Transparency International (TI) released its latest Corruption Perceptions Index (CPI) last month. A couple weeks back, in what has unfortunately become a necessary annual tradition, I posted a warning that one should not attach significance to short-term changes in any individual country’s CPI score. Today, I want to turn to another matter. In recent years, whenever TI releases a new edition of the CPI, the organization plays up certain themes or claims that, according to TI, the CPI reveals about corruption’s causes or impact. This year, one of the main themes in the report is the connection between corruption and campaign finance regulation. As this year’s lead TI press release on the CPI declares, “Analysis [of the data] shows that countries that perform well on the CPI also have stronger enforcement of campaign finance regulations.… Countries where campaign finance regulations are comprehensive and systematically enforced have an average score of 70 on the [100-point] CPI, whereas countries where such regulations either don’t exist or are so poorly enforced score an average of just 34 or 35 respectively.” (On the CPI, higher scores indicate lower perceived corruption.)
How did TI arrive at this conclusion? The report accompanying the CPI, and the longer research brief on this topic, give a bit more explanation. TI used another index, from the Varieties of Democracy (V-Dem) project, on “Disclosure of Campaign Donations.” The V-Dem index rates countries’ disclosure requirements for campaign donations on a 0-4 ordinal scale. TI took this scale, collapsed the 0 and 1 categories into one (allegedly for “data visualization purposes,” though I’m not sure what this means), and then calculated the CPI score for the countries in each of the four categories. The results:
- For those countries with a V-Dem disclosure score of 0/1 (no disclosure requirements or requirements that are partial and rarely enforced), the average CPI score was 34.
- For countries with a V-Dem score of 2 (uncertain enforcement of disclosure rules) the average CPI was 35.
- For countries with a V-Dem score of 3 (disclosure requirements exist and are enforced, but may not be fully comprehensive), the average CPI score was 55.
- Countries with a V-Dem score of 4 (comprehensive and fully enforced disclosure requirements) had an average CPI score of 70
That looks like pretty strong evidence that strong campaign finance disclosure rules are associated with lower corruption, and that’s certainly the story TI wants to tell. As the report puts it, “Unregulated flows of big money in politics … make public policy vulnerable to undue influence.” The research brief similarly explains, “Shedding light on who donates and how much, can expose the influence of money in politics and deter corruption and other pay-to-play situations.”
The claim may ultimately be correct, but on closer inspection, the evidence TI adduces in support of that claim is deeply problematic.
For starters, it seems that TI’s calculations include data for all countries, not just democracies with genuinely competitive multiparty elections. I’m not sure it’s so useful to include in the dataset countries like North Korea and Saudi Arabia (which, in case you’re curious, get a 0 on the V-Dem Disclosure of Campaign Donations Index) when trying to figure out how much campaign finance regulations affect corruption in democracies. For another thing, it’s well known that wealthy countries differ from poorer countries on a whole host of institutional and governance measures, including the CPI. And richer countries may look quite different, at least on average, with respect to their campaign finance systems as well. This is especially true insofar as the V-Dem campaign donation disclosure index captures not just the laws on the books, but how effectively these laws are enforced in practice; after all, lower-income countries generally have more difficulty enforcing their laws.
What one really should do, if one wanted to rigorously assess the hypothesis that the quality of campaign finance transparency regulation is negatively correlated with perceived corruption, is first limit the sample to genuine democracies, and then run a multivariate regression with a range of controls, including per capita GDP. I didn’t have the time to do anything that fancy, but I did spend an hour or so playing around with the data, making some really simple, back-of-the-envelope calculations. What I found, while certainly not decisive, is at least suggestive that the correlation is not nearly as robust as TI suggested.
Here’s what I did:
- First, I limited the sample to countries that were considered democracies according to the 2018 Polity IV index. (The 2018 index was the most recent one I could find with a quick search; I’m fairly confident the set of democracies didn’t look much different in 2019.)
- Then, I subdivided the sample to “high income countries,” defined by the World Bank as countries with a per capita gross national income over $12,376 in 2018, and other countries (lumping together low and middle income countries).
- Then, following TI’s approach, I sorted the countries in each of these two subsamples into bins based on each country’s 2017 V-Dem “Disclosure of Campaign Donations Score.” (I used the 2017 data because that was the data I could find online in Excel format. But I can’t imagine there are huge differences between this data and the most recent V-Dem data that TI used.) Again following TI, I combined the countries with a V-Dem score of 0 and those with a V-Dem score of 1, treating both as if they have essentially no meaningful disclosure requirements for campaign donations.
- I then calculated the average CPI score within each bin. (I dropped countries that didn’t appear in all four datasets, leaving me with 37 high-income democracies and 59 non-high-income democracies.)
