2nd Draft Not for Quotation! Comments Welcome! Public Sector Corruption and Gender Perceptions of Public Officials from Six Developing and Transition Countries Omer Gokcekus & Ranjana Mukherjee Second Draft: June 2002 Abstract: Based on 3,927 responses from surveying public officials in six developing and transition countries, this paper empirically shows that when the proportion of women is very low, increasing that proportion reduces the severity of corruption and increases the chances of its being reported. However, increasing women‘s proportion of public officials beyond a threshold value is counter-productive because that raises corruption and lowers its chances of being reported. 1 I. Introduction Corruption undermines economic development and damages social stability (Grindle, 1997; World Bank, 2000a and 2000b). Concern about corruption has intensified in recent years, and is accompanied by the effort to understand corruption. Bardhan, (1991), Gupta, Davoodi, and Alonso-Terme (1998); Klitgaard (1998), Mauro (1997); Tanzi and Davoodi (1997); and World Bank (2000) have focused on the economic, political, and historical factors underlying the persistence of corruption, its patterns and size; and its causes and consequences. Little is known about strategies that have been successful at reducing corruption. Could the proportion of women employees in a public organization be connected with the level of corruption in that organization? If women are under- represented in public organizations‘ employment, and if raising the percentage of women is associated with reduced levels of corruption, then actively promoting women‘s employment could be part of the World Bank‘s strategy to improve governance. Possible connections between gender and corruption have not been adequately investigated. The World Bank (2002, p. 9), while outlining a strategy for integrating gender into its work, notes, ―…a growing body of evidence suggests that gender equality in rights and resources is associated with less corruption and better governance….‖ The evidence cited in this World Bank report (Integrating Gender into the World Bank’s Work: A Strategy for Action) is based upon two studies. Lambsdorff (1999) lists the same two studies under corruption and gender in a detailed review of the available empirical research on corruption. In the first study, Dollar, Fisman, and Gatti (1999), using numerous behavioral studies, found women to be more trust-worthy and public-spirited than men. They found, in a large cross-section of countries, the greater the representation of women in parliament, the lower the level of corruption. In the second, Swamy, Knack, Lee and Azfar (2001) used several independent data sets to show that corruption is less severe where women comprise a larger share of labor force, and where women hold a larger share of parliamentary seats. 2 The two studies provide preliminary evidence of a commonly held belief that increasing women‘s representation might reduce corruption in an organization and its environment. The two previous papers found evidence of this in women‘s representation in parliament and in the labor force. The present paper examines whether this is applicable in public sector organizations. Is corruption in public sector organizations connected with the percentage of women employed in them? This investigation utilizes survey data from World Bank funded surveys (see www1.worldbank.org/publicsector/civil service/surveys.htm) of nearly 4 thousands public officials in 6 countries. It checks if there is a statistically significant connection between the percentage of women in public sector organizations on the one hand, and (i) severity of corruption, and (ii) its probability of being reported –on the other. Officials‘ responses to survey questions provide necessary information for checking the empirical relationship between gender and corruption at three different levels—across countries, within countries, and considering a pool of public organizations regardless of country. Results of the present investigation show that when public organizations have too few women (less than one-third), increasing the proportion of females reduces the severity of corruption. However, continuously increasing the proportion of females is counter- productive. After a certain threshold, further increasing the proportion of women increases corruption and reduces the chances of its being reported. The rest of this paper is organized into three main sections. Section one introduces the data set, the concepts and the corresponding indices, and discusses the reliability of self- reported perceptions of public officials. Section two discusses the three levels of analyses and presents the findings. Finally, section three contains concluding remarks, and a discussion of the policy implications of these findings. 