Adjusting Corruption Perception Index Scores for National Wealth

My post two weeks ago discussed Transparency International’s newly-released 2017 Corruption Perceptions Index (CPI), focusing in particular on an old hobby-horse of mine: the hazards of trying to draw substantive conclusions from year-to-year changes in any individual country’s CPI score. Today I want to continue to discuss the 2017 CPI, with attention to a different issue: the relationship between a country’s wealth and its CPI score. It’s no secret that these variables are highly correlated. Indeed, per capita GDP remains the single strongest predictor of a country’s perceived corruption level, leading some critics to suggest that the CPI doesn’t really measure perceived corruption so much as it measures wealth—penalizing poor countries by portraying them as more corrupt, when in fact their corruption may be due more to their poverty than to deficiencies in their cultures, policies, and institutions.

This criticism isn’t entirely fair. Per capita income is a strong predictor of CPI scores, but they’re far from perfectly correlated. Furthermore, even if it’s true that worse (perceived) corruption is in large measure a product of worse economic conditions, that doesn’t mean there’s a problem with the CPI as such, any more than a measure of infant mortality is flawed because it is highly correlated with per capita income. (And of course because corruption may worsen economic outcomes, the correlation between wealth and CPI scores may be a partial reflection of corruption’s impact, though I doubt there are many who think that this relationship is so strong that the causal arrow runs predominantly from corruption to national wealth rather than from national wealth to perceived corruption.)

Yet the critics do have a point: When we look at the CPI results table, we see a lot of very rich countries clustered at the top, and a lot of very poor countries clustered at the bottom. That’s fine for some purposes, but we might also be interested in seeing which countries have notably higher or lower levels of perceived corruption than we would expect, given their per capita incomes. As a crude first cut at looking into this, I merged the 2017 CPI data table with data from the World Bank on 2016 purchasing-power-adjusted per capita GDP. After dropping the countries that appeared in one dataset but not the other, I had a 167 countries. I then ran a simple regression using CPI as the outcome variable and the natural log of per capita GDP as the sole explanatory variable. (I used the natural log partly to reduce the influence of extreme income outliers, and partly on the logic that the impact of GDP on perceived corruption likely declines at very high levels of income. But I admit it’s something of an arbitrary choice and I encourage others who are interested to play around with the data using alternative functional forms and specifications.)

This single variable, ln per capita GDP, explained about half of the total variance in the data (for stats nerds, the R2 value was about 0.51), meaning that while ln per capita GDP is a very powerful explanatory variable, there’s a lot of variation in the CPI that it doesn’t explain. The more interesting question, to my mind, concerns the countries that notably outperform or underperform the CPI score that one would predict given national wealth. To look into this, I simply ranked the 167 countries in my data by the size of the residuals from the simple regression described above. Here are some of the things that I found: Continue reading

The New Corruption Perceptions Index Identifies Countries with Statistically Significant Changes in Perceived Corruption–Should We Credit the Results?

As most readers of this blog are likely aware, last month Transparency International (TI) released the 2017 edition of its important and influential Corruption Perceptions Index (CPI). As usual, the publication of the CPI triggered a fair bit of media coverage, much of it focused on how various countries ranked, and how individual country scores had changed from one year to the next (see, for example, here, here, here, and here).

There’s a lot to say about the most recent CPI—I may devote a post at some point to TI’s interesting decision to focus the press release accompanying the publication of the 2017 CPI less on the index itself than on the connection between (perceived) corruption and a lack of adequate freedom and protections for the media and civil society. But in this preliminary post, I want to take up an issue that regular GAB readers will know has been something of a fixation of mine in past years: the emphasis—in my view mostly misplaced—on how individual country CPI scores have changed from year to year.

In prior posts, I’ve raised a number of related but distinct concerns about the tendency of some commentators—and, more disturbingly, of some policymakers—to attach great significance to whether a country’s CPI score has gone up or down relative to previous years. For one thing, the sources used to construct the CPI for any given country may change from year to year—and adding or dropping an idiosyncratic source can have a substantial effect on the aggregate CPI score. For another, even when the underlying sources don’t change, we don’t know whether those sources are on the same implicit scale from year to year. And even if we put these problems to one side, a focus on changes in the final CPI score can sometimes obscure the statistical uncertainty associated with the estimated CPI—these scores can be noisy enough that changes in scores, even those that seem large, may not be statistically meaningful according to the conventional tests. Although TI always calculates statistical confidence intervals, in prior years these intervals have been buried in hard-to-find Excel spreadsheets, and the changes in CPI scores that TI highlights in its annual press releases haven’t always been statistically significant by TI’s own calculations. In an earlier post, I suggested that at the very least, TI should provide an easy-to-find, easy-to-read table assessing which changes in country scores are statistically significant at conventional levels, preferably over a 4-year period (as 1-year changes are both harder to detect if trends are gradual, and less interesting).

Apparently some folks within TI were thinking along similar lines, and I was pleased to see that in the 2017 CPI includes a reasonably prominent link to a spreadsheet showing those countries for which the 2017 CPI score showed a “statistically significant difference” from that country’s CPI score in each of five comparison years (2012, 2013, 2014, 2015, and 2016).

