The astounding figure Richard Rose and Caryn Peiffer report in their new book, Paying Bribes for Public Services, that almost one quarter of world’s population or 1.6 billion people, recently paid a bribe would suggest the answer to the question above is a resounding “No.” The 1.6 billion figure sounds so fantastically large that the suspicion arises that it is one of those gauzy numbers conjured up using shaky assumptions and questionable sources to capture headlines rather than advance learning. Yet recent research by the World Bank’ Art Kraay and University of Maryland Professor Peter Murrell shows that, if anything, the Rose and Peiffer 1.6 billion number is low.
Their figure is based on the most solid of evidence: interviews by phone or in-person where respondents are asked whether they had to pay a bribe to obtain a public service. Transparency International’s 2013 Global Corruption Barometer, a main source for the 1.6 billion number, is an example. Surveyors first ask respondents if they or anyone else in their household has had any contact in the past 12 months with anyone associated with any of eight government services: i) the education system, ii) the judiciary, iii) medical or health services, iv) the police, v) registry and permit services, vi) utilities, vii) tax collection or, viii) land service. If the answer is yes, the surveyor then asks:
In your contact or contacts have you or anyone living in your household paid a bribe in any form in the past 12 months?
What could be a more reliable way to gather evidence of bribery? Instead of asking what people think about bribery or what their perceptions of bribery or corruption are, they are asked about their own personal experience, or that of close relatives, with the crime of bribery. The rub comes with the last phrase in the preceding sentence: the respondent is being questioned about “the crime of bribery.”
In the 174 nations now a party to the United Nations Convention Against Corruption payment of a bribe is, or at least should be, a crime punishable by a fine, a prison sentence, or both. So the accuracy of the TI survey data and the data from the other surveys Rose and Peiffer use to compute the 1.6 billion figure depends upon citizens being willing to admit to a stranger, either in person or over the phone, that they have committed a crime. Until Kraay and Murrell’s work, corruption researchers have simply assumed people would not hesitate to report paying a bribe, for as Olken and Pandi say in their 2012 review of corruption research, in most countries little stigma attaches to paying a bribe. But whether a stigma attaches or not, and thus whether a citizen will confess they committed a crime to someone they have never met before, is a question of fact, and in social science, as in science generally, questions of fact are a subject for research not assumptions.
Kraay and Murrell research the subject building on techniques pioneered by an earlier generation of survey researchers. The latest version of their “Misunderestimating Corruption” paper is here, and Murrell’s English language summary here (Kraay and Murrell both being econometricians, much of the paper may be foreign to those not schooled in advanced math). Not surprisingly, they find that respondents’ reticence to admit to paying bribes varies from country to country. What is extraordinary is the effect this reticence can have on survey results: for some countries the actual level of level of bribery may be twice what survey show.
If that result were to hold across all countries, then the real number of people paying bribes across the world would be twice the 1.6 billion figure Rose and Peiffer report, or 3.2 billion. Since there are roughly 6.9 billion people alive today, that number might be plausible – at least in the eyes of some cynics. Excluding infants, half the world pays bribes and perhaps the other half collects them.
Taken together what the Kraay and Murrell and Rose and Peiffer work really show is that the academic and policy communities have a great deal to do before decision-makers and citizens can put much store in measures of corruption.
Important point, Rick. In light of your last post, it appears we’re confronting a couple of big, connected challenges: (1) gathering data and (2) ensuring that data is as accurate as possible. I initially thought there was still value in the sort of data you describe even if it wasn’t entirely accurate, particularly for comparative purposes. For example, regardless of whether studies underestimate the absolute number of people paying bribes, they can still assess whether the prevalence of bribe-paying is increasing or decreasing over time. Analyses can compare between countries in a similar manner.
But the Kraay and Murrell findings undermine this supposition and give rise to a host of questions. Differential reporting rates between countries make me a little skeptical of the validity of ranking those countries (although, to be fair, I do not know enough about, say, TI techniques and methodologies to pass much judgment). Even in-country analysis becomes problematic. Exogenous factors that have affected a respondent’s willingness to report could account for changes over time. And I imagine reporting could differ by sector, as well. I might be more inclined to admit I’d paid a bribe to a postman than to a judge.
I’ll say one last thing: I wonder if your analysis of “the _crime_ of bribery” could cut the other way. Rather than failing to report for fear of the consequences, might some people underreport because they do not believe their payments were illicit/abnormal (or even bribes)? My inclination is, mostly, “no” because I think the vast majority of people are aware they are overpaying for a benefit and are familiar enough with the corruption discourse to know it is a bribe. But I am thinking also of systematic bribe-taking when there is a set “price” for a free service (for example, health centers charging for malaria treatment that the government provides for free distribution).