3 II. Data Set and Indices Before analyzing the relationship between women officials‘ proportion in public organizations, the severity of corruption and its probability of being reported, it is necessary to explain how two appropriate corruption indices (severity, and reporting probability) were constructed, and why these indices were considered reliable measures. Responses of 3,927 officials from 90 organizations were utilized to construct corruption indices.1 Table 1 Data Set Summary Number of Number of Number of public Percentage of Country organizations women in the officials surveyed women surveyed sample 1 Argentina 370 5 155 42% 2 Bolivia 692 16 201 29% 3 Bulgaria 1,089 23 741 68% 4 Guyana 464 15 278 60% 5 Indonesia 640 15 166 26% 6 Moldova 672 16 296 44% Total 3,927 90 1,837 47% During each of the six country surveys, officials were first asked to assess the severity of corruption in their organizations as follows: Some consider that one of the main problems affecting the performance of the Government and its structures is corruption. How important is this problem in your organization? 1) Very important 2) Relatively important 3) Of little importance 4) Not important at all Officials were next asked about the chances of reporting corruption: In general, how often are cases of corruption reported to proper authorities? 1 Public officials surveys were administered in 16 countries. For the purpose of this investigation, only six countries were selected because they each contained almost identically worded questions probing the same issue. Sample distribution in all surveys closely followed actual distribution of officials (level, type of organization), and random stratified sampling was commonly employed. 4 1) Always 2) Frequently 3) Sometimes 4) Never Officials‘ responses to these corruption-probing questions were on closed-ended but varying descriptive scales—depending on the exact questions asked. For instance, Moldavian officials being asked about the severity of corruption were asked to choose among four responses ‗very important,‘ ‗relatively important,‘ ‗of little importance,‘ and ‗not important at all‘. Argentinean officials, on the other hand, were asked to choose among three choices whether corruption was a ‗significant problem,‘ ‗somewhat of a problem,‘ or ‗not much of a problem.‘ Similar minor variations appeared in questions about reporting cases of corruption. Moldavian officials were asked to choose between ‗always,‘ ‗frequently,‘ ‗sometimes,‘ ‗never‘. In Bulgaria, the available responses to whether corruption is reported were ‗yes‘ or ‗no‘. To measure all surveyed organizations‘ corruption indices on the same scale, officials‘ descriptive responses were converted to a numerical scale of 0 to 10,where 0 = good i.e. no corruption, and 10 = bad, i.e. highly corrupt. The following formulas were used to convert the different scales to a common scale: For questions with four response choices: 1 = very important, 2 = relatively important, 3 = of little importance, and 4 = not important at all, CORRUPTION INDEX organization-i = 40/3 – 10/3 (Average response organization-i). For questions with three responses 1 = significant problem, 2 = somewhat of a problem, or 3 = not much of a problem, CORRUPTION INDEX organization-i = 15 – 5 (Average response organization-i) Table 2 below summarizes surveyed officials‘ perceptions of the level of corruption and its probability of being reported. 5 Table 2 Officials‘ perceptions of severity and reporting of corruption Percentage of officials who reported Percentage of officials who reported Country that corruption is a serious / that corruption is not reported important problem All officials Female Male All officials Female Male 1 Argentina 62% 61% 63% 18% 20% 16% 2 Bolivia 88% 87% 88% 31% 28% 35% 3 Bulgaria 50% 53% 41% 73% 75% 70% 4 Guyana 32% 34% 28% 44% 44% 44% The survey administered to Indonesian 5 Indonesia 93% 93% 93% officials did not ask whether cases of corruption are reported to proper authorities 6 Moldova 73% 76% 69% 18% 17% 18% Reliability of using officials’ perceptions to measure corruption Table 2 above shows that women do not systematically under or over-estimate corruption. The biggest perception difference (12%) among genders was in Bulgaria. However, according to two-sided t-tests, these differences are not statistically significant even at a 0.10 level. Officials‘ subjective assessments based upon their perceptions of corruption were utilized in the present investigation to estimate public organization‘s severity and reporting of corruption. Perceptions are hard to measure and easy to challenge. Therefore, even before checking the relation between gender and corruption in public organizations, the reliability of using officials‘ perceptions of corruption were checked against another estimate of corruption that is not based on perceptions. Transparency International‘s (TI) Corruption Perception Indices (CPI) are composite, based upon assessments of business people and risk analysts, utilize 14 different data sources from seven different institutions, and omit countries where fewer than three reliable sources of data are available (Lambsdorff 2001, p.232-233.). Transparency International‘s 2001 Corruption Perception Indices were available for all surveyed countries in the present sample except Guyana. TI‘s widely acceptable 2001 CPIs were compared against each of the present study‘s surveyed organizations‘ corruption index. The comparison