What Location-Based Analysis of Governance Tells Us About the Demand Side of Corruption: Evidence from Bangalore, India. By Srikar Gullapalli 1

Transcription

1 What Location-Based Analysis of Governance Tells Us About the Demand Side of Corruption: Evidence from Bangalore, India By Srikar Gullapalli 1 Colgate University WORKING PAPER Do Not Cite or Quote Without Author s Permission Abstract Past literature on corruption focuses on supply-side causes for corruption incidence and its macrolevel consequences. This paper analyzes the demand-side of cross-sectional data on the quality of and corruption in government services collected using location-based citizen surveys in Bangalore, India. Data were collected in an exit-poll basis at the primary point of service provision for 8 government agencies. This reduces systematic biases and enables demand-side causal analyses. Summary statistics obtained delineate micro-level corruption trends especially important due to decentralization. I analyze a supply-side model using perception ratings as a proxy for institutional framework and confirm results of past literature, hence showing that the perception ratings are precise. I then find from a unique demand-side analysis of the effect of corruption on the quality of service that corruption is overpriced in the status quo hence calling into question the legitimacy of the greases-the-wheels hypothesis of corruption on the individual level. I also find increased speed of service, helpfulness and accessibility of officials each increases likelihood of citizen satisfaction. The paper concludes with incentive structures towards reform in this methodology. These results have external validity for urban governance contexts in India, while the model is exportable to other developing nations. JEL Classification: D73 C81 C83 C21 Keywords: Corruption, governance, location-based, Bangalore, market-price of corruption, crosssectional study, incentive structures 1 I am grateful to Prof. Michael Johnston for his continued support and mentorship throughout the duration of this research. I am also thankful to the British Council, the World Bank Institute, and the Lampert Endowment of the PPE program at Colgate University for generously sponsoring this research. Public Affairs Centre, Paraspara Trust, and Center for Public Policy at IIMB have all helped me collect data and validate my research. The views expressed here are my own and not necessarily those of the British Council, WBI, or Colgate University. All errors in this paper are my own. address for correspondence: sgullapalli@colgate.edu 1

2 List of Abbreviations: ICS: Indian Corruption Study CRC: Citizen Report Card BDA: Bangalore Development Authority BWSSB: Bangalore Water Supply and Sewerage Board BESCOM: Bangalore Electricity Supply Company Limited BBMP: Bruhat Bengaluru Mahanagara Palike PDS: Public Distribution System STROBE: Strengthening the Reporting of Observational Studies in Epidemiology RTO: Regional Transport Office OLS: Ordinary Least Squares RCT: Randomized Controlled Trial Introduction The primary point of interface between the government and the people is through the various public service agencies that dispense government services. This is the avenue for almost all petty corruption in government service provision. The recent national anticorruption protests in India were the result of an outpouring of anger against the grinding daily routine of petty corruption (Burke, 2011) in addition to the string of multi-million dollar corruption scams that were exposed. Petty corruption has been a mainstay of the national consciousness in India from the era of the license raj, with multiple Bollywood and South Indian movies, editorials, campaigns (political and non-political), and academic studies having focused on it. 2

3 The India Corruption Study (2010) conducted in villages in 12 states estimates that 8.57 million dollars was paid in bribes by rural households to just four government agencies over a period of one year. Clearly as Agarwal (2008) says while due attention continues to be devoted to macro issues, we cannot afford to any longer delay fully restoring efficient and honest functioning in the districts, where our people live and where all the vital development programs concerning their welfare are executed. The corruption in India is fragmented and often unpredictable due to significant decentralization in government processes (Bardhan, 1997) and hence micro-level data on the quality of government service provision is required to understand the many nuances of petty corruption, formulate effective reforms, and incentivize individuals (within and without the government) to push for these reforms. My work attempts to meet this burden with a unique approach of location-based data collection on the quality of government services and corruption in Bangalore, India, in an exit-poll fashion outside local branch offices of the government. There have been multiple past initiatives that have attempted to measure corruption in government services. The various iterations of the ICS have collected objective and perceptionbased data on corruption in a few government services in sample states across India. Since the ICS focuses primarily on corruption and doesn t measure the different aspects of the quality of citizens interaction with the government, it doesn t provide a holistic picture of petty corruption. It was also conducted on the macro-level and hence presents an evaluation tool for the state governments but not the individual offices which actually dispense services on the ground (ICS, 2011). The Citizen Report Cards (CRCs) initiated by Public Affairs Centre in Bangalore during the 90s evaluated multiple aspects of the quality of government services including corruption within the city of Bangalore. The CRCs found clear links between petty corruption and inefficient service provision in this more holistic approach (Paul, 2000). The CRCs heralded improvements in the quality of government services (Paul, 2000) (Ravindra, 2004) in Bangalore because of the greater 3

4 micro-level approach. These improvements were necessarily generic and still top-down because the CRCs are based on surveys that asked citizens for agency-wide feedback. Further, the CRCs and the ICS were both based on household surveys on government services. While household surveys provide an approximate snapshot of overall perception of government services, this method lends itself to perception bias if time has elapsed since the citizen received the service (Nicola, 2011) and it is virtually impossible to design a measurement approach that allows households to rank satisfaction unambiguously (Deichmann & Lall, 2003). This is especially true because citizens are continually exposed to the national consciousness against petty corruption discussed above and are hence prone to self-fulfilling prophecies and perception biases. Also the CRC surveys were 40 pages long 2, leading to potential fatigue bias (Hinkin, 1998). These are barriers for in-depth analyses of the data, since there is no way to reduce selection bias or perception bias when analyzing the effects of any one treatment or exposure in this context. At best, CRCs provide slightly biased summary data of the status quo. Finally, while the CRCs utilize a micro-approach, their data is not micro-oriented since only agency-wide data is collected as opposed to office-level data. My work draws inspiration from the empirical strategies described above but goes a step further in its approach. It is similar to these works in that it is a cross-sectional observational study with micro-level citizen-based surveys conducted to evaluate multiple facets of government service. The unique aspect of the approach discussed in this paper is that the field work in Bangalore involved surveys conducted at the primary point of the service provision by the government, similar to exit-polls. For instance, all data on police stations was collected by surveying citizens on their way out of the local branch offices of the police. These surveys evaluated the quality of service provision by asking each respondent a series of perception-based questions in a phenomenological 2 As mentioned by senior staff of PAC during a visit to their office in early August

