Chapter

3 Bureaucratic Corruption and the Rate of Temptation: Do Wages in the Civil Service Affect Corruption, and by How Much?

Author(s):
Sanjeev Gupta, and George Abed
Published Date:
September 2002
Share
  • ShareShare
Show Summary Details
Author(s)
Caroline Van Rijckeghem and Beatrice Weder 

The [Singapore] government believed that an efficient bureaucratic system is one in which the officers are well-paid so the temptation to resort to bribes would be reduced.

(Rahman, 1986, p. 151)

1. Introduction

The importance of adequate remuneration in ensuring an honest civil service is widely recognized in the policy debate.1 The issue of optimal government pay or its cost-effectiveness has not yet been settled, however, and a number of recent theoretical papers suggest that ensuring an honest civil service may be prohibitively expensive.2 This paper provides empirical estimates on the magnitude of the effect of civil-service wages on corruption, as a first step toward a cost-benefit analysis of civil-service wage increases.3

The models that predict eradication of corruption using wage policy to be expensive are in the spirit of the “shirking model” of Shapiro and Stiglitz (1984) and build on the early work of Becker and Stigler (1974). They assume that civil servants maximize expected income. Corrupt behavior, when detected, is penalized by job loss; hence, high pay constitutes an incentive to be less corrupt. However, when bribe levels are high or the probability of detection and fines low, these models predict that the wage necessary to eliminate corruption is high. Hence, it may be cost-effective for governments to pay “capitulation wages,” i.e. wages below reservation wages, that attract only the dishonest, rather than raise wages to the high levels required to deter corruption (Besley and McLaren, 1993).

There are, however, also reasons to expect that corruption can be eliminated at low wage levels. First, bribe levels in much of the civil service, such as in education, may be low, so that corruption may be eliminated at low wage levels even in shirking models. Second, delayed rewards, such as pensions, or the ability to obtain lucrative private-sector jobs after serving in the civil service can be used as an incentive to honest behavior, reducing the need to pay high wages during the working life.4 Third, the probability of detection and punishment of corruption may increase with the wage of civil servants because higher wages (fair treatment of civil servants) lead society to condemn corruption, again imparting greater effectiveness to wage policy.5

Finally, civil servants may not be motivated by greed, but may instead willingly forego opportunities for corruption, provided wages meet subsistence levels or are “fair.” This view is akin to the “fair wage-effort hypothesis” (Akerlof and Yellen, 1990) and is prevalent among experts on corruption who note a preference for honesty among civil servants.6,7Appendix B calibrates a theoretical model of corruption with shirking and the fair-wage hypotheses as special cases. It illustrates that the distinction between the “shirking view” and “fair-wages view” has important quantitative consequences for the strength of the relationship between wages and corruption: under fair-wage models, civil-service wages are an important determinant of corruption, whereas under shirking, civil-service wages may have to be raised to very high levels to eradicate corruption. It also shows that for sufficiently low wages (in conjunction with low bribe levels and or a high probability of detection), fair wages and shirking hypotheses are observationally equivalent.

From the above, it is clear that the effect of civil-service wage policy on corruption is an empirical question. An important obstacle for the empirical examination of the issue is, however, that a panel of data on civil-service wages is not available. We, therefore, assembled a new data set on civil service and manufacturing wages in which we took care to include only data of high quality and good comparability. Based on this data, we find evidence of a statistically and economically significant relationship between relative wages and corruption, in regressions based on country averages. While economically significant, the relationship, nevertheless, implies that a large increase in wages is required to eradicate corruption solely by raising wages, i.e. without accompanying policies to increase transparency and accountability in the civil service or in society as a whole. The relationship between wages and corruption is not found in regressions with country-fixed effects which points to ineffectual wage policy in the short run (we show that the power to reject the null hypothesis of no effect is low, however). Finally, case studies of pay reforms in tax administration suggest the existence of causality running from government wages to corruption.

Two related studies are those of Goel and Rich (1989) for the US and Rauch and Evans (2000) for a sample consisting mostly of developing countries. The research by Goel and Rich, which covers the period 1970-1983, suggests that the proportion of government officials convicted of bribery is negatively related to the difference between government wages and the average income of a private sector group of white-collar professionals (middle grade accountants). Rauch and Evans look into the factors affecting bureaucratic performance. We discuss their results in more detail in the text and use their wage data as an alternate data set when testing the robustness of our results.

The paper proceeds as follows. Section 2 discusses the data and the empirical strategy. Section 3 provides and interprets the panel-data results. It includes a discussion of a number of case studies as well. Section 4 draws out policy implications and concludes. Appendix A provides details on the econometric specification, whereas Appendix B presents a calibration exercise for shirking and fair-wage models, for use as a benchmark in interpreting the empirical results.

2. Empirical Strategy and Description of Data

The empirical estimates aim at testing whether there is a systematic relationship between wages in the civil service and corruption, both across countries and over time. We follow an estimation strategy consistent with the theoretical discussion in the introduction and in Appendix B. Our basic specification is of the following form:

where CONTROLS is a vector of variables that may impact on corruption. Theory suggests that such controls should include at least (i) proxies of the probability of detection (through internal or external controls on the bureaucracy); (ii) the penalty rate applied when detection occurs; (iii) the amount of distortions and opportunities for corruption in the economy; and (iv) other factors such as cultural determinants. Our choice of independent variables is detailed further in Van Rijckeghem and Weder (1997).

2.1 Description of Data

Our wage variable of choice is the ratio of government wages to manufacturing wages. The manufacturing wage is used as the comparator, as a measure of the opportunity cost which determines whether a worker is tempted to shirk (shirking hypothesis) or feels ill-treated (fair-wage hypothesis). Compared to GDP per capita, which is heavily influenced by the share of agriculture in GDP, the manufacturing sector has the advantage of being relatively comparable across countries in terms of skill-content.8 The skill-content in the manufacturing sector is probably lower than that in government, so that it should not be considered as a measure of alternatives available to government employees. The aim is only to have a consistent benchmark.9

To gain comparable data on wages in the civil service and in manufacturing, we assembled a new data set. The wage data cover 31 developing countries and low-income OECD countries10 over the period 1982–1994, gathered from IMF sources, statistical yearbooks, central bank bulletins, and the ILO yearbooks. Care was taken to ensure that only data of relatively high quality and good comparability was included.11

There is only one other source of data on relative wages, that of Rauch and Evans (2000), and we use this data in our regressions as well.12 They conducted a survey with experts in a number of less developed countries that included a question on the level of civil-service wages relative to comparable private sector employment. The correlation coefficient for the period-average relative wage data in our sample and the cross-section data of Rauch and Evans is 0.6.

