Columbia NCAER Conference on Trade, Poverty, Inequality and Democracy New Delhi March 31 April 1, 2011 Paper 7 India: Election Outcomes and Economic Performance * Poonam Gupta Indian Council on Research in International Economic Relations Arvind Panagariya Columbia University * This paper is a product of the Columbia Program on Indian Economic Policies at the School of International and Public Affairs, Columbia University. The program is supported by a generous grant from the John Templeton Foundation. The opinions expressed in the paper are those of the authors and do not necessarily reflect the views of the John Templeton Foundation.
India: Election Outcomes and Economic Performance Poonam Gupta Arvind Panagariya * This draft: March 8 2011 Abstract In this paper we provide the first analysis of the relationship of growth to election outcomes in India. Using a comprehensive data set consisting of all candidates contesting the election, we also provide the first systematic quantitative analysis of the 2009 Lok Sabha elections. Our key result is that superior growth performance at the level of the state gives a definite advantage to the candidates of the state incumbent party in the constituencies of that state. We offer two additional results: personal characteristics such as education and wealth have at most a small impact on election outcomes; and, at least in the 2009 election, incumbency at all levels candidate as well at the party level in the center and state government, contributed positively to election prospects of a candidate. * The authors are at Indian Council on Research in International Economic Relations, New Delhi and Columbia University, New York. They can be reached at pgupta@icrier.res.in and ap2231@columbia.edu, respectively. We would like to thank Laveesh Bhandari for sharing the data for social and economic indicators of the constituencies with us, the participants of the Annual Growth and Development Conference at ISI, Delhi for many useful comments, and to Ganesh Manjhi and Anjum Khalidi for excellent research assistance. Work on this paper has been supported by Columbia University s Program on Indian Economic Policies, funded by a generous grant from the John Templeton Foundation. The opinions expressed in the paper are those of the authors and do not necessarily reflect the views of the John Templeton Foundation. 1
Table of Contents 1. INTRODUCTION... 1 2. THE KEY RESULT: A QUICK PREVIEW... 5 3. THE RELEVANT LITERATURE... 7 4. SALIENT FEATURES OF THE 2009 ELECTION... 12 5. CHARACTERISTICS OF THE CONSTITUENCIES... 15 6. CHARACTERISTICS OF THE CANDIDATES... 18 6.1. WEALTH... 18 6.2. EDUCATION... 19 6.3. CRIMINAL CASES... 21 6.4. DISTRIBUTION OF CANDIDATES BY GENDER... 23 6.5. INCUMBENTS AND NON-INCUMBENTS COMPARED... 23 7. REGRESSION MODEL AND RESULTS... 25 7.1. THE EMPIRICAL MODEL... 25 7.2. SELECTION OF STATES... 29 7.3. DEFINING INCUMBENCY... 30 7.4. DEFINING THE ECONOMIC PERFORMANCE... 32 7.5. INTERACTION BETWEEN PERFORMANCE AND INCUMBENCY... 35 7.6. CANDIDATE SPECIFIC VARIABLES... 35 7.7. STATE SPECIFIC AND PARTY SPECIFIC VARIABLES AND FIXED EFFECTS... 35 2
7.8. SIZE OF THE PARTY... 36 7.9. PRINCIPAL REGRESSION RESULTS... 36 7.10. FURTHER ROBUSTNESS CHECKS... 45 8. CONCLUDING REMARKS... 49 3
1. Introduction Indian election results often spring surprises. It was particularly the case when the Bhartiya Janata Party (BJP), which led the National Democratic Alliance (NDA) government, unexpectedly lost the 2004 Lok Sabha election. Many critics of economic reforms celebrated the outcome as a vote against the reforms. 1 Since the NDA ally and Andhra Pradesh Chief Minister Chandrababu Naidu, who had been strongly identified with economic reforms, also suffered a bitter defeat in the state elections held simultaneously, this view gained additional currency. An alternative explanation offered for this outcome was the anti-incumbency factor. 2 This view assumed without offering the underlying reason that the Indian voters preferred change to status quo, and hence voted against the incumbent. But the outcome of the 2009 national elections seemingly went against this latter hypothesis: it returned the Indian National Congress, the main ruling party, to power with significantly larger number of seats in the Lower House of the Parliament. This time around, the state elections held in 2009 also returned the incumbent state governments in many states such as Andhra Pradesh, Orissa, Maharashtra and Haryana. These outcomes also seemed to contradict the incumbency disadvantage hypothesis. Election outcomes in India, thus, seem not to show a clear pattern in terms of incumbency disadvantage. At the same time, neither the government of the Congress led United Progressive Alliance (UPA) nor other governments have disavowed the reforms 1 Lok Sabha, translated as the House of People, is the lower house of the Indian Parliament. For purposes of elections to Lok Sabha, the country is divided into 543 constituencies, principally on the basis of population, with each constituency electing one member. Elections to the upper house, called Rajya Sabha, are indirect with the vast majority of its members elected by the state legislative assemblies. 2 The political-economy literature refers to this view as the incumbency disadvantage.