5 approach, and evaluated corruption by asking each respondent objective yes/no questions about whether he had just paid a bribe, used a middleman etc in a more ontological approach. The survey was short to reduce fatigue bias. And anyway since the surveys of randomly sampled respondents were conducted on location at the government office, the respondents were all faced with very similar circumstances ensuring that there was significant overlap between the exposure groups and control groups in the empirical models discussed later, hence reducing systematic biases in selection and enabling causal analysis without necessitating dozens of covariates. Further, the exit-poll approach minimizes perception bias significantly and accurately measures the immediate experience of the citizen as opposed to a general recognition of a macro trend across a government agency. Corruption is illicit and secretive and the primary challenge on the empirical side is measurement (Hanna, 2012). So to deal with that, I ask the citizens themselves whether they paid a bribe when acquiring the service that they came to the respective government office for. They have little incentive to lie in this regard, particularly as bribery is so common in the city, and since this methodology minimizes perception bias, it is as objective a measure as can be found. This, in addition to the location-based approach, allows for more nuanced analysis of specific data trends within each branch office of each agency. This enables greater bottom-up accountability and reforms (from the branch level up) in addition to more accurate top-down accountability and reforms (on aggregating the branch level data). It also lends itself to location-based dissemination of data, viz. as a datamap 3 of government services (Appendix B) in a tradition similar to Ushahidi. 4 The summary statistics provide an overview of the prevalence of corruption and the quality of government agencies on three levels: the individual branch office level, the agency-wide level, and the assembly constituency level. The latter level of analysis is obtained by clubbing together the 3 This datamap of Bangalore is going to be published and disseminated in multiple traditional media outlets including the Times of India, the Hindu, TV9, and News9 4 Ushahidi puts up crowd-sourced crisis information on maps. 5

6 individual branch office statistics in every constituency (Bangalore is divided into 28 assembly constituencies). By providing accurate location-based data on the micro-level administrative functioning of the government and the macro-level functioning of the political and administrative aspects, it increases the accountability of the government officer on the margin at a branch office, of the top bureaucrats in each agency, and of the elected leaders. The empirical framework in this research has two models. First, with the supply-side model, I measure the effect of the quality of the various facets of the government institutions that supply government services on corruption. I assume the perception ratings given by citizens are proxies for the real institutional structures and/or de facto norms of service and performance of government services and study the incidence of corruption as a function of these proxies. Since the results of this model are similar to results in literature on supply-side causes of corruption (Klitgaard, 1991) (Hanna, 2012) (Myint, 2000), I find that the perception ratings are precise and not biased by random measurement error. This leads to the second demand-side model where I analyze the effect of corruption on the experience of citizens who demand government services. This is a unique approach that tests the effect of corruption on the perceptions of speed of service, cost of service, and overall satisfaction with the service respectively. Since I record the perception ratings on the citizen s way out of the government office, these ratings actually serve as a function of his personal experience which is in turn partly based on whether he was involved in corruption. Huntington (1968) argues, in the early literature on corruption, that bribery greas(es) the wheels of bureaucracy, and hence as an extension, the citizens on the demand-side of government services have an incentive to be corrupt. This of course is balanced against the incentive of officials on the supply-side of government services who want to supplement their (sometimes very low) incomes. Just the existence of this micro-level demand-supply efficient-grease hypothesis propagates a false equivalence between the demand-side and supply-side in corrupt exchanges 6

7 over basic government services for citizens. To be sure, literature contrasts the efficient-grease hypothesis against the sand-in-the-wheels hypothesis on the macro-level in terms of aggregate efficiency in governance (Méon, 2008) and the long-term externalities of corruption on private investment, economic development and productivity on the macro-level (Mauro, 1995) (Ades, 1997) (Wei, 2000) (Lambsdorff, 2003), but it scarcely questions the efficient-grease hypothesis rigorously on the micro-level. Kauffman and Wei (1999) are perhaps unique in their research to test and disprove this hypothesis using firm-level data. In the case of the citizen-government relationship, there is little systematic rigorous analysis of the existence of collusion or coercion in the citizen-government relationship when it comes to government services and almost no citizenlevel data that studies the efficient-grease hypothesis for citizen interactions with the government, even though this is the level that matters most to the welfare of the people and that is most likely to incentivize long-term political reform to reduce corruption. This is the gap that this paper serves to fill. Certainly there are many cases where corruption is not based simply on a nonthreatening exchange between government official and citizen, for instance the official might link the corruption to the threat of violence, viz. extortion. In the following empirical framework, I control for whether the citizen felt threatened in his interaction with the officials (as part of the primary survey that this research is based on), and hence am able to study the efficacy of corruption in what seems to be a purely business exchange between the citizen and official in greasing-the-wheels. In this empirical framework, I run three logit models with marginal effects analysis for the supply-side model, and two OLS regressions and one logit estimation model with marginal effects analysis for the demand-side model to achieve the following results. The supply-side model confirms the results of past academic literature on corruption. Firstly, I find that bribery is significantly less likely to occur when processes are made simpler to 7

8 understand and hence more explicit, exactly as for instance, Oldenburg (1985) finds in his detailed case study. Secondly I find that making processes more organized and predictable significantly reduces the likelihood of the use of middlemen, which makes sense because as one would expect, middlemen thrive on uncertainty, and in fact often go to some lengths to create it or to sustain the appearance of uncertainty. Thirdly I find that people are significantly less likely to feel threatened if the government officers are more accessible to them, an intuitive result. The addition to existing literature that this paper makes is an analysis of the effect of corruption on the demand-side of the interaction, viz. on the citizen s perceptions of speed of service, cost of service, and overall satisfaction with the service. The primary counterintuitive result I find is that usage of middlemen and bribes both individually reduce speed of service perception in a significant manner ceteris paribus for citizens trying to get a government service. I infer that middlemen and bribes are overpriced in the current market. This fact is confirmed for middlemen by a second OLS regression which shows that using middlemen increases the overall opportunity cost of the corrupt interaction for every unit increase in speed. The lack of price adjustment shows us the primacy of these forms of corruption in getting government services and weakens the only real benefit of petty corruption that Leff (1964) and Huntington (1968) trumpet on the individual level. So this research shows that even seemingly quid-pro-quo corrupt exchanges on the microlevel between citizens and government are not nearly as equal or nearly as effective in improving the quality of service as the efficient-grease hypothesis implicitly leads us to believe. This could be because of many reasons, for instance the monopolization of the service by the official and the weak accountability mechanisms in the agencies as argued by Klitgaard (1991). I also find that paying bribes and feeling threatened each significantly decrease the likelihood of satisfaction with the government service, while increased speed, increased helpfulness, and increased accessibility of the officials each significantly increase the likelihood of 8

9 satisfaction. Hence these offer avenues that can better rebalance the citizen-official relationship to not just rebalance the quid-pro-quo corrupt exchanges in the short run, but also make them less likely to occur in the long run. In the last section, using these results I outline incentive structures for citizens and government officials, as to how the office-level and agency-level data will increase accountability and incentivize reform in the administration. But to ensure true long term reforms there needs to be a political stake as well. In this regard, I find that 56% of citizens say that the quality of government services in their constituency will affect their vote in the coming assembly elections; and people who paid a bribe are significantly more likely to say the same than those who did not. In addition, the overall quality of government services of each assembly constituency in Bangalore is available as a metric in the datamap to measure local elected leaders by, hence incentivizing elected leaders toward reform. This is especially important because as Bussell (2012) shows us, elected leaders are often the stumbling blocks for reform in the government services in India. The remainder of the paper is organized as follows: Section 2 discusses the methodology, Section 3 discusses the data, Section 4 presents the empirical models and estimation methods, Section 5 discusses the results of the empirical models, Section 6 concludes. 2. Methodology Common sense naturalistic observation is all that we have. It is the only route to knowledge noisy, fallible, and biased though it be (Campbell, 1988). Given how vagarious the results of anti-corruption reform in developing nations have been, observational studies with significant external validity, immediate applicability, and a detailed explication of context are an important component of any approach to formulate bottom-up anticorruption reforms. As Campbell (1988) tells us, observation is often the only route to knowledge when measuring the interrelationships between the citizen-government interaction, quality of 9

