The context surrounding the mid-1990s corruption eruption led to a series of changes which fundamentally altered our relationship with corruption and how we understand it, alongside signalling an unprecedented international effort to tackle it. It is during this period that we saw the development of both the CPI alongside the ambitious UNCAC. However, while this period led to a series of widespread, and indeed longstanding changes, there is relatively little research that explores in-depth the interaction between this mix of instruments and conventions developed during this period. It is within this context that this paper set out to explore what impact early ratification of the UNCAC had on a country’s CPI scores.
When one begins to explore the results of the OLS simple linear regression, outlined in Table 3 below, there are several interesting findings. In general, this paper finds that there is a statistically significant positive, but not incredibly substantive, relationship between the role played by early ratification of the UNCAC and a country’s CPI score. Across the various models employed we can see the impact of a one-unit increase in CPI corresponding to between a .025 (model 4) and .631** (Model 2 – significant to the 95% confidence level) higher average value in the scores of countries who ratified the UNCAC early versus those countries who did not ratify early. That the positive relationship exists across all models, with varying standard errors and varying levels of statistical significance, with the various control variables included, adds to the robustness of the findings that the relationship is positive between early ratification of the UNCAC with CPI scores; however, the effect is not substantive across all models.
If we focus firstly on models 1 and 2 contained in Table 3, i.e. those which include Protestantism and the Gini index as control variables, we can see that early ratification of the UNCAC is significant at the 90% (model 1) or 95% significance level (model 2) and that the relationship is positive, albeit not incredibly substantive. While a positive relationship is consistent throughout all of the models employed, models 1 and 2 are potentially interesting due in part to the statistically significant nature of their relationship with the dependent variable. Indeed, while there are a number of benefits to ascribing particular weight to statistical significance in a simple linear regression, we can assume that for a 95% confidence level there is, in fact, something interesting happening within the relationship between the dependent and independent variable that warrants a further exploration of the factors which contribute to this relationship. This is cognisant of the American Statistics Association’s statement that “[b]y itself, a p-value does not provide a good measure of evidence regarding a model or hypothesis. Researchers should recognise that a p-value without context or other evidence provides limited information. For example, a p-value near 0.05 taken by itself offers only weak evidence against the null hypothesis” (Wasserstein and Lazar, 2016, p. 132). From this perspective, this paper will seek to examine the context in which these results exist.
|Model 1||Model 2||Model 3||Model 4|
|of the UNCAC||(0.288)||(0.288)||(0.150)||(0.150)|
|Gross Domestic Product||0.00002||0.00002***|
|Purchasing Power Parity||(0.00001)||(0.00001)|
|Eastern Europe and||-0.039||-0.211||0.047||0.135|
|Post Soviet Union||(0.702)||(0.700)||(0.404)||(0.403)|
|Latin America and||-0.384||-0.121||-0.187||-0.273|
|North Africa and||-0.159||-0.107||-0.268||-0.398|
|The Middle East||(0.795)||(0.789)||(0.429)||(0.434)|
|South East Asia||0.043||0.316||0.293||0.115|
|Western Europe and||-1.099||-1.033||-0.371||-0.492|
|Residual Std. Error||2.702|
(df = 695)
(df = 696)
(df = 1926)
(df = 1912)
|Note:||*p<0.1; **p<0.05; ***p<0.01|
Table 3: Linear Regression output.
As the independent variable is a dichotomously coded dummy variable, we can understand that a one-unit change in the dependent variable, as epitomised through the coefficient in model 1 and 2 (0.545 and 0.631, respectively), tells us how the value of the dependent variable (CPI score) changes, on average, when the dummy variable switches from 0 to 1. Such a positive and significant relationship serves to indicate that the nature of the relationship between the dependent and independent variable, controlling for all of the other variables included in the regression, is one which requires further examination of the factors which could contribute towards this relationship. While the relationship does seem small, it does, however, fit in with a pattern that can be observed within other empirical studies that look at the impact policy interventions can have on an indices score. If we look at the work of Lars et al. (2010, p. 223) and Cole (2015, p. 71), for example, we can see that there is a comparable finding within the OLS linear regression analysis conducted by Cole (2015). While there are differences between this research and the work of Cole, the impact of a policy intervention on an index score remains comparable between the research conducted within this paper and within Cole (2015), thus building on previous literature to develop a more thorough analysis of the “treatment effect”, namely the impact policy interventions and conventions have on indices scores in general and the CPI in particular.
One factor which can perhaps go some way towards explaining the direction of the early ratification coefficient, alongside providing some insight into the scale of the coefficient, is the movement within the CPI. When one begins to examine the empirical results contained in this paper with the work of previous scholars who show the nature of movement within the CPI empirically, we can see that this movement, not only refers to movement within the CPI by rank (which can change depending on addition or exclusion of countries, amongst other superfluous changes), but, more importantly, also by score. Indeed, many scholars point to the low levels of internal movement being an enduring feature of the CPI (Andersson and Heywood, 2009, p. 754; Saha et al., 2012, p. 8; Zouaoui et al., 2017, pp. 90-91). This low level of internal movement is supported by the findings within this empirical analysis, which in turn is supported by the empirical research contained within Zouaoui et al. (2017). This particular facet of the results serves to confirm the hypothesis that early ratification of the UNCAC does have an impact on a country’s CPI score. Thus, it serves to confirm the hypothesis contained within this paper that there does exist a small positive and significant relationship between early ratification of the UNCAC and improvement within CPI scores. This is in addition to being compliant with the findings of other empirical analyses, by serving to illustrate that the rate of this change and indeed scale of it is, in fact, quite muted and not altogether conducive with the notion that early ratification led to a substantial change in a country’s CPI score.
