interventions. Indeed, these interventions are based on structural changes in the organization of the system and the ethical and cultural education of public officials and might, thus, entail long periods to display their results on the level of corruption. Nevertheless, a combination of different interventions proves to be more effective than single interventions. For example, policies guaranteeing impunity to officials or citizens who report corrupt practices (principal witness/leniency treatment) are more effective if associated with a high probability of audit than leniency alone. A low probability of detection can be compensated by the threat of high fines in reducing both the amount and the likelihood of bribe demands. To the contrary, a high probability of detection had no effect in the absence of severe sanction threats.The importance of the organizational and cultural environment in which the intervention is implemented clearly emerged in the literature. When possible, the characteristics of the settings where the interventions were developed were included in the meta-regression analysis (such as the level of income of the countries). When it was not possible to measure contextual factors and their interaction with the main intervention, a qualitative analysis was performed to reveal the complexities of these interactions.This additional analysis shows that the impact of the interventions was found to be affected by the likelihood of the continued interactions between bribe takers and givers, the amount and probability of fines, and the size of the bribe, among others. For example, reporting mechanisms and leniency policies increase their potential in combination with interventions that limit agent's exposure to one another - such as staff rotation. The success of audit risk on corruption is strongly dependent on the seriousness of the potential sanction and the probability that a sanction is applied. Some differences also emerge between high- and low-corruption countries regarding the effectiveness of anti-corruption interventions. For example, measures tending to increase social blame of corrupt practices work in low-corruption countries. Adding punishments in environments where actors' behavior is tightly monitored increases compliance, but more so in environments where corruption is the exception rather than the rule.In terms of implications for research, the fact that control and deterrence turns out to be more effective than organizational and cultural interventions in curbing administrative corruption confirms the importance of economic theories (and cost-benefit analysis). However, the meta-analysis also demonstrates the effectiveness of combining different types of interventions.This is true not only when combining policies reinforcing control and deterrence (monitoring frequency, detection probability and amount of fines), but also when policies based on organizational and cultural change are added (e.g., staff rotation and leniency). In particular, the role of moral levers in preventing corruption emerges, and especially the importance of strengthening professional identity and values in order to avoid conflicts between an individual's private interests and his/her public role. These results highlight the importance of going beyond economic models for explaining corruption, and considering the moral and cultural mechanisms underlying this phenomenon.It also emerges the need to understand how different forms of corruption operate in practice at macro- (cross-country), meso- (country/nation-state) and micro- (individual) level. In particular, individual-level factors, such as the strive for power, low self-control, loss aversion and risk acceptance would need to be addressed.It would be interesting to distinguish, when more experimental studies will be available, between top-down (from supervisors to officials) and bottom-up (from citizens to officials) interventions.From a methodological point of view, it could be tested whether the results change according to the types of games used as a basis for the corruption experiments (e.g., behavioral game theory, trust game, etc.) and according to the setting in which the experiment was conducted (e.g., context-free versus in-context presentation of experimental tasks).Considering the effect of sensitization messages in reducing bribery demand, we would encourage researchers to develop other corruption experiments that explore the impact of interventions in fostering professional self-identity, as well as the impact of organizational family culture on corruption. Furthermore, this review highlights the need for a comprehensive classification of anti-corruption policies that distinguishes interventions by type of corruption, risk factors, type of policy tool and administrative sector.