Measuring corruption : whose perceptions should we rely on ? Evidence from Iceland

The extent of corruption in Iceland is highly contested. International corruption measures indicate a relatively small amount of corruption while domestic public opinion suggest a serious corruption problem. Thus, uncertainty prevails about the actual extent of corruption and whose perceptions to rely on. This problem is relevant for corruption research in general. Perceptions are increasingly used as proxies for the actual levels of corruption in comparative research. But we still do not know enough about the accuracy of these proxies or the criteria they must meet in order to give dependable results. In fact, radical differences exist concerning evaluations of perceptions between those who believe in unbiased learning and those believing perceptual bias to be widespread. The purpose of this article is, therefore, to attempt to gauge which factors may influence how perceptions of corruption are shaped and why differences in corruption perceptions between different groups may be so pronounced. We present findings from original survey data from three parallel surveys – among the ‘public’, experts, and ‘municipal practitioners’ – conducted in Iceland in 2014. Expectations based on the perceptual bias approach are tested, indicating that perceptions may be affected by (1) information factors, (2) direct experience of corruption and (3) emotive factors. The validity of perception measures should be considered with this in mind. Domestic experts are likely to be well informed and avoid perceptual bias to a greater extent than other groups. Our examination of the Icelandic case suggests that the Corruption Perception Icelandic Review of Politics and Administration Vol. 12, Issue 2 (215–236) © 2016 Contact: Gunnar Helgi Kristinsson, ghk@hi.is Article first published online December 19th 2016 on http://www.irpa.is Publisher: Institute of Public Administration and Politics, Gimli, Sæmundargötu 1, 101 Reykjavík, Iceland Stjórnmál & stjórnsýsla 2. tbl. 12. árg. 2016 (215–236) Fræðigreinar © 2016 Tengiliður: Gunnar Helgi Kristinsson, ghk@hi.is Vefbirting 19. desember 2016 Birtist á vefnum http://www.irpa.is Útgefandi: Stofnun stjórnsýslufræða og stjórnmála, Gimli, Sæmundargötu 1, 101 Reykjavík DOI: http://dx.doi.org/10.13177/irpa.a.2016.12.2.2 This work is licensed under a Creative Commons Attribution 4.0 License. 216 STJÓRNMÁL & STJÓRNSÝSLA Measuring corruption: whose perceptions should we rely on? Evidence from Iceland Index (CPI) tends to underestimate corruption problems in ‘mature welfare states’, such as Iceland, whilst the general public tends to overestimate it.


Introduction
A controversy has raged about the levels of corruption in Iceland. 1 According to Transparency International's (TI) Corruption Perception Index (CPI), Iceland was ranked as the least corrupt country in the world in 2004 and 2005.Several authors have disputed the exactness of this evaluation, maintaining that cronyism, nepotism, clientelism and other forms of abuse of power probably played a role in the genesis of the financial crash in 2008 (e.g.Johnson et al. 2013;Gylfason 2014;Vaiman et al. 2011;Wade and Sigurgeirsdottir 2010), while others have argued practices such as cronyism indeed has had a long history in Icelandic politics (Kristinsson 2012).
Is the CPI seriously flawed in the case of Iceland?One recurring criticism is that the CPI to a large degree relies on perceptions of foreign business executives and assessments from experts, perceptions that generally are based from outside the country in question (see for example Thompson & Shah 2005;Arndt & Oman 2006;Andersson & Heywood 2009).In addition, these assessments tend to hinge on bribe-giving and bribetaking, i.e. not de facto measuring the somewhat broader phenomenon of corruption as captured by the standard definition employed by for example TI itself, i.e. 'the abuse of entrusted power for private gain'.Measures that primarily or only look at bribes, of course, tend to disregard more 'sophisticated' (e.g.Papakostas 2009) variations of illicit behavior associated with the concept of corruption, precisely behavior that is tied to cronyism, nepotism and clientelism and other forms of violations of the norm of impartiality (e.g.Rothstein & Teorell 2008).
Consequently, and perhaps unsurprisingly, scholars have demonstrated discrepancies between how international experts perceive and describe the problem of corruption in a given country, and how citizens living there actually perceive it, e.g.Lin and Yiu (2014) for Asia, and Razafindrakoto and Roubaud (2010) for sub-Saharan Africa.Taking Sweden as an example, Bergh et al (2016) highlight that despite the fact that Sweden is continuously ranked as one of the world's least corrupt countries, its citizens seem to believe that Sweden's corruption problems are considerably more widespread and serious than international indices like TI would lead us to believe; and this holds true particularly when compared to Denmark and Norway.
Where does this leave us?We agree with Teorell (2014) that perceptions of course are not perfect measures.But, perceptions are probably the best measure we got anddepending on the quality of the data -could in some cases be a good indicator of the extent of corruption.However, the question of whose perceptions should be consulted, and about what they should be asked, needs to be carefully considered.
Against this backdrop, it is pressing to analyze why mismatches in corruption perceptions among different categories of actors are observed; i.e. why we in some instanc-

STJÓRNMÁL
& STJÓRNSÝSLA es see pronounced differences between for example "international experts" and more country specific sources such as "ordinary citizens", "domestic experts/analysts" and "domestic politicians and civil servants".More precisely, the mismatches that have been observed make it interesting to ask which factors seem to influence how perceptions of corruption are shaped at the level of the individual within different groups.This brings us to the general question addressed in this article: whose perceptions should we rely on when assessing the extent of corruption in a given country, and what are the pros and cons of consulting different demographics?
The article proceeds as follows.We start out by summing up the main theoretical claims concerning the subject at hand and derive expectations from what has been dubbed "the perceptual bias approach", which basically suggests that public perceptions may be biased in a number of ways.We then go on describing our data and give arguments for why we have selected Iceland as our case, before presenting findings from original survey data which stem from three parallel surveys conducted in Iceland in 2014 in order to analyze our expectations.