10 government service, and corruption in an urban governance context. This fact is especially evident with the demand-side analysis in this research paper which was found on researching the data produced by the observational study that is the basis of this paper. But Campbell s caveats are entirely valid and observational studies are vulnerable to biases and lack of internal validity unlike experimental designs. This research paper offers a new kind of methodology for an observational study on demand-side effects of corruption in urban governance that uniquely minimizes these biases and maximizes internal validity, hence enabling us to analyze causal trends as well, bringing together as best as possible the benefits of observational study and quasi-experimental design. 2.1 Location-based Sampling The data is drawn from a survey of 2580 citizens across Bangalore in a location-based exitpoll basis that was conducted in June Specifically with police stations, sub-registrar offices, government hospitals, and the Public Distribution System (PDS), the surveys of citizens were conducted outside of the respective local branch offices. The first stratum in this stratified sampling approach was hence the local branch offices of these agencies. Then respondents were randomly sampled proportional to the total daily traffic observed at each branch office. With this approach, the downside was that there was an opening for harassment by officers who do not want to be rated. This was exactly what happened with police stations, as some constables threatened the surveyors with jail time for talking to citizens outside the police stations. For PDS, only a few sample slums around the city were chosen as a pilot to a more wide-ranging PDS study next year. With the four civic agencies, viz. BDA, BESCOM, BWSSB, BBMP the primary and most sustained point of service provision is at the household itself. For instance BESCOM sets up the electricity connection at one s house, provides one with electricity on a sustained basis at the house and when there is a problem, the officials often come to one s house to fix it. The sample of citizens 5 Surveys were conducted by the independent private sector surveying organization Team CVoter 10

11 that actually visited the branch offices for these four civic agencies was not representative of the range of services offered or the quality of service provided. These particular surveys were hence conducted at households, but to maintain the location-based approach and its benefits, I use a multi-stage random stratification process for these as well. The city was delineated into the 8 BBMP jurisdictional zones first. Then, proportional to the total number of wards in each of the BBMP jurisdictional zones, a few random wards were selected in each zone. Finally the right hand rule was used to randomly sample households within each ward to survey on all four civic agencies. This stratified sampling approach for both the office-based surveys and household surveys, minimizes the sampling error that exist in surveys which have a random citywide/statewide sample of households, since in that case the diversity in the group of respondents is unlikely to mirror the diversity of the people who actually use each of the government agencies above frequently. By conducting location-based surveys, I make no assumptions about which section of society is underrepresented or overrepresented when rating government services. Simply being on location takes care of proportional representation in the stratified sample, and as far as the demand-supply model for corruption in citizen-government interactions goes, only the population that actually shows up at the government offices is relevant to the analysis. 2.2 Minimizing Biases and Errors This location-based approach minimizes self-fulfilling prophecies and perception bias in the data because the respondents are commenting on their immediate experience as opposed to a general recognition of a macro trend across an agency or an interaction with the government they had weeks before. Nicola (2011) outlines how accuracy of survey recall data, especially perceptionbased ones, worsens over time because respondents resort to inference rather than memory. These biases are especially relevant in the aftermath of the national anticorruption protests which 11

12 were characterized by sweeping and impatient criticism directed at the State by citizens (Rajagopal, 2011) as opposed to clinical quantitative appraisals of corruption. This paper delves into questions of exogeneity and validity in both the methodology and data analysis to ensure the results thrown up are indeed true markers of the situation on the ground. It also conforms to the STROBE checklist which lays out a checklist of items that should be included in research articles based on observational studies (von Elm, 2007). There are 15 questions in the survey in addition to 5 demographic questions (Appendix A). The first set of questions specifically asks respondents for their ratings perception on 7 metrics of the quality of government service on a scale of The 1-10 rating scale is adopted to ensure sufficient variance in responses. Clear explanations are provided for the scale to reduce random measurement errors in the perception data, with 1-2 representing bad service and 9-10 representing excellent service. The second set of questions is of the objective yes/no type and asks whether the respondent used a middleman, paid a bribe, felt threatened, was satisfied overall. Then there are three questions related to their participation in the state assembly elections, followed by the demographic questions. These questions are very simple to grasp and clearly explained during the survey process and any disparities in perceptions or meanings of the questions are minimal. The analyses I conduct exploit the relationships between these individual sets of data. Since these are primary data, most potential confounding factors in the analysis of corruption in Bangalore have been accounted for in the primary survey itself. A pilot project was conducted in the summer of 2011 using this methodology specifically in Regional Transport Offices in Bangalore to test the validity of the construct. Extensive analysis was conducted on the RTO data to identify potential confounding factors for the next iteration, which in turn forms the basis of this paper. In addition I undertook detailed consultations with local governance NGOs such as Public Affairs Centre (of CRCs fame), Center for Public Policy at Indian Institute of Management Bangalore, and 12

13 with international organizations such as the World Bank Institute, British Council, and Transparency International, and I worked under the constant mentorship of Prof. Michael Johnston, author of Syndromes of Corruption 6, to ensure construct validity. Johnston and Kpundeh (2002) say that perception data are just estimates of how corrupt a society is thought to be and are thus open to influence and distortion. That s why in the survey model for this research paper, respondents are asked to answer all questions related to corruption in a purely ontological manner, viz. whether they personally just paid a bribe or used a middleman. But, since the government agencies are service-based agencies and interaction between citizens and the government is unavoidable, it is also crucial to consider the experience of individuals as can be gleaned from their perceptions. This is done through the first set of perception-based questions on the quality of government service. This is a democratic measure of service since it reinforces that each person holds the government to a different standard and that the objective of a service-based agency is to meet these standards to the satisfaction of each citizen, not to some objective measure. I devise a ratings algorithm 7 to obtain a single metric of the government service as a whole, and also to enable rankings and easy comparisons for both the public and policymakers. The ratings are used to make a datamap (Appendix B) of government services in Bangalore. The algorithm was formulated by conducting focus groups of citizens in Bangalore to prioritize the questions in the survey in terms of importance to them when getting a government service. Speed of service was selected unanimously as the top priority and hence it was weighted most heavily in the algorithm. This was also later confirmed on running regressions on the dataset. It was found that the effect of an increase in the speed of service rating significantly increased the likelihood of citizen satisfaction and was the largest of all the significant effects found. I believe that ratings and rankings are an effective and engaging way of revisiting corruption and transparency in government agencies. The This was done in collaboration with Prof. Dan Schult and Bharadwaj Reddy at Colgate University 13