It can also be seen that the spread of this positive change is not altogether evenly dispersed and that there is a considerable degree of regional variation at play, which may in fact point to a much more nuanced impact being felt across different regions. Of particular interest, however, is the impact early ratification appears to be having within Western Europe and North America, where across all models we can see a negative relationship existing. Whilst this relationship is not significant, it could, in fact, be an insight into what Garin (2014) refers to as the “ceiling effect”, wherein many countries within the region Western Europe and North America tended to score relatively high and in many cases are likely only to reduce their scores over the time period. Thus leading to a general reduction in scores due to the use of a more robust methodology being employed by the CPI, leading to lower absolute scores, while maintaining their higher relative positions to other countries. This theory is supported by the work of Zouaoui et al. (2017). That there is some degree of coalescence can also be seen when we begin to explore the impact being felt within the Sub-Saharan Africa region, with model 2 showing a statistically significant positive relationship at the 90% confidence level, which may lend credence to the notion of the inverse of the ceiling effect, namely the floor effect, having an impact with countries generally improving in score but having a limited effect in moving a country’s rank within the table, in line with the findings contained within Zouaoui et al. (2017).
Of the other interesting relationships amongst the control variables within model 2, we can see the substantial and significant relationship when one controls for the human development index. Countries with higher HDI scores have better scores with regards to perceptions of corruption. This is not altogether surprising; indeed, a number of scholars point to the importance of economic development in controlling corruption (Gupta et al., 2002; Dreher and Herzfeld, 2005). However, a particular novelty of this index being shown to have a positive relationship with regards to perceptions of corruption, is an ancillary insight offered by this paper and points to a need for further research on usage of the HDI when controlling for corruption.
It is also possible that a causal explanation as to the improvement in scores over this time could in part be due to what was identified by Cole (2015): “The worldwide increase in corruption scores beginning in the mid-1990s can be attributed, at least partially, to international efforts to fight corruption”. Namely, that the increased attention paid to corruption – as demonstrated through the use of a longer corruption index, with data beginning from the early 1980s – alongside standardised definitions and more significant sanctions for engaging in corruption, led to a short-term increase in perceptions of corruption. We could perhaps view the improvement in scores in those countries who were early ratifiers of the UNCAC as a solution to the information paradox, wherein “[i]ncreasing the quantity and quality of information regarding undesirable practices often gives the impression that those practices worsened, when in fact underlying conditions may have stabilized or even improved”. It is through this that we could view those countries which were early ratifiers as potentially benefitting from a greater perception of controlling corruption and thus benefit from improved scores within the CPI. The information paradox, and indeed the lukewarm levels of adoption of the policy provisions contained within the UNCAC, could lead to a situation wherein those countries who ratified early could benefit from overcoming the information paradoxes. Indeed, ratification could lead to a perception of greater commitment to control corruption than what actually exists whilst at the same time not facing any enforcement or monitoring of their adoption of UNCAC provisions prior to the development of a review mechanism in 2009.
A serious challenge which faces this analysis is the lack of an effective method in which to quantify what in fact ratification signifies, due to the review mechanism being developed in such a way so as to eliminate rankings between countries. Thus, the quantitative limitations of the data available (United Nations Office of Drugs and Crime, 2011) led to a situation of taking all ratifications as being equal in intent when evaluating the significance of ratification and what it means in the context of differing national objectives and expectations of what ratification would come to mean. Indeed, the rather limited effect of early ratification could in part be due to the lukewarm commitment to adopting the provisions into domestic policy (Hechler et al., 2011, p. 7). Further research would thus be required to explore how far countries which ratified the convention have come in adopting the provisions of the convention into domestic policy.
Alongside this, whilst some countries may have benefitted from an early ratification bounce of their CPI scores, we could be seeing a more sanitised impact due in part to countries post-2009 being more aware of the implications of ratification and their obligations within the review mechanism and because of some inspection of their enforcement of the provisions of the convention. This could lead to countries post-2009 doing better than one would expect in controlling corruption, thus limiting the average difference between early and late ratifiers of the conventions and producing results that point towards a relatively small impact.
Another factor, which appears to be having a limiting influence, is the inclusion in particular of the Gini variable in models 1 and 2. As can be seen in Appendix 2, data availability on Gini is relatively sparse by comparison to other variables. The nature of this variable and its ironically unequal data availability could mean that it is reducing the number of democracies included in the regressions in model 1 and 2, thus causing a higher coefficient between the dependent and independent variable. While this does not change the nature of the relationship in models 3 and 4, it does serve to curtail further the impact of early ratification alongside removing the significance of the relationship. This could mean that the relationship may well remain positive but not entirely significant nor substantive.
 Appendix 1 presents the Variance Inflation Factors for predictors and control variables across all models.
 While the adjusted R2 values seem to signify a strong fit, this is likely not the case as much of this variation is largely due to the inclusion of the lagged value across all models.
 Due to the lack of available data, it was decided to exclude both Protestantism and Gini variables within model three and model four to explore the impact of an increased number of observations (which effectively doubles). However, the effect is quite muted.
 A ‘treatment effect’ is the average causal effect of a binary (0–1) variable on an outcome variable of scientific or policy interest. The term ‘treatment effect’ originates in a medical literature concerned with the causal effects of binary, yes-or-no ‘treatments’, such as an experimental drug or a new surgical procedure (Newey, 2007)
 The ceiling effect is said to occur when participants’ scores cluster toward the high end (or best possible score) of the measure/instrument.