How Does Electoral Competition Affect Bureaucratic Performance Part 3

6 Empirical Results We begin by providing the OLS estimates for the effect of electoral competition on time taken to sanction the projects when the measure of competition is victory margin in the previous election. Table 6 gives the results of this specification. Column 1 shows that as margin of victory increases by 1%, the time taken to sanction the projects reduces by about a day. Now, the 10th percentile victory margin distribution is at 1.8% while the 90th percentile is at 22.6%. Back of the envelope calculations suggest that moving from one end of the distribution to the other, a fall in electoral competition reduces the number of days to sanction by about 16 days 17 . Hence, bureaucrats perform better when electoral competition is low. As we discussed before, past victory margin may not be a good predictor the extent of competition faced by the politician in the next election. Moreover, there are concerns about reverse causality when one uses this measure. We now present the results from the difference-in-difference specification. It provides us with the causal effect of change in politician’s probability of winning on bureaucrats’ performance. 6.1 Results from Difference-in-Differences Strategy Consider the following model: yibpct = γ0 + γ1 ∗ T reatmentc ∗ P ost + γ2 ∗ T reatmentc + γ3 ∗ P ost (18) + πp + φb + τt + ψ + ibpct where, yibpct denotes the time taken to sanction project i by bureaucrat b when paired with politician p in constituency c in year t. πp , φb , τt and ψ are fixed effects as defined before. The variable ‘Treatment’ takes value one if the announcement of delimitation affected the particular constituency while ‘post’ is a dummy variable denoting the period after the delimitation (2008-2009). Table 7 gives the results for specification 18. The difference-in-differences estimate γ1 is positive and statistically significant at 10% l.o.s. It shows that as the probability of winning goes to zero, the average time taken to sanction a project increase by about 10 days. Given 17 Assuming a linear relationship between victory margin and time taken to sanction. 24 that the mean of the dependent variable is 73 days, it means that sanctioning time increases by approximately13%. In column 2, we use the cost of project as a measure of the size of project. The results remain unchanged. 6.2 Falsification Tests for DID Strategy One key assumption of DID estimation is that the trends in outcomes of interest would be the same in both the groups in absence of the shock, and it is the new information that induced a deviation from the common trend. One way to check this is to compare the trends before the shock took place. This can be carried out formally by creating a fake shock prior to the actual shock and estimating model with the same treatment and control groups. If pre-trends in time taken to sanction are the same, then the difference-in-differences estimates with the ‘fake’ shock should be zero. We create such a fake shock occurring in December of 2005, two years prior to the actual shock. The treated and control groups remain the same. Table 8 provides the results of this specification. As we can see, the magnitudes of the difference-in-differences estimator has gone down substantially. Moreover, the coefficient is insignificant. This is comforting and suggests that we can assume that pre-shock trends in times taken to sanction were same across the control and treated groups. 6.3 Isolating the Mechanisms While the DID estimate provides us with the causal effect of politician’s probability of winning on bureaucrats’ performance, it does not by itself help identify the causal mechanism. As discussed before, this is because when the shock occurs, both re-election concerns as well as access to dynamic incentives shut down. In order to identify which mechanism is driving the results, we now estimate the following model: yibpct = β0 + β1 ∗ Strongholdc + πp + φb + τt + ψ + ibpct (19) where stronghold is a dummy variable that takes value one if the same party won over all four elections between 1989-1999. πp and φb are politician and bureaucratic effects respectively. τt corresponds to year fixed effects and ψ denotes project type fixed effects. Column (1) of Table 9 provides the estimates of equation 19. The average time taken to sanction a project is significantly lower by a magnitude of about nine days if a constituency is 25 a party stronghold compared to when its not. Comparing this to the mean of the dependent variable, we find that this constitutes about an 11% decrease in approval speed. In column 2, we add the size of the project which is meant to capture how big the project is. Column (3) controls for politicians experience and shows that more experienced the politician is, the faster the projects are sanctioned. This does not change the effect of being a stronghold on the speed of approval. Column (4) controls for whether the incumbent is from the same party as the leader of the state and (5) includes party fixed effects. The results remain the same: when competitive pressures are very low, the projects are sanctioned significantly faster. What does this tell us about the potential mechanisms? From table 2 we see - that we can reject the hypothesis that pure re-election concerns drive the results. The results are consistent with dynamic incentive mechanism: when the probability of winning is high, the politicians can credibly use dynamic contracts to incentivize the bureaucrat. In competitive areas, on the other hand, promise of future rewards are less credible. 