let alone reverse them. Seen this way, the election outcomes in India remain something of a puzzle. In this paper, we take the first stab at a systematic quantitative analysis of the determinants of election outcomes in India using the data for 2009 national elections. Our analysis focuses on the personal characteristics of the candidates such as their wealth and education levels and the role incumbency may play at the level of the candidate as well as parties in power at national and state levels. Most importantly, we ask whether growth at the state level has a perceptible impact on victory prospects of the candidates contesting on the ticket of the party in power in the state. We ask whether the candidates of the main ruling party at the center and state enjoy an advantage in states experiencing superior growth outcomes and suffer a disadvantage in states with poor growth outcomes. Given the relative ease of gathering the candidate-specific data for more recent elections, our analysis focuses on the latest 2009 parliamentary election. The 2009 election is of interest in its own right as well since, like the 2004 election, it too carried a large element of surprise. Given the general disarray in both the Congress-led UPA, which ruled during 2004-09, and the BJP, the main opposition party, predictions of the election results varied widely from marginal victories for the UPA and NDA to the emergence of a Third Front consisting of a group of the left-of-center parties. Yet, defying all forecasts, the Congress greatly increased its tally from 145 to 206 seats and comfortably formed government with a group of smaller parties. To carry out our analysis, we assemble a large new data set covering all 8,071 candidates that contested the 2009 election. The data set includes several relevant characteristics of all candidates, their party affiliation, their incumbency status as 2
candidates, the incumbency status of their parties at the center and in the state in which their constituencies are located, and the relative growth rates of various states. The candidate specific information includes gender, education, wealth and criminal record of the candidates and is compiled from the affidavits that the Election Commission requires each candidate to file with his or her nomination. 3 Our main results may be summarized as follows. First, the 2009 Parliamentary election shows very strongly that controlling for other relevant determinants of election outcomes, the advantage enjoyed by candidates of the incumbent party over those of nonincumbent parties is greater in a state that grows faster than the national average than that enjoyed in a state that grows at the national average. Symmetrically, the advantage enjoyed by candidates of the incumbent party over those of non-incumbent parties is smaller in a state that grows slower than the national average than that enjoyed in a state that grows at the national average. The larger the deviation from the national growth rate, the larger is this effect in either direction. Second, on average, incumbency at all levels was helpful in winning the 2009 election. That is to say, on average, incumbent candidates enjoyed an advantage over non-incumbent candidates, candidates of incumbent national parties over those of non-incumbent national parties and candidates of a state incumbent party within the constituencies of that state over those of nonincumbent state parties. This incumbency effect could be due to a variety of reasons such as the incumbent candidates and parties having more resources to spend on election 3 The Election Commission requires each candidate contesting an election for the Lok Sabha, Rajya Sabha, or state assemblies to file these affidavits since 2002. The first general election when the affidavits were submitted is 2004. 3
campaigns, better name recognition or even being more charismatic. 4 Our results here do not separate the pure incumbency effect on which a great deal of the political science literature focuses from other effects that may be associated with the attributes of and resources available to incumbent candidates. Finally, we also find that on average, more educated and wealthier candidates have a better chance of victory. These advantages turn out to be far more important in the states exhibiting low growth and indeed become statistically insignificant in states exhibiting high growth rates. The idea pursued here is similar to the one proposed in an op-ed article in Wall Street Journal by Bhagwati and Panagariya (2004). Commenting on the trend that shows that anti incumbency seems to have become more dominant in Indian elections since 1991, they propose that in more recent years voters have started taking into account the economic performance to decide whether to vote in favor of or against the incumbents. Whereas in earlier years during the 1950s through the mid 1980s when the overall economic performance in general was not impressive, people saw no perceptible change in their lives, which led them to turn extremely pessimistic in so far as their economic fortunes were concerned. Resigned that a significant change was impossible, their voting decision was perhaps based on other factors, which often resulted in the incumbent Congress Party being voted back to power. With the high growth of the 1980s and thereafter, when incomes began to grow at higher rates on a sustained basis and poverty began to decline, people s aspirations were fundamentally altered: having experienced change for the better, they wanted more of it and sooner than later. And if a current government would not deliver it, they would look for another one. Thus Bhagwati and 4 In future work we plan to include the length of the incumbency to see whether an incumbency fatigue sets in after a long spell of incumbency. 4