Perspectives on corruption perceptions: what should we expect?
If we plan to use perceptions as an instrument gauging corruption, we need to address the question of validity.Perceptions can either be taken at face value, i.e. as more or less accurate reflections of reality, or they can be subjected to critical evaluation as potentially biased and hence more or less inaccurate.These two positions have a distinguished pedigree relating to fundamental issues of epistemology and the philosophy of science.In the present context, however, we wish to escape as far as possible the broader philosophical context and concentrate on the matter as it relates to corruption.In line with the approach adopted by e.g.Gerber and Green (1999) and Fischle (2000) -who incidentally advocate rather different positions -we distinguish between the theory of unbiased learning and that of perceptual bias.
According to proponents of unbiased learning, experience is a relatively sound source of knowledge and people can, for the most part, be relied on to perceive things accurately given fairly basic conditions.While early voting research in the US seemed to fundamentally challenge this position (Campbell et al. 1964), subsequent research is more ambivalent (e.g.Nie et al. 1976).Gerber and Green (1999) argue that, generally speaking, there is surprisingly little evidence for selective perceptions of the kind reported e.g. in the American Voter study and in the U.S. context: Democrats, Republicans and Independents perceive current events in a similar manner.
When it comes to corruption research, Transparency International (2010) has conducted analyses in order to demonstrate that the Corruption Perception Index does not diverge all too much from how ordinary citizens perceive problem.In line with this, Holmberg (2009) has reported strong rank order correlations between public and expert perceptions of corruption, and Charron (2016) finds evidence of strong international correspondence between country rankings based on corruption perceptions and corruption experiences.Charron concludes that 'strong counter-evidence is found to the

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland prevailing pessimistic claims in the literature -the consistency between actual reported corruption, as well as citizen and expert perceptions of corruption, is remarkably high and such perceptions are swayed little by 'outside noise'' (compare also e.g. the literature review in Lin and Yu 2014: 144).
While the correlations between public and expert perceptions suggest that they are to some extent exposed to similar factors, it is entirely possible that one group nonetheless has systematically exaggerated perceptions of corruption compared to the other.High correlations do not preclude such a possibility.Imagine a four case scenario where experts and public rate corruption on a ten-fold scale in the following way: Thus 10:60, 20:70, 30:80 and 40:90.Pearson's correlation would yield a perfect r = 1.00*** despite a substantial and systematic difference in expert and public perceptions where the public perceives far greater corruption than the experts.
The perceptual bias position suggests that perceptions can be flawed and manipulated in a number of ways.It is well known from different strands of social research that surveyed perceptions may differ considerably from objective indicators.Examples include institutional performance (Flynn 2007) and crime perceptions, where there seems to be a strong tendency among survey respondents to overestimate the threat from crime (e.g.Baier et al. 2016).This has influenced research in sociology and social psychology, indicating that perceptions are in many cases selective or biased.According to research on motivated reasoning and related topics, individuals react to information on the basis of prior affect and partisanship, and may hence be susceptible to attribution errors in detecting deceptions (e.g.Burden & Hillygus 2009;Fischle 2000;O'Sullivan 2003).
Therefore, a large number of factors may influence perceptions independently of the actual level or direct experience of corruption.Donchev and Ujhelyi (2014) maintain that corruption experience is weakly related to corruption perceptions in international research while economic development, democratic institutions and Protestant traditions may systematically lead to lower perceptions of corruption.Factors which are likely to affect perceptions of corruption within a single system include partisanship and ideology, socioeconomic status and political involvement.
According to this, direct experience of corruption may give a better indication of its actual levels than beliefs and perceptions because it is more objective.Respondents reporting perception may be influenced by a number of factors, but when asked about facts (actual experience) the scope for subjective evaluations is more limited (although, of course, it may still well exist).Rose and Mishler (2007) point out, in the case of Russia, that there is a "big gap between the 86 percent who perceive most public officials as corrupt and the 23 percent who say their household has paid a bribe in the past two years' concluding that 'neither the payment of bribes nor the number of contacts with public officials has a significant effect on the perception of corruption" (p. 1).Hence, "individual and aggregate perceptions of corruption are not a surrogate indicator for actual corruption in Russia" (p.20) they maintain and 'a bad press and bad mouthing are more important than bad experiences in determining perceptions of corruption'.According to the perceptual bias approach we should therefore not expect close cor-

