E-Government, Internet Adoption, and Corruption: An Empirical Investigation

World Development Vol. 57, pp. 114–126, 2014 Ó 2013 Elsevier Ltd. All rights reserved. 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

http://dx.doi.org/10.1016/j.worlddev.2013.12.005

E-Government, Internet Adoption, and Corruption: An Empirical Investigation NASR G. ELBAHNASAWY * Kent State University, USA Mansoura University, Egypt Abstract. — This study empirically investigates the impact of e-government and internet adoption on curbing corruption, by utilizing a large panel dataset. The results reveal that e-government is a powerful tool in reducing corruption—via telecommunication infrastructure and the scope and quality of online services—which is strengthened by greater internet adoption. The interaction effects between egovernment and internet adoption suggest both as complements in anti-corruption programs. A dynamic panel data model that addresses the endogeneity problem and considers corruption persistency is employed. Results of panel Granger causality tests indicate a unidirectional causality from e-government to corruption, while a bilateral causality between internet adoption and corruption. Ó 2013 Elsevier Ltd. All rights reserved. Key words — corruption, e-government, internet adoption, law enforcement

1. INTRODUCTION

public officials. A few other studies have attempted to estimate the impact of internet adoption on corruption. For instance, while Andersen (2009) found support for the view that e-government is a useful tool in reducing corruption, he claims that the internet use is more important than e-government in fighting corruption. Likewise, Goel, Nelson, and Naretta (2012) argue that the internet reduces corruption by enhancing the access to information and the speed of information dissemination, which raises the level of corruption awareness and elevates the detection risk of corrupt behavior. However, it can also be claimed that the internet may possibly increase corruption perception by disseminating information about corrupt acts from individuals’ perspectives—such as amateur video uploads—which gives rise to the feeling that ‘‘everyone is corrupt” and hence encourages persistent corruption. Andersen (2009) stresses the need for macro evidence on the effect of e-government on corruption since it is also possible that e-government may cause corruption to migrate elsewhere in the economy. Hence, e-government may turn out to be ineffective on the macro level, even though micro-level evidence suggests that it is effective in combating corruption. In the same line, Ojha, Palvia, and Gupta (2008) emphasize the need for empirical research evidence on the effects of e-government on anti-corruption efforts since the majority of current research is theoretical and descriptive. In a similar vein, although it recently supports e-government as a component of anti-corruption programs, the UNDP asserted that much of the evidence that links e-government to corruption reduction is anecdotal (United Nations Development Program, 2006, p. 4). This paper attempts to correct this shortcoming by empirically investigating the role that e-government and the level of internet adoption play in reducing corruption, both in

Many policymakers and various international organizations who are committed to promoting sustainable development and global economic growth have recently embraced the view that e-government would play a substantial role in the battle against corruption (Bhatnagar, 2003; Organization for Economic Co-operation and Development (OECD), 2005; United Nations Development Program (UNDP), 2006, 2008). E-government is alleged to lower the interaction between government officials and citizens, and hence diminish the discretionary power of officials. It may also enhance accountability and transparency by disseminating a greater quantity and a higher quality of information in the economy, which incites citizens and businesses to question arbitrary decisions and unreasonable procedures. Thus, e-government may possibly eliminate many opportunities for corruption (OECD, 2005; Piatkowski, 2006). Accordingly, it can combat corruption and hence boost economic growth, particularly in developing countries where corruption appears to be the single greatest obstacle to economic development, as identified by the World Bank (2001). This is because corruption deteriorates national institutions, erodes the incentive system meant to maintain economic growth and sustainable development, escalates economic inefficiency, and impedes the success of United Nations Millennium Development Goals. 1 However, it can also be argued that e-government may lead corrupt public employees to learn ways to beat the new e-government systems, as these systems can have weaknesses that enable corrupt behavior to continue and may even grow faster (Bhatnagar, 2003). Thus, e-government can be unsuccessful tool in curbing corruption. 2 Yet, to the best of my knowledge, the empirical evidence of the impact of e-government on corruption is scant. Only a handful of case studies using the micro-level data report some effect of e-government implementation on corruption reduction. Examples of those studies—where e-government is shown to reduce corruption—include Kim, Kim, and Lee (2009) and Chawla and Bhatnagar (2004). 3 Conversely, using five case studies, Heeks (1998) argues that information technology may even create new opportunities for corruption of

* I would like to thank Emmanuel Dechenaux, Charles Revier, and three anonymous referees for their valuable comments and suggestions, which substantially improved the paper. I also extend my thanks to Shawn Rohlin and Eric Johnson. All remaining errors or omissions are mine. Final revision accepted: December 19, 2013. 114

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION

developed and developing countries. Understanding the linkage between e-government and corruption would enable using e-government more effectively in anti-corruption efforts. Furthermore, it would allow giving considerable priority to curbing corruption in defining the vision and priority areas of e-government, especially in developing countries. Hence, e-government can be used as a tool to fight corruption and to promote economic development, along with enhancing the quality of services provided to citizens and advancing good governance. This work also examines the interaction effects between e-government and internet adoption on curbing corruption. A comprehensive panel data set was compiled, which consists of 160 countries, covering the period from the year 1995–2009. The study employs a unique measure of e-government from the United Nations covering various dimensions of e-government that has never been utilized before in prior research on corruption. To estimate the effects of e-government and internet adoption on corruption, besides a random effects analysis, this paper adopts a dynamic panel data model, in order to deal with the potential endogeneity problem and to consider the inertia inherent in corruption (Mishra, 2006; Tirole, 1996). The direction of causality between e-government and corruption, and between internet adoption and corruption, is examined by employing panel Granger causality tests. The results from the panel Granger causality tests suggest that greater e-government reduces corruption, but not the other way round, while causality between internet adoption and corruption runs in both directions. The empirical findings highlight the importance of e-government in the battle against corruption. From a policy perspective, e-government is a powerful tool in anti-corruption efforts that needs to be recognized by policymakers. This finding is quite robust to different model specifications and various measures of law enforcement. The driving force of e-government influence on depressing corruption is the telecommunication infrastructure, besides the scope and quality of online services. On the contrary, the impact of the extent of internet adoption on corruption reduction is ambiguous and seems to be sensitive to model specification, in contrast to findings from prior research. Nevertheless, the interaction effects between e-government and internet adoption suggest that a greater number of people with access to the internet strengthens the positive impact of e-government on curbing corruption, implying a complementary relationship between both policy tools that should be regarded in anti-corruption efforts. Other new findings reveal that inflation variability and trade protectionism do not impact corruption, unlike the findings of prior literature. Furthermore, the persistency of corruption seems highly likely. The remainder of this paper is organized as follows. Section 2 provides the theoretical background and related literature. Section 3 describes the empirical methodology and data used for empirical analysis. Section 4 proceeds to the results, while Section 5 concludes and discusses the policy implications of the results. 2. THEORETICAL BACKGROUND AND RELATED LITERATURE E-government can be defined as the use of information and communication technology (ICT) by the government in order to work more effectively, share information, and deliver better services to the public (United Nations Development Program, 2006). 4 It refers to the delivery of federal and local