Data on corruption is available in the form of an index based on surveys by ICRG, a private international investment-risk service. This has become the standard data set used in the empirical growth literature.13 The variable attempts to capture the extent to which “high government officials are likely to demand special payments” and “illegal payments are generally expected throughout lower levels of government” in the form of “bribes connected with import and export licenses, exchange controls, tax assessment, police protection, or loans.”14 While one could question the comprehensiveness of the index as Political Risk Services (various issues) focuses on assessing the business climate for international businesses only, the index is reasonably consistent across countries and time, as it is produced by a single organization, which presumably instructs its country specialists uniformly as to how to rate countries.15

As proxies of the probability of detection, we use an index of “quality of the bureaucracy” and an index of the “rule of law” available from the same source.16 We also proxy for societal pressures using an index of “political rights and civil liberties,” a simple average of the index of political rights and the index of civil liberties compiled by Freedom House (various issues) and published in Freedom in the World. This index captures factors such as the right to vote, the right to organize political parties, fair elections, meaningful representation by elected representatives, freedom of the press, freedom of assembly and demonstration, an independent judiciary, and the absence of political terror and torture. We also include PPP adjusted per capita GDP and secondary school enrollment (from the Summers-Heston v. 5.6 data set) as proxies for external controls, in general, under the assumption that the higher the social pressures and institutional safeguards against corruption, the higher the general level of income and education.

The penalty rate is probably the instrument governments use most frequently when attempting to combat corruption. In an optimizing model, the penalty rate can directly substitute for wage increases and sufficiently high penalties lead to the eradication of corruption even in the presence of low-detection probabilities. However, cross-country data on statutory penalty rates are not available; therefore, this variable is not included in the empirical analysis.

In addition to these variables, the literature has suggested a number of other determinants of corruption. For instance, in a recent set of papers, Ades and Di Tella (1997, 1999) have shown that the degree of competition and industrial policy have a significant effect on corruption.17 In their view, this occurs through the rents, which absence of competition and active industrial policies create, by way of more profitable or favored domestic firms, which bureaucrats and politicians then extract. Kaufmann (1997) has found a very strong correlation between bribery to public officials and “regulatory discretion” for a sample consisting mostly of Latin American and Asian countries, using survey responses by businesses. We use the black market premium for foreign exchange as an indicator of the level of distortions in the economy, following common practice in the growth literature.18 Presumably, economies experiencing exchange controls are subject to a number of other controls generating shortages and providing opportunities for corruption.

In addition to the variables discussed above, corruption is often attributed to purely cultural factors or lack of leadership. For instance, Tanzi (1994) argues that the absence of a culture of arm’s-length relationships may lead to corruption becoming ingrained and systemic. Lee (1986, p. 97) suggests that a culture of bureaucratic elitism may lead to a dissociation of civil servants with the rest of society and breed corruption. Alternatively, the level of education of civil servants can be a factor which reduces corruption. Such cultural factors and leadership are inherently difficult to measure; they are not included in the empirical analysis for lack of plausible proxies. Finally, Shleifer and Vishny (1993) suggest that more ethnically diverse countries are prone to a disorganized form of corruption and Mauro (1995) finds evidence of a link between an index of “ethnolinguistic fractionalization” and corruption.19 We include this variable in our regressions.

2.2 Econometric Issues

Having discussed our choice of variables, we now turn to estimation issues. One important issue here is the choice of estimator. As is well-known, unbiasedness of the OLS, “between” (i.e. regressions based on country means), and “random effects” estimators requires that the independent variables not be correlated with the country-specific effects (often reflecting omitted variables) subsumed in the error term. Even “fixed effects” (i.e. including country dummies) estimation is biased for short panels, in the absence of strict exogeneity (e.g. when the estimating equation is dynamic). The econometrician’s, solution to these problems is often to first difference the data and estimate a relationship for the first differenced data using instrumental variables.

We are reluctant to apply this procedure here, however, for two reasons. First, our measure of relative wages exhibits very little variation over time. The share of the variation within countries in the total variation in relative wages is 18% for our 31-country sample. Hence, the information content of the data corresponds in large part to the crosscountry variation in the data.20 Estimation based on first differencing (or fixed effects) would, therefore, suffer from low power compared to estimation based on country means or OLS.

Second, the timing with which wage policy and other variables affect corruption could be subject to long lags because of institutional inertia and societal attitudes or uncertainty as to the permanence of the changes. Wage increases might not produce lower corruption contemporaneously, whereas a sustained policy of high civil-service pay could over time produce lower corruption. The lack of variation in the corruption index we use (the coefficient of variation over time averages 0.2 for the countries in our 31-country sample) indeed suggests that corruption may be generated by a moving average process with long lags. If an inappropriate lag structure is specified for the estimating equation, first differencing of the data (or using fixed effects) produces inconsistent estimates of the long-run relationships between the variables. Estimation based on a cross-section or OLS also provides inconsistent results. Estimation based on cross-country means (where the raw data cover a reasonably long time span), on the other hand, provides a consistent estimate of the long-run relationships in the data. That is, the estimate is robust to dynamic mis-specification (Pesaran and Smith, 1995, p. 88).

In view of the issues of power and uncertainty surrounding the correct lag specification, the strategy adopted in this paper is to estimate a number of estimators, including “between” and fixed effects. As we will see, the estimates vary depending on the technique used. This could be a sign of mis-specification, resulting either from a mis-specified lag structure in the fixed effects estimation or from omission of correlated country-fixed effects (variables correlated with wages) in the between estimation. Under the first interpretation, the fixed effects results pertain to the contemporaneous effect of wages on corruption and the “between” results to the long-run relationship between wages and corruption (see Appendix A). Under the second, less benign, interpretation, the significant finding on wages in “between” regressions could be spurious, and the fixed effects estimator would be preferable. It is, however, hard to think of omitted correlated country-fixed effects, as we include a large number of control variables, covering internal and external controls, as well as distortions.21

Estimates are derived after weighing the data by a function of the number of observations to correct for heteroscedasticity resulting from using an unbalanced panel (see Appendix A) and “White-correction” of the standard errors to ensure robustness of the standard errors to heteroscedasticity.22 Only countries with at least five consecutive observations are included in the regression. In our sample, 88% of the countries have seven or more consecutive observations, while 47% of the countries have 10 or more consecutive observations.

A final econometric issue pertains to the direction of causality. Two arguments for reverse causality could be made: (1) corrupt countries have poor tax collection and, therefore, are constrained to pay less well; and (2) corrupt countries pay less purposefully arguing that civil servants already have incomes from corruption. This issue is difficult to investigate econometrically due to the absence of instruments for civil-service wages that are both highly correlated with wages but uncorrected with corruption, and can probably be addressed only through case studies.23

3. Results

3.1 Regression Analysis

We find a close negative association between relative civil-service wages and corruption across the developing and lower-income OECD countries in our data set. Figure 1 shows the scatterplot of relative wages and corruption. It shows a negative association between the variables, when no other factors are controlled for.

Figure 1.Correlation Between Relative Civil Service Wages and Corruption

Notes: Only countries for which at least five observations are available are included. Corruption index is ICRG index minus 6. Countries include: Argentina, Bolivia, Botswana, China, Colombia, Costa Rica, Egypt, El Salvador, Ghana, Greece, Guatemala, Hong Kong SAR, India, Jordan, Kenya, Korea, Mexico, Morocco, Panama, Peru, Portugal, South Africa, Singapore, Spain, Sri Lanka, Suriname, Thailand, Trinidad and Tobago, Turkey, Uruguay, Zimbabwe.