STJÓRNMÁL
& STJÓRNSÝSLA respondence between experience of corruption and perceptions of corruption among the public.Some smaller groups, however, may have an information advantage compared to the general public.These are the (somewhat imprecisely and loosely termed) 'experts' on which expert surveys build.While the general public may typically lack the information and skills required for accurate evaluation, smaller groups of experts or persons with extensive experience in the relevant fields (e.g.elites) can provide a more accurate account.As Lin and Yiu (2014) argue: through education, training, and experience, experts generally have specialized knowledge in some particular subject and can hence act as more reliable sources of information than the public; for instance, when it comes to gauging the level of corruption within a country.
Accordingly, expert surveys have gained increasing currency in political science in recent years, often -as Clinton and Lewis (2007, 4) point out -providing means of "assessing quantities that are not easily quantifiable".In the case of corruption research, in fact, expert surveys have become the standard tool for measuring corruption and comparing levels of corruption cross-nationally.The advantages of employing expert opinions, however, still begs the question of which experts, chosen from where, are relevant for what types of questions.
To take but one example, most famously, TI's Corruption Perception Index is basically a composite measure based on expert perceptions.While the accuracy of expert measures is in some cases contested (see for example Anderson & Heywood 2009), a perhaps more damaging criticism is aimed at the choice of experts and what questions they are presented with, rather than the core idea of asking the people who should know.Foreign businesspeople, on which TI partly relies, may not be well suited for gauging corruption levels in a given country, and confining the search with bribes may blind researchers to other more subtle, or 'sophisticated', forms of corruption.
At any rate, according to Teorell (2014, 67), there is no research -at least to his knowledge, as he puts it -suggesting that expert evaluations are systematically biased in the same manner as public opinion.In advanced countries, Maeda and Zigfeld (2015) argue, social position may affect perceptions of corruption.High-income and highly educated citizens tend to perceive lower levels of corruption compared with other citizens (p.5).Blais et al. (2015) also come to the conclusion in most of their cases that the better informed citizens perceive less corruption than less informed ones.Hence, we expect that the general public diverges considerably from groups with an information advantage in its perceptions of corruption; and we expect the groups to diverge in two ways in less corrupt countries: • The public is likely to perceive higher levels of corruption than elite groups.
• Differences in perceptions in countries with lower levels of corruption are likely to be particularly large with regard to the more serious forms of corruptioni.e. the general public will perceive it to be higher compared to groups with an information advantage.

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland Additionally, research in the perceptual bias tradition suggests that a whole range of emotive factors are likely to influence corruption perceptions.This includes a number of partisan and ideological factors.Thus, Blais et al. (2015, 6) note that partisans of incumbent parties perceive less corruption while non-partisans and less informed perceive more.Voters tend also to be more tolerant towards corruption cases when they affect their own parties than others (Anduiza et al. 2013).We therefore expect emotive factors, in particular ideological ones -i.e.whether or not one supports the insider/ruling parties or not -are likely to influence perceptions of corruption.Basically, then, there are three expectations we have on our empirical material, which we will study below: 1. Contrary to the unbiased learning approach, but in line with the perceptual bias approach, we expect not to find any strong connection between experience of corruption and perceptions of corruption among the public.Something else, besides experience, explains how the majority of the public assesses the problem of corruption.2. Related to the expectation above, we also expect that the general public diverges considerably from 'experts' (i.e.people with an information advantage) in its perceptions of corruption; and we will see the divergence manifested in higher general levels of corruption perceived by the public especially with regard to the more serious forms of corruption.3. We expect emotive factors, and in particular whether or not one supports the insider/ruling parties or not, will influence perceptions of corruption: outsiders will perceive the problem to be worse, than supporters of the majority/insider parties.
Grounded in an analysis of these expectations here below, we will sum up with a tentative evaluation of whose opinions about what one ought to pay closer attention to if one is really interested in the level and seriousness of the corruption problem within a country.

Case selection, data and approach
In our quest to initiate an informed discussion on what factors shape perceptions of corruption, we have collected data from one of the world's most homogeneous countries -Iceland.Theoretically, we believe homogeneity is crucial when addressing the subject at hand, since homogeneity suggests that different segments of the population are likely to be subject to similar media coverage, influences and experiences of corruption.This means that our research design does not have to take into account the potential bias created by radically different experiences, cultures or scandals revealed by the media.Firstly, there are no substantive ethnic, religious or linguistic minorities in Iceland, although a trickle of immigration in the last few decades has modified the uniquely high level of cultural homogeneity to a certain extent.According to the Global Gender Gap

STJÓRNMÁL
& STJÓRNSÝSLA report, Iceland also was the most gender equal country in the world in 2015 (1/145) (World Economic Forum 2015) and the Gini-index, according to World Bank estimates, is among the lowest, indicating a similar level of income equality as in the other Nordic countries. 2 Importantly for our purposes, media literacy is high.According to Gallup polls, a large proportion of the population reports using major media on a regular basis. 3Given the small size of the media market there are only two major newsrooms serving the television and radio market.The situation for printed media is not much different.Although the political forces regularly attempt to influence the media, the agenda and style of reporting is basically similar. 4Hence, most Icelanders will have approximately the same information about most known scandals, and in addition, it is highly unlikely that media users in Iceland experience radically different versions of scandals and corruption through media reporting.We therefore argue that Iceland´s homogenous character makes it well suited for the analysis of perceptions of corruption among different segments of the population within one and the same country.
Our research design is focused on three parallel surveys conducted at the end of 2014 among groups which may be assumed to be exposed to different types of influence concerning corruption perceptions in an otherwise homogenous population.In these, we targeted three demographics that we argue to be theoretically relevant for the purposes at hand which for reasons of simplicity are called: the 'public', the 'experts' and the 'municipal practitioners' (the last group being a combination of local government officials and politicians).These three groups of respondents were chosen to reflect what we believe being different levels of knowledge, experience and exposure to perceptual bias.Knowledge can be assumed to be high among 'experts', and relatively high also among 'practitioners' -although perhaps more limited in scope.Compared to these groups, we should expect knowledge among the 'public' to be at a lower level.
Direct experience of corruption is likely to vary with the actual amount of corrupt activity taking place.The public is likely to have direct experience of corruption in highly corrupt systems where corruption is 'systemic', but less so in low-corrupt settings where corruption has an 'isolated' character (e.g.World Bank 1997).The type of corruption one experiences may also be different.The same goes for experts, although experts with close associations to government are likely to come into contact with (or know about concrete cases of) corruption more commonly than the public.Practitioners are also likely to have more direct experience of corruption than the public.
As regards sources of bias, these may affect all three groups, while a priori we should expect the experts to be least affected, everything else being equal, given their higher level of knowledge and experience of government.The public, with smaller knowledge and less direct experience (assuming a low corruption case) may be more subject to biased perceptions than the experts.The practitioners, despite considerable firsthand knowledge, may be more biased than the experts given their close association, and probably closer emotional attachment, to the political system and local government.This could lead them to focus on justifications of questionable practices at the expense of