6.4 Promotions: Direct Evidence for Dynamic Incenitives Until now we have been talking about future rewards without saying what they actually are. In this section we explore the nature of incentives provided to the bureaucrats. Since the bureaucrats we look at are career civil servants, their wages are determined exogenously. The only way the politician can control the bureaucrat is by reassignment to other districts or a plum job at the time of promotion. We now show that it is the promise of favorably affecting the promotions that induces bureaucrats to perform better. Using the career histories, we construct a promotion timeline for the bureaucrats. As we mentioned before, we look at bureaucrats when they are collectors because this is the period in which they are responsible for approving and executing the public goods. This corresponds to roughly years 5-16 of their careers. In their 9th year, they are promoted to the next level but there is no screening: everyone is promoted 18 . This is basically a hike in their salaries. It is in the 13th and the 16th years that the promotions have a bite. The politicians can potentially influence bureaucrats’ job assignments once they are up for these promotion. We construct a variable that measures how far the bureaucrats are away from the promotions. For this exercise, we only consider bureaucrats who are in years 10-16 years of their career since the 9th year promotion is “non-functional”. The variable “three years to promotion” means they are either in the 10th year of their career or just crossed 13th year. Two 18 See http://persmin.gov.in/DOPT/EmployeesCorner/Acts_Rules/IASPromotionGuideLines.pdf 26 years and one year from promotion are defined similarly. Our first hypothesis is that when bureaucrats are up for promotion, the performance should improve. Table 10 looks at effect of distance from promotion screening on approval times. Column (1) shows that compared to three or more years, the time taken to sanction is about 14 days less when the bureaucrat is up for promotion within the year. This magnitude is 20% of the mean of dependent variable. This result remains robust to controlling for project characteristics. To see how the promotions tie up with the dynamic incentives, we now perform an empirical test. Suppose we are in time t = 0 and elections are to be held in period t = 2. Let B1 denote a bureaucrat who is up for promotion in t = 1, i.e. before the elections. Let B2 denote a bureaucrat who is up for promotion in t > 2. How should B1 and B2 perform? The answer rests in whether we are in a stronghold (S) or a non-stronghold (N). Let us first consider the B1 bureaucrats. These bureaucrats are going to be up for promotion before the next elections. This means that both the stronghold as well as non-stronghold politicians will be in office at the time of promotion. If favorable promotions are what the bureaucrats care about, then we should not see any difference in the time taken to approve for B1 bureaucrats across strongholds and non-strongholds. Now, let us consider B2 bureaucrats who would be up for promotion after the elections. The politicians in strongholds are likely to be re-elected and hence would be able to affect the reassignment of the bureaucrats. On the other hand, politicians in non-strongholds are less likely to be in office in the next term and hence have a lesser ability to influence their bureaucrat’s promotions. We should therefore find that B2 bureaucrats perform better for stronghold politicians. Table 11 presents the results of this exercise. The base category consists of bureaucrats who are up for promotion within a year in non-strongholds. The first row in panel A shows that the time taken to approve for B1 bureaucrats in strongholds is the same as time taken for B1 bureaucrats in non-strongholds. If we compare the time taken for B2 bureaucrats across these constituencies, we find that time taken to approve is about 14 days less in strongholds. This is 19% of the mean dependent variable. Panel B provides the p-value for this difference. These results provide direct evidence for dynamic incentives: there is no difference in performance when both stronghold and non-stronghold politicians are likely to in office at time of promotion screeinging (before next elections). However, there is a significant difference in performance when one type of politician is likely to be in office to influence promotions and 27 the other type of politician is not likely to be in office. An important point to note is that in addition to providing direct evidence for dynamic incentives, these results also help eliminate other alternative explanations for why, on an average, the projects are approved faster in strongholds. For example, one might expect politicians in strongholds to be of higher ability. That could mean that these politician monitor the bureaucrat more that the non-stronghold politicians and therefore we see better performance in strongholds. However, potential difference in politician ability cannot explain the promotion results. The only explanation that is consistent with promotion results is that politician’s ability to provide incentives to bureaucrat matters for their performance. 