Table 1 presents the results of regressions based on country averages, with different sets of control variables. The relative wage, in a regression with real GDP per capita in PPP terms, is a highly significant explanatory variable (column 1). Real GDP per capita is a comprehensive measure of development, which is positively correlated with law and order, quality of the bureaucracy, civil rights, etc. However, it is not statistically significant in this regression.24

Table 1.Determinants of Corruption. Regression Based on Country Averagesa,b
(1) Parsimonius specification(2) Basic specification(3) Rauch- Evans data(4) Full specification
Constant6.395.146.108.86
(3.28)(18.05)(17.42)(3.34)
Relative wages−1.04−0.50−0.33−0.64
(−5.37)(−3.07)(−1.90)c(−2.43)
GDP per capita−0.29−0.51
(−1.11)(−1.73)
Secondary schooling0.01
(0.77)
Law and order−0.38−0.31−0.57
(−2.04)(−1.99)(−3.45)
Quality of bureaucracy−0.24−0.54−0.07
Quality of bureaucracy(−1.93)(−4.09)(−0.59)
Political rights and civil liberties−0.09

(−0.63)
Ethnographic fractionalization−0.001
(−0.19)
Black market premium−0.001
(−0.62)
Korea dummy1.74
(6.62)
Singapore dummy0.58
(0.85)
No. of countriesd31313528
Adjusted R20.410.660.750.74

t-Stalisties below estimate.

For countries with five or more consecutive observations. Based on weighted least squares, where weights are a function of the number of observations (see text). Standard errors are White-corrected for heteroscedasticity.

When estimated with the method of least absolute deviations to reduce the role of outliers, the t-statistic is 3.6.

Based on developing and low-income OECD countries. Number of countries varies because of availability of the data.

t-Stalisties below estimate.

For countries with five or more consecutive observations. Based on weighted least squares, where weights are a function of the number of observations (see text). Standard errors are White-corrected for heteroscedasticity.

When estimated with the method of least absolute deviations to reduce the role of outliers, the t-statistic is 3.6.

Based on developing and low-income OECD countries. Number of countries varies because of availability of the data.

Next, we replace real GDP per capita with “law and order” and “quality of the bureaucracy.” These are more precise measures of internal and external controls, which affect the probability of detection. Relative wages continue to be significant.25 As for economic significance, the coefficient on relative wages equals 0.5, indicating that a 1-point change in relative wages (say from one to two times the manufacturing wage) leads to a 0.5-point reduction in the corruption index.

We then add a number of variables one at a time to the previous “basic” specification, from a list including real GDP per capita, secondary school enrollment, political rights and civil liberties, an index of ethnographic fractionalization, and the black market premium. The relative wage continues to be statistically or marginally significant when these variables are added.26

As an additional test of robustness, we re-estimate the regression in column 2 using the rating for the level of relative wages of higher officials in economic agencies contained in the Rauch and Evans (2000) data set.27 We find a significant or marginally significant relationship for relative wages, the index of law and order and the quality of the bureaucracy for the 35 countries in the Rauch-Evans data set (column 3).28 The role for relative wages appears somewhat more important than we found earlier. Column 3 indicates that an increase in rating from a “1” to a “4”—or an increase in wages of higher officials in the economic agencies relative to those of private-sector managers from less than 50% to comparable levels—reduces the corruption index by 1 point (3 X 0.33). Thus, the Rauch-Evans data set supports the existence of a significant relationship between civil-service wages and corruption.

Column 4 includes all variables in one regression. In addition to GDP per capita, enrollment rates, rule of law, and quality of the bureaucracy, it also includes the black market premium, “political rights and civil liberties,” “ethnolinguistic fractionalization,” as well as a dummy variable for Singapore to test whether this outlier influences the results significantly (Singapore has both very low corruption and very high civil-service pay). The black market premium is not significant in this regression (and it has the “wrong” sign), while the Singapore dummy is insignificant.29 Korea is an outlier in this regression, with higher corruption than expected given its characteristics.30

Table 2 gives an impression of the economic significance of our results. It shows the effect of a one standard deviation change on the corruption index, based on the coefficients in the full specification (column 4 in Table 1). We find that the index of rule of law and the relative wage have the largest impact on corruption.

Table 2.Economic Significance of Results
Standard deviationaCoefficient in full specification in between regressionEffects of one standard deviation change on corruption index
Corruption index1.25
Index of law and order1.33−0.57−0.76
Ratio of civil-service wage to manufacturing wage0.670.64−0.43
Log of real GDP per capita in constant US$0.72−0.51−0.37
Secondary school enrollment21.12−0.01−0.21
Index of political rights and civil liberties1.510.090.14
Index of quality of bureaucracy1.41−0.07−0.10
Black market premium (percent)b287.000.00020.06
Index of ethnolinguistic fractionalization28.80−0.001−0.03

Pertains to sample for which all variables are available (227 observations).

Pertains to within regression coefficient.

Pertains to sample for which all variables are available (227 observations).

Pertains to within regression coefficient.

Table 3 turns to the time-series dimension of the data, and reveals no evidence of a within-country effect of wages. It also shows that the high significance of the “rule of law” variable pertains to the crosscountry variation in the data. The “quality of the bureaucracy” and the black market premium are significant in the within regressions, though the latter is not economically significant (Table 2). This suggests that higher pay (and/or improved rule of law) does not lead to lower corruption in the short run. However, the power to reject the null hypothesis of no effect is low given the low variance of relative wages (and rule of law) within countries.

Table 3.Determinants of Corruption Index: Within-Country EstimationaFor countries with five or more consecutive observations. Standard errors are White-corrected for heteroscedasticity.Fixed effects are not reported.
Constant
Ratio of civil-service wage to manufacturing wage−0.01
(−0.48)
Log of real GDP per capita in constant US$−0.31
(−0.97)
Secondary enrollment rate−0.02
(−2.11)
Index of law and order0.11
(1.35)
Index of quality of bureaucracy−0.39
(−5.85)
Index of political rights and civil liberties0.05
(1.46)
Black market premium (percent)0.0002
(3.69)
Number of observations236
Number of countries29
Adjusted R20.89

t-Statistics below estimate.

t-Statistics below estimate.

Table 4 provides information on the simple correlation between our measure of relative wages and the other independent variables included in the regressions. To the extent that there is an important correlation, it is possible that relative wages also operate through additional channels to reduce corruption. The correlations with the quality of the bureaucracy and the rule of law are of particular interest. These are quite high (48% and 41%, respectively, for the full sample), indicating that relative wages could influence corruption through these channels. This would be consistent with the view that wages have an effect on the tolerance for corruption on the part of the civil service itself, the judiciary, and society at large.