Your point about differential willingness to admit bribing a judge versus a postman or woman worries me too. One thing I had thought experience surveys might be useful for was identifying sectors or ministries for priority attention. Fear of reprisal may make respondents more reluctant to report bribing a judge than someone working for the postal service. If true, this would render the surveys much less informative than one would like.
Fear of reprisal may also explain why the Ugandan Inspectorate of Government’s Fourth Annual Report on corruption trends shows that while the surveys continue to report the police are perceived to be the most corrupt entity, they are not the agency against which the most complaints are lodged.
This is really interesting, Rick. But if the survey simply asks people if they or anyone in their household has paid a bribe, is it really fair to consider that then attributable to the person, as well? For example: suppose that someone was living in a household with extended family members, or 10 people in total. If one person paid a bribe, each of the 10 people would answer ‘yes’ to that question. Depending on household size, shouldn’t that 1.6 billion number actually be somewhere around 320 million instead? Otherwise the statistic should be roughly “1.6 billion people lived in households that recently paid bribes.” The larger number might not overestimate the number of people affected by corruption, but it would overestimate the number (and therefore perhaps dollar figure) of bribes paid.
The discussion in the comments about differential willingness to admit to various types of bribes is really interesting. There should still be a way to get accurate information with slightly more complex polling techniques. Professor Daniel Corstange at Columbia has been working with a method that could help here. Suppose for a moment that people are willing to admit to all types of bribery except bribery of the judiciary. You then give people a list of public actors (police, teachers, postal workers etc…) and simply ask them to tell you how many on the list they have bribed. If you give half your sample a list that includes judges and half a list that does not, you can calculate the number of people bribing judges without having to ask anyone to specifically admit to it.
At Transparency International, we are always looking for ways to improve our data collection practices to ensure that we are getting accurate data to inform our anti-corruption advocacy messages. We do take the problem of under-reporting seriously, and draw on our National Chapters, and country experts, to advise on the local context or potential questionnaire translation issues. In addition, conforming with survey good-practice, respondents are informed about their right to anonymity and confidentiality, and are assured that the interviewers are not from the government or a government agency.
A few important points raised here in the comments have already been reflected in the next round of the Transparency International Global Corruption Barometer (GCB) survey, currently in field, and due to be published in 2016. We are partnering with the AfroBarometer, and we are now using their survey wording of “How often, if ever, did you have to pay a bribe, give a gift, or do a favour” for a number of different public official. By asking not just about ‘bribes’, respondents may feel more comfortable answering in the affirmative, and this will also allow us to include non-financial forms of public sector corruption in our calculations.
It’s important to note that the Rose and Pfeifer have already assessed false reporting in their book across the various barometer surveys in three ways, and did not find much evidence for false reporting:
1. Comparing the ‘don’t know’ and ‘refused’ rates for the bribery questions. They find that on average in each of the barometers only a small percentage of respondents refuse to give an answer or say that they don’t know.
2. Comparing the bribery rate of those who think that paying a bribe is wrong compared to those who think that paying a bribe is not wrong. One would expect that those who think that the act of paying a bribe is wrong, would be less likely to report that they have done this. Rose and Pfeifer find no statistically significant difference in the LiTS in the bribery reporting rate between those who think that the paying a bribe is wrong vs those who think that it is acceptable, and only a small difference in the AfroBarometer results.
3. Comparing the ‘don’t know’ and ‘refused’ rates for the bribery questions of those who have a lower level of education to those who received more education. They find that the Eurobarometer survey does show a higher proportion of ‘don’t knows’ or ‘refused’ answers between those with lower level education and those with higher level education.
That said, comparing the results at the total barometer survey level is likely to mask the cross-cultural differences (including national or sub-national political, social and historical context) which impact false reporting rates. Where sample sizes are robust, national level analysis could be run or where national sample sizes are too small (which is likely for sub-samples of bribe payers), grouping small numbers of countries which share political, social and historical similarities would allow for an assessment of false reporting.
This blog makes a big claim about the Kraay and Murrel paper: “If that result were to hold across all countries”… As the majority of the Kraay and Murrel data comes from a Gallup World Poll in 9 Asian countries plus 1 comparator from the Americas (Peru), the results are of very limited use for drawing global conclusions at this stage. Reticence and guilt on issues of corruption are likely to vary in other parts of the world, and it would be important to first repeat the study in Africa, the Middle East, Europe & other countries in the Americas before extrapolating results to the global population.