showed that public 6 organizations‘ corruption as calculated in the present investigation from officials‘ survey responses is almost perfectly correlated with Transparency International‘s estimate (Figure 1). The estimated correlation coefficient is 0.99, which is statistically significant at a 0.01 level, according to a two-sided t-test. Transparency International found overall corruption to be high in Indonesia and Bolivia, and low in Bulgaria. Officials in Bolivian and Indonesian public organizations have rated severity of corruption in their organizations as very high; while officials in Bulgaria estimated their own organizations‘ corruption as less severe. Figure 1 Public officials perceptions of corruption versus TI-2001 Corruption Perception Indices Percentage of officials that reported corruption to be a 100% y = 1.30 -0.20x Indonesia R2 = 0.97 90% Bolivia significant problem 80% Moldova 70% Argentina 60% Bulgaria 50% 40% 1 2 3 4 Transparency International's 2001 Corruption Perceptions Index (10 = very little corrution, 0 = highly corrupt) III. Main Findings The association between gender and corruption in public sector organizations was assessed, using three different bases of analyses to avoid possible biases. The connection between gender and corruption (severity and reporting probability) was checked: first, 7 across the six sampled countries; next, within each surveyed public sector; and finally, within a common pool of public organizations regardless of country. At each level of analysis, appropriate statistical tests were performed to determine the statistical significance of the associations. A. Cross-country Connections Figure 2 presents the relationship between public officials‘ estimate of corruption in the public sector and the percentage of female officials employed in sampled organizations in six countries. It shows that, in general, public sector corruption declines with rising percentage of female employees. Among the sampled public sectors, Bolivia and Indonesia had a comparatively small percentage of females, and high corruption. Figure 2 Association of corruption severity with percentage of females employed in public sector, r= 0.9 100% Percentage of officials who reported that Indonesia corruption is a significant problem Bolivia 75% Moldova Argentina 50% Bulgaria Guyana 25% 25% 50% 75% Percentage of female officials in public sector Bulgaria and Guyana had a large percentage of females, and less severe corruption. However, increasing females‘ percentage beyond 60% proves counter-productive by raising the severity of corruption. 8 At cross-country level, the next relationship examined was that between the percentage of female officials employed in the public sector and percentage of (all) officials reporting that corruption is not reported. Figure 3 below presents the relationship between corruption -reporting indicator and percentage of female officials in the public sector in five countries. (As mentioned earlier in Table 2, the survey in Indonesia did not ask officials about reporting corruption.) When the percentage of females is low (less than one-third), increasing the percentage of females generally raises the likelihood of corruption reporting. However, raising the percentage of females beyond 40% reduces the likelihood of corruption being reported. Figure 3 Association of reporting corruption with gender in public sector organizations, r=0.78 100% Percentage of officials who reported that corruption is not reported 75% Bulgaria 50% Guyana Bolivia 25% Moldova Argentina 0% 25% 50% 75% Percentage of female officials in public sector Taken together, cross country empirical analysis suggests that having too few or too many women is associated with an increase of the severity of corruption and a drop in reporting. 9 B. Within-country Findings Aggregate country pictures could be considered unreliable because they consider averages of corruption indices of all surveyed public sector organizations in the country, and each of those organizations could have widely varying levels of corruption and proportions of female employees. Focusing on averages without taking standard deviations into consideration might be misleading. Therefore, it is a useful exercise to check whether, within each country, corruption of public sector organizations is associated with the percentage of female officials employed in the organization. To do this, surveyed organizations in each country were divided into two groups, HIGH and LOW, depending on whether the percentage of female officials employed in that organization were high or low. The classification of HIGH or LOW was not absolute, but relative to the average of surveyed public organizations‘ employment for that country. Next, each public sector‘s average percentage of females was calculated. Organizations with the percentage of females at least one standard deviation above the country‘s average were included in HIGH group; similarly, organizations with percentage of females at least one standard deviation below the country‘s average were included in LOW group. Both groups‘ index of corruption severity and corruption reporting were