Table 4.Data Descriptions and Simple Correlations (Number of Observations in Parenthesis)Data correspond to developing and low-income OECD countries for which five consecutive observations are available.
IntersectionFull Sample
Simple correlation withSimple correlation with
VariableMeanStandard deviationCorruptionRelative wagesMeanStandard deviationCorruptionRelative wagesRauch- Evans
Corruption index2.761.2513.131.211
(227)(227)(227)(897)(897)(897)
Ratio of civil-service wage to manufacturing wage1.150.67−0.4711.220.67−0.461
(227)(227)(227)(227)(473)(473)(292)(473)
Log of real GDP per capita in constant US$8.020.72−0.550.437.640.85−0.360.470.40
(227)(227)(227)(227)(1684)(1684)(686)(440)(716)
Secondary school enrollment55.4421.12−0.340.2941.5424.24−0.210.310.20
(227)(227)(227)(227)(1405)(1405)(706)(432)(596)
Index of law and order2.811.33−0.660.312.741.19−0.590.410.34
(227)(227)(227)(227)(907)(907)(897)(301)(450)
Index of quality of bureaucracy2.641.41−0.660.402.691.32−0.620.480.35
(227)(227)(227)(227)(907)(907)(897)(301)(450)
Index of political rights and civil liberties3.461.51−0.280.094.421.77−0.240.080.06
(227)(227)(227)(227)(1948)(1948)(884)(473)(777)
Index of ethnolinguistic fractionalization36.8528.800.09−0.2244.5129.620.23−0.22−0.23
(227)(227)(227)(227)(1774)(1774)(854)(437)(801)
Black market premium (percent)48.40287.000.13−0.07110.931352.45−0.07−0.110.14
(227)(227)(227)(227)(144)(144)(615)(433)(657)
Memorandum item
Rauch-Evans, Index of ratio of top civil-service wage to private equivalent1.740.66−0.410.60
(801)(801)(440)(317)

Finally, we test the sensitivity of the results to the presence of omitted variables by conducting Leamer’s Extreme Bounds Analysis, i.e. we add a number of “free” regressors (all possible combinations of up to three out of five additional regressors, or 25 combinations) to our basic specification and evaluate the robustness of the results to the inclusion of these variables. The “fixed” variables consist of relative civil-service wages, “law and order,” and a dummy variable for Korea (see above). The “free” variables consist of GDP per capita, “quality of the bureaucracy,” “political rights and civil liberties,” the black market premium, and a dummy variable for Singapore. We find that zero lies outside of the “extreme bounds,” defined as the lowest value of “the estimate minus two standard deviations” (−1.6) and the highest value of “the estimate plus two standard deviations” (0.1) and conclude that the results are robust to EBA.31

To sum up, a robust relationship between corruption and relative civil-service pay appears to exist across countries. The cross-country results are invariant to specification, to the exclusion of the city-states Hong Kong SAR and Singapore, and to the use of an alternate data set. The results are also economically significant. They imply that an increase in civil-service pay from 100% to 200% of the manufacturing wage is associated with an improvement in the corruption index of at least 0.5 points of the index, and more if the indirect effects of wages through the quality of the bureaucracy and rule of law are included.

3.2 Implications for the Shirking vs. Fair-Wage Hypotheses

What light do these results shed on the validity of the shirking vs. fair-wage hypotheses? First, adherence to the shirking view would predict a relatively weak relationship between wages and corruption, as long as the probability of detection of corruption is low and bribe levels are high relative to wages, a situation that appears characteristic of developing countries. As a corollary, wages at which corruption can be eliminated will generally be high under the shirking hypothesis.

The estimated regression equations can be used to calculate the relative wage at which corruption is brought down to the level of the higher-income OECD countries (a corruption index equal to zero). Substitution of the value of zero for the corruption index in our regression result (we use the “full specification” in Table 1) indicates that this value ranges from 2 to 8, depending on the values of the control variables. The upper range is difficult to reconcile with the conceivable level of fair wages, casting doubt on the fair wage-corruption hypothesis. It is, however, important to recall that few countries had high civil-service wages so that the hypothesis that corruption is (close to) zero for that wage range cannot be tested directly; the results rely instead on extrapolation of a linear relationship estimated for low relative wages. And as shown in Appendix B, behavior under the fair-wage hypothesis can “collapse” to that under the shirking hypothesis when wages are low if bribe levels are also low and/or the probability of detection and penalties are high.32

At the same time, the values of 2–8 are low compared to what one might expect if the shirking hypothesis is in fact is true. In India, some say—though this is not more than a guess and does not apply uniformly to all parts of government—that the probability of detection is 10%, the probability of charges being brought (conditional on detection) is 10%, and the probability of punishment (conditional on charges being brought) is also 10%. This implies an unconditional probability of punishment of 0.1%.33 India ranks about average in terms of our empirical proxies for internal and external controls.34 At a probability of detection of 0.1%, the stylized shirking model described in Appendix B predicts that wages would have to be equal to as much as 19 times the private sector equivalent under the best of circumstances (i.e. when there is a credible commitment to pay sufficiently high pensions), or much above what our empirical results predict.35 In sum, the results of this section indicate that neither the shirking nor the fair-wage hypothesis appear to hold in pure form.

3.3 Case Studies on Pay Reform in Tax Administrations

We now address the possibility that the cross-country correlation need not reflect a causal link from government wages to corruption using case studies. As noted above, the fact that corrupt countries tend to have poor budgetary performance and face strong budgetary pressures, or may subscribe to the view that civil servants already earn sufficient income from corruption, may lead them to pay less well.

Evidence from six case studies (Ghana, 1983; Peru, 1991; Uganda, 1992; Zambia, 1994; Kenya, 1995; and Tanzania, 1996) conducted by the authors point to a strong improvement in revenue collection following pay reforms in tax administration.36 These countries introduced pay reforms in the context of the introduction of independent new revenue authorities and experienced increases in the tax revenue to GDP ratio of up to 3–4% of GDP, amounts that exceed those experienced by other low-income countries engaged in stabilization efforts.

Of course, not all of the increases in the tax ratios can be attributed to pay reforms. Tax rate and base improvements as well as improvements in tax administration are likely to account for part of the improvements in revenues. In order to isolate the impact of pay reforms which went hand in hand with other administrative improvements as well as changes in tax rates and exemptions, we also compared revenue developments with those in low-income countries engaged in an IMF program. The low-income countries with IMF programs are a natural standard for comparison. A study of 21 low-income countries with ESAF programs with the IMF in the late 1980s indicates that these countries experienced improvements of 0.6% of GDP on average in their tax revenues over the course of their (usually 3-year) programs. Of these, 11 experienced improvements of over 1% of GDP.37 As this improvement in revenues falls much short of the increase in revenues experienced in countries that introduced pay reforms simultaneously with the introduction of independent revenue authorities, this suggests that pay reforms played an important role in generating increased revenues.

4. Policy Implications and Conclusion

Theory is ambiguous in its implications for optimal wage policy. Various mechanisms through which concepts of fairness can affect corruption suggest that the relationship between civil-service wages and corruption may be stronger, and wages at which no corruption occurs lower, than predicted by models postulating self-interested behavior (i.e. shirking models), especially when bribe levels are high and the probability of detection of corrupt acts is low. This result has obvious implications for optimal wage policy, whether the government’s goal is to maximize social welfare or its cost-effectiveness. If the “fair wage-corruption” hypothesis holds, paying wages that ensure low corruption may not necessarily be very costly (though this would depend on civil servants’ standards of fairness), and paying wages that ensure an honest civil service may be cost-effective.

We have presented suggestive evidence of a negative relationship between civil-service pay and corruption. The cross-country or “between” regressions for a sample of 31 developing countries indicate that increasing relative pay from 1 to 2 is associated with an improvement in the corruption index on the order of 0.5 points excluding any indirect effects of wages on the quality of the bureaucracy and the rule of law, and more including indirect effects. Quasi-eradication of corruption is associated with a relative wage of two to eight times the manufacturing wage, not taking into account indirect effects. This is more than one would expect if the fair-wage hypothesis held in pure form, but less than predicted by the shirking hypothesis, suggesting that neither hypothesis holds in pure form.