115

governmental information and services to citizens, businesses, and governmental agencies using the internet or any other digital means. On the other hand, corruption is usually defined as ‘‘the misuse of public power, office, or authority for private gains” (UNDP, 2006, p. 2, 2008, p. 18). Corruption is likely to reduce economic growth by depressing investment; distorting public spending and the allocation of resources; weakening public institutions, contract enforceability, and propertyrights; increasing income inequality and poverty; allocating talented people to rent-seeking activities rather than to more productive ones; and by raising economic inefficiencies, unpredictability of polices, and political instability (Gupta, Davoodi, & Alonso-Terme, 2002; Mauro, 1995, 2002; Murphy, Shleifer, & Vishny, 1991). Corruption can be viewed as a problem of asymmetric information and incentives that can be explained by a simple principal-agent-client model. In this model, the elected government officials representing the state and its citizens (the principals) are unable to supply public services to citizens (the clients) and that they have to employ public servants (agents) on their behalf to adequately deliver those services to citizens (Klitgaard, 1988). But, due to asymmetric information, whereby agents know more about the administration than both of the principals and the clients, the agents may act opportunistically in their own interest to take advantage of the entrusted power by engaging in corrupt acts—such as bribery, extortion, fraud, nepotism, and embezzlement (Lio, Liu, & Ou, 2011; United Nations Development Program, 2008)—and that corruption arises, particularly in the existence of lack of accountability. The illicit rent-seeking behavior would be exacerbated with greater monopoly power granted to agents over clients; greater discretionary power delegated to agents; and with weak accountability of agents to the principal. 5 Hence, to combat corruption, it would be vital to restructure the principal-agent-client connection by lessening the amount of discretionary power entrusted to agents and enhancing the accountability of agents to the principal (Klitgaard, 1988). In this regard, e-government can be seen as an effective tool in restructuring the principal-agent-client relationship to reduce corruption by expanding access to information; simplifying rules and procedures and making them more transparent; providing detailed data on transactions and hence easing the process of tracking actions and decisions made by agents; enhancing the questionability of their unreasonable actions; reducing their discretionary power by standardizing the delivery of services; and promoting accountability (Bhatnagar, 2003, p. 30). In addition, it allows maintaining full data on transactions, which increases the rate of detection of corrupt acts. 6 Therefore, e-government may generally create disincentives for government officials to engage in corrupt behavior by increasing the probability of exposure. 7 However, there has been little research on the empirical evidence of these possible effects of e-government on corruption. At the microeconomic level, Kim et al. (2009) have investigated the development of an anti-corruption system in the Republic of Korea—called online procedures enhancement for civil application (OPEN)—which enables citizens to monitor the progress of their applications in 54 common procedures. They find that corruption was reduced in Seoul Metropolitan Government where this system was implemented. 8 Using a panel dataset consisting of 70 countries and covering the period of 1998–2005 for three independent variables (gross domestic product (GDP) per capita, education, and internet users), Lio et al. (2011) have found that the internet adoption reduces corruption, although the magnitude of the effect is quite small. Furthermore, they argue

116

WORLD DEVELOPMENT

that the causality between internet adoption and corruption runs both ways. Goel et al. (2012) have used internet data on the number of internet hits on corruption and bribery that Google and Yahoo search engines would produce at a certain point of time in a given country to claim that, in their cross-sectional regressions, greater corruption awareness— measured by the internet hits (per capita) about corruption—may act as a deterrent of corruption. Andersen (2009) has applied ordinary least squares (OLS) and two-stage least squares (2SLS) to a dataset consisting of 149 countries with two time observations (1996; 2006) to find support to the view that e-government is a useful tool in reducing corruption, but in nonOECD countries. 9 He also argues that the internet use is more important than e-government in corruption reduction, although he is unable to provide conclusive evidence for this claim due to data limitations. 3. EMPIRICAL METHODOLOGY AND DATA To estimate the effects of e-government and internet adoption on corruption, this study utilizes a large panel data set that consists of 160 countries and for the period 1995–2009. This data set is an unbalanced panel since some of the countries in the sample have different number of time series observations. (a) Random effects models There is no widely agreed upon proper regression model for the analysis of corruption due to the lack of strong theoretical framework for corruption (Seldadyo & de Haan, 2006). Nevertheless, the random effects model (RE) is commonly used by researchers for the analysis of panel data on corruption, since the fixed effects estimator (FE) will include too many country dummies when the number of countries in the sample (N) is large. This may aggravate the multicollinearity problem among explanatory variables and lead to a serious loss of degrees of freedom (Baltagi, 2008, pp. 14–15; Wooldridge, 2011, pp. 321–334). In addition, Judge et al. (1985, pp. 489–491) suggested that the random effects estimators are more efficient than the fixed effects estimators when N is large and T (the number of years) is small, assuming that the other assumptions of the random effects model hold. Hence, the following benchmark cross-country random effects panel data model is estimated. Corruptioni;t ¼ a þ X 0i;t b þ li þ mi;t

ð1Þ

where Corruptioni,t refers to corruption in country i in year t; i = 1, . . ..N (N denotes the total number of countries); t = 1, . . ., T (T denotes the total number of years); Xi,t is the vector of explanatory variables for country i in year t; li denotes unobservable time-invariant country-specific effect and accounts for any country-specific effect that is not included in the regression; and mi,t indicates the remainder disturbance in this one-way error component model. The li are assumed to be random and independent of the mi,t, and that li  IIDð0; r2l Þ and mi  IIDð0; r2m Þ. Furthermore, the Xi,t are independent of the li and the mi,t for all i and t. In this analysis, the Corruption Perceptions Index (CPI) by Transparency International—a widely used index to measure corruption—is utilized. The CPI is a composite index that assesses the perceived level of public-sector corruption in 180 countries, based on various expert and business surveys. 10 It ranks countries on a scale from zero to 10, where higher score indicates lower perceived corruption, which I have adjusted so

that a higher score represents higher perceived corruption in order to ease interpretation of the results. The vector of explanatory variables in Eqn. (1) includes e-government and internet adoption, besides other controls. To measure e-government, this paper uses the e-government readiness index, composed by the United Nations (UN). The UN e-government readiness index—also known as the UN e-government development index (EGDI)—is a comprehensive measure of the ‘‘willingness and capacity of national administrations to use online and mobile technology in the execution of government functions. It is based on a comprehensive survey of the online presence of all 192 Member States.” 11 It ranges from zero to one, where higher scores denote better e-government. 12 The EGDI is a weighted average of three different dimensions of e-government: the scope and quality of online services; the telecommunications infrastructure; and human capacity. 13 Each of these dimensions of e-government is a composite index of some other various measures. The online services index assesses the level of web content accessibility in each country, according to the Web Content Accessibility Guidelines of the World Wide Web Consortium. This index is based on a four-stage model of online service maturity: the emerging online presence with simple websites; the enhanced information services with the deployment of multimedia content and two-way interaction; the online provision of transactional services; and the connected services where government websites communicate with citizens using interactive tools. 14 The telecommunication infrastructure index is based on five indicators: the number of personal computers; internet users; telephone lines; mobile cellular subscriptions; and fixed broadband subscribers (each is per 100 persons). 15 Similarly, the human capital index is made up of the adult literacy rate and the gross enrollment ratio of the primary, secondary, and tertiary education. 16 On the other hand, the internet adoption is measured by the number of internet users (people with access to the worldwide network) per 100 populations, which is obtained from the World Bank’s World Development Indicators (WDIs). The hypotheses tested here are that more extensive e-government and greater internet adoption in a country would act as corruption deterrent. Regarding other controls to include in the vector of regressors, there is no broadly accepted theory on the determinants of corruption that may guide the selection of those regressors in Eqn. (1). However, among those that have been found to be robust determinants of corruption is GDP per capita, which is the most consistent finding of empirical studies on corruption determinants. Rich countries are able to allocate more resources to fight and prevent corruption. Hence, poor countries are perceived to be more corrupt than rich countries (Elbahnasawy & Revier, 2012; Jain, 2001; Serra, 2006; Treisman, 2007). Data on GDP per capita are adjusted for purchasing power parity (also measured in constant purchasing power; in 2005 international dollars) and come from the World Bank’s WDIs. Treisman (2000, 2007) argues that openness to international trade, measured by the share of imports in GDP, is among the most important predictors of corruption. Greater openness to international trade increases market competition, and hence discourages rent-seeking behavior of corrupt officials by reducing the monopoly power of domestic producers. Dutt (2009) also finds that trade protectionist policies would lead to higher corruption. Unlike prior studies that used some sort of measures for trade volume, this work uses a direct measure of trade protectionism; the weighted mean applied tariff rate, from the World Bank’s WDIs. 17