Here’s what I found:
- For the high-income democracies, almost all of the countries get either a 3 or a 4 on the V-Dem campaign donations disclosure index. Three of them (Hungary, Panama, and Sweden) get a 2, and one (Switzerland) gets a 0. (No high-income democracies get a V-Dem score of 1.) Switzerland, the only country in the 0/1 category, has a 2019 CPI score of 85. For the three countries that get a V-Dem score of 2, the average CPI score is 55. If we lump these four countries into a single category (that is, high-income democracies with a V-Dem campaign donation disclosure score below 3), the average CPI is about 63; this average combines two very-high-income countries with very high CPI scores (Switzerland and Sweden, both of which get an 85 on the CPI) and two countries that just barely qualify as high income, with much lower CPI scores (Panama and Hungary, with CPI scores of 36 and 44, respectively). The average CPI score for high-income democracies with a V-Dem score of 3 is 67, while the average CPI score for high-income democracies with a V-Dem score of 4 is 69. So, the average CPI score for high-income democracies with weaker campaign donation disclosure requirements (a V-Dem score below 3) is 63, the average CPI for those with disclosure systems that are enforced though not fully comprehensive (V-Dem 3) is 67, and the average CPI for those with comprehensive systems (V-Dem 4) is 69. There’s a slight upward trend, but not one that any reasonable person would treat as meaningful.
- For the non-high-income democracies, the average CPI score for the countries with no meaningful campaign donation disclosure regulation (a V-Dem score of 0 or 1) was approximately 38, the average CPI score for countries with a V-Dem score of 2 was 35, and the average CPI score for countries with a V-Dem score of 3 was approximately 39. So, for non-high-income countries with anything less than fully comprehensive disclosure systems, there doesn’t seem to be any meaningful difference in perceived corruption. For non-high-income countries with the maximum V-Dem campaign donation disclosure score of 4, the average CPI is higher, coming in at 54. But, importantly, there were only four countries in this category (Bhutan, Brazil, Costa Rica, and Georgia), raising questions about the extent to which this is a real result or an anomaly.
Putting this all together, it’s really hard to see clear prima facie evidence for the claim that the strength or weakness of campaign finance donation disclosure rules has a strong association with perceived corruption. Wealthy democracies generally have strong campaign finance disclosure systems and good CPI scores. Moderate differences in the comprehensiveness of campaign donation disclosure rules (such as would be captured by the distinction between a 3 and a 4 on the V-Dem scale) have no meaningful correlation with CPI scores. And those few high-income democracies with V-Dem disclosure scores below 3 appear to have abnormally high or abnormally low CPI scores, without much difference in the average score relative to other high-income democracies. As for non-high-income democracies, the average CPI scores are roughly the same for the sets of countries with no meaningful disclosure rules (V-Dem 0 or 1), those with unclear enforcement (V-Dem 2), and those with disclosure requirements that are enforced but not comprehensive (V-Dem 3). A handful of countries (Bhutan, Costa Rica, and Georgia) boost the average CPI of those non-high-income democracies with comprehensive disclosure regimes (V-Dem 4), but it seems like a huge stretch to make aggressive causal claims on the basis of three countries.
I want to re-emphasize two things I said earlier in this post. First, the claim that weak campaign finance regulations are associated with higher corruption might well be true; indeed, I’m inclined to believe it probably is true, all else equal. Second, to really test whether there’s a genuine association here, one could and should run multivariate regressions with a range of control variables. I have not had the time to do so, and I am not making any strong assertions about what the data does or doesn’t show. But I do think that in this part of its report TI is being a bit sloppy, and seems to be more interested in finding data to back up a preconceived narrative than to actually analyze the data carefully to see what hypotheses are supported.
This is a great conversation, and I agree, that the analysis done needs to be extended to make it more credible. We did just that with my coauthor, Luciana Cingolani, looking at changes to campaign finance regulations over time within 28 EU countries and linked that to public procurement corruption. So we use objective data on both sides. The effect is nill…
See here: http://www.govtransparency.eu/index.php/2016/06/13/breaking-the-cycle-how-not-to-use-political-finance-regulations-to-counter-public-procurement-corruption/
As always, good work Matthew. There may also be some endogeneity here given the range of generic questions the CPI uses. That is why the piece by Fazekas and Cingolani noted in the comment above is important since it focuses on a specific type of corruption, procurement corruption. It also builds on the point made by Heywood and others about the need to disaggregate the many forms of corruption to explore how they interact.
Professor, thank you for this post! The correlation / causation mistake TI is making here is fairly misleading, at least from a statistical inference standpoint. If anything, I suspect reverse causality is at play- countries with lower levels of corruption are likely better able to put strong campaign finance disclosure laws in place. Instead of separating the countries into bins, I’d be curious to see what the results look like after controlling for income as well as the strength of democracy (maybe from the Freedom House reports?)
I also think the focus on disclosures, although an important tool, might be misplaced. Take a country such as India, which has fairly strong disclosures (at least for candidates), yet large amounts of corruption that prop up and result from the electoral system. The disclosures themselves aren’t enough- it is the tracking and focus of those disclosures that counts. The effectiveness of disclosures is contingent on the larger sensitivity to corruption, and it might be that third variable that is captured in the correlation.