Some caution is needed in drawing policy implications or carrying out cost-benefit analysis based on our cross-country regression analysis. First, the existence of a relationship cannot be confirmed based on regressions with country-fixed effects (the “within” estimates), which means that the cross-country results could reflect a spurious correlation notwithstanding our attempts to control for other factors. This also means that higher pay does not lead to lower corruption in the short run.38 Second, the cross-country correlation need not reflect a causal link from government wages to corruption. The fact that corrupt countries tend to have poor budgetary performance and face strong budgetary pressures, or may subscribe to the view that civil servants already earn sufficient income from corruption, may lead them to pay less well. In view of the findings from case studies that point to increases in revenues following pay reforms in tax administration, we believe that, these caveats notwithstanding, our results support the presumption that an active wage policy can help in fighting corruption. This is not to say that other instruments are not important. Indeed the results of the paper suggest that strengthening the rule of law, in particular, will also have beneficial effects on corruption.

Appendix A. Econometric Specification

Supposing the dynamic model underlying the relationship between x and y is as follows (where i denotes the country, t denotes time, Ti0+1, denotes the first period for country i, and Ti1 the last period):

Aggregating over country observations, we obtain:

Removing and adding xi0 and adding and removing xi1, we find:

From this it follows that: (1) a regression on country averages will yield the long-term coefficient (β0 + β1), and (2) that the error term exhibits heteroscedasticity, since (ignoring end effects for simplicity) the variance of the error term is:

Appendix B. Calibration Exercise for Shirking and Fair-Wage Models

This section calibrates a theoretical model of corruption with shirking and the fair-wage hypotheses as special cases. The section has two purposes. First, it illustrates that corruption is much more responsive to wages under fair wages and shows that civil-service wages would have to be raised to very high levels to eradicate corruption if the shirking hypothesis in fact holds. Second, it makes the case that the empirical estimate for wages at which corruption is eliminated is an upper bound because fair-wages and shirking hypotheses are observationally equivalent for certain parameters. The shirking model—which is a variant of the well-known Becker-Stigler model—is presented mainly to facilitate comparison with the fair wages model.

A.1. The Shirking Hypothesis: Corruption in a Maximizing Framework

The analysis draws heavily on Becker and Stigler (1974), with the exception that corruption is modeled as a continuous variable rather than a binary variable. Government employees are assumed to maximize the present discounted value of a stream of expected income. In so doing, they balance the benefits from corrupt behavior against the penalties when caught and punished. These penalties are assumed to include dismissal (with a cost equal to the wage differential with the private sector plus bribes foregone) and other penalties. In a multi-period model, the present discounted value (PDV), of expected income, in the last period of employment, period T, can be expressed as follows:

Here PDV is the present discounted value of expected income. The equation expresses expected income as a weighted average of the income when corruption is not detected and when it is detected. When corruption is not detected, income equals income from bribery, CB, where C is the number of corrupt acts and B is the level of the bribe, plus the government wage, Wg. Wg can be thought of as including the government pension. When corruption is detected, the wage, pension, and bribery income is foregone and income equals the private sector wage, Wp, minus penalties or jail terms, f. This formulation assumes for simplicity that the probability of detection equals pC, or the probability of detection for an individual corrupt act, p, times the number of corrupt acts, C.39 Except for C, all variables are assumed to be exogenous. Furthermore, all variables, except for C and Wg are assumed constant over time.

Taking the first derivative of PDV with respect to C and solving gives the amount of corruption that a civil servant who maximizes expected income would chose to engage in, in the last period of employment:

In this formulation, government wage policy has an effect on corruption. However, high wages are not necessary for low corruption, if the government can manipulate p and f at will.40

The government can ensure that corruption equals zero by providing the wage for which C = 0 in the above equation. We label this condition the No Corruption Condition (NCC), in parallel with the No Shirking Condition in the efficiency wage literature:

Note from this NCC that very high wages are required in order to eradicate corruption when the probability of detection is low and/or the level of bribes is high. Solutions for earlier periods are obtained by backward induction (see Becker and Stigler, 1974). PDV in the second to last period of employment can be expressed as follows:

where r represents the discount rate. Assuming the NCC holds in the last period, PDVT–1 is obtained by setting PDVT equal to Wg from Eq. (1b). Maximizing PDVT−1 subject to C and then equating C to zero provides the NCC for the second to last period, conditional on the NCC holding in the last period:

Note that when the NCC is not expected to hold in later periods, higher wages than indicated by Eq. (2) have to be paid in earlier periods. In practice in developing countries, the NCC is not expected to hold in later periods; that is, future wages are likely to be such that corruption “pays” (i.e. future wages are below the PDV that civil servants can obtain when corrupt). Therefore, the results given by Eq. (2) provide a lower bound to the actual wage required to eradicate corruption.

A calibration can illustrate the magnitudes involved. Assuming that the probability of detection and punishment of an individual act of corruption is 0.1%,41 that penalties consist of dismissal, foregone bribes, and additional penalties of 150, that the private sector wage equals 100, that the bribe level equals 20, and that the discount rate is 10%, one finds that corruption is eradicated when government wages equal 200 times the private sector wage in the last period of employment and 19 times in all earlier periods. If wages fall short of 200 times the private sector wage in the last period of employment, a factor greater than 19 is required in order to eradicate corruption in earlier periods.

A.2. The Fair-Wage Hypothesis: Corruption in a Satisficing Framework

Individual behavior may not be appropriately described by the optimizing framework laid out above. According to the psychological literature on “fair wages,” “workers who do not receive a fair wage […] may change actual effort. […] or their perceived level of remuneration (by redefining the nonpecuniary terms of the job)” (Akerlof and Yellen, 1990, p. 257).42

What we will call the “fair wage-corruption” hypothesis is the hypothesis that workers choose levels of C in an attempt to reach an expected income level EI = W*.

The solution for C, as we shall see, is a function of government wages, Wg, relative to the fair wage, W*.43 The “fair” wage could be determined according to a variety of mechanisms: wages of peers within or without the place of employment, societal expectations, the market clearing wage, subsistence requirements, the status of civil servants, etc.44 Solving for C and choosing the negative root, one obtains:45

where:

Setting C = 0, we find that the NCC is simply:

This means that a higher W* implies a higher government wage is necessary to eliminate corruption, as one would expect. Under certain circumstances—high W*, low Wg, low B, high p or f—D is negative and there is no solution for C (i.e. the solution is imaginary): targeted income cannot be reached whatever the level of corruption. In this situation, one would expect the civil servant to engage in the number of corrupt acts that maximizes expected value, while also reducing effort, say according to effort = f(EI/W*).46 That is, one would expect civil servants to act as maximizers even when they are truly satisficers. Thus, satisficing behavior can be observationally equivalent to maximizing behavior. This observational equivalence (which is more likely to occur at low civil-service wages) implies that it may be impossible to test the fair-wage and shirking hypotheses empirically. Linear extrapolation based on observations in the low-wage range may produce a higher intercept (i.e. wages at which corruption is zero) than the true intercept. Under these circumstances, micro data on corruption incidence in the parts of public administration that offer high bribes or have a low probability of detection would be necessary to distinguish the two efficiency wage theories empirically.

References

    AcemogluD.VerdierT.2000. The choice between market failures and corruption. American Economic Review90 (1) 194211.

    AdamolekumL.JahO. (Eds.) 1997. Enclave Approach to Tax and Customs Administration in Africa: Three Country Studies. MimeographWorld BankWashington, DC.