Panagariya (2004) propose that in more recent years economic performance has become an important determinant of the way voters behave, and it perhaps explains why anti incumbency has become a more prominent feature of election outcomes. 5 We offer a more detailed discussion of the relevant literature in Section 3. The paper is organized as follows. In Section 2, we offer a quick preview of our main result. In Section 3, we discuss the literature on elections in general and that on elections in India in particular. In Sections 4 we describe some salient features of the 2009 election. In Sections 5 and 6, we summarize the relevant characteristics of the constituencies and candidates, respectively. In Section 7, we present the empirical model and regression results and in Section 8, we conclude the paper. 2. The Key Result: A Quick Preview We find it useful to give a preview of our main result at the outset. This requires us to define the incumbent party at the state level and to define the relative economic performance of the states and divide the states into high- and low-growth states. We define as the incumbent party the main ruling party in power (or two main parties if they shared power) in 2007 and the preceding two or more years. This means that if a state legislative assembly election is held in 2008 or 2009 and the government changes hands, the outgoing party is still considered the incumbent in that state for purposes of the 2009 national elections, which were held in April and May of that year. To group the states on the basis of growth performance, we first identify 19 major states counting Delhi as a state and excluding the union territories, seven northeastern 5 Linden (2004) suggests that the proliferation of parties in recent years and increased competition may have further contributed to voters voting for alternative parties and against the incumbents. 5
states, Sikkim, Karnataka and Jammu and Kashmir. The reasons for the exclusion of these states are discussed in Section 7. But to ensure that the trends reported here are not influenced by the choice of states, we later carry out various checks and find the results to be robust. We calculate the average growth rates in these 19 states between 2004-05 and 2008-09 and rank them in declining order of the growth rates. 6 This allows us to divide the states into three groups of roughly equal number of states exhibiting high, medium and low growth rates. Figure 1: Proportion of the Candidates of the Incumbent Party in the State Winning the National Election (sorted by the states growth rates) 90 80 70 60 50 40 30 20 10 0 Seats won as % of Seats Contested by Incumbent Party (at state level) High Growth Medium growth Low Growth 6 Years 2004-05 and 2007-08 and other similarly expressed periods refer to India s financial year, which begins on April 1 and ends on March 31. Therefore, 2004-05 stands for the period from April 1 2004 to March 31, 2005. 6
Armed with this classification of the states and the definition of the incumbent party, we can ask the following key question: what proportion of the candidates fielded by the state incumbent party in the Lok Sabha constituencies located in that state won the national election? The outcome is depicted in Figure 1. Remarkably, incumbent parties in the high-growth states won 85 percent of the seats they contested. In contrast, those in medium and low growth states could win only approximately 52 and 40 percent of the seats contested, respectively. This strong relationship between growth performance and election outcomes handsomely survives every model modification we consider in our regression analysis in Section 7. 3. The Relevant Literature A large body of the literature on electoral competition developed in the context of the western democracies employs the principal-agent framework and focuses on how the desire to win elections conditions the behavior of politicians. This literature asks how political incumbents might try to maximize their chances of reelection through tax and expenditure policies favorable to their constituencies, cast legislative votes that conform to the ideological make-up of their constituencies and exchange political favors for campaign contributions. 7 Given that our objective is to study the determinants of electoral outcomes rather than incumbent behavior to maximize the chances of electoral victory, this literature is at best indirectly relevant to our work. 7 For example, Rogoff and Sibert (1988) and Alesina and Rosenthal [1989] analyze the use of fiscal and monetary policy actions and Besley and Case (1995) of tax-expenditure choices by incumbents to gain electoral support. Levitt and Poterba (1994) study the effect of Congressional Representation on state economic growth. Levitt (1994), Baron (1989) and Snyder (1990) examine the response of politicians to campaign contributions. Lee (2001) provides additional references. 7
A different strand of the literature examines whether incumbency by itself is an asset or liability in elections. This literature is closer to our paper in that it focuses on the determinants of election outcomes but it is somewhat narrowly focused on the identification of the incumbency advantage. The literature stems from the fact that higher unconditional probability of victory of an incumbent over non-incumbents may be the result of selection bias and therefore need not represent incumbency advantage per se. Conversely, a lower unconditional probability of victory of the incumbent may not represent incumbency disadvantage. Incumbents may win more frequently simply because they happen to be better candidates or have more resources to spend on campaigns. Alternatively, if incumbents lose more frequently than non-incumbents, this may be simply because they fail to keep a number of inconsistent promises made in the prior election or because they prove themselves to be inept during their term. Therefore, the observed frequencies of losses and wins by incumbents are by themselves insufficient to isolate the effect of incumbency. The most compelling approach to identifying the impact of incumbency is regression discontinuity, which tries to identify incumbents and non-incumbents who are otherwise identical in all respects and compares their probabilities of victory in election. 8 The literature that is closest to our work is the one that looks at the impact of economic growth on the reelection prospects of the incumbents. The earlier work on this issue included papers by Fair (1978), Lewis-Beck (1988), Powell and Whitten (1993), and Alesina and Rosenthal (1995). Their results showed that economic growth had no 8 An excellent example of this analysis is Lee (2001). A vast body of political science literature is devoted to the analysis of the incumbency effect in election outcomes. For example, see Erikson 1971, Collie 1981, Garand and Gross 1984, Jacobson 1987, Payne 1980, Alford and Hibbing 1981, and Gelman and King 1990 and Lee 2001. 8