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION

Elbahnasawy and Revier (2012) find that the rule-of-law, measured by the rule-of-law index (RL), is among the strongest candidates for corruption determinants. Strong rule-of-law reduces the probability that corruption occurs. The rule-of-law also captures the quality of institutions (Goel et al., 2012). Nevertheless, Treisman (2007) has cautioned that the rule-of-law is a subjective concept and both of the ruleof-law index and corruption indicators may measure the same underlying perception when these indicators are taken from the same survey or rating agency. Yet, he has emphasized the importance of studying the impact of the rule-of-law on corruption. To add to the line of this investigation, this study uses a new proxy for law enforcement; the property-rights index (PR) from Heritage Foundation. The property-rights index measures ‘‘the ability of individuals to accumulate private property, secured by clear laws that are fully enforced by the state. It measures the degree to which a country’s laws protect private property-rights and the degree to which its government enforces those laws. It also assesses the likelihood that private property will be expropriated and analyzes the independence of judiciary, the existence of corruption within judiciary, and the ability of individuals and business to enforce contracts.” 18 The property-rights scores range from zero to 100, with higher scores denoting greater certainty of the legal protection of property in a country, and hence stronger law enforcement. 19 In the absence of the ability to have a precise measure of the amount of law enforcement in a country, the use of alternative measures of law enforcement will be useful tests of the validity of the findings. Hence, to check the robustness of the findings, the study also applies the rule-of-law index, from the World Bank’s Worldwide Governance Indicators (WGIs), used in previous research. 20 The rule-of-law measures ‘‘perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property-rights, the police, and the courts, as well as the likelihood of crime and violence” (Kaufmann, Kraay, & Mastruzzi, 2010, p. 4). It ranges from 2.5 to 2.5, where higher values imply better rule-of-law, and thus more effective law enforcement. The rule-of-law index is a weighted average of a large number of individual data sources (incorporated into a single aggregate measure by using unobserved components model), with greater weights given to sources with higher correlation with each other. 21 Another robust predictor of corruption in prior literature is inflation. A higher and more variable inflation leads to greater complexity in monitoring government spending and public contracts, leading to higher corruption (Braun & Di Tella, 2004; Treisman, 2007). Hence, countries with higher and more variable inflation have greater corruption. The inflation rate is obtained from the International Monetary Fund’s World Economic Outlook (WEO) database. 22 An additional determinant of corruption in previous research is press freedom. Better press freedom enhances transparency and elevates the risk of corrupt acts, amplifying the cost of corrupt behavior and leading to lower corruption (Freille, Haque, & Kneller, 2007; Kolstad & Wiig, 2009; Treisman, 2007). Data on press freedom come from the Freedom House that has constructed the index of press freedom since 1980, which provides ratings for countries based on their degrees of press freedom. The freedom of the press index ranks countries on a scale ranging from zero (best press freedom) to 100 (worst press freedom), which I have rescaled to ease interpretation so that higher values denote greater press freedom. 23 This paper also contributes to the literature investigating causes of corruption by including the scope of a country’s urbanization. The intuition behind the scope of urbanization

117

in a country is that a rural population is likely to be less knowledgeable about the process of government bureaucracy, and hence is more tolerant of questionable public officials behavior (Elbahnasawy & Revier, 2012). In contrast, with a greater share of rural population, the level of transactions in a country would be lower, and hence the opportunity for public officials rent-seeking behavior is reduced. In addition, it is highly likely that ‘‘everybody knows everybody” in rural population, suggesting that the cost of corrupt behavior of public officials is greater since corruption in this case would have a ‘‘stigma” as it implies that the pubic officer will be reputable in the community as ‘‘the one that can be bought.” Therefore, a priori, a greater percentage of rural population in a country would increase corruption if the former effect is dominant, and would reduce corruption when the latter effects are stronger. The source of the percentage of population that is rural is the World Bank’s WDIs. Since the dataset contains many more observations for some of the regressors than for others, Eqn. (1) has three various specifications. The first specification controls for GDP per capita, law enforcement, internet adoption, rural population, inflation, and press freedom. The number of observations drops significantly in the second specification as it adds e-government to explanatory variables. The third specification is the unrestricted specification that adds the tariff rate to the previous set of regressors, where the number of observations declines. In addition, to analyze the impact of each subcomponent of e-government independently on corruption reduction, the fourth specification disaggregates e-government in the second specification to its constituent parts; the online services component; the telecommunication infrastructure; and the human capacity. Moreover, the fifth specification examines the possible interaction effects between e-government and internet adoption on combating corruption. 24 The Breusch–Pagan Lagrange multiplier (LM) test for random effects is statistically significant at the 1% level in each specification. Hence, the null hypothesis that there is no significant unobserved heterogeneity across countries is rejected, concluding that the random effects model is appropriate. Both of the Baltagi–Wu test statistic for zero first-order serial correlation (Baltagi, 2008; Baltagi & Wu, 1999, pp. 97–98) and the Bhargava, Franzini, and Narendranathan (1982) Durbin–Watson statistic reject the null hypothesis of no first-order serial correlation. (b) Panel Granger causality tests Corruption may also impede the introduction of information and communication technology (ICT) and internet adoption programs, diminishing their potential success. Therefore, reverse causality is likely (Lio et al., 2011). To empirically investigate the causal relationship between each of the regressors of e-government and internet adoption and corruption, the study employs panel Granger causality test which utilizes both of the cross-sectional and time-series data, and thus is more efficient than solely utilizing the time-series data (Dumitrescu & Hurlin, 2012; Granger, 1969; Justesen, 2008). The following autoregressive models are estimated. Corruptioni;t ¼ w0 þ

p p X X aj Corruptioni;tj þ bj X i;tj þ ui;t j¼1

j¼1

ð2Þ

118

X i;t ¼ U0 þ

WORLD DEVELOPMENT p X j¼1

kj X i;tj þ

p X dj Corruptioni;tj þ mi;t

ð3Þ

j¼1

where Xi,t denotes e-government in country i in year t in the first set of regressions that investigates the causal relationship between e-government and corruption, while it refers to internet adoption in country i in year t in the second set that examines the direction of causality between internet adoption and corruption; Xi,tj indicates the past values of e-government and internet adoption respectively (the initial conditions); and ui,t and mi,t are the error terms which are assumed to be distributed normally and independently with ui;t  IIDð0; r2u Þ and mi;t  IIDð0; r2m Þ. The system generalized method of moment (GMM) estimator, developed by Arellano and Bover (1995), that employs an instrumental variables technique is used to estimate Eqns. (2) and (3), since pooled OLS, the fixed effects, and the random effects estimators are biased and inconsistent due to the existence of the lagged dependent variable. The null hypothesis is that each of e-government and internet adoption does not Granger cause corruption and vice versa. Before testing for causality, Holtz-Eakin, Newey, and Rosen (1988) underline the importance of testing for the appropriate lag length, particularly in short panels. Otherwise, misleading results on causality can be obtained. They also suggest that the total time periods in the regression should be at least as large as twice the lag length. Using the Akaike information criterion (AIC) to choose the optimal lag length, and taking into consideration the limited number of time periods in the data set, I choose one lagged term for corruption and internet adoption in Eqns.(2) and (3), when testing for Granger causality between e-government and corruption. By the same technique, I include up to a two-lag length for corruption and a lag length of one for internet adoption in Eqns. (2) and (3), when testing for the causal impact between internet adoption and corruption. In this case, each equation will have two specifications to estimate. (c) Dynamic panel data model Corruption seems to have inertia (corruption is a function of its initial level), as suggested by past research (Andersen, 2009; Lio et al., 2011; Mauro, 2002). These lags of corruption may occur due to psychological reasons resulted from the force of habit. For instance, corrupt officials may not stop engaging in corrupt behavior immediately following improvements in egovernment or internet adoption, although the cost of corrupt acts becomes higher. The process of behavioral change may involve even higher cost (in terms of significant drop in the benefits side of corrupt acts) for those corrupt officials who become accustomed to the supernormal income from rentseeking opportunities. Another possible interpretation of corruption inertia is the lag in the introduction of new reforms in a country and their actual implementation by public officials. Including lagged corruption in the model has also the advantage of controlling for observed and unobserved historical factors that influence its current level, in addition to controlling for all time-invariant structural characteristics that influence current perceived corruption (Andersen, 2009; Wooldridge, 2011, pp. 371–374). Furthermore, the random effects model assumes exogeneity of all explanatory variables with the random country effects. However the disturbances contain unobservable, time-invariant, country effects that may be correlated with the regressors. The dynamic panel analysis allows for such endogeneity of all explanatory variables with the country effects by employing the instrumental variables technique (Baltagi, 2008, pp. 17–22; Mundlak, 1978). Therefore, to check

the validity of the random effects findings, addressing the endogeneity problem, and examining the persistency of corruption, the following dynamic panel data model is utilized. Corruptioni;t ¼ WCorruptioni;t1 þ X 0i;t b þ li þ mi;t