    • Search Google Scholar
    • Export Citation

    AdesA.Di TellaR.1997. National champions and corruption: some unpleasant interventionist arithmetic. Economic Journal107 (443) 10231043.

    • Search Google Scholar
    • Export Citation

    AdesA.Di TellaR.1999. Rents, competition and corruption. American Economic Review89 (4) 982994.

    AkerlofG.A.YellenJ.1990. The fair wage effort hypothesis and unemployment. Quarterly Journal of Economics105 (2) 255283.

    AkerlofG.A.YellenJ.1994. Gang behavior, law enforcement, and community. In: AaronH.J.MannT.E.TaylorT. (Eds.) Values in Public Policy. BrookingsWashington.

    • Search Google Scholar
    • Export Citation

    AlfilerM.C.P.1986. The process of bureaucratic corruption in Asia: emerging patterns. In: CariñoL.V. (Ed.) Bureaucratic Corruption in Asia: Causes Consequences and Controls. NMC PressQuezon City, Philippines pp. 1568.

    • Search Google Scholar
    • Export Citation

    AndvigJ.C.MoeneK.O.1990. How corruption may corrupt. Journal of Economic Behavior and Organization13 (1) 6376.

    BardhanP.1997. Corruption and development: A review of issues. Journal of Economic Literature35 (3) 13201346.

    BeckerG.S.StiglerG.J.1974. Law enforcement, malfeasance, and the compensation of enforcers. Journal of Legal Studies2119.

    BesleyT.McLarenJ.1993. Taxes and bribery: the role of wage incentives. Economic Journal103 (416) 119141.

    BrunettiA.KisunkoG.WederB.1998. Credibility of rules and economic growth: evidence from a worldwide survey of the private sector. World Bank Economic Review12 (3) 353384.

    • Search Google Scholar
    • Export Citation

    CariñoL.V. (Ed.) 1986. Bureaucratic Corruption in Asia:Causes Consequences and Controls. NMC PressQuezon City, Philippines.

    ChandS.K.MoeneK.O.1999. Controlling fiscal corruption. World Development27 (7) 11291140. [Reproduced as Chapter 4 in this volume—Ed.]

    • Search Google Scholar
    • Export Citation

    CharapJ.HarmC.2000. Institutionalized corruption and the kleptocratic state. In: MenardC. (Ed.) Institutions Contracts and Organizations: Perspectives from New Institutional Economics. Edward ElgarCheltenham, UK pp. 188209. [Reproduced as Chapter 6 in this volume—Ed.]

    • Search Google Scholar
    • Export Citation

    CoplinW.D.O’LearyM.K.SealyT.1993. A Business Guide to Political Risk for International Decisions. Political Risk ServicesSyracuse, NY.

    • Search Google Scholar
    • Export Citation

    DumontR.1979. Remuneration and corruption levels in French-speaking Africa. In: EkpoM. (Ed.) Bureaucratic Corruption in Sub-Saharan Africa: Toward a Search of Causes and Consequences. University Press of AmericaWashington, DC.

    • Search Google Scholar
    • Export Citation

    FehrE.TyranJ.1996. Institutions and reciprocal fairness. Nordic Journal of Political Economy.

    FehrE.KirchsteigerG.RiedlA.1993. Does fairness prevent market clearing? An experimental investigation. Quarterly Journal of Economics108 (2) 437459.

    • Search Google Scholar
    • Export Citation

    FlattersF.McLeodW.B.1995. Administrative corruption and taxation. International Tax and Public Finance2 (3) 397417.

    Freedom House (various issues).Freedom in the World: The Annual Survey of Political Rights and Civil Liberties. Freedom HouseNew York.

    • Search Google Scholar
    • Export Citation

    GoelR.RichD.1989. On the economic incentives for taking bribes. Public Choice61 (3) 269275.

    GouldD.J.1980. Bureaucratic Corruption and Underdevelopment in the Third World: The Case of Zaire. PergamonNew York.

    HaqueN.SahayR.1996. Do Government wage cuts close budget deficit?IMF Staff Papers43 (4).

    IsraelA.1987. Institutional Development: Incentives to Performance. Johns Hopkins Univ. PressBaltimore.

    KahnC.M.SilvaE.C.D.ZiliakJ.P.2000. Performance-Based Wages in Tax Collection: The Brazilian Tax Collection Reform and its Effects. MimeographTulane University.

    • Search Google Scholar
    • Export Citation

    [KaufmannD.1997. Corruption: the facts. Foreign Policy107 (Summer) 11431—Ed.]

    KlitgaardR.1989. Controlling Corruption. Univ. California PressBerkeley.

    KlitgaardR.1991. Adjusting to Reality Beyond State Versus Market in Economic Development. ICS PressSan Francisco.

    KnackS.KeeferP.1995. Institutions and economic performance: crosscountry tests using alternative institutional measures. Economics and Politics7 (3) 207227.

    • Search Google Scholar
    • Export Citation

    LeeR.P.L.1986. Bureaucratic corruption in Asia: the problem of incongruence between legal norms and folk norms. In Cariño (Ed.) Bureaucratic Corruption in Asia: Causes Consequences and Controls. NMC PressQuezon City, Philippines pp. 69107.

    • Search Google Scholar
    • Export Citation

    LindauerD.NunbergB. (Eds.) 1994. Rehabilitating Government: Pay and Employment Reform in Africa. World BankWashington.

    LuiF.T.1986. A dynamic model of corruption deterrence. Journal of Public Economics231 (2) 122.

    MauroP.1995. Corruption and growth. Quarterly Journal of Economics110 (3) 681712.

    MauroP.1998. Corruption and the composition of government expenditure. Journal of Public Economics69 (2) 263279. [Reproduced as Chapter 10 in this volume—Ed.]

    • Search Google Scholar
    • Export Citation

    MookherjeeD.PngI.P.L.1995Corruptible law enforcers: how should they be compensated?Economic Journal105 (428) 145159.

    MyrdalG.1968. Asian Drama: an Inquiry into the Poverty of Nations. Twentieth Century FundNew York.

    NashashibiK.GuptaS.LiuksilaC.LorieH.MahlerW.1992. The Fiscal Dimension of Adjustment in Low-Income Countries. IMF Occasional Paper No. 95. International Monetary FundWashington.

    • Search Google Scholar
    • Export Citation

    PesaranM.H.SmithR.1995. Estimating long-run relationships from dynamic heterogeneous panels. Journal of Econometrics68 (1) 79113.

    • Search Google Scholar
    • Export Citation

    Political Risk Services various issues.International Country Risk Guide. Political Risk ServicesNew York.

    RahmanA.T.R.1986. Legal and administrative measures against bureaucratic corruption in Asia. In: Cariño (Ed.) Bureaucratic Corruption in Asia: Causes Consequences and Controls. NMC PressQuezon City, Philippines pp. 109162.

    • Search Google Scholar
    • Export Citation

    RauchJ.E.EvansP.B.2000. Bureaucratic structure and bureaucratic performance in less developed countries. Journal of Public Economics75 (1) 4971.

    • Search Google Scholar
    • Export Citation

    Rose-AckermanS.1999Corruption and Government: Causes Consequences and Reform. Cambridge Univ. PressCambridge, UK.