6.5 Party Stronghold vs Politician Stronghold Till now we have defined stronghold as those constituencies where the same party had been in power for a decade before our time period of study. If we think about the dynamic incentives mechanism, what matters is the relationship between the politician and the bureaucrat, not the bureaucrat-party relationship. If the results are indeed driven by relationship between politician and bureaucrat, then, these results should remain the same when we define stronghold as politician stronghold rather than party stronghold. In our sample, within the party stronghold constituencies, 68% of them had the same politician win all four terms. Table 12 shows the results of the earlier analysis with politician stronghold dummies. Column (1) shows that in a politician stronghold constituency, time taken to approve projects is about 14% less than the average. This magnitude is similar to one obtained in column (5) of table 9. Column (2) examines the promotion results using the politician stronghold dummy. As before, consider bureaucrats two years away from next election. B1 are those bureaucrats who will be screened for promotion before the elections and B2 are ones that will be screened after the elections. The results show that there is no statistical difference between the time taken to approve by B1 bureaucrats between strongholds and nonstrongholds. However, B2 bureaucrats take 33 days less to approve projects for stronghold politicians compared to non-stronghold politicians. These results are similar to the ones obtained in table 11, but the effects are much stronger. 6.6 Diluted Incentives: Role of Number of Bureaucrats per Politician As we mentioned before, the administrative boundaries do not perfectly overlap with the electoral boundaries. This creates a situation where at any given time, one politician can 28 be working with multiple bureaucrats. In our dataset, there are 386 constituencies and 694 district-constituency pairs. Out of the 386 constituencies, 136 overlap with two districts and 80 constituencies overlap with three or more districts. The rest have only one bureaucrat per politician. We interact the stronghold dummy with number of bureaucrats per politician. The hypothesis is that the more the number of bureaucrats, the worse the performance should be. To see why this is so, let us go back to our main mechanism. The politicians use current and future rewards to incentivize bureaucrats to put in higher effort levels. We can think of the number of rewards as being fixed. This is a valid assumption in our institutional setup since the contracts are implicit in nature are based on relationship between the politician and bureaucrats. These relational contracts require repeated interactions between the politicians and bureaucrats. When there are multiple bureaucrats per politician, it becomes costlier for the politician to implement a high effort level from all bureaucrats. In such a case, since the rewards per bureaucrat reduces, the incentives of bureaucrat to perform get diluted. Hence, we should see that projects take longer to get approved as the number of bureaucrats increase. We now estimate the following model: yibpct = θ0 + θ1 ∗ Strongholdc + 3 X θj ∗ Strongholdc ∗ jBureaucratsc + ibpct (20) j=2 where j refers to the number of bureaucrats. The hypothesis is that θ1 < 0 and θ1 < θ2 < θ3 . Table 13 presents the results of the above specification. As we can see, when there is one bureaucrat per politician, the projects are sanctioned 15 days faster in strongholds than nonstrongholds. However, when we compare the performance of bureaucrats in strongholds with one bureaucrat per politician with strongholds with multiple bureaucrats, the performance is worse when there are multiple bureaucrats. This is especially true when there are three or more bureaucrats. Hence, we cannot reject the hypothesis that θ1 < θ2 < θ3 . 6.7 Speed of Approvals and Project Types Our results show that politicians in strongholds are able to get the bureaucrats to speed up the sanctioning of projects. A valid concern is that politicians in low competition areas may 29 be systematically choosing to implement their pet projects faster. In this section we show that there is no ‘project bias’. Before we proceed to look at specific project types, we first address the question of what the optimal time taken to sanction a project should be. The recommendation of Ministry of Statistics and Programme Implementation is that the decision to approve the project must be completed within 45 days. This threshold is supposed to be an upper bound to how long sanctions can take and is applicable to all types of projects. In order to get a better understanding of the implementation process, we visited 17 districts in India. In each district, we talked to the district magistrate, the director of the development authority and four engineers. Engineers were asked the exact steps that are involved in the sanctioning process and how many days it generally takes. Based on the interviews, we find that the sanctioning process entails the following steps: Step 1: Politician sends project recommendation to bureaucrat. Step 2: Bureaucrat chooses implementing agency and sends proposal to chief engineer. Step 3: Chief engineer