Very interesting. Thanks for the thorough and thoughtful response. I may try to weigh in later with more thoughts on the data issue, but I did want to make one observation about the second of the three points from Rose & Pfeifer that you highlight:
R&P report that those who believe paying bribes is morally wrong do not report a lower rate of bribery than those who say that paying bribes is not morally wrong. You and R&P infer from that that under-reporting is likely not a problem, on the logic that under-reporting bias would disproportionately affect those who think bribery is immoral. (That may or may not be the case — if the reason for under-reporting is so-called “social desirability bias”, rather than internal shame, there might not be a difference in under-reporting. But put that to one side.) If we believe this result, there’s another quite striking implication: that moral values have no impact on behavior in this context — such that changing people’s values (so that they come to see bribery as wrong) would not actually have any impact on their propensity to bribe. If we believed this, it would further imply that attempts to change moral values — say, by education or public campaigns — is a waste of time and resources.
Do we believe that? Does TI believe that? If not, doesn’t that in turn suggest that we/you don’t actually place much stock in the result that bribery rates are for the most part unaffected by moral values? And if you don’t place much stock in that result for purposes of assessing whether we should try to influence public attitudes toward bribery, doesn’t that imply that we shouldn’t place much stock in it as a response to the under-reporting concern either? After all, I’d assume we need to be consistent on this point.
(A slight variant on the same point: Suppose those who believe bribery is immoral DO report paying bribes at a much lower rate. That would not necessarily be evidence of reporting bias; it could just be evidence that those who see bribery as wrong in fact pay fewer bribes.)
Thanks to Coralie Pring of TI for the update on the status of the 2016 Global Barometer survey. It is reassuring to learn that TI is aware of the under-reporting issue and taking measure to address it. Let’s hope the 2016 Global Barometer report contains a discussion of the adjustments made to reflect under-reporting.
I confess that I did not find Professors Rose and Peiffer’s arguments that survey respondents were not substantially under-reporting the payment of bribes persuasive. They advance three:
1. The simplest way for a survey respondent to conceal the payment of a bribe, they assert, is to reply “don’t know” or refuse to give any answer, and because the surveys show few “don’t knows” or refusals to answer responses, under-reporting is not a problem. Can people credibly tell a surveyor that they do not know whether they ever paid a bribe? Common sense suggests the easiest way to conceal payment of a bribe is to say “’no” falsely.
2. Their second argument is that the evidence shows that people who think bribery is wrong bribe as much as people who think it okay to bribe. But what do one’s personal views about the morality of bribery have to do with a reluctance to report paying a bribe? Rose and Peiffer seem to be assuming that those who think bribery is wrong are more reluctant to admit paying a bribe than those who don’t. What evidence supports this assumption?
This second argument also depends upon an antecedent condition: that the decision to pay a bribe is under the control of the would-be bribe payer. If both those who think it wrong to pay a bribe and those who think it okay are free to decide in any given instance whether to pay or not pay, then one would expect those who believe bribery wrong to pay fewer bribes. But what if a bribe is being demanded in return for treating one’s sick child? Or avoiding being jailed on trumped up charges? Would those with moral objections to bribery be willing to let their child suffer? Go to jail? Maybe a pure Kantian would, but what about the other 99.9 percent of the population?
3. The third Rose-Peiffer argument why under-reporting of bribery is not significant rests on the finding from the surveys that the percentage of briber payers does not differ by level of education. Both highly educated and poorly educated report bribe at the same rate. They couple that finding with this one: ”There is some evidence in Western nations that the under-reporting of socially undesirable activities varies with education.” They then make this stretch, hence “more educated people will be less likely to report paying bribes” (p.47). Since more educated don’t, at least in Euro-barometer surveys, the authors conclude under-reporting is not an issue. This conclusion rests on one paper (“some evidence”) that under-reporting varies by educational level drawn from one region. It also rests on the assumption that when confronted with a demand for a bribe, the would-be payer has the option to refuse.
All three Rose-Peiffer arguments depend on what to me appear to be shaky assumptions of fact. Of course, as the original post notes, in science one does not argue about facts, one conducts research to test their validity. If the original post and the many insightful comments it has drawn prompts such research, it will have served its purpose.
More on willingness to report and education
The Limitations of Education for Addressing Corruption — Lessons from Attitudes Towards Reporting in Papua New Guinea, Grant Walton and Caryn Peiffer
June 1, 2015
Crawford School Development Policy Centre Discussion Paper 39
Educated citizens are often considered more likely to report corruption; this belief shapes anti-corruption campaigns. However, we know little about how other factors may interact with education’s impact on willingness to report corruption. This paper examines data from a household survey undertaken in Papua New Guinea. We find that when respondents were better educated and believed corruption would be addressed by the government, they were more willing to report various types of corruption to officials. However, the positive effects of education on willingness to report corruption are significantly diminished when citizens lacked trust that authorities would address corruption.
Pingback: In Bribery Experience Surveys, Should You Control for Contact? | GAB | The Global Anticorruption Blog