ð4Þ

where li  IIDð0; r2l Þ and mi  IIDð0; r2m Þ are both independent of each other and among themselves. The OLS estimator is biased and inconsistent because of the presence of lagged corruption among regressors, which is correlated with the error term. The fixed effects and the random effects estimators are also biased in this dynamic panel data model. 25 Arellano and Bond (1991) have suggested a generalized method of moment (GMM) procedure to estimate this model by utilizing the orthogonality conditions that exist between the lagged values of the dependent variable and the disturbances mi;t , to obtain additional instruments. 26 The GMM estimator uses the lagged values of the endogenous explanatory variables as instruments to address the endogeneity problem. Arellano and Bover (1995) and Blundell and Bond (1998) have developed a unifying GMM framework for looking at efficient IV estimators. The Arellano–Bover/Blundell–Bond two-step system GMM estimator is utilized to estimate Eqn. (4), using orthogonal-deviations transformation technique (which removes the fixed effects—the time-invariant country characteristics that are correlated with the regressors—and minimizes data loss) and applying the robust standard errors proposed by Windmeijer (2005). 27 The vector of explanatory variables in the first specification includes lagged corruption, GDP per capita, law enforcement, internet adoption, rural population, inflation, and press freedom. The second specification adds e-government to corruption controls, where the number of observations in this dynamic panel setting drops. The third specification examines the influence of each sub-component of e-government on corruption, while the fourth specification investigates the interaction effects between e-government and internet adoption on corruption reduction.

4. ESTIMATION RESULTS (a) Estimation results of the cross-section random effects models Table 1 shows the results of the cross-section random effects panel data models. Columns (1) to (5) reflect the estimation results of the five specifications discussed above, where the property-rights index is used to measure law enforcement. Heteroskedasticity-robust standard errors are used to deal with heteroskedasticity. To compare the results of the random effects models with those obtained from the fixed effects models used in a few prior work, Column FE (1) employs the fixed effects estimator without controlling for e-government, while Column FE (2) takes into consideration the impact of e-government. GDP per capita is statistically significant at the 1% level in all equations. Thus, poor countries are perceived to have higher corruption than rich ones. Each one-standarddeviation increase in GDP per capita (a rise of 12,739 per capita international dollars) lowers perceived corruption by 0.04–1.01 points in the corruption perception index, holding everything else constant. Likewise, law enforcement—measured by the propertyrights index—is statistically highly significant in all specifications, except in specification (1) where it is insignificant. Holding everything else equal, a one-standard-deviation

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION

119

Table 1. Estimation Results of the Cross-country Random Effects Panel Data Models (Law Enforcement: PR) FE FE (1) Intercept GDP per capita Law enforcement

1.645 (1.728) 0.172*** (0.037) 0.011 (0.01)

E-government

RE FE (2) *

4.419 (2.30) 0.035* (0.021) 0.004 (0.007) 1.690** (0.861)

(1) ***

8.965 (0.463) 0.079*** (0.018) 0.006 (0.007)

(2)

(3) ***

10.363 (0.369) 0.041*** (0.008) 0.030*** (0.004) 2.087*** (0.653)

(4) ***

10.272 (0.434) 0.035*** (0.007) 0.039*** (0.005) 1.216** (0.595)

Human capacity Online services Telecom infrastructure Internet users

***

9.406 (0.453) 0.033*** (0.008) 0.028*** (0.004)

0.461 (0.429) 0.616** (0.244) 2.936*** (0.731) 0.006* (0.004)

Inflation Press freedom

0.003 (0.005)

0.024*** (0.006)

0.013*** (0.003)

0.016*** (0.003)

0.180*** (0.037) 0.002 (0.002) 0.008 (0.008)

0.103** (0.048) 0.001 (0.007) 0.010 (0.010)

0.023*** (0.006) 0.004** (0.001) 0.025*** (0.007)

0.006 (0.004) 0.004 (0.006) 0.009** (0.004)

0.006 (0.005) 0.001 (0.007) 0.007* (0.004) 0.003 (0.013)

0.008* (0.004) 0.003 (0.006) 0.007** (0.003)

0.009*** (0.003) 0.023* (0.013) 0.008* (0.004) 0.004 (0.006) 0.008** (0.004)

0.415 17.08*** 2050

0.546 3.19*** 716

0.60 703.83*** 2050

0.827 968.31*** 716

0.853 1064.5*** 541

0.834 1256.2*** 716

0.831 1085.81*** 716

Tariff rate R2 (overall) Wald X2 Total panel obs.

10.23*** (0.377) 0.039*** (0.007) 0.029*** (0.004) 2.123*** (0.651)

0.001 (0.006)

Interaction effect Rural population

(5)

Notes: Robust standard errors are in parentheses. * Denotes statistical significance at the 10% level. ** Denotes statistical significance at the 5% level. *** Denotes statistical significance at the 1% level.

increase in the strength of property-rights (a 23.6 point increase in the property-rights index) reduces perceived corruption by 0.66–0.92 of a point. Strikingly, E-government reduces perceived corruption in all models. This result is statistically significant in all equations. Holding everything else constant, a one-standard-deviation increase in e-government (a 0.20 of a point increase in e-government index) decreases perceived corruption by 0.25–0.43 of a point. It seems that part of the correlation between GDP per capita and corruption operates via e-government (the estimated coefficient of GDP per capita falls as e-government is added to the regression), while e-government reinforces the impact of law enforcement on corruption (the estimated coefficient of law enforcement rises as we control for e-government). For the constituent parts of e-government, human capacity is statistically insignificant, while the scope and quality of online services and the telecommunication infrastructure have significant influence on corruption reduction. The scope and quality of online services is significant at the 5% level; each one-standard-deviation increase in the scope and quality of online services (a 0.24 point increase in online services index) reduces perceived corruption by 0.15 of a point, everything else equal. Similarly, the telecommunication infrastructure is statistically highly significant, where each one-standarddeviation rise in telecommunication infrastructure (a 0.22 point increase in telecommunication infrastructure index) results in a drop in corruption by 0.65 of a point, all else equal.

Thus, the telecommunication infrastructure and the scope and quality of online services components of e-government are deriving the results of the influence of e-government on corruption reduction. The coefficient of internet adoption is highly significant in all equations (except in specification (4); significant at the 10% level). Hence, an expansion of the internet adoption in a country lowers perceived corruption. A one-standard-deviation increase in the access to the worldwide network (increasing the number of internet users by 21 per 100 populations) reduces perceived corruption by 0.19–0.5 of a point. With regard to the interaction effects between e-government and internet adoption, the interaction term is statistically significant at the 10% level with a negative sign. In this equation, both e-government and internet adoption have significant impact on corruption reduction, and their interaction augments that favorable influence on reducing corruption. Hence, e-government and internet adoption have complementary interaction effects on curbing corruption. Rural population adversely impacts corruption in specifications (1) and (4) where its coefficient is statistically significant. The rural-population coefficient holds a positive sign in all equations, indicating that a higher percentage of population that is rural increases perceived corruption. Similarly, the coefficient of inflation variability retains a positive sign in all equations, but statistically significant only in specification (1), denoting that countries with higher inflation variability have higher perceived corruption.

120

WORLD DEVELOPMENT

Moreover, strengthening freedom of the press in a country seems to depress corruption. The coefficient of press freedom is significant in all equations with a negative sign. Conversely, the tariff rate does not have significant influence on perceived corruption. As a robustness check, I re-ran the above regressions using the rule-of-law index from the World Bank—rather than the property-rights index—to control for the perception about the strength of law enforcement in a country. As Table 2 shows, the impact of GDP per capita, law enforcement, and e-government on reducing perceived corruption is still statistically significant. Similar to the results in Table 1, human capacity is insignificant, while the scope and quality of online services and the telecommunication infrastructure are both highly significant with negative signs. The interaction term is significant at the 1% level and retains the same negative sign. On the other hand, the internet adoption significantly lowers perceived corruption only in specifications (1) to (3), while it is statistically insignificant in the other equations. Moreover, neither freedom of the press nor inflation variability significantly influences corruption in any equation. Additionally, rural population increases perceived corruption only in model (1), while tariff rate is still insignificant. (b) Results of the panel Granger causality tests The results of the panel Granger causality tests are shown in Table 3. The Arellano–Bover system GMM estimator is employed to estimate Eqns.(2) and (3). In each specification, the Sargan test for over-identification rejects the null

hypothesis (at 1% level) that over-identifying restrictions are valid. The Wald tests show that the coefficients of lagged e-government in Eqn. (2) are jointly statistically significant at the 1% significance level, while the coefficients of lagged corruption in Eqn. (3) are statistically insignificant even at the 10% level. These results suggest that the direction of causality is from e-government to corruption and there is no reverse causality from corruption to e-government. The estimated coefficient of lagged e-government is highly significant with a negative sign, denoting that higher level of e-government in the past reduces current corruption. In contrast, Wald tests indicate that the coefficients of lagged internet adoption in Eqn. (2) and the coefficients of lagged corruption in Eqn. (3) are jointly statistically different from zero at the 1% level in each specification. Therefore, causality between internet adoption and corruption runs in both directions. The estimated coefficients of lagged internet adoption and lagged corruption are statistically highly significant with negative sign. Hence, greater internet adoption in the past lowers current corruption. Likewise, past corruption impedes current internet adoption. Thus, the problem of reverse causality should be addressed in the empirical work on the impact of internet adoption on corruption. (c) Estimation results of the dynamic panel data model Table 4 shows the results of the dynamic panel data model. Columns (1) to (4) represent the estimation results of the four