I’ve still got some criticisms and concerns, which—in the spirit of constructive engagement—I’ll turn to in just a moment. But before getting to that, let me pause to note my admiration for TI as an organization, and in this case its research department in particular, for constantly working to improve both the CPI itself and how it is presented and interpreted. It’s easy for folks like me to criticize—and I’ll continue to do so, in the interests of pushing for further improvements—but it’s much more challenging to absorb the raft of criticisms from so many quarters, sift through them, and invest the necessary time and resources to adapt and adjust from year to year. So, in case any folks at TI are reading this, let me first acknowledge and express my appreciation for how much work (often thankless) goes into the creation and continued improvement of this valuable tool.

Having said that, let me now proceed to raising some comments, questions, and concerns about TI’s claims about countries that appear to have experienced statistically meaningful changes in their CPI scores over the last five years. Continue reading

The Bayesian Corruption Index: A New and Improved Method for Aggregating Corruption Perceptions

As most readers of this blog are likely aware, two of the most widely used measures of corruption perceptions—Transparency International’s Corruption Perceptions Index (CPI) and the Worldwide Governance Indicators (WGI) corruption index—are composite indicators that combine perceived corruption ratings from a range of different sources (including private rating agencies, NGOs, international development banks, and surveys of firms and households). The CPI takes a simple average of the available sources for each country; the WGI uses a somewhat fancier “unobserved component model” (UCM) which assumes that each source’s score is a noisy signal of the “true” level of perceived corruption; the UCM differs from a simple average in a few ways, perhaps most notably by giving less weight to “outlier” sources, though in practice the WGI and CPI are highly correlated, and the WGI’s creators report that the results for the WGI turn out not to change very much if one takes a simple average rather than using the WGI.

These composite indicators have a number of well-known problems, which I won’t bother going into here. Rather, the main purpose of this post is to introduce readers to an alternative index, developed by Samuel Standaert at Ghent University, which he calls the “Bayesian Corruption Index” (BCI). Standaert introduced the BCI in a 2015 article, but so far as I can tell it has not attracted much attention. The BCI certainly doesn’t solve all the problems of the traditional aggregated corruption perceptions indicators (more on this below), but it’s definitely an improvement, and deserves wider use. Let me first say a bit about how the BCI differs from the WGI, why I think it’s an advance over the WGI and CPI, and what some of its limitations are. Continue reading

The 2016 CPI and the Value of Corruption Perceptions

Last month, Transparency International released its annual Corruption Perceptions Index (CPI). As usual, the release of the CPI has generated widespread discussion and analysis. Previous GAB posts have discussed many of the benefits and challenges of the CPI, with particular attention to the validity of the measurement and the flagrant misreporting of its results. The release of this year’s CPI, and all the media attention it has received, provides an occasion to revisit important questions about how the CPI should and should not be used by researchers, policymakers, and others.

As past posts have discussed, it’s a mistake to focus on the change in each country’s CPI score from the previous year. These changes are often due to changes in the sources used to calculate the score, and most of these changes are not statistically meaningful. As a quick check, I compared the confidence intervals for the 2015 and 2016 CPIs and found that, for each country included in both years, the confidence intervals overlap. (While this doesn’t rule out the possibility of statistically significant changes for some countries, it suggests that a more rigorous statistical test is required to see if the changes are meaningful.) Moreover, even though a few changes each year usually pass the conventional thresholds for statistical significance, with 176 countries in the data, we should expect some of them to exhibit statistical significance, even if in fact all changes are driven by random error. Nevertheless, international newspapers have already begun analyses that compare annual rankings, with headlines such as “Pakistan’s score improves on Corruption Perception Index 2016” from The News International, and “Demonetisation effect? Corruption index ranking improves but a long way to go” from the Hidustan Times. Alas, Transparency International sometimes seems to encourage this style of reporting, both by showing the CPI annual results in a table, and with language such as “more countries declined than improved in this year’s results.” After all, “no change” is no headline.

Although certain uses of the CPI are inappropriate, such as comparing each country’s movement from one year to the next, this does not mean that the CPI is not useful. Indeed, some critics have the unfortunate tendency to dismiss the CPI out of hand, often emphasizing that corruption perceptions are not the same as corruption reality. That is certainly true—TI goes out of its way to emphasize this point with each release of a new CPI— but there are at least two reasons why measuring corruption perceptions is valuable: Continue reading