calculated. Figure 4 shows that in four of the six sampled public sectors—Argentina, Bulgaria, Guyana, and Indonesia, organizations with lower-than-average females had higher corruption than organizations with higher-than-average proportion of females. These differences are statistically significant at a 0.01 level in Guyana; at a 0.12 percent level in Argentina; 0.14 level in Bulgaria; and at a 0.27 level in Indonesia. Gender had the biggest impact on public sector corruption in Argentina and Guyana—2.19 and 2.07 standard deviations respectively. On the other hand, in Bolivia and Moldova, corruption difference according to females‘ percentage is small and not statistically significant. Testing within-country differences between LOW and HIGH groups in these five countries showed that the corruption severity difference is a statistically significant one at a 0.12 level. 10 Figure 4 Severity of corruption in public organizations with higher or lower than average female officials Severity of corruption (in standard deviations from the 2 1.5 1.36 1 0.80 0.59 0.53 0.61 national average) 0.5 0.14 0 Argentina Bolivia Bulgaria Guyana Indonesia Moldova -0.5 -0.45 -0.49 -1 -0.67 -0.83 -0.87 -1.5 -1.54 -2 LOW HIGH Figure 5 below shows that in Bolivia (with the second-smallest percentage of female officials in the sample, 29%), the chances of reporting corruption were higher in HIGH public organizations. In the remaining sampled public sectors that already had more than 40% women officials – i.e. Argentina (42%), Moldova (44%), Guyana (60%) and Bulgaria (68%), the chances of reporting corruption were lower in the HIGH group of organizations. 11 Figure 5 Association of chances of reporting corruption with gender in public organizations Chances of reporting corruption (in standard deviations from the 1.2 1.05 1 0.85 0.8 0.6 0.38 national average) 0.4 0.23 0.25 0.2 -0.02 0 -0.2 Argentina Bolivia Bulgaria Guyana Moldova -0.4 -0.29 -0.37 -0.6 -0.57 -0.57 -0.8 LOW HIGH Taken together, within country findings are that in Bolivia (29% women) and Moldova (42% women), raising the proportion of women improves the severity of corruption. The chances of corruption reporting improved with higher percentage of women only in Bolivia that had less than 30% female officials. C. Common pool of organizations regardless of country To eliminate cultural or any other country-specific bias, the 90 surveyed public sector organizations from the six countries were first placed in a common pool, and next divided into four quartiles according to the percentage of female officials employed. The bottom quartile i.e. the 22 public organizations with fewest females were labeled as LOWEST; and the top quartile i.e. the 22 organizations with the highest percentage of female officials were labeled as HIGHEST. 12 Figures 6 presents the average corruption and reporting corruption scores for LOWEST and HIGHEST groups. For the LOWEST and HIGHEST groups, the average standardized corruption scores were 0.18 and -0.39 respectively. The difference in corruption severity is 0.57 standard deviation. Figure 6 Organizations with the highest and lowest percentage of women had the lowest and highest corruption severity 0.3 deviations from the national average) 0.18 Corruption severity (in standard 0.2 0.1 0 LOWEST HIGHEST -0.1 -0.2 -0.3 -0.4 -0.39 -0.5 Organizations in the highest and lowest quartiles 13 On the other hand, the average standardized reporting-corruption scores are 0.49 and -.25 for the LOWEST and HIGHEST agencies, respectively (Figure 7). The difference is a Figure 7 Organizations with the highest and lowest percentage of women had the highest and lowest chances of corruption being reported 0.6 0.49 standard deviations from the national Chances of corruption reporting (in 0.5 0.4 0.3 average) 0.2 0.1 0 LOWEST HIGHEST -0.1 -0.2 -0.3 -0.25 Organizations in the highest and lowest quartiles 0.74 standard deviation, and as is summarized in Table 3, the corruption difference between LOWEST and HIGHEST groups is statistically significant at a 0.01 level. Table 3 summarizes that the corruption difference between LOWEST and HIGHEST groups of organizations is statistically significant at a .06 level Table 3 Pair-Wise t-test for Differences in Means of Level of Corruption and Reporting Corruption Grouping LOW HIGH │t-values│ Level of Corruption 1.71 Within-country 0.32 -0.46 (0.12) 1.87 Pooled-organization 0.18 -0.39 (0.06) Reporting Corruption 1.77 Within-country 0.37 -0.19 (0.11) 2.63 Pooled-organization 0.49 -0.25 (0.01) Two-tailed tests, assuming equal variances, (p-values in parentheses). 