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland principles of impartiality, or alternatively, have incentives to paint a cleaner picture of what is going on at the local level: because of their status as 'insiders' they could be suspected to have a stake in the issue at hand, and hence prone to underreport the frequency of corruption.Data was obtained by running three parallel surveys in Iceland 2 -29 Dec. 2014 using, for the most part, the same battery of questions.The three groups surveyed were firstly a stratified random sample of panel respondent contacted by the Social Science Research Institute of the University of Iceland, representing the general public.The number of respondent was 960 and net response rate 66%.The second group of respondents included elected representatives and top administrators in municipalities ('practitioners') with over 2 thousand inhabitants.Respondents were 208 (a response rate of 70%) of whom 60% were elected representatives and 40% administrators.The final group of respondents represents a group which comes as close to being expert respondents as possible.Members of the Public Administration Association in Iceland (where a master's degree in public administration is an entry requirement) were approached and answers received from 79 respondents, which amounts to a net response rate of 66%.

Results and analysis
Our dependent variable is perceptions of corruption.Perceptions of corruption were measured in an identical manner among the public, the experts and the practitioners.

STJÓRNMÁL & STJÓRNSÝSLA
First, respondents were asked about perceptions of different types of corruption in local government, on the assumption that respondents among the public are more likely to have direct experience of local rather than national level corruption.Later in the questionnaire respondents were asked about their perceptions of the general level of corruption among politicians, public employees and in local governments.In tables 1 and 2 we show how the response was distributed.
Rather than using the word corruption (or related concepts), we opted for describing the activity in question. 5The purpose was to, as far as possible, avoid the strong normative connotations often associated with the concept of corruption.We have no definitive way of deciding at this point if the figures in table 1 should be considered high or low.However, for a country that generally speaking and historically has been considered a low-corruption case, some of the figures are surprisingly high.The most common form of corruption in local government according our respondents is favoritism in public appointments, which 72% think is common or very common.Fraud scores lower (57% think it is common + very common), followed by bribes (51% common + very common), embezzlement (34% common + very common) and extortion (25% common + very common).The most surprising figure is perhaps the one concerning bribes, which in the Icelandic public debate are generally not considered to be particularly common and very few court cases exist where public officials have been charged with or found guilty of accepting bribes.
A different item in the questionnaire concerned more general evaluations of corruption in Iceland, among politicians, public employees and more specifically, in the municipalities.Again, the figures must be viewed as surprisingly high, considering that Iceland has traditionally been viewed as one of the world´s least corrupt countries in international indices.Corruption is thought to be relatively common among politicians (69% rather or very common) but a sizeable majority of respondents also thinks it is rather or very common in local government (62%) and among public employees (57%). 6

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland

Direct experience and corruption
We expect corruption perceptions to be influenced by a number of factors apart from direct experience.Direct experience is therefore expected to be only moderately correlated to perceptions.
Respondents among the public were asked about their experiences of corruption, i.e. if they had personal experience, knew someone with personal experience, if they had heard of such things or knew of them through the media.(Questions were asked concerning each type of corruption separately so the figures in table 3 do not add up to 100%).A striking difference appears between the perceptions of corruption reported in table 2 and personal experience in table 3. Whereas 51 -72% of respondents believed favoritism, fraud and bribes to be common, only 7 -19% report personal experience of such activities.In the case of bribes, for example, 51% of respondents believe them to be rather or very common while 93% have no personal experience of such activities.And importantly, considering only those who believe bribes to be common or very common, only 15% of them have any personal experience of them whatsoever.
To test the relationship between experience and perception of corruption we ran correlations between dummy variables representing the different types of experience (0 = no experience, 1 = experience) and the different types of corruption.The relationships between personal experience and perceptions of corruption turned out to be weak (r between .16 and .25).Knowing someone with personal experience gave slightly stronger correlations (r between .22 and .31)but still, the figures are far from convincing.Having heard about corruption or knowing of it through the media basically gave no meaningful correlations (r was from -.11 to .16).
To test the relationship of experience with corruption perceptions still further, we constructed a fivefold scale to measure the 'directness' of corruption experiences.According to this measure 0 means no experience of corruption, 1 only media experience, 2 that the most direct experience was hearing about corruption, 3 means knowing some-

STJÓRNMÁL
& STJÓRNSÝSLA one with personal experience and 4 having personal experience.Again, the results were less than convincing: while the relationships were statistically significant for all types of corruption, they were on the whole weak, giving r between .32 and .37.Such weak correlations (explained variance around or below 6%) give -we maintain -little reason to believe that direct experience, generally speaking, has an important impact on corruptions perceptions among the public.This supports the first expectation we formulated, i.e. that direct experience of corruption is not going to explain corruption perceptions among the general public, whose perceptions are likely to be influenced by a range of other factors.