significant impact on the reelection prospects of the incumbents except in the US, where the effect was positive. The more contemporary work on this is by Brender and Drazen (2008). This paper uses cross-country data distinguishing between developed and developing countries and between established and newer democracies. The authors find that economic growth increases the reelection prospects of incumbents in the developing but not developed countries. We address this issue at a more detailed level and in the specific context of India where incumbency can be defined at different levels. 9 We exploit the variation in economic growth across Indian states and measure the extent to which economic performance at the state level determines the election outcome. We also do a more detailed analysis of incumbency by distinguishing incumbency at the candidate level, at the party level in the national government and at the party level in the state government. In addition, we are able to assess the effect of individual characteristics on electoral outcomes. Since the work we do here is conducted at the candidate level it is less susceptible to endogeneity problems. A final important and interesting feature of our analysis is that it relies on data from states that belong to a single developing country, which has been a democracy throughout its history since the independence. Therefore, neither the distinction between developing and developing countries nor the length of democracy, critical to the conclusions of Breder and Drazen, plays a substantive role in our analysis. Thus, we are able to show that even when all units of analysis (Indian states) are developing and have the same uniform record of democracy throughout the relevant 9 Bhalla has frequently relied on economic performance as a tool of forecasting the election outcomes. For example, see Bhalla (1999, 2009). 9
history, differential growth outcomes can have significant influence on the election outcomes. In the Indian context, the literature on the incumbency advantage or disadvantage is relatively new. In an as yet unpublished paper, Linden (2004) uses the regression discontinuity approach and finds that prior to 1991, incumbents had enjoyed an advantage over non-incumbents. But beginning in 1991, this relationship reversed with incumbents suffering a disadvantage. For the elections from 1991 to 1999, he estimates that on average incumbents were 14 percentage points less likely to be elected than comparable non-incumbents. 10 He reaches this conclusion by comparing the probabilities of victory of candidates in an election that had barely won (incumbents) to those of the candidates who barely lost (non-incumbents) the prior election. The underlying assumption is that the candidates that just win and those that just lose an election are identical in all respect and any advantage or disadvantage to a victorious candidate (incumbent) in the following election must result from incumbency. While Linden (2004) studies incumbency disadvantage at the level of the candidate, a number of descriptive analytic studies following the 2004 election have focused on the disadvantage arising from association with an incumbent party. Panagariya (2004), and Yadav (2004) note that on average the state ruling parties performed poorly in the 2004 national elections in the constituencies located in their own states but with one major exception: candidates of parties that had defeated the party in power in a state election held just prior to the national election did well in the latter as well. Yadav characterizes the one to two-year period between the state and national elections when the state ruling 10 Uppal (2005) also finds that incumbency has hurt the candidates in recent Indian elections. 10
party has just come in power as the honeymoon period during which the latter s candidates (i.e., candidates of the recently empowered incumbent party in the state) enjoy a positive advantage. Panagariya (2004) states, The results [of 2004 Parliamentary elections] broadly reflect an anti-incumbency vote principally at the state level. Even where antiincumbency explanation does not apply, the state-level politics rather than a rural-urban split remains the decisive factor. Until recently, Rajasthan, Madhya Pradesh and Chhattisgarh had Congress governments, which had pursued policies centered on rural development, primary education and health. Nevertheless, in the state-level elections in December 2003, the Congress governments in all three states lost by landslides to the BJP and its allies. In the current parliamentary elections, all three states voted overwhelmingly for the BJP and its allies. In the December 2003 state elections, the Congress had managed to retain power in Delhi and it swept there in the parliamentary elections as well. Ravishankar (2009) carries out a quantitative analysis of the prospects of victory for the incumbent candidates of the main party in power relative to the incumbent candidates of the main opposition party using the national and state election data from 1977 to 2005. Because her analysis is strictly restricted to incumbent candidates, it does not compare incumbent and non-incumbent candidates. She finds that setting aside the parties in their honeymoon period, incumbent candidates of the main party in power in both national and state elections face higher probability of loss in their reelection bids than the incumbent candidates of the main opposition party. Ravishankar (2009) also finds a cross effect flowing from party incumbency at the national level to state elections and vice versa. 11