14 ratings obtained from the algorithm are discussed in the summary statistics but are not used in the empirical models. Hence a more rigorous description of the algorithm is not required for this paper. 3. Data The total sample consists of 2580 surveys. 893 surveys were conducted at 50 sub-registrar offices, 452 surveys were conducted at 20 government hospitals, 952 surveys were conducted at 54 police stations, and 215 household surveys were conducted in 8 zones. The average age of the respondents was 34 years, 79.95% of the respondents were male, 29% had a college degree, and 36% had a monthly income greater than Rs (approx. $181). Overall approx. 33% reported not being satisfied with the service they received, 38% were asked to pay a bribe, 39% used a middleman to get a service, 15% felt threatened while getting a government service, and the average rating of all government services in Bangalore is 4.9/10. In line with past estimates and expectations, corruption was found to be highest in sub-registrars where 45% of respondents paid a bribe, followed by police stations where the figure was 42% and BBMP where it was 38%. Middlemen were most frequently used in sub-registrars where the figure was 44%, followed by police stations where the figure was 41.2% and government hospitals where it was 35%. Also as expected, people felt most threatened when at a police station, with the figure at 22%. This number is probably higher in reality because the citizens who felt the most threatened were scared to talk to surveyors right outside the police station for fear of reprisal by the police officers. This was followed by government hospitals where 20% felt threatened on location, which makes sense since government hospitals and police stations are where citizens are the most vulnerable. Respondents were most dissatisfied with the BBMP, with 47.27% reporting they weren t satisfied with the service they got. This was followed by police stations where 36.7% were dissatisfied and BWSSB where 36.5% were dissatisfied. All these results are reported in Table 1. 14

15 BESCOM had the best average rating with 5.5/10, while police stations had the worst average rating with 4.7/10. The average ratings along with the individual best and worst rated branch offices of each agency are reported in Table 2. The Yelahanka assembly constituency was found to have the most corrupt government services, with 95.8% of respondents having paid a bribe. Pulakeshinagar was the worst-rated constituency with a rating of 3.59/10 (edging out Yelahanka s rating of 3.6/10), while BTM layout was best-rated with a rating of 5.94/10. As discussed above, these are metrics that can be used to judge the record of elected leaders in enforcing discipline, accountability, and dialogue in government services. The ratings and corruption statistics in all 28 assembly constituencies are reported in Table 3. The ratings and corruption statistics in each of the four parliamentary constituencies are reported in Table 4. These summary statistics repudiate the belief that everyone in government is corrupt. By providing more accurate location-based data accountability is increased at the level of the branch offices of each agency, while still maintaining anonymity of the government officials and citizens. 4. Empirical Models and Estimation Methods 4.1 Assumptions Since respondents are surveyed at the physical location of the government services, the control group and the treatment group have an almost complete overlap of circumstances such as presence of middlemen, number of corrupt officials, real time speed of service etc, surrounding the government service. In fact, the primary difference between citizens is in their perception of these circumstances, and while the distributions of these perceptions are not precisely normal, histogram analyses show they are close to normal for each of the perception-based variables. The implicit assumption that I am making with respect to the 8 government agencies above is that people make the final choice to indulge in corruption at the location of the government service (and don t selfselect beforehand), and where there are factors or circumstances that might threaten or coerce 15

16 them into indulging in corruption, those factors are present at the location of the government service. This is a reasonable approach to take with bribes, since bribes are paid directly to the government official on location (and not faxed in beforehand for instance). With middlemen, it is slightly harder to make this assumption. That is why for the purposes of this research, I have picked government agencies where I found, after extensive groundwork, that middlemen and touts primarily offer their services right outside the government office (such as sub-registrars, government hospitals, BBMP etc), or where the middleman is one of the lower level officials of the bureaucracy that provides the government service (such as BESCOM, BWSSB, BDA, police stations etc). This ensures the decision is largely made within the overlap of circumstances above. There is also a potential for bias in that citizens might be predisposed to corruption because they are rich and can afford it, because they are young and impatient, are not educated etc. To deal with this, I control for a variety of demographic factors, including among other things, age, monthly family income and education levels. But more importantly, even if a citizen is somewhat predisposed to corruption due to factors beyond all the demographic information I control for, since the final choice to indulge in corruption is made on location at the government office, at worst (for my analysis) the circumstances at the government office were not good enough to reduce this predisposition to corruption and in fact were bad enough to reinforce these predispositions, and at best the predisposition is removed completely by the circumstances at the government office. Finally, a potential bias is that citizens who paid a bribe or felt threatened at the government office might be afraid to respond to the survey. The latter was seen especially at police stations. First of all, the entire survey process was entirely anonymous to minimize this fear. Second, the recent national anti-corruption protests and the success of such initiatives as ipaidabribe 8 have amped up the anger and willingness of citizens to report on corruption, with one

17 citizen at a government hospital even taking the surveyor straight to the room where the citizen was asked to pay a bribe and directly pointing at the officials and the amounts involved. But this bias still exists as was noticed outside of police stations where a few citizens who felt threatened refused to respond to the survey for fear of reprisal. If anything, this bias paints a slightly rosier picture of corruption in the real world, in the summary statistics and analyses in the next section. This methodology is better than purely household recall surveys like the CRCs, often conducted much after the service was received, aimed at rating agency-wide services as opposed to an office-level service (Ravindra, 2004). In the case of the purely household recall surveys it is almost impossible to ensure overlap between the control group and treatment group because it is impossible to ensure that they were exposed to similar circumstances. Given that nearly 40% of people used a middleman and nearly 40% of people paid a bribe, the estimates obtained in the models are precise since the treatment and control groups are almost the same size. But even after causation can be established the direction of causation still needs to be ascertained for the demand-side model. To do this I assume that, since a respondent s perceptions of the quality of service are recorded as he s exiting the government office, the perception ratings he gives are a result of the corruption variables. Also I assume that whether a citizen feels satisfied with the service overall respectively is a result of all the other factors of the government service. Hence the direction of causation I see for the demand-side model is as follows: Corruption variables Ratings for perception variables Overall satisfaction with the service. The latter assumption is a reasonable one to make. But there is a potential objection to the former assumption, viz. that there is a possibility of a perception bleed, where the citizen s prior perceptions of the government service affect the perception ratings he gives after getting the government service. These prior perceptions which influence his final perception ratings might have made him indulge in corruption, hence rendering ambiguous the first arrow above (though 17

18 not reversing it). I acknowledge this bias does exist, but still note that the location-based exit-poll approach is the best possible approach to minimize this bias, because it is conducted immediately after the citizen s experience which minimizes the effects of past generalized perceptions. Also more importantly as already noted above, the perception bleed will only be significant if the citizen s past perceptions are reinforced by his current service experience, in which case the perception bleed doesn t significantly bias the perception ratings of the quality of service recorded. 4.2 Regressions I investigate two sets of regressions. The first set consists of supply-side regressions, where the perception ratings given by citizens for the quality of the various aspects of the government services are assumed to be rough estimates of the real quality of the institutional structures. In these regressions, the corruption variables are used as regressands and their incidence is analyzed as a function of the quality of institutional structures. This is the approach used by previous empirical research such as Andrei (2009) with more objective factors. Since the results of these regressions match the results of previous scholarly work such as Klitgaard (1991) and Paul (2000), I assume citizen s perceptions are precisely measured and do not have significant random measurement error. This legitimacy is essential to the second set of regressions. The second set consists of the demand-side regressions. Since the surveys are conducted in an exit-poll fashion when the citizen is on his way out of the government office, the perception ratings that I record are a direct result of the experience he has just had and hence these ratings are shaped by whether he used a middleman or paid a bribe. In these regressions the corruption variables are regressors. This allows us to test the efficiency-grease hypothesis on the level of the citizen, and whether, in the absence of threats and coercion, corruption is based more on the demand-side or supply-side of government services. After all, most corruption literature implicitly accepts the claim that corruption greases the wheels on the individual level. The literature only 18