The public compared to groups with information advantage
The second expectation we formulated stated that we expect the public to diverge considerably from groups which can be assumed to have an information advantage in that the public is likely to overestimate the amount and the seriousness of corruption taking place.We used two elite groups for comparisons with the general public, on the one hand 'experts' (members of the Public Administration Association) and 'municipal practitioners' (local officials and council members).Comparisons between the three groups in table 4 are presented on the basis of the proportion in each group claiming corruption to be rather or very common.Table 4 reveals a substantial difference between public perceptions of corruption on the one hand, and experts and practitioners on the other.In all cases the public is considerably more prone to believe that corruption is common compared to the elite groups which we assume to have an information advantage.Thus, members of the public are three times as likely to believe that bribes are common compared to the experts and more than twice as likely to believe embezzlement is common.With regard to favoritism, fraud and extortion the difference is smaller, but still considerable in all cases.The difference between the public and practitioners is even more pronounced.Members of the public are more than 25 times as likely to believe bribes to be common as practitioners and over 11 times as likely to think the same of embezzlement.In the case of extortion, fraud and favoritism the difference is less striking but still very con-

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland siderable.Practitioners are in all cases less likely to believe corruption is common than the experts.We shall return to the difference between experts and practitioners below.However, suffice to say here, is that the fact that practitioners are less prone to see corruption than the two other groups could be interpreted in view of the fact that they can be regarded as insiders, in a sense making statements about their own backyard, and hence not necessarily with them having a 'information advantage'.An interesting feature of table 4 is that although there are marked differences in how widespread beliefs in corruption are between the three groups, some trends and similarities are to be observed as well.Among all three groups, favoritism is the most commonly perceived form of corruption in Iceland, followed by fraud.Bribes, embezzlement and extortion are believed to be less common among all the surveyed groups, although the belief that bribes are common is particularly common among the public.This indicates that while the tendency for exaggerated beliefs in corruption may affect the groups differently, they are nonetheless subject to some common influence which affects the relative spread of such beliefs, i.e. which type of illicit behavior is most common as well as rarest in Iceland.
Part of our second expectation was concerned with the seriousness of corruption.According to this, the public should be more prone to believe that serious corruption takes place than the other groups.We have no absolute standard against which we can measure the seriousness of corruption.One way, however, to assess the seriousness of corruption in a country where the rule of law is well established is to consider its legality or illegality.Two of the types of corruption which were considered in our questionnaire are unequivocally illegal, i.e. bribes and embezzlement.The three remaining types are sometimes illegal and usually ethically questionable, but may not always imply law breaking in the strictest sense, at least in the manner we put the questions.Threats in the case of extortion may be implicit or hinted at rather than explicit.Avoiding uncomfortable information is often a question of putting the right spin on an issue rather than violating the public information act.And interfering with public appointments to non-political positions is an art form well established in Icelandic politics and administration without necessarily involving law-breaking. 7In this sense, bribes and embezzlement are more serious forms of corruption than the other three.According to the second part of our second expectation, we should expect the difference between public perceptions and the other two groups to be greater with regard to the more serious forms of corruption than with regard to the less serious ones.
Looking back at table 4, the pattern seems broadly consistent with the second part of our second expectation.The public is 1.2 -1.7 times more likely to think fraud, extortion and favoritism are common than the experts while it is 2.3 -3.0 times more likely to think embezzlement and bribes common.A similar pattern holds for the practitioners.The public is 3.4 -3.7 times more likely than the practitioners to think that favoritism, fraud and extortion are common and it is 11.3 -25.5 times more likely to think that embezzlement and bribes are common.
Thus, we maintain, our findings support of the perceptual bias support.The ublic is

STJÓRNMÁL & STJÓRNSÝSLA
more likely to think that corruption is common than groups with an information advantage, and the difference gap grows the more serious forms of corruption we consider. 8 We have, of course, no conclusive way of establishing that the expert and practitioner perceptions are more accurate than those of the public.This is simply an assumption on which our research design is based.There are however two strong arguments for taking their perceptions more seriously.In the first place there is hardly any doubt that these groups know more about what goes on in the murky corners of the political system than the general public.They are likely to have both greater general knowledge of how the system works and greater personal experience -first or second hand. 9Secondly, the perceptions of the experts and practitioners are to a far greater extent based on personal experience than those of the public.In table 4 above, we found only very modest correlations between the direct experience of the public of corruption and its perceptions of corruption.In table 5 we produce comparable figures for the experts and practitioners.While we would expect the experts primarily to have second hand knowledge of corruption it seems that personal experience of the more serious types is likely to influence their perceptions significantly.This may mean that they are skeptical concerning corruption claims unless they have first-hand experience of it.The practitioners, on the other hand, tend to base their corruption perceptions more on direct experience than voters except in the cases of embezzlement.The deviation in the case of embezzlement remains a puzzle to a certain extent, but it should be kept in mind that according to both the experts and practitioners, embezzlement is relatively rare.