Once again, setting aside the parties in their honeymoon period, incumbent candidates of the main party in power at the center face a higher probability of defeat than the incumbent candidates of the main opposition party at the center. Symmetrically, incumbent candidates of a party in power in a state face a higher probability of defeat in the national election than the incumbent candidates of the main opposition party within that state. A key shortcoming of Ravishankar (2009) is that it excludes non-incumbent candidates. If the incumbency effect is associated with the party in power, there is no reason why it would not apply to non-incumbent candidates contesting the election on the incumbent party s ticket. Our data set, though confined to the 2009 national elections, includes all candidates and therefore allows for more complete test of the incumbency effect at the level of the party. 4. Salient Features of the 2009 Election To provide some background, Table 1 reports the broad results of the elections held in 1999, 2004 and 2009. It shows that the national parties numbering six or more have won only a little more than two-thirds of the seats in each of the three elections. 11 As a result, the party winning the largest number of seats has fallen well short of the majority so that each government has been based on a multi-party coalition. Because the party 11 India has more than one thousand registered political parties. These are divided into national parties, state parties and other (unrecognized) parties. Any registered party that lacks the status of state or national party is an unrecognized party. The Election Commission (EC) confers the status of state party on any party that meets certain thresholds in terms of votes received and seats won in an election. A state party acquires monopoly on the use of its party symbol in the state. A party qualifying as state party in four states gets the national status and then has the monopoly over the use of its election symbol over the entire country. It is not unusual for parties to lose the national status if they lose the qualifications for it. 12
with the second most seats ends up in the opposition, state parties, which together account for approximately 30 percent of the seats acquire great importance. Table 1: Broad Results of the National Elections in 1999, 2004 and 2009 Party 1999 2004 2009 National Parties 369 364 376 Indian National Congress 114 145 206 Bharatiya Janata Party 182 138 116 Bahujan Samaj Party 14 19 21 Nationalist Congress Party 9 9 Communist Party of India 4 10 4 Communist Party of India (Marxist) 33 43 16 Rashtriya Janata Dal 24 4 State Parties (E.g., DMK, TDP, SP, TC) 158 159 146 SP 36 23 JD (U) 8 20 AITC 2 19 DMK 16 18 BJD 11 14 Shiv Sena 12 11 AIADMK 0 9 TDP 5 6 JD(S) 4 3 Other (unrecognized) Parties 10 15 12 Independent candidates 6 5 9 TOTAL 543 543 543 Led by the Bhartiya Janata Party (BJP), the National Democratic Alliance (NDA) had ruled from 1999 to 2004. Counting on its popularity at the time, it called for an early 13
election. But, the BJP suffered major losses shrinking its seats from 182 to 138, whereas the INC, the Congress Party, improved its tally significantly from 114 to 145 seats, though still well short of the 272 seats necessary to form a government. 12 But remarkably, it was successful in cobbling together a coalition that came to be known as the United Progressive Alliance (UPA). The UPA government successfully served its entire term until 2009. At one level, it could be argued that neither the decline in the seats held by the BJP from 182 to 138 nor the rise in the seats held by the Congress from 114 to 145 represented a major shift away from the incumbent towards the opposition. Yet, given the expectations of a clear mandate in favor of a very popular Prime Minister, the media uniformly described the outcome as a decisive vote against the incumbents. The 2009 national election was different from the 2004 election in one fundamental sense: it returned the main ruling party, the Congress party, INC, to power with a larger number of seats as well as with a larger victory margin. Beating even the most optimistic predictions, the Congress increased its tally yet again from 145 to an impressive 206 seats. The Marxist Communist Party suffered the worst losses shrinking from 43 to 16 seats. The BJP also declined from 138 to 116 seats. Thus between 1999 and 2009, the Congress and the BJP had more or less exchanged their positions. Among the national parties, the Marxist Communist Party of India, the Communist party of India, and the Rashtriya Janada Dal (RJD) suffered the largest losses besides the BJP. RJD even lost its status as a national party after the elections. Among the state parties, Samajwadi Party suffered the largest losses. Those making major gains other than the Congress were the Congress ally in West Bengal All India Trinamool Congress and 12 Virmani (2004) offers an analysis of the voter s behavior in the 2004 election. 14
opposition parties JD (U) and AIADMK. The other myriad state parties and unrecognized parties broadly maintained their positions. One immediate reaction to the results in the press was that incumbency had helped rather than hurt in this election, though some observers did question this conclusion. However our observation is that rather than a simple incumbency factor relevant at the central government level the picture seems to be more nuanced, and one in which the incumbency seems more relevant at the state government level and it is the interplay between state incumbency and economic growth that seems to be a determining factor in election outcomes. This idea is related to the Bhagwati and Panagariya (2004) hypothesis that the electorate rewarded the ruling party in a performing state while punishing that in a non-performing state. And as observed by Panagariya (2009) the outcome in the 2009 election seems consistent with this idea, e.g. he points out that the national incumbent, the Congress party, could win only nine out of 72 seats in the states of Bihar, Orissa and Chhattisgarh, which had performing non-congress governments. On the other hand, Delhi and Andhra Pradesh had performing Congress chief ministers and the party respectively bagged seven out of seven and 33 out of 42 seats in those two states. In Rajasthan, the Congress had trounced out an unpopular BJP chief minister less than six months prior to the national elections. In the national election, it went on to win 20 out of 25 seats in that state. We investigate these ideas systematically in the rest of this paper. 5. Characteristics of the Constituencies Table 2 shows the salient features of the constituencies based on all constituencies in the country. Because these features hardly move when we consider only the 19 states on 15