19 attempts to show that corruption leads to inefficiency at the macro level. I test that claim and hence test perhaps the last remaining pillar of legitimacy for corrupt practices with two OLS regressions that regress the log of speed of service perception and the log of cost of service perception against the corruption variables. I also run a logit regression with the dummy variable that asks if the citizen is satisfied overall with the government service against the corruption variables. On conducting the The Breusch-Pagan test, it was found that there is heteroskedasticity in the data and hence only robust standard errors are used for the analysis. The first set of regressions, viz. the supply-side model, consists of three logit models estimated as the following: (1) where i is the individual respondent. This logit regression is run three times with the dummy variables for whether a middleman was used (logit 1.1), whether a bribe was paid (1.2), and whether the citizen felt threatened as the respective regressands. O is the vector of the yes/no objective questions that are not the regressand, and is the vector of corresponding coefficients. P is a vector of all perception-based questions in the survey except the cleanliness question (which I later use as an instrumental variable to test endogeneity of the variables I am concerned with using the Hausman test). I analyze P as a proxy for the real institutional structures in the agencies for the purposes of this model. X is a vector of other covariates in the regression including government agency, location of survey, gender, age, family income, education level, and social group. I use marginal effects to analyze the relationships between the covariates and the regressand. These logits allow us to find important predictors of primary forms of corruption. The second set of regressions, viz. the demand-side model consists of the following: (2.1) 19

20 Here O is the vector of the yes/no objective questions except the dummy variable for whether the respondent was satisfied overall with the service, because as already outlined in my temporal assumptions for causation, the answer to that question is large dependent on the regressand itself here. This regression includes the interaction variable to ensure that the difference in speed for those who used a middleman and paid a bribe and those who paid a bribe without using a middleman is accounted for, especially since this interplay is often symbiotic. The primary analyses from this regression come from the vector and coefficients. (2.2) Since the opportunity cost variable consists of both money spent and time spent to get a service, I infer that the cost variable is a function of the speed of service variable, and hence I add the two interaction terms with the coefficients to help analyze the effect of speed of service on the cost of service for those who used a middleman and those who paid a bribe respectively. Speed of service and the cost of service are specifically analyzed in the OLS regressions above because these two variables were consistently rated as the most important factors in judging a government service by focus groups. These regressions allow us to look at the impact of corruption on these specific factors especially given the temporal gap outlined in the assumptions. Finally, I run the logit model: (2.3) to test the effect of the corruption variables on a citizen s overall satisfaction with the government service. 5. Results: Using Durbin-Wu-Hausman chi-sq tests I find that the primary regressors analyzed in the logit models and OLS regressions above are exogenous. Depending on the regression, the 20

21 instruments that I use for the test are whether someone voted in the last elections, their age, and how clean the respondent perceived the office to be. This shows internal validity in the data. 5.1 Supply-Side Model Results Table 5 gives the estimation results of marginal effects for the three logistic regressions in the supply-side model. These logistic regressions analyze the perception ratings given by the citizens as proxies for the quality of the institutional mechanisms of the government agency. They confirm previous results in corruption literature and can point policymakers to important pressure points for reform. Causality is not as assured as in an RCT, but as already explained, an RCT would not be able to model the many complexities of real citizen/official encounters on the ground and the behaviors of each respective group. The data in this research are still found to be internally and construct-wise valid though, and hence these results are representative of real causal trends. Using marginal effects in the logistic regression 1.1, I find that an increase in the rating of how organized the process was leads to a significant fall in the likelihood of the usage of a middleman. The effects of other perception-based variables are statistically insignificant. The former observation offers an avenue for reform to fight the presence of middlemen in government. An example for making the process more organized is to put up detailed citizen charters which consist of clearly outlined procedures for all services on location at the office. People who felt threatened were 16% more likely to use a middleman than those who did not, as one would expect. In the logistic regression 1.2, I find that an increase in the rating of how simple the process was leads to a significant fall in the likelihood of the payment of a bribe similar to the results of the case study conducted by Oldenburg (1985) and the exhaustive list of anti-corruption reforms compiled by Hanna (2011). Interestingly the effect of helpfulness of the government officer and the accessibility of the government officer ratings individually was statistically insignificant. This shows us that there is at least some degree of complicity by the citizen in the process of corruption since 21

22 the lack of simplicity of the process makes the citizen more likely to pay a bribe even with a more helpful and accessible government officer. Of course this complicity is also partially coerced by the government, since the government is responsible for instituting processes that are not simple. Hence future reforms that lead to greater simplicity and reduced uncertainty in what should be quite straightforward processes of providing government services are likely to reduce the likelihood of bribery. Bangalore One centers, a one-stop shop to pay most bills for government services, have made the process of this payment significantly simpler and have led to a fall in corruption in Bangalore (Bussell, 2012). Similar initiatives and greater e-governance penetration aimed at making the process simpler are hence the most effective ways forward to reduce bribery. People who used a middleman are 42% more likely to have paid a bribe than those who were not, and people who felt threatened are 12% more likely to pay a bribe than those who didn t feel threatened, again both expected results. There is no significant effect at the 10% significance level of different social and economic groups on the occurrence of bribery. In regressions 1.1 and 1.2, I find that, on controlling for demographic factors such as education, family income, and social group, citizens who voted in the last assembly elections are significantly less likely to use a middleman and less likely to pay a bribe respectively. This implies that citizens at least have some role to play in reducing corruption, because citizens who are not apathetic toward the process of government and who actually take the effort to take part in the processes of government are less likely to allow corruption to occur. In the logistic regression 1.3, I find that an increase in the rating of how accessible were the government officers leads to a significant fall in the likelihood of feeling threatened. This is also backed up by considerable anecdotal evidence collected during the survey process. People consistently complained that they didn t know who to go to, and when they did know and were subsequently threatened by the official they interacted with, there was no clear escalation 22