Do emotive factors influence perceptions of corruption?
If public perceptions of corruption have little to do with direct experience and seem rather exaggerated compared to better informed groups, the question arises: how does the public form their beliefs about corruption?Research cited above indicates that emotive factors may play a role and our third expectation states that emotive factors, including ideological ones, probably influences perceptions of corruption.To study this,

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland respondents from the general public sample were asked about voting intentions, using standard questionnaire items from the Social Science Research Institute.To indicate closeness to government the parties were ranked according to their share of government power since the turn of the century.The party with the longest periods in government is the Independence Party (IP), followed by the Progressive Party (PP), the Social Democratic Alliance (SDA), the Left Greens (LG) and "others" (an array of populist forces mainly but not exclusively on the centre-left).In each case we get a broadly similar pattern.Supporters of 'insider' parties (those who have had more share of governmental power after the turn of the century) perceive smaller corruption than supporters of 'outsider' parties.Thus, supporters of the Independence Party in all cases perceive a smaller amount of corruption than those of all the other parties while supporters of new or non-represented parties (the 'other' category) perceive the greatest amount of corruption (tying with the Left Greens in the case of favoritism).This is irrespective of the majority in the particular municipality in which the question was asked, which yielded no significant results.If we perceive insider-outsider status as an ordinal scale and correlate it with perceptions of overall political corruption in Iceland we get a Spearman's rho of -.38 which indicates a stronger relationship than obtained through the analysis of corruption experience among voters.Ideological or partisan factors, in other words, provide an equally or even slightly stronger account of corruption perceptions than experience of corruption.
This simple test gives only a rough indicator of the relationship between emotive factors and perceptions of corruption.The data lacks further information concern-

STJÓRNMÁL
& STJÓRNSÝSLA ing such factors but our analysis seems to indicate a promising path for future research.The question remains, however, if only the general public is receptive to emotive factors or if they may affect expert groups as well.As noted above there are considerable variations in the perceptions of corruption among our experts and our practitioners.The experts from the public administration association perceive a greater amount of corruption at the local level than the practitioners.It is important to select expert groups for survey purposes so as to minimize the amount of bias likely to affect their evaluations.One indicator that a bias exists may be the presence of large variations in the experts' evaluations.Experts who disagree wildly lack credibility.With this in mind it is interesting to compare the answers of our two groups concerning the spread of corruption in Iceland.The picture we get is that in all cases between 46% and 67% agree on a single alternative, which in all cases is 'rather rare'.This seems to indicate a relatively broad consensus

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland among experts that corruption is rather rare.If we add 'rather rare' and 'hardly takes place' we get 53% to 87%, i.e. somewhere between a clear and an overwhelming majority.The evaluation of corruption we get through this data is very unlike the one examined in table 2 above, according to which 57% to 69% of the public believed corruption to be rather or very common.While we take this to indicate that expert opinion goes against the mainstream evaluation among the public, it should nonetheless be noted that a sizeable minority of the experts believe corruption to be rather widespread, especially among politicians and in local government.
Another cause for concern is the difference between the experts and the municipal practitioners.Whether the estimates of the experts or the practitioners are more accurate is difficult to establish.The evaluations of the practitioners in the case of municipal government is suspect, of course, given their close association with it.In a sense, the practitioners are answering here as 'insiders' and can be suspected of having a stake in portraying their own backyard as less corrupt than they actually believe it is.Having a political or administrative career in municipal government may hence hold back these respondents' statements about the spread of corruption, or, alternatively make them more tolerant of what they may perceive as the practical necessities of local governance; hence, not admitting that corruption is actually taking place.Yet, one should also note that practitioners not only think corruption less widespread at the municipal level but also among politicians and civil servants more generally.Table 7 suggests that there is not a great deal of difference between the perceptions of municipal politicians and administrators, although greater belief among administrators in corruption among politicians is to some extent telling, while politicians are more likely to suspect civil servants of corruption.Thus, insider groups may wish to downplay

STJÓRNMÁL
& STJÓRNSÝSLA the corruption going on in their own backyard.Both groups, however, are clearly less inclined to believe in widespread corruption compared to the experts.A contributing factor is probably that separation in the tasks of politicians and administrators at the local level are less well established than in the national administration.Only the highest level administrators were included among our respondents, a group which is likely to be well integrated in political life, and in many cases includes non-elected mayors, who are usually closely associated with the political majority.