which our regression analysis focuses, we do not show them separately for these latter states. A total of 8,071 candidates contested the 2009 election. Of these, as many as 3,825 or 47.4 percent were independent, another 30 percent affiliated with the national or regional parties and the rest belonged to the unrecognized parties. In the 2009 election, in all 372 parties fielded one or more candidates. Party affiliations in general, especially with a national or state party, mattered most: candidates with a party affiliation accounted for more than 98 percent of the top four candidates and for the majority of the winning candidates. 534 winning candidates out of a maximum possible of 543 had some party affiliation. Only nine winning candidates had contested as independents. Table 2: Description of the Constituencies across All States Average Minimum Maximum Number of Voters 1,319,916 45,981 2,343,012 Number of Candidates 15 3 43 Voter Turnout (%) 59.4 25.6 90.4 Votes obtained by the Top Candidate (%) 43.9 21.3 78.8 Votes obtained by the Second Candidate (%) 34.3 8.7 48.7 Victory Margin 9.7.04 70.1 Votes Obtained by the top four candidates (%) 93.8 62.6 100 The average number of candidates per constituency was 15 with the maximum and minimum number of candidates in any constituency being 43 and 3, respectively. Remarkably, as the latter figure indicates, there was not a single constituency with direct election between two candidates. Countrywide, 59.4 percent of the voters turned up to vote. The maximum turnout was 90.4 percent (in Tamluk constituency in West Bengal) and the minimum 25.6 percent (in Srinagar constituency in Jammu and Kashmir). 16
Constituencies near the higher limit were in West Bengal followed by the North Eastern and Southern states (Andhra Pradesh, Kerala, Tamil Nadu). Those near the lower end were in the states of Jammu & Kashmir, Bihar, Uttar Pradesh and Rajasthan. On average, the winning candidates secured about 44 percent of the votes casted, and the second candidate from top obtained about 34 percent of the total votes. Ms Sushma Swaraj of the BJP won with the highest proportionate majority in Vidisha constituency of the Madhya Pradesh, claiming 78.8 percent of the total votes casted. The simple average of the percentage point victory margins across all constituencies was 9.7 percent and the margins ranged from 0.04 to 70 percentage points. Figure 2: Percentage Votes Obtained by Candidates at Different Ranks Percent of Votes Received 0 20 40 60 80 0 10 20 30 40 Candidates ranked, 1=winner The top four candidates summing to 2,170 out of a total of 8,071 candidates accounted for the bulk of the votes polled in most constituencies. In aggregate, these 17
candidates accounted for more than 90 percent of the total votes polled. This can be gleaned from the fact that the density of votes in Figure 2 is heavily concentrated in the first four candidates. In view of this distribution of votes polled, we will frequently limit the sample to top four candidates in our regressions in Section 7. 6. Characteristics of the Candidates We now turn to a consideration of some key characteristics of the candidates relevant to our analysis. These relate to: wealth, education, criminal cases distinguished by the seriousness of the charges, gender, and incumbency status. We provide the data for all candidates, the top four candidates the victorious candidates. Because the average hardly move between all states and 19 largest states, in the following, we confine our presentation to the former. 6.1. Wealth Table 3 provides the distribution of candidates by wealth across five different wealth categories. For each wealth category, column IV shows the percentage of candidates in the top four candidates and column V of those winning the election. Two features of the table stand out. First, candidates from all wealth categories are able to participate in elections and make it to the list of the top four candidates; nearly half of the top four candidates come from the lowest wealth category of 5 million rupees or less. The system does seem to offer an opportunity to run for election without regard to wealth status. Second, the unconditional probability of victory rapidly rises with wealth: while ¾ of the candidates belong to the bottom two wealth categories, only a little more than a quarter of the elected candidates come from the latter. Alternatively stated, 56 percent of the 18
winning candidates possess at least INR 10 million in declared wealth (and perhaps much more in reality). The contrast is brought out most sharply by a comparison of unconditional probability of victory of a candidate in the highest wealth category (6.7 percent) to that of the lowest wealth category (0.4). It bears cautioning, of course, that no causal relationship between wealth and election outcome can be drawn from these data. Wealth can very well be positively correlated with other attributes defining a good candidate in the eyes of the electorate. Table 3: Distribution of Contesting and Winning Candidates According to Wealth (Candidates in All Indian States) Wealth Category Wealth (rupees million) Number of Candidates (% of total) Number of top Four Candidates Number of Candidates Winning Probability of Being in the Top Four Candidates I II III IV: (II/I)x100 Probability of Victory V: (III/1)x100 1 0-0.5 3,176 (39%) 274 14 8.6 0.4 2 0.5-5 2,835 (35%) 642 134 22.6 4.7 3 5-9 700 (8.7%) 329 89 47 12.7 4 9-50 896 (11%) 629 194 70.2 21.7 5 50-higher 464 (5.8%) 296 112 63.8 24.1 Total All wealth categories 8071 2170 543 26.9 6.7 6.2. Education Next, we consider the distribution of candidates by education level. Once again, we identify five education levels, the lowest one being no formal education and the highest one a post-graduate or higher or a technical degree. Table 4 reports the frequency distribution and unconditional probabilities of being in the top four and the winning 19