23 mechanism to the higher officials. This problem is especially accentuated in police stations and government hospitals where citizens are in substantially more vulnerable situation. Hence the first reform required to make people feel more secure at government offices is a clear escalation mechanism that is implemented effectively and immediately. Also interestingly, I find that people in the higher monthly income group of income between Rs and Rs (most of middle-class India) were significantly more likely to feel threatened than people in all other income groups. On the other hand people in a low income group with a monthly income between Rs 6000 and Rs were significantly less likely to feel threatened. The latter trend is probably because a low income group s tolerance for threats and corruption in government offices is higher than that of the middle class group, a trend that has been found in the CRCs as well (Paul, 2000), or perhaps because threatening poor people is less rewarding for government officials. Finally in my analysis of the supply-side model I find in regression 1.1 and 1.2 that on regressing the two corruption variables against each of the demographic variables (gender, age, education level, monthly family income, and religion & caste ), there is only one significant effect out of the 10 possible effects I consider in that people who pay bribes (just bribes, not middlemen) are more likely to be men. This effect is small (about 5%), and so I ignore it for the analysis of the demand-side model where the payment of a bribe and the usage of middlemen respectively are used as treatments. Since there is largely no significant difference in the composition of the control group and treatment group, the results of the demand-side model, where I test the effect of corruption on the citizen s perception of the quality of service, are valid. Further this evidence supports the argument that corruption is systemic rather than a function of individual demographic or cultural trends. Finally, this evidence further strengthens the justification for using the perception ratings given by citizens as a proxy for the real quality of institutional structures in government, given that demographic factors don t significantly affect the behavior of citizens. 23

24 5.2 Demand-Side Model Results Table 6 gives the results of the logistic regression and two OLS regressions in the demandside model. Perhaps the most counterintuitive result in regression 2.1 is that people who use middlemen and people who pay bribes each have significantly lower speed of service ratings than people who did not use middlemen and did not pay bribes respectively. It is important to note before jumping to conclusions that people who use middlemen or pay a bribe will expect a higher real time speed of service, and hence are likely to have a lower speed of service perception than people who don t use middlemen or pay bribes even if both groups have exactly the same real time speed of service. Keeping this in mind, the primary explanation for this trend is that the markets for bribes and middlemen are overpriced in the status quo. The primary goal for the individual in using bribery and middlemen is to make things move faster, except in cases where he is actively threatened. Since these OLS regressions control for the threat factor, it is clear that the individual s expectations of increased speed from bribes or middlemen are not being met. These increased expectations might arise only from the simple qualitative act of paying a bribe or using a middleman, and not be tied to the exact monetary cost of the bribe or the middleman, in which case the mere existence of middlemen and bribes are unjustified; or as the cost of the bribe and the middleman goes up, the expectations of speed might likewise increase. Either way this implies necessarily that the market for bribes and middlemen as they exist are overpriced, both quantitatively and/or qualitatively. This fact is confirmed by looking at regression 2.2 where the log of the overall cost perception variable is the regressand. The overall cost perception consists of two forces, viz. the actual money and time spent by the citizen in getting the service. In the context of middlemen and bribes, the actual money spent definitely goes up and the time spent by the citizen in getting the service is supposed to go down. 24

25 The coefficient on the speed of service perception variable is significant and negative at -.02, which tells us that without middlemen and without bribes, as speed of service increases, the cost perception of the service falls by 2%. Now when we consider the interaction term middleman*speed of service, the coefficient is significant and positive at Hence when a middleman is used the effect of an increase of 1 in the speed of service rating on the cost of service rating is (-0.02) = 0.02, or 2%. So middlemen increase the monetary cost to an individual to a larger extent than the corresponding reduction in the time spent by the individual to get the service, viz. middlemen are overpriced. A similar confirmatory analysis for bribes is not possible because the coefficient on the interaction term of bribe*speed of service is not statistically significant even though it is positive and the coefficient on the variable for bribery is statistically significant. This result is extremely important because the argument that petty corruption, whether in the form of middlemen or bribes, greases the wheels and increases the individual s personal utility is seen as one of the few saving graces of corruption. This research considerably weakens the strength of this argument by showing that when considered on a qualitative level or on a quantitative level, middlemen and bribes are overvalued and their actual efficacy is over-perceived. It might be argued that middlemen and bribes are being used by citizens primarily for greater convenience and nothing else. But the OLS regressions have a proxy for convenience in the perception variables for How simple was the process? and How organized was the process? On the other hand, initiatives such as Bangalore One make the process simpler and more organized and actually eliminate the need for middlemen, hence using these variables as proxies for convenience helps control for these different conceptions of convenience in the regressions. Logistic regression 2.3 regresses the dummy variable for overall satisfaction against the corruption variables and other covariates. As expected (and as also described in pre-survey focus 25

26 groups), an increase in the speed of service rating leads to a significant increase in the likelihood of the respondent being satisfied with the service overall. Increases in the helpfulness of the officer rating and the accessibility of the officer rating also each led to a significant increase in the likelihood of the respondent being satisfied with the service. This is backed up by anecdotal evidence, with respondents reporting that government officers who were courteous and helpful made them feel much more content on their way out of the government office even in the absence of any subsequent increase in the speed of service. As expected, people who paid a bribe were significantly less likely to say they were satisfied with the service and people who felt threatened were significantly less likely to say they were satisfied with the service. People who had a college degree were 27% more likely to say they were dissatisfied with the government service, probably because of the higher standards that educated citizens hold the government to. The summary statistics in Section 3, the regression results, and the model itself all have significant external validity for other cities in India, since on the supply side, government structures in Bangalore are similar to these cities, and on the demand side, the population of Bangalore, which has migrants from all over India, represents a crude cross-section of the nation. Further since this model derives its primary analyses from data given by citizens on the ground, it is a model that is easily exportable to urban contexts in other developing nations as well. 6. Discussion Griliches (1985) has said, standard economic theory while couched in micro language, (is) not truly micro-oriented. This methodology changes that. The micro-level, location-based data collection approach outlined in this paper analyzes the complexity of group behaviors in the real world in a topic that has historically been misdiagnosed and where the success of reforms have been erratic, viz. corruption. It reveals relationships between multiple dimensions of the citizen s 26

27 interaction with the government in a real world context and avoids the all too common error of looking at macro-level supply-side factors and deriving incentive-based reforms from those. For instance, most academic literature often automatically assumes that middlemen and bribes have a positive impact on the speed of service for the citizen at the micro level and often narrow their search to finding evidence aimed at showing that corruption leads to inefficiency at the macro level. So it is especially hard for citizens to weigh out this positive impact of corruption which is usually not contrasted with the impact of not being corrupt on the individual level, and the long-term costs distributed over time and a wide population. The demand-side model analyzed in this paper shows that middlemen and bribes are overpriced in the status quo and their impact on the speed of service is often over-perceived. Reforms packaged with the subsequent inference that bribes aren t going to make things go easier for you are more likely to leverage a citizen s selfinterest and inspire the citizen s participation in bottom-up reforms than inefficient top-down reforms sold on the premise that middlemen and bribes cause macro-level implications for balance in the economy. As Bussell (2012) shows, often these top-down reforms fail in real-world implementation because elected leaders block the reforms from being implemented as they are involved in extracting rent from inefficient government services as well. Hence for true long-term change, a bottom-up change in the behavior of the citizen and government official at the margin is a necessary component. Indeed due to the nature of the microlevel, location-based data collection, this methodology lends itself to analyses that will matter to the individual at the margin, similar to the petty corruption is overpriced analysis. More demand-side analysis is hence more likely to leverage the self-interest of the citizen than existing literature. My research has a real-world implementation approach and that is why I undertook a crosssectional study with a location-based approach, viz. to understand, analyze, and account for the complex interplay of multiple factors while simultaneously generating data that is most likely to 27