Conclusions
The debate about the pros and cons of international indices and their uses of expert perceptions captures a dilemma about how corruption could and should be measured.While some scholars highlight the advantages of employing expert perceptions, others have demonstrated a marked difference between the views of experts on the one hand, and the views of the public on the other.
On a general level, this controversy frames the overarching theme we have been interested in: Whose perceptions about what should be taken into account when we are interested in gauging to what extent corruption constitutes a problem in a given country?More precisely, we wanted to make an attempt to disentangle what factors seem to shape how corruption perceptions are formed by opening up one so called 'low corrupt country' and see what influences how different segments of the demographics within the one and same country view the problem.
So, where to look?In recent years, the case of Iceland has been highlighted as an example that illustrates how badly calibrated corruption indices may be -judged to be the world´s least corrupt country in 2004-2005 by Transparency International's CPI, then plummeting heavily first after the financial crisis in 2008 (and the debate that then followed suit concerning whether corruption and related practices may have played a role in generating the economic crash).We argue that Iceland has important advantages if one wants to understand which factors shape perceptions of corruption.The country is homogeneous in so many theoretically relevant respects while the extent of corruption is highly contested.In order to gauge what factors shape perceptions of corruption, we focused on three different groups of actors within the Icelandic demographic: 'the public', 'the experts', and 'the practitioners'.
Increasing use of perception measures as indicators of corruption calls for a critical evaluation of the factors which may influence or bias results.Our focus was on three such factors.In the first place we show that perceptions of corruption in Iceland vary considerably according to the level of information which respondents are likely to possess.Thus, respondents among the general public were much more prone to believe corruption to be widespread than groups which ex ante can be argued to have a substantial information advantage.Although we noted some differences among the expert groups, a sizeable majority believed corruption to be rather rare, contrary to prevailing views among the public.
In some cases, of course, lack of systematic or generalized knowledge among the public may be compensated for by a great amount of direct experience of corruption.

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland Direct experience is a far more objective indicator of corruption than perceptions.In the Icelandic case, however, the vast majority of respondents among the public has no personal experience of corruption, hence, their beliefs that corruption is so widespread must be based on something else.Although our data on emotive factors is limited, we found evidence that support for insider vs. outsider parties was significantly related to perceptions of corruption -in fact the effect was stronger than obtained by any measure of experience of corruption.And although we have not been able to analyze it with the data we have at hand here, maybe we are be witnessing something similar to what van de Valle (2008) found in the Belgian case, i.e. that perceptions of corruption seem to be embedded in general attitudes towards government, and this indicator therefore primarily measures general attitude of distrust towards the administration, and therefore should not been used as an indicator of actual corruption.
At the outset we contrasted the two conflicting perspectives of unbiased learning and perceptual bias and formulated expectations derived from the latter.These were in the first place that direct experience of corruption would prove a weak predictor of corruption perceptions; secondly that public perceptions would deviate considerably from those of groups with an information advantage and thirdly that emotive factors were likely to influence perceptions of corruption significantly.All three expectations were confirmed by our data, lending support to the perceptual bias perspective concerning public perceptions of corruption.This indicates that public perceptions of corruption need to be approached with caution as indicators of corruption.Dismissing the unbiased learning perspective altogether, however, may be premature.Our data does not contain a comparative dimension.Some research has revealed comparative evidence of a relationship between public corruption perceptions and corruption experiences (Charron 2016) as well as between public and expert perceptions (Holmberg 2009).This could indicate that even if public perceptions may be exaggerated they may nonetheless reflect underlying factors which are related to experience and expert evaluations.In our data we found that even if public perceptions are likely to be exaggerated they are nonetheless similar to the evaluations of the expert groups concerning the relative frequency of the different types of corruption.While we cannot be sure if this underlying factor is the actual level of corruption or perhaps some feature of the public debate, the fact that the pattern among municipal practitioners -who are both experienced and clearly skeptical concerning public wisdom on municipal corruption -is broadly similar to the one among the public may offer a significant clue.
The relevance of our study concerns not only the two different approaches of unbiased learning and perceptual bias but also the methodological one of whose perceptions we should look for as proxy measures for corruption.Given a choice between expert evaluations and public perceptions our conclusion is in line with the prevailing view that expert perceptions are to be preferred.But expert perceptions can clearly differ.In the Icelandic case there exists long-standing skepticism concerning the evaluations of foreign businesspeople on which TI relies a lot.They are unlikely to have substantial first-hand knowledge of different types of corruption in Iceland and even less likely

STJÓRNMÁL & STJÓRNSÝSLA
to have a balanced theoretical overview.In the choice of expert groups, we should ask three questions.Are they likely to have the broadly based knowledge and conceptual understanding in order to make proper evaluations?Are they likely to have direct experience of corruption which might compensate for lack of general knowledge?Are they likely to be affected by emotive factors of any kind, ranging from ideological factors to self-justification.In our research design we selected two elite groups with different characteristics in this respect.The 'experts' (members of the Public Administration Association) are likely to have theoretical knowledge and good conceptual understanding but may vary in experience.They are, moreover, not likely to be especially ideologically motivated.The second elite group, the municipal practitioners, consist of council members and executives of the municipalities.They are likely to be less theoretically proficient than the experts but with greater experience of the actual levels of corruption.However, their answers may be more influenced by emotive factors than those of the expert given their greater involvement with political life and perhaps a tendency for selfjustification given their proximity with municipal governance.Given a choice between the experts and municipal practitioners the expert perceptions therefore seem a better choice as an indicator of the actual level of corruption.
Our data allows us to present a more nuanced view of corruption in Iceland than hitherto available through sledge hammer statistics of the kind presented by TI or through public perceptions of corruption.Whereas placing Iceland as the least corrupt state in the world is likely to underestimate both various types of corruption taking place and the overall level, the deeply pessimistic view obtained through public perceptions is likely to be an overstatement as well.If we go by the groups most likely to have good general knowledge and first-hand experience, we come to the conclusion that corruption is rare but still clearly discernible.Less serious types of corruption, such as favoritism in public appointments and failure to disclose information, are more common than more serious forms such as extortion, bribes and embezzlement.Nonetheless, it should be noted, that a sizeable minority of the experts still believes corruption to be common, especially in the case of favoritism and fraud.
We believe that the results that have been presented here are all but trivial, and they could potentially have important policy implications.Many international aid-donors use indices as a guide when conditioning developmental aid (e.g.Kurtz & Schrank 2007;Andvig 2005).In line with some previous studies, our findings imply that measures of corruption seem to have their intrinsic problems, suffering from perceptual bias, adverse selection of experts, and also some conceptual fuzziness -i.e.mainly focusing on bribegiving and bribe-taking.Perhaps it is the case that, at least when it comes to the developed world and so called 'mature welfare states', that neither the foreign experts (who tend to underestimate the domestic problems) nor the domestic public (which tends to overestimate the domestic problems), are the right demographics to consult; but rather we should turn to some mix of domestic, well-educated experts that are presented with questions about different forms of corrupt/illicit behavior, in order to get a more calibrated and nuanced picture of the problem at hand.