candidate. Three features of the table are noteworthy. First, contrary to the common impression, most candidates contesting elections have some formal education. Indeed, the vast majority of those contesting have at least gone through the middle school. Table 4: Distribution of Contesting and Winning Candidates According to the Level of Education (Candidates in All Indian States) Education Education level Number of candidates Number of top Four candidates Number of winners Probability of being in top 4 Probability of victory Category I II III IV: (II/I)x100 V: (III/1)x100 0 No Formal Education 134 5 0 3.7 0 1 Up to Class V 964 106 15 11 1.6 2 Middle or High 2,665 495 104 18.6 3.9 School 3 Undergraduate 1,623 603 157 37.2 9.7 4 Post Graduate or 1,984 875 260 44.1 13.1 higher or technical Total All education levels 7370 2084 536 28.3 7.3 The second point to note from Table 4 is that while the proportion of those with an undergraduate or higher degree is approximately half among those contesting, it is more than 80 percent among those winning. Four out of every five members in the 2009 Lok Sabha boast of an undergraduate or higher degree. At the other extreme, while approximately 100 candidates with no formal education contested elections, reflecting the participatory nature of India s democracy, none actually won. Finally, the unconditional probability of getting elected consistently rises with the education level. The biggest jump takes place as we move from high school to a college degree. We remind, however, that as in the case of wealth, this fact need not reflect 20
causation if education is correlated with other factors that make a candidate attractive to the electorate. 6.3. Criminal Cases Perhaps the most interesting characteristic relates to criminal cases pending against the contesting and winning candidates. Table 5 documents the relevant data. In constructing the table, we identify five categories based on the number of pending cases against a candidate. Two features of the table stand out. First, a significant number of candidates approximately 14 percent have criminal cases pending against them; and even a larger percentage of elected members of Lok Sabha approximately 30 percent have one or more criminal cases registered against them. Even if we exclude the candidates with just one case since the prospects of frivolous cases are high against those in politics, more than 80 current members of Lok Sabha, accounting for approximately 15 percent of all members, have two or more criminal cases pending against them. Second, somewhat disconcertingly, the within group probability of victory is higher for the candidates with a large number of cases pending against them. We note that closer examination leads to a more nuanced picture from the one emerging from the aggregate data shown in Table 5. The criminal charges against the candidates range from the benign such as participation in rallies declared unlawful to more serious ones such as murder. To understand the true picture, we must disaggregate the data further. To economize on space, we relegate this task to an appendix available upon request from the authors. Here we simply note that the probability of a serious crime by a contestant rises with the number of criminal cases registered against him or her. Whereas only 40 percent of the candidates with one case registered against them had 21
been accused of a serious crime, nearly 90 percent of those with 10 or more pending cases had one or more serious criminal charges against them. Table 5: Distribution of Contesting and Winning Candidates According to Criminal Cases (Candidates in All Indian states) Crime Category Number of Criminal Cases Number of Candidates Number of top Candidates Number of Winners Probability of Probability of Being in Top Victory 4 candidates I II III IV: (II/I)x100 V: (III/1)x100 0 0 6,894 1,578 381 22.9 5.5 1 1 627 286 76 45.6 12.1 2 2 to 4 392 212 59 54.1 15.1 3 5 to 9 92 61 16 66.3 17.4 4 >10 44 30 11 68.2 25 Total All crime categories 8049 2167 543 26.9 6.7 Detailed examination also reveals shows some concentration of criminal cases against candidates by state. The state that tops the chart is Bihar with more than a quarter of the candidates having at least one criminal case and more than 17 percent of the candidates with at least one serious charge against them. Other states, which exhibit large proportions of candidates with criminal cases, include Jharkhand, Orissa, Uttar Pradesh, Gujarat, West Bengal and Maharashtra. We have an interesting case in Kerala where 22 percent of the candidates have criminal cases registered against them but the vast majority of them involve benign charges such as participation in demonstration; in only a third of the cases involve serious crimes. 22
6.4. Distribution of Candidates by Gender Table 6 reports the gender distribution of contesting and victorious candidates. The data show the expected pattern. Much fewer women than men contest election. This also translates in many more male members in the Lok Sabha. One mildly interesting feature is that the unconditional probability of a woman winning the lection is higher than that of a man. Table 6: Gender Composition of Contesting and Winning Candidates (All States) Gender Total Candidates Top Four Candidates Number of Winning Candidates Probability of being in Top Four Probability of Victory I II III IV: (II/I)x100 V: (III/I)x100 Women 556 183 58 32.9 10.4 Men 7,515 1,987 485 26.4 6.5 Total 8071 2170 543 26.9 6.7 6.5. Incumbents and Non-incumbents Compared We may now compare the incumbent and non-incumbent candidates within the populations of all, top four and victorious candidates. As table 7 shows, while nonincumbent candidates far outnumber incumbent ones, virtually all of the latter are among the top four. With 15 candidates per constituency contesting election on average, it should be no surprise that even if half of the incumbents were voted out, the unconditional probability of their victory relative to non-incumbents would be very high. Therefore, losses to a large number of incumbents are quite consistent with the incumbents having a strong showing in a statistical sense. In a similar vein, even as the main parties such as the Congress and the BJP might experience a decline in their tally of 23
seats, the statistical probability of their candidates winning would still remain very high relative to the rest of the main parties taken together. Thus in our regression we include a control variable representing the size of the party. Table 7: Incumbents among all contestants, top four and winners (All States) Incumbency Status Total Candidates Top Four Candidates Number of Winning Candidates Probability of being in top four Probability of victory I II III IV: (II/I)x100 V: (III/I)x100 Incumbents 387 376 184 97.2 47.5 Non incumbents 7684 1,794 359 23.3 4.7 Total 8,071 2170 543 26.9 6.7 Table 8: Average of the Characteristics Across Various Candidates (All States) Characteristics All Candidates Top four candidates Winning candidates Age 46 51 53 Wealth Category (Average wealth in rupees) 2.1 (17 mln INR) 3 (41 mln INR) 5 (59 mln INR) Criminal Record (probability) 14 27 30 Serious Crime (Probability) 7.4 13 13.8 Member, national party (probability) 20 60 69 Member, state party (probability) 9 19 27 Male (probability) 93 92 89 Education (category) 2.6 3.1 3.2 Incumbent (probability) 4.8 17.3 34 Next, in Table 8, we provide the average of each characteristic across all candidates and the winning candidates. If we could construct a winning candidate with these average characteristics, he would be a wealthy male (with mean assets worth 59 million 24