Table 2. Estimation Results of the Cross-country Random Effects Panel Data Models (Law Enforcement: RL) FE

Intercept GDP per capita Law enforcement

RE

FE (1)

FE (2)

(1)

(2)

(3)

(4)

(5)

0.228 (1.645) 0.149*** (0.035) 0.201 (0.285)

2.194 (2.573) 0.024 (0.022) 0.525 (0.340) 2.856*** (0.919)

8.14*** (0.425) 0.037*** (0.012) 0.767*** (0.168)

8.661*** (0.473) 0.029*** (0.007) 1.083*** (0.159) 2.772*** (0.735)

8.239*** (0.625) 0.023*** (0.006) 1.301*** (0.163) 2.008** (0.966)

7.833*** (0.571) 0.021*** (0.007) 1.047*** (0.165)

8.479*** (0.464) 0.025*** (0.007) 1.08*** (0.154) 2.785*** (0.707)

E-government Human capacity Online services Telecom infrastructure Internet users

0.215 (0.464) 0.847*** (0.285) 2.939*** (0.987) 0.003 (0.005)

0.006 (0.006)

0.004 (0.005)

0.03*** (0.006)

0.008** (0.004)

0.009*** (0.003)

0.209*** (0.032) 0.001 (0.003) 0.003 (0.009)

0.148** (0.060) 0.006 (0.005) 0.003 (0.013)

0.016*** (0.005) 0.002 (0.002) 0.007 (0.005)

0.004 (0.004) 0.004 (0.004) 0.0004 (0.004)

0.003 (0.004) 0.006 (0.006) 0.001 (0.004) 0.004 (0.014)

0.006 (0.004) 0.005 (0.004) 0.001 (0.004)

0.0002 (0.004) 0.044*** (0.014) 0.006 (0.004) 0.004 (0.004) 0.001 (0.004)

0.443 18.04*** 1668

0.539 4.42*** 765

0.68 1454.96*** 1668

0.813 1157.6*** 765

0.848 1281.88*** 562

0.821 1618.99*** 765

0.825 1561.92*** 765

Interaction effect Rural population Inflation Press freedom Tariff rate R2 (overall) Wald X2 Total panel obs.

Notes: Robust standard errors are in parentheses. ** Denotes statistical significance at the 5% level. *** Denotes statistical significance at the 1% level.

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION

121

Table 3. Results of Panel Granger Causality Tests Direction of causality

Lags

Wald tests

Obs.

Countries

Instruments

E-government ? Corruption Corruption ? E-government Internet adoption ? Corruption

[1/1] [1/1] [1/1] [2/1] [1/1] [1/2]

X2(1) = 42.70*** X2(1) = 1.17 X2(1) = 147.85*** X2(1) = 145.22*** X2(1) = 595.69*** X2(1) = 742.09***

654 326 2205 2085 2191 2071

164 163 167 167 167 167

26 5 107 106 107 107

Corruption ? Internet adoption ***

Denotes statistical significance at the 1% level.

Table 4. Estimation Results of the Dynamic Panel Data Models (Two-step System GMM, Orthogonal Deviation) Law enforcement (PR) (1) Intercept Lagged corruption (1) GDP per capita Law enforcement

***

2.364 (0.284) 0.763*** (0.026) 0.014*** (0.003) 0.005** (0.002)

E-government

(2) ***

5.421 (0.679) 0.479*** (0.062) 0.025*** (0.006) 0.021*** (0.004) 0.882** (0.372)

Human capacity

***

4.87 (0.611) 0.484*** (0.060) 0.016*** (0.006) 0.019*** (0.004)

0.004** (0.002)

0.004 (0.003)

0.235 (0.249) 0.244 (0.172) 2.024*** (0.488) 0.003 (0.003)

0.003* (0.002) 0.0002 (0.001) 0.006*** (0.002) 2075.27***

0.002 (0.003) 0.006 (0.005) 0.003 (0.002) 371.17***

0.051 111 0.022 156 12.65 1974

0.304 52 0.165 154 4.65 716

Online services Telecom infrastructure Internet users

(3)

Inflation Press freedom F-Test Hansen test of overid. (p-value) Number of instruments Diff.-in-Hansen test GMM (p-value) Number of countries Average obs. per country Total panel obs.

***

4.686 (0.621) 0.511*** (0.054) 0.021*** (0.006) 0.015*** (0.004) 0.923*** (0.350)

(5) ***

1.596 (0.255) 0.751*** (0.03) 0.002 (0.003) 0.441*** (0.07)

(6) ***

3.894 (0.579) 0.535*** (0.063) 0.019*** (0.004) 0.581*** (0.109) 1.108*** (0.353)

(7) ***

3.396 (0.483) 0.541*** (0.059) 0.013*** (0.003) 0.549*** (0.119)

(8) 3.342*** (0.538) 0.55*** (0.058) 0.011* (0.006) 0.599*** (0.094) 0.903*** (0.323)

0.003** (0.002)

0.001 (0.003)

0.004 (0.005) 0.004 (0.005) 0.002 (0.002) 363.73***

0.009 (0.005) 0.053** (0.023) 0.006** (0.003) 0.01* (0.006) 0.003 (0.002) 448.89***

0.240 (0.227) 0.291* (0.148) 1.826*** (0.488) 0.007*** (0.003)

0.003 (0.002) 0.001 (0.002) 0.001 (0.002) 2009.82***

0.0004 (0.002) 0.002 (0.003) 0.0002 (0.002) 486.88***

0.001 (0.002) 0.001 (0.004) 0.0004 (0.002) 554.25***

0.013*** (0.004) 0.054** (0.021) 0.004** (0.002) 0.001 (0.003) 0.001 (0.002) 630.45***

0.222 54 0.139 154 4.65 716

0.576 52 0.549 154 4.65 716

0.002 96 0.001 159 10.48 1667

0.271 52 0.253 158 4.84 764

0.286 54 0.240 158 4.84 764

0.535 52 0.603 158 4.84 764

Interaction effect Rural population

Law enforcement (RL) (4)

Notes: Windmeijer-corrected cluster-robust errors are in parentheses. The null hypothesis in the Hansen test of over-identified restrictions is that all instruments are valid. The null hypothesis in the difference-in-Hansen tests of exogeneity of GMM instruments is that instruments are exogenous. * Denotes statistical significance at the 10% level. ** Denotes statistical significance at the 5% level. *** Denotes statistical significance at the 1% level.

specifications of Eqn. (4) discussed above, when the property-rights index is used to reflect the strength of law enforcement, while Columns (5) to (8) illustrate the estimation results of these four models when law enforcement is measured by the rule-of-law index, to check the robustness of the results. The effect of the past level of corruption is statistically significant at the 1% level with a positive sign in all models. Therefore, corruption seems to have inertia, and that part of present corruption attributes to its initial conditions.