assigns a junior engineer to go and inspect the site. Step 4: Junior engineer inspects and submits a feasibility report. Step 5: Chief engineer prepares budget estimate and technical feasibility report and sends it back to bureaucrat. Step 6: Bureaucrat reviews structural and financial feasibility and approves the projects. The bureaucrat (district magistrate) monitors the entire process of sanctioning. She can ask for a status update on whether project sites have been visited, whether the feasibility report has been submitted, etc. If the engineers are delaying the process and the bureaucrat does not keep tabs on the procedures, then administrative sanctions get delayed as well. Once the initial engineering reports are submitted, how long the bureaucrat takes to review the proposal is entirely in her control. According to the engineer’s responses, all the above steps can be followed within approximately 30 days.19 Hence, the median upper bound from the engineer’s responses is 15 days lower than the official guideline of 45 days. These benchmarks suggest that projects that are 19 The median number of days for the entire sanctioning process as reported by the engineers is 30 days while the average is 38 days. 30 sanctioned above 45 days are delays. We now examine how being a stronghold affects the probability that projects are delayed. Column (1) of Table 14 looks at the probability of a project being sanctioned within 30 days. As we can see, the probability is 7% higher for stronghold constituencies. Moreover, column (3) shows that probability that the projects take more than 45 days to sanction is 5% less in stronghold. There is no difference between strongholds and non-strongholds in the probability of a project being sanctioned between 30-45 days. All specifications control for project type. These results suggest that the distribution of time taken to sanction shifts to the left for strongholds: the probability of delays is significantly lower and the probability of sanctioning projects within 30 days is also significantly higher. Table 15 looks at these probabilities across project types. Panel [A] shows that the probability of the projects being sanctioned within 30 days in strongholds is significantly higher for road, water, and irrigation as well as health projects. Also, from panel [C], we can see that these projects are less likely to take more than 45 days to be approved. These results show that all types of projects are approved faster in strongholds. 7 Concluding Remarks In the existing empirical literature, politician-bureaucrat interactions have mostly been analyzed with respect to re-election concerns. This paper examines a hitherto unexplored mechanism that connects electoral competition to bureaucratic performance: access to dynamic contracts. Using a richly constructed data from India and developing a novel empirical methodology to isolate these mechanisms, we show that politicians in low competition constituencies can get bureaucrats to perform better by using long-term contracts. We also show that the future rewards used to incentivize bureaucrats take the form of promotions. While this paper focused on the variations within bureaucrat performance, an equally important issue is the selection of bureaucrats itself: how does electoral competition affect allocation of bureaucrats across different politicians? Although, in India, the initial appointment of bureaucrats to various posts and districts is done by an independent organization headed by the top bureaucrats within a state, politicians can influence the career paths of individual bureaucrats. The results of this paper suggest that bureaucrats respond to politician’s incentives when they are up for promotion. Moreover, Iyer and Mani (2012) show that when there is political turnover, the probability of bureaucratic turnover increases significantly. I study the bureaucrat selection issue in a parallel work-in-progress. 31 With respect to the role of electoral competition in public goods provision, this paper focused on the implementation problem. An equally important issue is how electoral pressures affect the allocation of public goods. In a related paper, Nath (2014a), I examine this allocation problem. Using household survey data from India, I first analyze how stated preferences vary across wealth classes. Then I use the dataset on MPLAD local public goods to show that in absence of re-election concerns, politicians are most likely to spend on projects that are desired by the rich. These results contribute not only to our understanding of how politician-bureaucrat interactions affect policy implementation, but also to the role of political competition. On one hand, it is argued that in autocracies, the politicians are able to get things done because of centralized power, while in democracies - there are too many political constraints which slows thing down. On the other hand, it is argued that democracies are more redistributive. The results of this paper taken together with insights from Nath (2014a) highlight this tradeoff: democratic pressures lead to higher redistribution but also more delays in service delivery. 32 Figure 1: Electoral Constituency Boundaries and Victory Margins for 1999 Elections Figure 2: Overlap Between Electoral Constituencies and Administrative Districts Figure 3: Overlap Between Boundaries: Zooming In