28 incentivize citizens, government officials, and elected leaders at the margin to improve governance. Some of these incentive structures are outlined here: Citizens: Regressions 1.1 and 1.2 show, citizens who are more involved in the processes of government (using whether someone voted in the last local elections as a proxy) are less likely to be corrupt. Hence it is important to incentivize greater participation of this group toward the goal of meaningful long-term change in governance. The individual ratings for each branch office and the corresponding datamap empower citizens with new information on their locality to demand specific changes from their local elected leaders and administration. Increased awareness of corruption and inefficiency levels in other better performing districts and awareness of the current overpriced market for corruption decreases a citizen s tolerance for corruption. Already in a pilot in the Byatarayanpura constituency in Bangalore (a low-performing constituency with a rating of 4.57/10), the dissemination of this datamap has led to citizens organizing and signing petitions that demand greater dialogue and accountability from the government on the constituency level. Citizens are also able to make better personal decisions, viz. they can demand non-zone restricted governmental services from less corrupt and more efficient zones. More importantly the analysis of the unique demand-side model done in this research is much more effective at leveraging a citizen s self-interest (as already discussed) than abstract macro-level supply-side research. The movement against corruption does not only need to be based on civic virtue and hence suffer from the subsequent collective-action problems. This demand-side analysis shows that not only are citizens paying for services that are their right and ought to be free, they re also getting significantly less than what they re paying for when they do indulge in corruption. People not indulging in corruption seem to be better off so corruption is clearly not inevitable. Government Officials: The corruption cycle begins with the bureaucrat paying a rent to be placed in desirable posts where there is a lot of recognition and investment and hence a greater 28

29 scope for corruption (Wade, 1985). In the short run, with the office-level location-based data, bureaucrats should get immediate recognition, promotion opportunities, salary raises, and credit for verifiably not being corrupt and performing better. This would also force bureaucrats who are already in desirable posts, and hence potentially blasé about reform, to now continually meet better standards created by lower placed officials or officials in other zones. The Department for Personnel and Administrative Reforms in the Karnataka state government is fully collaborating with me to implement reforms based on the analyses in this research paper and to ensure that good bureaucrats get the recognition they deserve. This department initiated the Sakaala initiative which rewards officials in a few select districts for meeting turn-around deadlines on the maximum number of applications (Rajneesh, 2012). Sakaala has already seen a virtuous cycle in the behavior of these officials, hence confirming that rewards which lead to greater recognition and promotions incentivize government officers to provide better service overall. It is important to note here that the incentive structures for individual government officials do not actualize unless the government itself is involved in providing rewards for wellperforming officials. This is because of two reasons: first, any outside publicity for good performance by government officials is necessarily ephemeral and hence not much of an incentive for long-term change. Second, in my case study this summer, which involved interviews of multiple bureaucrats in four departments the MS Building (which holds the central bureaucracy offices for the entire state of Karnataka), I find that bureaucrats are mostly publicity-shy, in some cases even actively averse to it. To many of these bureaucrats, internal reward mechanisms are the only catalyst to better performance. Another important condition to be fulfilled before these incentive structures for officials can be realized is that a staunchly reformist official has to be at the vanguard of this movement. This past summer, most of my initial meetings with top level government officials, for instance with the 29

30 Urban Affairs Advisor to the CM and the Director General of Police, were stonewalled. The Director General of Police even refused to entertain discussion on the topic because he felt that it brought bad publicity to the Police department. This is a very real obstacle to reform in that top officials are unwilling to accept the problems in the status quo that need to be fixed, probably because said top officials often have vagarious short-term postings (Wade, 1985) and hence are not willing to accept the department s foibles in the short-run and are not interested in developing reforms for the longrun. The enthusiastic involvement of Dr. Rajneesh, a reformist official at the Department of Personnel and Administrative Reforms, who is also the mission director of the Sakaala initiative mentioned above (Shenoy, 2012), was crucial to the initiation of discussion with the government and the organizing of a workshop with the government agencies included in this survey. Elected Leaders: When it comes to elected leaders, not only did 56% of the people surveyed say that the quality of government services will affect their vote in the coming elections, people who paid a bribe were also significantly more likely than people who did not pay a bribe to say that quality of government services will affect their vote. Bussell (2012) outlines how elected leaders and the compulsions of coalition politics have been stumbling blocks of reforms in government services. But the statistics above seem to be strong incentives for politicians to enact the reforms outlined in this paper especially once this data is disseminated widely. I assumed that local elected leaders on the constituency level would use this data as an incentive to improve their ratings and implement reforms before the coming assembly elections because it is rational given the above statistics. But I was stonewalled by these elected leaders. For instance, the local MLA at Byatarayanapura constituency refused to discuss the results of his constituency, refused to offer any reforms for the coming weeks or discuss any past reforms he might have instituted related to governance, and refused to accept that there were structural problems in the governance of his constituency (Byatarayanapura was ranked 27 th out of the 30 30

31 constituencies covered in this study) (ByreGowda, 2012). This particular MLA was educated in the US, is considered a clean politician, and is seen as the vanguard of a youth revolution in Indian party politics, and hence he should ve been a soft target for the incentive structures for elected leaders outlined. So MLAs in Bangalore are not very likely to get involved in reform for its own sake and hence should not be depended on as primary drivers of reform in government services, a fact that is also generalized by Bussell (2012). But clearly, as Bussell (2012) also tells us, elected leaders have to be part of the solution or they will be part of the problem. I see three alternatives to consider to activate these incentive structures among elected leaders: First, much greater and more relentless publicity of this data should be conducted right before the assembly elections (when it is part of the public debate, the opposition has the incentive to take advantage of it if the ratings are low, and the sitting leader has the incentive to take advantage of it if the ratings are high). The next iteration of this survey will be conducted just before the assembly elections next year enabling panel data analysis and offering a longitudinal metric for each local elected leader s record on government services right before the elections to attempt this. Second, the same analysis can be done on the party-level instead of the level of the individual elected leader. This could create state-wide dialogue even in a non-election year as any spin on this in the media can result in the painting of the governing party or the opposition party as anti-governance and anti-progress which can potentially lead to the fall of the government or the strengthening of it. This is not possible on the level of the individual elected leader. Third, a reformist elected leader in the top echelons of government should be found to activate these incentive structures on the constituency level. There should also be central and state government grants toward gathering and publishing such data regularly in the public interest. This paper owes a debt to the precedent set by the Citizen Report Cards, India Corruption Studies, and other such studies on corruption in government services in India. But inherent in the 31