Remarkably, e-government lowers perceived corruption in all models, similar to the outcome from the random effects analysis. This result is statistically significant at the 1% level in each equation, but significant at the 5% level in specification (2). A one-standard-deviation increase in e-government (a rise of 0.20 of a point in e-government index) reduces perceived corruption by 0.18–0.22 of a point, all else equal. Like the results in Tables 1 and 2, the telecommunication infrastructure

122

WORLD DEVELOPMENT

component of e-government is statistically highly significant in both specifications, while human capacity is insignificant in each specification. The scope and quality of online services reduces corruption in specification (7), whereas it is insignificant in specification (3). The influence of the internet adoption on corruption is ambiguous, similar to the results in Table 2. The sign of the coefficient of internet adoption changes from negative to positive in specifications (7) and (8). Yet, the interaction term is still statistically significant (at the 5% level) in each specification and retains its negative sign, implying that the internet adoption strengthens the favorable impact of egovernment on curbing corruption, even if the internet adoption by itself may be statistically insignificant or its coefficient may hold a positive sign. This still supports the complementary effects of e-government and internet adoption on combating corruption in Tables 1 and 2. The impact of law enforcement—as measured by either the rule-of-law index or the property-rights index—on perceived corruption is statistically highly significant in all regressions with a negative sign. Each one-standard-deviation increase in the strength of law enforcement measured by the property-rights index (a 23.6 point increase in the property-rights index) decreases corruption by 0.12–0.50 of a point, while corruption is reduced by 0.45–0.61 of a point for each one-standard-deviation increase in the strength of law enforcement measured by the rule-of-law index (a 1.01 point increase in the rule-of-law index), holding everything else constant. GDP per capita reduces corruption in all specifications, except in specification (5) that does not control for e-government when the rule-of-law index is used as a measure of law enforcement. As one accounts for the impact of e-government in the regression, the estimated coefficient of GDP per capita increases and turns out to be statistically significant. Similar to the results from the random effects analysis, the coefficient of rural population is significant only in regressions (1), (4), and (8), and retains its positive sign in all regressions. A higher percentage of population that is rural induces more perceived corruption. On the contrary, inflation variability does not influence corruption in any equation. Likewise, press freedom does not seem to lower perceived corruption in any equation, except in specification (1) where it is statistically highly significant. 5. DISCUSSION AND CONCLUDING REMARKS Various anti-corruption organizations have recently encouraged the implementation of e-government in anti-corruption programs. However, those organizations—along with some researchers in the social sciences—have lately stressed the need for macro empirical evidence on the impact of e-government on corruption, since the majority of current research in this regard is theoretical and the existent empirical evidence is based on few case studies on the micro-level. There has also been little research on the influence of internet adoption on corruption, whereas a small number of studies argue that the internet use is more important tool than e-government in corruption reduction. This study contributes to the existing literature by empirically investigating the impact of e-government and the level of internet adoption on curbing corruption by utilizing a large panel data set. The study uses a unique measure of e-government from the United Nations that covers multiple dimensions of e-government. This work also examines the impact of each dimension of e-government independently on curbing corruption and investigates the interaction effects between e-government and internet adoption on combating

corruption. Furthermore, the study explores the direction of causality between e-government and corruption, in addition to the causality between internet adoption and corruption, by employing panel Granger causality tests. A dynamic panel data model that addresses the endogeneity problem and takes into consideration the persistency of corruption is utilized to estimate the effects of e-government and internet adoption on corruption, in addition to the random effects estimators. This paper provides new evidence on the causal relationship between e-government and corruption, which conclusively runs from e-government to corruption, but not the other way round. Additionally, the results of panel Granger causality tests indicate that causality between internet adoption and corruption is bi-directional. The empirical results reveal that e-government is a powerful tool in curbing corruption. Part of the correlation between GDP per capita and corruption seems to operate via e-government. In addition, e-government reinforces the influence of law enforcement on corruption reduction. These findings are quite robust to different model specifications and various measures of law enforcement. Another new finding is that the driving force of the favorable impact of e-government on curbing corruption is the telecommunication infrastructure, which is a quite robust result, in addition to the scope and quality of online services. The human capacity component of e-government does not seem to impact corruption under any specification, which is also a robust result. Therefore, e-government reduces corruption by expanding the access to information and raising the level of corruption awareness, which increases transparency and improves accountability. On the contrary, unlike the findings from previous research, the effect of the level of internet adoption on corruption is ambiguous and seems to be sensitive to model specification. Yet, the level of internet adoption reinforces the impact of egovernment on curbing corruption. This study provides robust evidence for interaction effects between e-government and internet adoption on corruption reduction, suggesting that egovernment and internet adoption should be considered as complements, rather than substitutes, in anti-corruption programs. Another finding is that inflation variability and trade protectionism do not seem to impact corruption, unlike the findings of prior research. Furthermore, corruption seems to be persistent and is a function of its initial level. In sum, the impact of e-government on curbing corruption is clearly positive, while mere internet adoption has an ambiguous effect. This work reveals that e-government is a useful instrument in anti-corruption efforts, which needs to be recognized by policymakers. In some countries, however, e-government services may be offered online but the official process to obtain these services still requires citizens to physically meet with public officials where the administrative procedures have been undergone little or no change, and hence the benefits of e-government will be minimal (UNDP, 2006). Therefore, the legal framework and laws have to be reviewed to be consistent with e-government to make it more effective. These laws have to ensure transparency, free access to information, and the ability to track actions and decisions back to the individual public officers. In this case, e-government would create a larger impact on reducing corruption. Furthermore, similar to other policy measures introduced by the government, e-government would be more effective in curbing corruption with a greater willingness and commitment of the government to truly address and combat corruption dilemma, which can be considered in future research.

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION

Yet, developing countries may need adequate infrastructure to support e-government applications. They may also need to allocate more resources to train officers and citizens on e-government services, to increase people’s participation in e-government. As noted by UNDP (2006), e-government can be initially implemented in corrupt departments to specifically target lower corruption. These issues require further research. Moreover, future research can inves-

123

tigate the roles that alternative forms of e-government play in combating corruption and the extent to which democracies are likely to create these forms of e-government, in addition to exploring the determinants of e-government adoption across countries. Future research can also examine the impact of e-government on various forms of corruption independently, when disaggregated consistent data on corruption become available.

NOTES 1. Transparency International (2005) has declared that the ‘‘Millennium Development Goals are unreachable without commitment to fighting corruption.” For more details, see http://archive.transparency.org/news_room/latest_news/press_releases/2005/14_09_2005_mdg. 2. The Pacific Council on International Policy (2002, p. 10) has alerted that, although e-government may offer a weapon against corruption, it may also allow corruption to continue when officials leading technologyempowered processes find new rent-seeking opportunities. In this case, e-government will simply lead to an inter-generational shift in corruption toward the younger officials (the more tech-literate officials). 3. For more examples, see Andersen (2009) and United Nations Development Program (2006). 4. West (2005, p. 1) defines e-government as ‘‘the public sector use of the internet and other digital devices to deliver information and democracy itself.” It can also be defined as ‘‘the process of connecting citizens digitally to their government in order that they might access information and services offered by government agencies” (Lau, Aboulhoson, Lin, & Atkin, 2008, p. 89). 5. As Klitgaard (1988, p. 75) put it, corruption = monopoly + discretion  accountability. 6. Yet, since corruption can take many various forms, e-government is likely to have different impact on various forms of corruption. The focus of this paper would be on the aggregate level of perceived corruption. 7. For additional literature surveys on how e-government may promote anti-corruption efforts, see Bertot et al. (2010, pp. 265–266). 8. For more examples of individual case studies, see Wescott (2003), World Bank (2004), Chawla and Bhatnagar (2004), UNDP (2006); and Andersen (2009). 9. Andersen (2009) has used the control of corruption index, from the World Bank, to measure corruption. To measure e-government, he utilized an e-government index that measures only one subset of e-government; the internet-based e-government. This index ranges from zero to 100—where higher score denotes greater e-government—and was compiled by a research team during June and July 2006, supervised by Darrell M. West of Brown University. In his analysis, for e-government variable, Anderson used only 2006 time observation, since the index he used is available only for this year.