    Schiavo-CampoS.de TommasoG.MukherjeeA.1997. An International Statistical Survey of Government Employment and Wages. World Bank Policy Research Working Paper No. 1806.

    • Search Google Scholar
    • Export Citation

    ShapiroC.StiglitzJ.E.1984. Equilibrium uemployment as a worker discipline device. American Economic Review74 (3) 433444.

    ShleiferA.VishnyR.1993. Corruption. Quarterly Journal of Economics108 (3) 599617.

    TanziV.1994. Corruption, Governmental Activities, and MarketsIMF Working Paper No. 94/99.

    TanziV.ZeeH.H.1995. Human Capital Accumulation and Public Sector GrowthIMF Working Paper No. 95/95.

    Van RijckeghemC.WederB.1997. Corruption and the Rate of Temptation: Do Low Wages in the Civil Service Cause Corruption?IMF Working Paper No. 97/73.

    • Search Google Scholar
    • Export Citation

    WeiS.1997. How taxing is corruption on international investors?Review of Economics and Statistics82 (1) 111.

    WeiS.2000. Natural Openness and Good GovernmentNBER Working Paper No. 7765.

Reprinted from Journal of Development Economics, Caroline Van Rijckeghem and Beatrice Weder, “Bureaucratic corruption and the rate of temptation: do wages in the civil service affect corruption, and by how much?” Vol. 65, pp. 307-331 (2001), with permission from Elsevier Science.This is a condensed version of an IMF Working Paper (WP/97/73). The views expressed herein are those of the authors and do not necessarily represent those of the International Monetary Fund. The authors would like to thank Alberto Alesina, Ernst Fehr, Aart Kraay, Nadeem Haque, Paolo Mauro, Susan Rose-Ackerman, Ratna Sahay, Vito Tanzi, three anonymous referees, and participants at seminars at the University of Basel, the LACEA meetings, and Bogazici University for useful suggestions. Manzoor Gill, Philip Polsky, and Stacy Maynes provided excellent research assistance.
1See Myrdal (1968), Cariñio (1986), Israel (1987), Klitgaard (1989), Tanzi (1994), and Lindauer and Nunberg (1994). There is also an increasing recognition—based on recent research—that net benefits of reducing corruption are positive in terms of investment and growth (see Mauro, 1995, 1998; Wei, 1997; Brunetti et al., 1998).
2Notably, Besley and McLaren (1993) and Flatters and McLeod (1995). Some exceptions are Haque and Sahay (1996), who argue that raising wages could be cost-effective by ensuring better human capital in tax administration, and Mookherjee and Png (1995), who conclude that bribes may be less efficient than bonus payments in encouraging effort among law enforcers. See Bardhan (1997) and Rose-Ackerman (1999) for overviews of the corruption literature.
3We thank Shanta Devarajan for suggesting cost-benefit analysis of civil-service-pay reform as a topic for research, as opposed to simply documenting the existence of a negative relationship between pay and corruption.
5Tanzi (1994), for example, notes that “unrealistically low wages always invite corruption and, at times, lead society to condone acts of corruption.”
6See, e.g. Peter N.S. Lee (quoted by Alfiler, 1986, p. 66), who notes that “it is not a question of paying sufficient salary to make a man incorruptible, but rather of not paying salary on which a man is encouraged to be corrupt in order to meet his reasonable commitments.” See also Klitgaard (1991).
7An additional channel would be that high wages help attract and retain a workforce which includes better, i.e. less corrupt workers.
8It is only natural, and not an indication of good pay, to have a high government wage relative to GDP per capita in less developed countries, given the predominance of agriculture. Hence, one would not want to use GDP per capita as the comparator.
9Data limitations preclude us from capturing certain components of civil-service pay, such as the relative stability of government employment, and deferred benefits, such as pensions. The unemployment rate and education level of civil servants affect the opportunity cost in the shirking model; however, data are not available to permit one to correct the opportunity cost accordingly. Also, benchmarks other than the alternative wage in the private sector may be relevant for corruption behavior in the fair-wage model, but are difficult to capture. These benchmarks could include past wages, wages in civil services abroad, an “adequate” standard of living, etc.
10We exclude high-income OECD countries, where relative civil-service wages are low; however, this probably only reflects the high supply of skilled labor, assuming government is the skill-intensive sector (for a theoretical exploration of the link between relative wages in government and development, see Tanzi and Zee, 1995). To decide which sample to use, we ran regressions of relative wages on GDP per capita and secondary education. For a sample which includes all OECD countries, these development indicators have significant negative coefficients. On the other hand, the coefficients are insignificant (and positive) for the more homogeneous sample of developing countries and low-income OECD countries. This suggests that a sample excluding the high-income OECD countries might be appropriate.
11The data must, nevertheless, be interpreted with care as the definitions for government wages are not always comparable across countries. The scope of government also varies across countries and time (for example, after privatization of services formally provided by general government), causing further problems for comparisons. For some countries, the government wage bill was divided by government employment to obtain the average wage, but only if there was a clear indication that the numerator and denominator had the same coverage (i.e. both included or excluded military personnel, casual employees, personnel in semi-autonomous bodies, such as universities and hospitals, whose wage expenditures are often financed through transfers from the central government rather than recorded directly as government wages in the budget). Definitions also differ with respect to inclusion of in-kind benefits (housing, cars, etc.) and allowances. The exact sources of the data are described in detail in IMF Working paper no. 97/73, which can be downloaded at http://www.imf.org.
12Schiavo-Campo et al. (1997) provide cross-country data for 1 year for a large sample of countries, but concentrate on government wages relative to GDP per capita, not relative to manufacturing wages.
13Published by Political Risk Services in the International Country Risk Guide (ICRG), for the period 1982–1995. This data set was assembled by the IRIS Center (University of Maryland) from hard copies of the International Country Risk Guide. The index ranges from 0 to 6 in the original data, with 0 indicating the highest corruption and 6 the lowest. Most of the industrial countries and a few developing countries, such as Singapore, have a value of 6 for the index. For clarity, we redefine the ICRG corruption index, so that an increase in the index indicates worsening corruption (with 0 indicating the lowest level of corruption and 6 the highest).
15Coplin el al. (1993) describes how country specialists are selected.
16Quality of the bureaucracy measures “autonomy from political pressure” and “strength and expertise to govern without drastic changes in policy or interruptions in government services” as well as the existence of an “established mechanism for recruiting and training.” Rule of law reflects the degree to which “citizens of a country are willing to accept the established institutions to make and implement laws and adjudicate disputes” and the presence of “sound political institutions, a strong court system, and provisions for an orderly succession of power.”
17The degree of competition is measured by an index of market dominance, of the effectiveness of anti-trust regulations, and of political or natural barriers to trade, while industrial policy is measured by indices of the extent to which procurement is open to foreign bidders and of the extent to which all enterprises are treated equally (most data are from the Global Competitiveness Report). We could not include these variables as they were not available for a large number of countries in our sample.
18The sources for the black market and the official exchange rate are International Currency Analysis, World Currency Yearbook (New York, various issues, December figures) and the IMF’s International Financial Statistics (Washington, various issues), respectively.
19This index measures the probability that two randomly selected persons from a given country do not belong to the same ethnolinguistic group. We would like to thank Paolo Mauro for sharing this data.
20This applies to most independent variables. The shares of the within-country variation is 13% for the “rule of law,” 15% for “quality of the bureaucracy,” 18% for “political rights and civil liberties,” 2% for real GDP per capita, and 69% for the black market premium.