STJÓRNMÁL & STJÓRNSÝSLA
Measuring corruption: whose perceptions should we rely on?
Evidence from Iceland

Notes
1 The project was part of the Power and Democracy project at the University of Iceland.The authors are greatful for the comments of two anonymous reviewers on an earlier version of the paper.2 http://www.indexmundi.com/facts/indicators/SI.POV.GINI/rankings 3 http://gagnatorg.gallup.is/survey/data/list/cat_id:2293 4 Iceland switched from party controlled press and public radio and television under political scrutiny during the 1980s and 1990s to a more commercialized system.The political forces remain highly sensitive to the management of public radio and television as well as ownership of private media, cf.Kristinsson 2012, 196-197. 5 Our basic approach was to use non-technical terms describing the conduct in question in order to minimize the risk of different groups understanding the questions in different ways.In the case of extortion the Icelandic version was "að stjórnmálamenn eða opinberir starfsmenn í sveitarfélögum á Íslandi láti undan hótunum af einhverju tagi", for embezzlement we used "nýti aðstöðu sína til að draga til sín fjármuni eða afla tekna umfram það sem þeim ber í launagreiðslur", bribes was "þiggi greiðslur eða hlunnindi fyrir að hygla þeim sem slíkt býður", fraud was "leyni mikilvaegum upplýsingum eða gefi vísvitandi villandi upplýsingar til að forðast gagnrýni", and favoritism "mismuni aðilum við ráðningar í störf vegna pólitískra tengsla, kunningskapar eða fraendsemi".6 A noticeable feature of the corruption perceptions indicated in table 2, is that compared to the ICENES study, conducted only a year-and-a-half earlier, perceptions of corruption seem to have increased markedly.To us it seems highly unlikely that corruption or corruption related events can explain this variation between the two surveys to a satisfactory degree.For instance, although not a corruption scandal proper, our survey took place well before the exposures of the Iceland-related Panama papers-scandal in the spring of 2016.We believe that the most likely explanation for the increase in 2014 is that the questions were asked at the end of the questionnaire after respondents had been subjected to a large number of questions on corruption which were likely to focus their attention on the problem.This may indicate that perceptions of corruption are volatile although it is of course not conclusive evidence.7 The power of politicians to make appointments to many administrative positions is uncontested and a certain amount of discretion is legally recognized.But they also have considerable power in shaping the whole process, including the timing of advertisements, how evaluation of candidates takes place and the precise qualifications asked for when positions are advertised.For further discussion, see Kristinsson (2012).8 One could raise objections to our interpretations and pose alternative explanations as to why 'experts' and 'practitioners' perceive less corruption than does the public.Iceland is a small country, our samples in these two groups are rather small, and the survey asks about controversial questions with potential adverse consequences for this groups of respondents.If respondents do not trust the anonymity of the survey, they may have moderated their answers, stating lower levels of corruption than they de facto think there is.However, we do not believe in such an alternative explanation.
Taking part in the surveys was entirely voluntary.If respondents felt they had anything to fear, they would not have taken part.Additionally, the survey was carried out by a highly respected agency, the Social Science Research Institute, which is known for discretion and confidentiality.Furthermore, the experts were not asked any background questions except for gender, which makes triangulation impossible.Concerning the other group -the practitioners -they did get background questions, but the sample contains a much greater number of persons so that finding out about individual respondents would have been impossible given the small number of background questions asked (gender, educational level, size category of municipality, elected representative or not).If these respondents were concerned about traceability they could skip individual questions.9 We cannot exclude the possibility that the 'outsiders' may regard themselves as sufferers of corruption and hence likely to take it more seriously than others.Our data, however, shows hardly any

STJÓRNMÁL & STJÓRNSÝSLA
effects of income and very little of education on public perceptions of corruption.Moreover, as our data indicates, most people in Iceland have never -in their lives if we take the answers literally -come into direct contact with the different forms of corruption we ask about.Seeing them as especially victimized by corruption therefore seems rather difficult to establish.

Figure 1 .
Figure 1.Perceptions of corruption among supporters of different parties (0 low, 5 high)

Table 1 .
Public evaluations: "If you think about the municipalities in Iceland" …

Table 2 .
Public evaluations: "How common do you think corruption is in Iceland among…"

Table 3 .
Experience of corruption among voters (% of respondents reporting experience)

Table 4 .
Perceptions of corruption among public, experts and practitioners (% claiming corruption to be rather or very common).

Table 6 .
How widespread is corruption in Iceland?Expert and practitioner evaluations

Table 7 .
Evaluations of corruption among local politicians and administrators (%)