32 location-based methodology and the processes by which this particular research was carried out are factors that significantly reduce selection bias, perception bias, fatigue bias, and random measurement errors which are present in previous studies. As a result this research collects more accurate data and can operate in a framework where regression and causal analyses are valid, hence giving us not just a set of observations about the status quo (like most previous studies in this field) but also a definite set of academically rigorous reforms that can and should be implemented to deal with the often misunderstood forces that keep corruption alive. The primary problem with the Anna Hazare-led anti-corruption protests was that condemnation of government was sweeping and the proposed reform in the Jan Lokpal was utterly simplistic. As Rajagopal (2011) says Attention (was) not paid to procedures whereby grievances can be adjudicated; to assume that a new law or a new leader can solve the problems is too optimistic. The research methodology discussed in this paper enables bottom-up reform by creating a more accurate quantitative awareness of corruption in government services which is experience in specific ways by citizens and officials on the ground, while creating incentive structures for all the groups involved to reduce corruption, while also enabling the causal analyses required to formulate future reforms, some of which are outlined in this paper. References Ades, A. and R. Di Tella The new economics of corruption: a survey and some results. Political Studies. 45 (special issue) Agarwal, S.K Towards Improving Governance. New Delhi: Academic Foundation. Bardhan, Pranab Corruption and Development: A Review of Issues. Journal of Economic Literature. September. Vol. XXXV Burke, Jason Corruption in India: All your life you pay for things that should be free. The Guardian. August 19. Viewed August 20 th Read online at 32

33 Bussell, Jennifer Corruption and Reform in India: Public Services in the Digital Age. Cambridge University Press. ByreGowda, Krishna Personal Interview at the office of the Member of Legislative Assembly for Byatarayanapura constituency. Conducted on 19 th August Campbell, Donald T Methodology and Epistemology for Social Sciences: Selceted Papers. University of Chicago Press. Carlson M, Morrison S Study Design, Precision, and Validity in Observational Studies. Journal of Palliative Medicine : doi: /jpm Deichmann, Uwe and Somik V. Lall Are you satisfied? Citizen feedback and delivery of urban services. Washington DC: Development Research Group, World Bank. Griliches, Zvi Data and Econometriciains The Uneasy Alliance. American Economic Review. May. Vol 75. No Hanna, Rema and Abhijit Banerjee and Sendhil Mullainathan Corruption. MIT Dept of Economics Working Paper Series. Working Paper Hanna, R., Bishop, S., Nadel, S., Scheffler, G, Durlacher, K The effectiveness of anti-corruption policy: what has worked, what hasn t, and what we don t know a systematic review. Technical report. London: EPPI-Centre, Social Science Research Unit, Institute of Education, University of London. Hinkin, T.R A brief tutorial on the development of measures for use in survey questionnaires. Organizational Research Methods, Huntington, S.P Political order in changing societies. New Haven, Yale University Press. Indian Corruption Study (ICS) New Delhi: CMS Research House Johnston, Michael and Sahr J. Kpundeh The measurement problem: A focus on governance. Forum on Crime and Society, December, Vol. 2, No.1. Kaufmann, Daniel and Shang-Jin Wei. 1999). Does grease money speed up the wheels of commerce? NBER Working Paper Available online at Klitgaard, Robert Controlling Corruption. University of California Press. Lambsdorff, J.G How Corruption Affects Productivity. Kyklos. Vol. 56, No

34 Leff, N.H Economic development through bureaucratic corruption. American Behavioral Scientist. Vol. 8, No Reprinted in Political corruption: A handbook, A.J. Heidenheimer, M. Johnston and V.T. LeVine (eds.). Oxford, Transaction Books, , Mauro, P Corruption and Growth. Quarterly Journal of Economics. Vol. 110, No Méon, Pierre-Guillaume & Laurent Weill "Is Corruption an Efficient Grease?" World Development, Elsevier. Vol. 38, No March. Myint, U Corruption: Causes, Consequences and Cures. Asia-Pacific Development Journal. Vol 7, No 2. December. Nicole, Francesca de and Xavier Gine How accurate is recall data? Evidence from Coastal India. Northeast Universities Development Consortium Conference Available online at: Oldenburg, Philip Middlemen in Third-World Corruption: Implications of an Indian Case. World Politics. July. Vol. 39. No. 4. Paul, Samuel and Sita Sekhar Benchmarking Urban Services The Second Report Card on Bangalore. Bangalore: Public Affairs Centre. Rajagopal, Arvind Visibility as a Trap in the Anna Hazare Campaign. Economic and Political Weekly. Vol XLVI, No 47. Rajneesh, Shalini Personal Interview at the Department for Personnel and Administrative Reforms. Conducted on 30 th July Ravindra, Adikeshavalu An Assessment of the Impact of Bangalore Citizen Report Cards on the Performance of Public Agencies. Evaluation Capacity Development Working Paper Series. No 12. Shenoy, Jaideep Sakala applications are disposed off timely in Udupi. The Times of India. Available online at von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, et al The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for reporting observational studies. PLoS Med 4(10): e /journal.pmed Wade, Robert The Market for Public Office: Why the Indian State is not Better at Development. Ch. 87 (pp ) in Stephen P. Osborne (ed), Public Management: Policy making, ethics and accountability in public management. Routledge Publishers. Wei, Shang-Jin Local corruption and global capital flows. Brookings Papers on Economic Activity

35 Appendix A Please talk about your personal experiences with these agencies. B0. What service did you come here to get? On a scale of 1 to 10, with 1 being the least and 10 being the highest, please rate the following: QB1 QB2 QB3 QB4 QB5 QB6 QB7 How simple was the process? How organized was the process? How fast did you get what you came for? How helpful were the government officers? How accessible were the government officers? How costly was the process to you personally (both in terms of money and time spent)? 1 is least expensive and 10 is most expensive How clean were your surroundings? Please answer these questions with Yes/No: QE1 QE2 QE3 QE4 QE5 Did you use a middleman for the service? 1. Yes 2.No Did you pay a bribe? 1. Yes 2.No Did you feel threatened in any way? 1. Yes 2.No Were you satisfied with the service provided by the government office? 1. Yes 2.No Did you vote in the last MLA (assembly) elections in 2008? 1. Yes 2.No QE6a QE6b Will you vote in the coming 2013 MLA (assembly) elections? 1. Yes 2.No If yes, does the quality of government services in your constituency influence your vote? 1 Yes, significantly 2 Yes, to some extent 3 Not very much 4 Not at all Demographics 35

36 D1 D2 D3 GENDER (NOTE, DO NOT ASK, CODE ONE ANSWER) 1. Male 2.Female Actual Age (ASK Actual AGE) Education: Highest attained CODE ONE ANSWER 1 Non-formal education (can read and write) 2 Up to primary level 3 Up to High school 4 Higher Secondary 5 Up to Graduation 6 Up to Post-graduate 7 Professional 8 Illiterate (can't read or write) D4 Monthly Family Income (CODE ONE ANSWER) 1 Less than Rs Rs Rs ,000 4 Rs 10,000-20, ,000-50, ,000-1,00,000 7 More than 1,00,000 8 Can't Say D5 Social Group (CODE ONE ANSWER) 1 SC (Scheduled Caste/Dalits) 2 ST (Scheduled Tribes) 3 OBC (Other Backward Classes) 4 UCH (Upper Caste Hindus) 5 Muslim 6 Christians 7 Sikhs 8 Others 9 Can't Say Appendix B (Datamap) None of the ratings top 6/10, so the color distribution (still being tinkered) only has shades of red. 36

37 37