21Only a few of the relevant variables can not be captured, namely pensions, the size of penalties, cultural factors, and “leadership.” Pensions are in theory correlated with relative wages if governments provide delayed rewards in lieu of current ones, under the mechanism identified by Becker and Stigler (1974); however, this mechanism hardly seems relevant in practice in developing countries. The remaining variables do not appear to be related to relative wages a priori and are, therefore, not problematic.
22Specifically, correction for heteroscedasticity on account of differing numbers of observations across countries is carried out in three steps. First, between estimation is carried out on the raw data. Second, the squared residuals from this regression are regressed on a constant and the inverse of the number of observations available in a country. Third, the raw data is divided by the square root of the fitted values of the previous regressions (i.e. the estimated country-specific error variances), and between estimation carried out, with White-correction, using this weighted data.
23Acemoglu and Verdier (2000) suggest another reason why the trade-off between wages and corruption may not be identified: if governments set efficiency wages, they choose just one point on the trade-off schedule, and the trade-off itself cannot be identified. We believe, however, that because of budgetary and political constraints, governments cannot follow optimal pay policy, so that variations in such constraints allow us to trace out the trade-off between wages and corruption. Furthermore, variations in preferences can also help identify the trade-off. One source of such variations is the degree of natural openness in an economy, with more open economies preferring lower corruption because of the presumed high sensitivity of international trade and investment to corruption (Wei, 2000). Evidence presented by Wei indeed suggests that naturally open economies exhibit less corruption. Furthermore, Wei, using our data set, shows that open economies appear to pay higher civil-service wages, presumably in an effort to combat corruption.
24This appears to reflect the presence of outliers. When estimated with the technique of least absolute deviations, instead of OLS, real GDP per capita is statistically significant.
25Note that this is a strong test because the two control variables—rule of law and quality of the bureaucracy—are from the same expert survey as the corruption index and are, therefore, highly correlated, leaving little variation to be explained by relative wages.
26Results are not reported and are available upon request. When estimated by the technique of least absolute deviations (LAD), which reduces the role of outliers, the coefficient on the relative wage is always statistically significant.
27Specifically, Rauch and Evans define the ratings for the salaries of higher officials in economic agencies, compared to those of private sector managers with roughly comparable training and responsibilities, as follows: 1 = less than 50%; 2 = 50–80%; 3 = 80–90%; 4 = comparable; 5 = higher.
28When estimated with LAD to reduce the role of outliers, the t-statistic on relative wages is 3.6 rather than 1.9. When real GDP per capita is the control variable, instead of rule of law and quality of the bureaucracy, the t-statistic is 3.0 under LAD. Rauch and Evans (2000) report an insignificant effect of wages on corruption; however, as our regressions show, this appears to reflect the fact that in Rauch-Evans, the dependent variable is defined as a weighted average of (ratings for) the level and trend in relative wages (rather than just the level of relative wages).
29Regressions, which exclude the city-states Singapore and Hong Kong SAR, give almost unchanged results, confirming that the relationship between wages and corruption is not driven by outliers with high wages and low corruption.
30Some observers have noted that Korea’s civil service could actually be less corrupt than the ICRG index indicates, the relatively high rating for corruption reflecting political corruption, rather than bureaucratic corruption. Without the dummy variable for Korea, the t-statistic on the relative wage variable is still reasonably high, at 1.75.
31Results are not reported and are available upon request.
32Intuitively, when the targeted income level is unreachable because bribe levels are too low or the probability of detection too high, it is reasonable to expect civil servants to attempt to get as close to their target as possible, which is equivalent to maximizing expected income (i.e. the shirking outcome).
33We thank Amitabha Mukherjee for bringing this quote to our attention.
34India rates 3.1 on the quality of bureaucracy (compared to 2.8 for the entire sample) and 3.4 for the rule of law (compared to 2.9).
35This back-of-the-envelope calculation of the efficiency wage might be an overestimate, as it does not take into account informal punishments, i.e. those not involving formal charges being brought. These punishments would involve (1) stigma; (2) reduced promotion possibilities; and (3) payments to supervisors to avoid punishment. Whether these factors are important or not would depend on how generally accepted corrupt behavior is. When corruption is generally accepted, supervisors turn a blind eye even when they are not corrupt, so that these factors may not come into play.
36Pay of professionals increased from US$50 monthly to US$ 1,000 after the introduction of an independent revenue authority in Peru (SUNAT). Pay increases were important in the other countries as well, with the exception of Zambia, where pay increases were limited to expatriate staff. See Adamolekum and Jah (1997). In some countries, such as Ghana, bonuses were introduced in addition to pay reforms (Chand and Moene, 1999). For an exploration of the effects of the introduction of performance-based wages in the Brazilian tax administration, see Kahn et at. (2000).
37See Nashashibi et al. (1992), pp. 40–41.
38As explained in Section 2, the fact that the within estimates are not significant could reflect a lack of power to reject the null and the omission of dynamics. This is because the variance of relative civil-service wages is low in our sample and because corruption may not respond to pay increases, except when pay increases are sustained (in that case, no effect would be detected in country-fixed effects regressions of corruption on contemporaneous wages).
39This is a good approximation of the true probability of detection of at least one act (when these probabilities are independent), or 1 − (1 − p)c, for low p and C. Note also that the assumption of a fixed probability of detection abstracts from the possibility of multiple equilibria. See Lui (1986) and Andvig and Moene (1990) for models of multiple equilibria and Bardhan (1997) for a review of this literature.
40It is interesting to note that in reality p and f tend to be endogenous, rendering the task of controlling corruption through p or f, but without using wage policy, difficult. Thus, cooperation with enforcement agencies is diminished (p is low) when civil servant wages are low. Similarly, high penalties (f) tend not to be enforced by the courts, that is a high f leads to low p. For example, Thailand had the death penalty for corruption at one point, but this penalty was never imposed. See Akerlof and Yellen (1994) for a model where cooperation with the police in controlling gangs (p) depends on penalties (f).
41The choice of p = 0.1 reflects anecdotal information for India (see Section 3), which rates about average on p in our sample using our empirical proxies for internal and external controls.
42See Akerlof and Yellen (1990) for a review of sociological evidence. Fehr et al. (1993) provide experimental evidence suggesting that wages motivate effort even when there are no penalties for shirking, though the effect is not very large. Fehr and Tyran (1996) interpret this finding in terms of “reciprocity,” i.e. they see a desire on the part of workers to reward well-paying employers.
43Charap and Harm (2000) see predatory states actually setting up systems with low civil-service wages to force civil servants into corruption, so as to create a system of rents.
44See Dumont (1979) for some insights on this for the case of Sub-Saharan Africa.
45The positive root involving the same expected income; however, more corruption theoretically constitutes a second solution, but this solution is ignored here.
46The incorporation of both corruption and effort in one model fits the stylized facts described in the corruption literature, see, Gould (1980, p. 71) who explains that “The civil servant who does not wish–or does not have the opportunity–to steal, or whose corruption is not sufficiently remunerative, may engage in another behavior strategy: taking on a second job, moonlighting.”

    Other Resources Citing This Publication