my data set to be time consistent with other variables. Similarly, egovernment scores for 2009 in my data set are from the 2010 e-government index, since it was issued in December 2009. 13. Accordingly, the EGDI = (0.34  online service index) + (0.33  telecommunication infrastructure index) + (0.33  human capacity index). United Nations (2010, p. 109). 14. For full description of each stage, the corresponding survey, and the construction of the index value for a given country, see United Nations (2010, pp. 95, 110–112). 15. Each of these indicators is normalized by subtracting the lowest value for any country in the survey from the indicator’s value for a given country and dividing by the difference between the highest and lowest values of any country in the survey. For a given country, the telecommunication infrastructure index value is then the simple arithmetic mean of each of the five normalized indicators (United Nations, 2010, p. 113). 16. Both indicators are normalized by the same method of normalizing the telecommunication infrastructure sub-indicators. The human capacity index value for a particular country is then the weighted arithmetic mean of these two normalized indicators, where weights assigned for adult literacy and gross enrollment ratio are 0.6667 and 0.3333, respectively. For more details, see United Nations (2010, pp. 109–113) and United Nations (2008, pp. 12–18). 17. The weighted mean applied tariff rate is the average of effectively applied tariff rate, weighted by the product import shares corresponding to each partner country. For more details, see http://search.worldbank.org/ data?qterm=tariff%20rate. 18. Heritage Foundation (2010, p. 465). 19. The index of economic freedom is composed of 10 equally-weighted components of economic freedoms. One of these components is the property-rights (a composite of quantifiable measures of different sources), which is used in this study. Another component is the freedom from corruption, where scores for this freedom are derived primarily from corruption perception index by Transparency International. The property-rights index and the freedom from corruption are two different indicators and are not supposed to measure the same thing. For description of the methodology and sources of each component of the economic freedom, see Heritage Foundation (2010, pp. 457–468).

10. For more details, see Transparency International (2009).

20. Available online at: http://info.worldbank.org/governance/wgi/index.asp.

11. United Nations (2010, p. 109).

21. For more details, see Kaufmann et al. (2010).

12. The UN e-government index has been issued in 2003, 2004, 2005, 2008, and 2010. The 2008 e-government index was issued in January 2008. Hence, it is more appropriate to make it the 2007 e-government index in

22. It measures the percentage change in the consumer price index, expressed in end of the period (not annual average data). See http:// elibrary-data.imf.org/.

124

WORLD DEVELOPMENT

23. For the methodology used to construct this index, see Freedom House (2013, pp. 35–38). 24. I thank the anonymous referees for suggesting this implementation. 25. For more details, see Baltagi (2008, pp. 147–148).

26. For more discussion on the GMM estimator, see Arellano (2003). 27. The system GMM estimator is designed for panels with few time periods and large individuals, and with independent variables that are not strictly exogenous. See Roodman (2006).

REFERENCES Andersen, T. B. (2009). E-government as an anti-corruption strategy. Information Economics and Policy, 21(3), 201–210. Arellano, M. (2003). Panel data econometrics. Oxford University Press. Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies, 58(2), 277–297. Arellano, M., & Bover, O. (1995). Another look at the instrumental variable estimation of error-components models. Journal of Econometrics, 68(1), 29–51. Baltagi, B. H. (2008). Econometric analysis of panel data. Wiley. Baltagi, B. H., & Wu, P. X. (1999). Unequally spaced panel data regressions with AR (1) disturbances. Econometric Theory, 15(6), 814–823. Bertot, J. C., Jaeger, P. T., & Grimes, J. M. (2010). Using ICTs to create a culture of transparency: E-government and social media as openness and anti-corruption tools for societies. Government Information Quarterly, 27(3), 264–271. Bhargava, A., Franzini, L., & Narendranathan, W. (1982). Serial correlation and the fixed effects model. The Review of Economic Studies, 49(4), 533–549. Bhatnagar, S. (2003). E-government and access to information (pp. 24–32). United Nations, Global Corruption Report 2003. Available from http://unpan1.un.org/intradoc/groups/public/documents/apcity/ unpan008435.pdf. (Retrieved October 1, 2013). Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), 115–143. Braun, M., & Di Tella, R. (2004). Inflation, inflation variability, and corruption. Economics & Politics, 16(1), 77–100. Chawla, R., & Bhatnagar, S. (2004). Online delivery of land titles to rural farmers in Karnataka, India. The international bank for reconstruction and development, the World Bank. Available from http://info.worldbank.org/etools/docs/reducingpoverty/case/96/fullcase/India%20Bhoomi%20Full%20Case.pdf. (Retrieved October 1, 2013). Dumitrescu, E. I., & Hurlin, C. (2012). Testing for Granger non-causality in heterogeneous panels. Economic Modelling, 29(4), 1450–1460. Dutt, P. (2009). Trade protection and bureaucratic corruption: An empirical investigation. Canadian Journal of Economics/Revue canadienne d’e´conomique, 42(1), 155–183. Elbahnasawy, N. G., & Revier, C. F. (2012). The determinants of corruption: Cross-country-panel-data analysis. The Developing Economies, 50(4), 311–333. Freedom House (2013). Freedom of the press 2013. Available from http:// www.freedomhouse.org/sites/default/files/FOTP%202013%20Booklet% 20Final%20Complete%20-%20Web.pdf. (Retrieved August 15, 2013). Freille, S., Haque, M. E., & Kneller, R. (2007). A contribution to the empirics of press freedom and corruption. European Journal of Political Economy, 23(4), 838–862. Goel, R. K., Nelson, M. A., & Naretta, M. A. (2012). The internet as an indicator of corruption awareness. European Journal of Political Economy, 28(1), 64–75. Granger, C. W. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica: Journal of the Econometric Society, 37(3), 424–438. Gupta, S., Davoodi, H., & Alonso-Terme, R. (2002). Does corruption affect income inequality and poverty?. Economics of Governance, 3(1), 23–45. Heeks, R. (1998). Information technology and public sector corruption. Information systems for public sector management working paper series no. 4. Manchester, UK: Institute for Development Policy and Management. Available from http://unpan1.un.org/intradoc/groups/ public/documents/NISPAcee/UNPAN015477.pdf. (Retrieved October 1, 2013). Heritage Foundation (2010). 2010 index of economic freedom. Available from http://www.heritage.org/index/download. (Retrieved November 20, 2012).

Holtz-Eakin, D., Newey, W., & Rosen, H. S. (1988). Estimating vector autoregressions with panel data. Econometrica: Journal of the Econometric Society, 56(6), 1371–1395. International Monetary Fund (2012). World economic outlook databases. Available from http://www.imf.org/external/ns/cs.aspx?id=28. (Retrieved December 15, 2012). Jain, A. K. (2001). Corruption: A review. Journal of Economic Surveys, 15(1), 71–121. Judge, G. G., Hill, R. C., Griffiths, W. E., Lutkepohl, H., & Lee, T. C. (1985). Introduction to the theory and practice of econometrics (2nd ed.). New York: John Wiley & Sons. Justesen, M. K. (2008). The effect of economic freedom on growth revisited: New evidence on causality from a panel of countries 1970–1999. European Journal of Political Economy, 24(3), 642–660. Kaufmann, D., Kraay, A., & Mastruzzi, M. (2010). The worldwide governance indicators: Methodology and analytical issues. World Bank policy research working paper (no. 5430). Kim, S., Kim, H. J., & Lee, H. (2009). An institutional analysis of an egovernment system for anti-corruption: The case of OPEN. Government Information Quarterly, 26(1), 42–50. Klitgaard, R. (1988). Controlling corruption. University of California Press. Kolstad, I., & Wiig, A. (2009). Is transparency the key to reducing corruption in resource-rich countries?. World Development, 37(3), 521–532. Lau, T. Y., Aboulhoson, M., Lin, C., & Atkin, D. J. (2008). Adoption of e-government in three Latin American countries: Argentina, Brazil and Mexico. Telecommunications Policy, 32(2), 88–100. Lio, M. C., Liu, M. C., & Ou, Y. P. (2011). Can the internet reduce corruption? a cross-country study based on dynamic panel data models. Government Information Quarterly, 28(1), 47–53. Mauro, P. (1995). Corruption and growth. The Quarterly Journal of Economics, 110(3), 681–712. Mauro, P. (2002). The persistence of corruption and slow economic growth. IMF working paper WP/02/213. Available from http://www. imf.org/external/pubs/ft/wp/2002/wp02213.pdf. (Retrieved February 20, 2012). Mishra, A. (2006). Persistence of corruption: Some theoretical perspectives. World Development, 34(2), 349–358. Mundlak, Y. (1978). On the pooling of time series and cross section data. Econometrica: Journal of the Econometric Society, 46(1), 69–85. Murphy, K. M., Shleifer, A., & Vishny, R. W. (1991). The allocation of talent: Implications for growth. The Quarterly Journal of Economics, 106(2), 503–530. OECD (2005). Good practice paper on ICTs for economic growth and poverty reduction. The DAC Journal, 6(3), 27–95. Ojha, A., Palvia, S., & Gupta, M. P. (2008). A model for impact of e-government on corruption: Exploring theoretical foundations. Critical thinking in e-governance. In: The 6th international conference on e-government (ICEG 2008) (pp. 160–170). New Delhi, India. Available from http://ww.csi-sigegov.org/critical_pdf/18_160-170.pdf. (Retrieved October 1, 2013). Pacific Council on International Policy (2002). Roadmap for e-government in the Developing World: 10 questions e-government leaders should ask themselves. The working group on e-government in the Developing World. Available from http://www.itu.int/wsis/docs/ background/themes/egov/pacific_council.pdf. (Retrieved October 1, 2013). Piatkowski, M. (2006). Can information and communication technologies make a difference in the development of transition economies?. Information Technologies & International Development, 3(1), 39–53. Roodman, D. (2006). How to do xtabond2: An introduction to difference and system GMM in Stata. Center for Global Development working paper (p. 103).