rupees and median assets worth 12 million rupees) in his mid 50s with at least an undergraduate degree. He would come from one of the main political parties. There is a 30 percent chance that he would have at least one criminal case against him and a 15 percent chance that he will have 2 or more criminal cases against him, and a 14 percent chance that the case would involve a serious crime. There is also 34 percent probability that he had served as an MP in the previous parliament. 7. Regression Model and Results We now turn to a quantitative analysis of the election results using data at the candidate level. We first outline the empirical model we employ followed by the discussion of various implementation issues such as the selection of states, definition of incumbency and the measurement of economic performance and candidate characteristics. 7.1. The Empirical Model Because we only observe the number and percentage of votes obtained by each candidate and the outcome of the elections rather than the direct behavior of the voters, we use a latent variable approach to model the voter s decision to vote. We include three kinds of variables in the model: candidate specific characteristics, party specific attributes including their ideologies and performance of the incumbent candidates and their parties. Consider each of these in turn. First, the voters voting behavior may have preference for certain types of candidates, leading them to vote on the basis of their characteristics; e.g., some voters may prefer a more educated candidate over a less educated one, or for a woman candidate, some voters may not want to vote for a candidate who has criminal or 25
serious criminal charges against him, etc. Wealth can be an important factor affecting the voter s behavior as it may directly have a bearing on the voter s preference for a certain candidate, and can perhaps also be used, through general or targeted election expenditure to influence the voter s behavior. Second, the voters may have some ideological leanings, leading them to favor one party over the other parties. To the extent that these preferences may differ across states and parties, we control for state fixed effects and party fixed effects in the regressions. In some specifications we control for fixed effects varying over state-party. Finally, besides the observable characteristics of the candidates, and their ideological preferences, voters may also infer about the competence of the candidates or the parties by examining the performance record of the incumbent candidates or their parties. However, there is very little information by way of judging the past performance of the incumbent candidates. In the absence of more direct information, the voters might draw inferences about the competence of a candidate through observed attributes such as experience, gender, education, wealth, party affiliation, criminal record etc., the variables which we include in the model. 13 Since judging the competence of an individual candidate is difficult, the voters might take into account the performance of the party at the state as a proxy for the collective competence of the party. We measure incumbent party s performance by the growth record of the state. 13 In principle, an incumbent candidate s performance can be judged by variables such as the presence in parliament, the participation in debates through the number of questions asked, frequency of visits to the constituency and efforts to assist the constituency members through government funded projects and programs. One could also look at the amount spent and the quality of work done using the funds available under the MPLAD scheme. But these data are difficult to come by. Furthermore, a casual look at the data on the disbursements of funds under MPLAD suggests that there is little variation in the amount spent across the members of parliament. The number of projects undertaken runs into hundreds, and again one cannot judge their quality based on the description of these projects. 26
Formally, we represent the probability of winning elections as follows: Prob(Y c,s,p =1)=G(Candidate characteristics, Party characteristics, Candidate and state performance) Or, more specifically, (1) Prob(Y c,s,p =1)=α + β c X c + β p X p + β s X s + γ c Incumbency c + γ s Incumbency s + γ n Incumbency n + δ c Incumbency c *Economic Performance+ δ s Incumbency s *Economic Performance + δ n Incumbency n *Economic Performance Here Y c,s,p refers to the election outcome of candidate c, belonging to party p, who is contesting from state s. It is a binary variable taking the value of 1 in case of victory and 0 in case of defeat. X c is a vector of candidate-specific variables such as wealth, education, gender, number of criminal cases pending and candidate level incumbency. X p, likewise, is a vector representing party-specific variables, and X s, is a vector representing state specific variables. Besides controlling for selected variables for state or party specific effects, we include state fixed effects and party fixed effects in the regressions. The next three variables measure the effect of incumbency at candidate, state, or national level, respectively; and the final three variables measure the state s economic performance has on the probability of victory of the incumbent candidate, the candidate of the national incumbent party and the candidate of the state incumbent party. The coefficient δ s measures the advantage an incumbent enjoys over nonincumbents in a fast growing state relative to a reference unit of analysis. For example, suppose we represent economic performance by the difference between the annual growth rate of the state and the national average growth rate. Then the coefficient 27