Are Aggregate Corruption Indicators Coherent and/or Useful?: Further Reflections

Last week, I used Professor Michael Johnston’s recent post on the methodological and conceptual problems with national-level perceived corruption indicators as an opportunity to respond to some common criticisms of research that relies on these indicators. In particular, I have frequently heard (and interpreted Professor Johnston as advancing) two related criticisms: (1) composite indicators of “corruption” are inherently flawed because “corruption” is a multifaceted phenomenon, comprised of a range of diverse activities that cannot be compared on the same scale, let alone aggregated into a single metric; and (2) corruption is sufficiently diverse within a single country that it is inappropriate to offer a national-level summary statistic for corruption. (These points are related but separate: One could believe that corruption is a sufficiently coherent concept that one can sensibly talk about the level of “corruption,” but still object to attempting to represent an entire country’s corruption level with a single number; one could also endorse the idea that national-level summary statistics can be useful and appropriate, even when there’s a lot of intra-country variation, but still object to the idea that “corruption” is a sufficiently coherent phenomenon that one can capture different sorts of corruption on the same scale.) For the reasons I laid out in my original post, while I share some of the concerns about over-reliance on national-level perceived corruption indicators, I think these critiques—if understood as fundamental conceptual objections—are misguided. Most of the measures and proxies we use in studying social phenomena aggregate distinct phenomena, and in this regard (perceived) corruption is no different from war, wealth, cancer, or any number of other objects of study.

Professor Johnston has written a nuanced, thoughtful reply (with a terrific title, “1.39 Cheers for Quantitative Analysis”). It is clear that he and I basically agree on many of the most fundamental points. Still, I think there are still a few places where I might respectfully disagree with his position. I realize that this back-and-forth might start to seem a little arcane, but since so much corruption research uses aggregate measures like the Corruption Perceptions Index (CPI), and since criticisms of these measures are likewise so common, I thought that perhaps one more round on this might not be a bad idea.

Let me address the two main lines of criticism noted above, and then make some more general observations. Continue reading

The Level-of-Aggregation Question in Corruption Measurement

Recently I learned that CDA Collaborative (a nonprofit organization that works on a variety of development and conflict-resolution projects) has launched a new blog on corruption. Though it’s a new platform, they already have a few of interesting posts up, and it’s worth a look.

While I’m always happy to advertise new platforms in the anticorruption blogosphere, in this post I mostly want to focus on the first entry in the CDA’s new blog, a post by Professor Michael Johnston entitled “Breaking Out of the Methodological Cage.” It’s basically a critique of the anticorruption research literature’s alleged (over-)reliance on quantitative methods, in particular cross-national regression analyses using country-level corruption indices (such at the Corruption Perceptions Index (CPI) or Worldwide Governance Indicators (WGI) graft index). There are some things in Professor Johnston’s post that I agree with, and much that I disagree with. I want to focus on one issue in particular: the question of the right unit of analysis, or level of aggregation, to use when attempting to measure corruption.

Professor Johnston has two related complaints (or maybe two variants on the same underlying complaint) regarding these national-level perceived corruption measures. First, he complains these “[o]ne dimensional indices tell us … that corruption is the same thing everywhere, varying only in amount[.]”  In other words, corruption indices lump a whole bunch of disparate phenomena together under the same umbrella term “corruption,” ignoring the internal diversity of that category. Second, he contends that “relying … on country-level data is to assume that corruption is a national attribute, like GDP per capita” when in fact “corruption arises in highly specific processes, structural niches, and relationships.” Corruption, he explains, is not an attribute of countries, but of more specific contexts, involving “real people … in complex situations[.]”

Respectfully, I think that these points are either wrong or irrelevant, depending on how they are framed. Continue reading

Larger Governments Have Less Corruption (Part 2 – Possible Explanations)

In my last post, I argued that the familiar hypothesis—advanced by Gary Becker and others—that big governments are associated with more corruption is inconsistent with the available cross-country empirical evidence. In fact, though the results of different studies are not entirely consistent, the weight of the evidence seems to suggest that (controlling for other possible correlates), countries that have larger governments—defined primarily as those that have higher levels of government spending as a percentage of GDP—have lower levels of perceived corruption, as measured by the familiar indexes, such as Transparency International’s Corruption Perceptions Index (CPI). Again, there are some questions about the robustness of this negative correlation—some studies find that it is statistically significant, while others do not—but there’s enough supporting evidence that I think it’s fair to (tentatively) treat this correlation as genuine.

Perhaps in hindsight this shouldn’t be so surprising. Putting aside multiple regression and other fancy statistical techniques, if one just eyeballs the CPI “league table,” it’s clear that the group of countries that consistently score near the top of the rankings include lots of countries—particularly countries in Northern and Western Europe—with quite large governments (such as Denmark, Sweden, Belgium, Norway, the Netherlands, Finland, and Iceland), while the bottom of the CPI list includes countries with very small governments. (Even if one excludes barely functioning states, like Somalia, the bottom group in the CPI includes small-government states like Bangladesh, Cambodia, Haiti, Russia, and the Central African Republic). Of course, this by itself doesn’t tell us much, especially given the well-established correlation between GDP and the government spending/GDP ratio—but, again, multiple regression techniques that control for GDP and other factors show that the positive correlation is genuine, and the handful of favorite examples often trotted out to suggest that small governments are the key to lower corruption (like Singapore and Hong Kong) are in fact statistical outliers.

So let’s assume that, as most studies seem to show, there’s a negative correlation between the government spending/GDP ratio and perceived corruption. What’s the explanation for this?

The short answer is that I don’t know, and I’m not aware of any research that really nails this down. But here are a few possibilities, some cribbed from existing papers, others based on my own wild speculations: Continue reading