E-GOVERNMENT, INTERNET ADOPTION, AND CORRUPTION: AN EMPIRICAL INVESTIGATION Seldadyo, H., & de Haan, J. (2006). The determinants of corruption: A literature survey and new evidence. In: European public choice society conference (pp. 20–23). Turku, Finland. Serra, D. (2006). Empirical determinants of corruption: A sensitivity analysis. Public Choice, 126(1–2), 225–256. Tirole, J. (1996). A theory of collective reputations (with applications to the persistence of corruption and to firm quality). The Review of Economic Studies, 63(1), 1–22. Transparency International (2005). Available from http://archive.transparency.org/news_room/latest_news/press_releases/2005. (Retrieved August 10, 2012). Transparency International (2009). Corruption perception index 2009, methodological brief. Available from http://archive.transparency. org/policy_research/surveys_indices/cpi/2009/methodology. (Retrieved September 20, 2012). Treisman, D. (2000). The causes of corruption: A cross-national study. Journal of Public Economics, 76(3), 399–457. Treisman, D. (2007). What have we learned about the causes of corruption from ten years of cross-national empirical research?. Annual Review of Political Science, 10, 211–244. United Nations Development Program (2006). Fighting corruption with egovernment applications. APDIP e-note 8. Available from http:// www.unapcict.org/ecohub/resources/apdip-e-note-8-fighting-corruptionwith-e. (Retrieved October 1, 2013). United Nations Development Program (2008). Tackling corruption, transforming lives: Accelerating human development in Asia and the Pacific. New Delhi, India: Macmillan Publishers. United Nations (2008). UN e-government survey 2008: From e-government to connected governance. New York: Department of Economic and Social Affairs. United Nations (2010). United Nations e-government survey 2010: Leveraging e-government at a time of financial and economic crisis. New York: Department of Economic and Social Affairs. Available

125

from http://www.unpan.org/egovkb/global_reports/08report.htm. (Retrieved July 25, 2012). Wescott, C. (2003). E-government to combat corruption in the Asia pacific region. In: The 11th international anti-corruption conference. Seoul, Republic of Korea. Available from http://unpan1.un.org/ intradoc/groups/public/documents/apcity/unpan035450.pdf. (Retrieved October 1, 2013). West, D. M. (2005). Digital government: Technology and public sector performance. Princeton University Press. Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step GMM estimators. Journal of Econometrics, 126(1), 25–51. Wooldridge, J. M. (2011). Econometric analysis of cross section and panel data (2nd ed.). MIT Press. World Bank (2001). Available from http://www.worldbank.org/html/ extdr/thematic.htm. (Retrieved May 15, 2012). World Bank (2004). Making services work for the poor. World development report 2004. World Bank (2012). World development indicators. Available from http://www.data.worldbank.org/indicator/all. (Retrieved December 1, 2012). World Bank (2012). Worldwide governance indicators. Available from http://info.worldbank.org/governance/wgi/index.asp. (Retrieved June 30, 2012).

APPENDIX A See Tables 5–7.

Table 5. Summary Statistics of the Variables Included in the Study Variable

Mean

Std. dev.

Min.

Max.

Obs.

Corruption perception index GDP per capita (thousands) Property-rights index Rule-of-law E-government Human capacity Online services Telecom infrastructure Internet users Rural population Inflation Press freedom Tariff rate

8.016 11.043 50.363 0.113 0.428 0.765 0.331 0.201 14.03 45.461 10.13 52.007 8.763

2.801 12.739 23.605 1.012 0.202 0.209 0.243 0.22 20.989 23.251 29.084 24.323 6.934

1 0.151 5 2.691 0 0 0 0 0 0 11.864 0 0

11 84.043 95 1.964 0.927 0.999 1 0.86 92.181 92.8 549.206 95 100.57

2510 2388 2193 1840 820 820 820 820 2376 2505 2395 2454 1501

Table 6. The Correlation Matrix for the Variables of the Models (1) (1) Corruption perception index (2) GDP per capita (3) Property-rights index (4) Rule-of-law (5) E-government (6) Human capacity (7) Online services (8) Telecom infrastructure (9) Internet users (10) Rural population (11) Inflation (12) Press freedom (13) Tariff rate

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

1 0.682 1 0.665 0.690 1 0.768 0.771 0.876 1 0.810 0.738 0.740 0.807 1 0.551 0.537 0.484 0.575 0.803 1 0.711 0.597 0.653 0.700 0.900 0.578 1 0.866 0.847 0.812 0.863 0.892 0.624 0.755 1 0.697 0.701 0.543 0.710 0.854 0.609 0.717 0.951 1 0.512 0.671 0.482 0.562 0.684 0.601 0.591 0.651 0.486 1 0.161 0.145 0.133 0.204 0.280 0.194 0.232 0.316 0.141 0.079 1 0.555 0.479 0.687 0.733 0.626 0.434 0.568 0.654 0.480 0.353 0.164 1 0.509 0.528 0.419 0.537 0.683 0.619 0.552 0.653 0.519 0.419 0.045 0.459

(13)

1

126

WORLD DEVELOPMENT Table 7. List of Countries Included in the Analysis

Albania Algeria Angola Argentina Armenia Australia Austria Azerbaijan Bahrain Bangladesh Belarus Belgium Belize Benin Bhutan Bolivia Bosnia and Herzegovina Botswana Brazil Bulgaria Burkina Faso Burundi Cambodia Cameroon Canada Central African Republic Chad

Chile China Colombia Congo, Democratic Republic Congo, Republic Costa Rica Coˆte d´Ivoire Croatia Cyprus Czech Republic Denmark Dominica Dominican Republic Ecuador Egypt El Salvador Equatorial Guinea Eritrea Estonia Ethiopia Fiji Finland France Gabon Gambia Georgia Germany

Ghana Greece Grenada Guatemala Guinea Guyana Haiti Honduras Hong Kong Hungary Iceland India Indonesia Iran Iraq Ireland Israel Italy Jamaica Japan Jordan Kazakhstan Kenya Kuwait Kyrgyzstan Laos Latvia

Lebanon Lesotho Liberia Libya Lithuania Luxembourg Macedonia Madagascar Malawi Malaysia Mali Malta Mauritania Mauritius Mexico Moldova Mongolia Morocco Mozambique Myanmar Namibia Nepal Netherlands New Zealand Nicaragua Niger Nigeria

Note: Some countries have missing observations for some variables in some years.

Available online at www.sciencedirect.com

ScienceDirect

Norway Oman Pakistan Panama Papua New Guinea Paraguay Peru Philippines Poland Portugal Qatar Romania Russia Rwanda Saudi Arabia Senegal Serbia Seychelles Sierra Leone Singapore Slovakia Slovenia South Africa South Korea Spain Sri Lanka Sudan

Suriname Swaziland Sweden Switzerland Syria Tajikistan Tanzania Thailand Timor-Leste Togo Trinidad and Tobago Tunisia Turkey Turkmenistan Uganda Ukraine United Arab Emirates United Kingdom United States Uruguay Uzbekistan Venezuela Vietnam Yemen Zambia