14 To control the effect of country specific variables‘ and organizational functions‘ effect on corruption, a regression analysis was conducted. Dummy variables were created for transition economies (two out of six in our sample of countries) and policy implementing agencies (the most common type of organization surveyed). The table below presents the estimation results. Table 4 The OLS regression estimates Coefficients Standard Error t Stat Intercept 7.776 1.067 7.288 Transition -0.802 0.569 -1.408 POI 1.731 0.522 3.316 R-female -9.139 5.014 -1.823 (R-female)2 11.682 5.557 2.102 R2 0.149 # of observations 89 Transition countries had a relatively lower level of corruption than other countries in our sample; and policy implementing organizations had a higher level of corruption than organizations with other functions. Even controlling for these factors, increasing the percentage of females up to 40% level reduces the level of corruption, and increasing females beyond this level (of 40%) raises the severity of corruption. In other words, having too few or too many women is associated with an increase of the severity of corruption. 15 Concluding Remarks Empirical analysis provided evidence Table 4. Proportion of women in work force that there is a statistically significant % of women in % of women at Country Name relationship between corruption and labor force ministerial level Argentina 32 8 gender. Across countries, within Bolivia 38 6 public organizations in a particular Bulgaria 48 NA Guyana 34 NA country, or considering public Indonesia 40 3 Moldova 49 0 organizations irrespective of country, Source: World Development Indices, 2001 higher percentage of female employees is connected with lower corruption in public organizations and increases the probability of corruption instances being reported. Statistics in Table 4 indicate that countries with a low proportion of women in the work force (Argentina, Guyana, Bolivia and Indonesia) may benefit from raising the proportion of women. It could also be argued that a very low percentage of women in public sector employment is the result of the tendency for clientilistic, traditional systems to exclude certain groups. Equally, causality could in fact the other way round, in that corruption causes imbalanced Figure 8 Severity of corruption Figure 9 Chances of reporting corruption 1.20 1.20 Percentage of officials who reported significant 1.00 1.00 Percentage ogf officials who reported that 0.80 0.80 corruption would not be reported 0.60 0.60 0.40 0.40 corruption 0.20 0.20 0.00 0.00 0% 10% 20% 30% 40% 50% 60% 70% 80% -0.20 0% 10% 20% 30% 40% 50% 60% 70% 80% -0.20 -0.40 -0.40 -0.60 -0.60 -0.80 -0.80 Percentage of females in public Percentage of females in public organizations -1.00 organizations 16 representation of men and women in the public sector. The findings of this investigation have led to the simplistic recommendation that continuously increasing the percentage of females in public sector organizations can reduce corruption until if all employees were women, corruption would be eliminated. However, empirical evidence, as is summarized in Figures 8 and 9, suggests a point of flexion. Increasing the percentage of females beyond a certain point (around 45%) reduces the likelihood of its being reported, and increasing females beyond 70% again raises the severity of corruption. For the Bank‘s work in client countries that have a low percentage of women in the public sector, actively promoting women‘s recruitment to raise that percentage to about 40 may help reduce corruption. On the other hand, in countries with more than 50% women employees in the public sector, actively promoting the recruitment of men may help reduce corruption. 17 References Bardhan, P. 1997. ―Corruption and Development: A Review of Issues.‖ Journal of Economic Literature 25: 1320-1346. Dollar, David, Raymond Fisman, and Roberta Gatti. 1999. Are Women Really the “Fairer” Sex? Corruption and Women in Government. The World Bank Policy Research Report on Gender and Development Working Paper Series, No 4. Washington D.C.: The World Bank. Grindle, Merilee (editor). 1997. Getting Good Government: Capacity Building in the Public Sector of Developing Countries. Boston: Harvard Institute for International Development. Gupta S. H. Davoodi, and R. Alonse-Terme. 1998. ―Does Corruption Affect Income Inequality and Poverty?‖ IMF Working Paper 98/76. Washington, D.C.: International Monetary Fund. Klitgaard, R. 1998. ―International Cooperation Against Corruption.‖ Finance and Development, 35(1). Lambsdorff, Johann Graf. 1999. Corruption in Empirical Research-A Review. The 9th International Anti-Corruption Conference: The Papers. Durban, South Africa. www.transparency.org/iacc/9th_iacc ___________________ 2001. ―Transparency International 2001 Corruption Perceptions Index‖ in Robin Hodess with Lessie Banfield and Toby Wolfe eds., Global Corruption Report 2001, Berlin, Germany: Transparency International. Mauro, Paolo. 1997. Why Worry About Corruption, IMF Economic Issues Series No 6. Washington, D.C.: International Monetary Fund. Swamy, Anand, Steve Knack, Young Lee, and Omar Azfar. 2001. ―Gender and Corruption.‖ Journal of Development Economics 64(1): 25-55. Tanzi, V. and H. Davoodi. 1997. ―Corruption, Public Investment, and Growth.‖ IMF Working Paper 97/139. Washington, D.C.: International Monetary Fund. 1997. World Bank. 2000a. Reforming Public Institutions and Strengthening Governance. Washington D.C.: The World Bank. __________ 2000b. Anticorruption in Transition: A contribution to the policy debate. Washington D.C.: The World Bank. __________ 2002. Integrating Gender into the World Bank’s Work: A Strategy for Action. Washington D.C.: The World Bank.