Histogram of the 2014 BCI values
Significance of changes over time and
differences between countries
the WGI and BCI indicators, the values of the BCI index can be
used to compute the significance of changes in the level of
corruption over time or differences between countries.
The reason is that estimating the BCI index results not just in
a point estimate (BCIi,t) of the level of corruption
for each country i and year t. Instead it
returns thousand of draws from the entire distribution of this
estimate. The significance can easily be computed using these
draws. If less than 5% of all draws, the level of corruption of
country A is lower than that of B, we can state that B’s level
of corruption is significantly lower at the 5% significance
For example: in 2017, the level of corruption of Andorra (19.9)
is more than to 7 percentage-points higher than that of Denmark
(12.6) but this difference is not significant (at the 5% level).
In contrast, Japan’s score (18.2) is lower than that of Andorra
but because it is measured with less uncertainty it is found to
be significantly more corrupt than New Zealand.
Every year, the Corruption Perception Index will publish a
ranking of countries based on their level of corruption.
However, these rankings have been criticized for being very
sensitive to the smallest of differences in the actual scores of
countries. To address this problem, the ranking of the countries
in the BCI dataset only uses these significant differences. A
country will have a higher rank if, and only if, it is
significantly more corrupt than at least one country with a
lower ranking. This allows for a more meaningful comparison of
the level of corruption between countries than a simple ranking
would allow. Only if a decrease in corruption is large enough,
will a country actually achieve a higher ranking.
A dataset containing a thousand random draws of the BCI values
is posted on the website, allowing users to perform their own
analyses. The article “Divining the level of corruption”
(Standaert, 2014) discusses various ways in which these draws
can be used to make robust inferences. Desbordes and Koop (2015)
discuss an alternative (frequentist) way in “The known unknowns
Funding provided by the European Union's Horizon 2020 research and
innovation program under the Marie Sklodowska Curie grant agreement No
665501 with the Research Foundation Flanders (FWO) and the Belgian