ISSN (Print) - 0012-9976 | ISSN (Online) - 2349-8846

A+| A| A-

The Cost of Ruling: Anti-Incumbency in Elections

"Anti-incumbency" is the most frequently cited reason for why ruling parties face poor odds of getting re-elected in India. Drawing on the comparative politics literature and using electoral data from 1977 to 2005, this paper analyses the performance of ruling parties in national and state elections in India. The findings are that incumbent members of Parliament from national ruling parties and legislative assembly members from state ruling parties are less likely to win than incumbents from the opposition when they come up for re-election. The paper also measures the "honeymoon period" effect, namely, the advantage that candidates from the state ruling party enjoy in national elections that are held early in the state government's term and candidates from the national ruling party enjoy in state elections. India's patronage-based democratic system and federal structure creates incentives for voters to favour the same party for national and state office and coordinate their votes. However, the honeymoon period is short-lived, and the positive effect turns into a negative penalty within two years of a party's term in office.

SPECIAL ARTICLE

The Cost of Ruling: Anti-Incumbency in Elections

Nirmala Ravishankar

“Anti-incumbency” is the most frequently cited reason for why ruling parties face poor odds of getting re-elected in India. Drawing on the comparative politics literature and using electoral data from 1977 to 2005, this paper analyses the performance of ruling parties in national and state elections in India. The findings are that incumbent members of Parliament from national ruling parties and legislative assembly members from state ruling parties are less likely to win than incumbents from the opposition when they come up for re-election. The paper also measures the “honeymoon period” effect, namely, the advantage that candidates from the state ruling party enjoy in national elections that are held early in the state government’s term and candidates from the national ruling party enjoy in state elections. India’s patronage-based democratic system and federal structure creates incentives for voters to favour the same party for national and state office and coordinate their votes. However, the honeymoon period is short-lived, and the positive effect turns into a negative penalty within two years of a party’s term in office.

Research was made possible by grants from the South Asia Initiative at Harvard University. The author thanks Karthik Muralidharan for his collaboration on the data collection efforts and Gary King, Jorge Dominguez and Devesh Kapur for their helpful comments and suggestions.

Nirmala Ravishankar (nirmalar@u.washington.edu) is with the Institute for Health Metrics and Evaluation, University of Washington, Seattle, United States.

I
ndian voters routinely vote ruling parties out of office. Journalists and political commentators have coined the term “anti-incumbency factor” to describe this regular outcome of national and state elections. Election reporting in the popular media has become increasingly focused on determining whether there is an anti-incumbent mood amongst voters. Reports of election results and subsequent poll analysis are framed in the context of whether or not ruling parties managed to survive the anti-incumbent vote. In the most recent state elections, for example, the Bharatiya Janata Party (BJP) is said to have fallen victim to the anti-incumbent vote in Rajasthan, but successfully countered it in Madhya Pradesh. In Delhi, it was counting on an anti-incumbent vote against the ruling Congress Party to regain power after two terms in opposition, but failed.

The cost of ruling faced by national and state ruling parties is the focus of this study. It draws on the comparative politics literature as well as studies of Indian electoral politics to formulate hypotheses about the anti-incumbency factor in Indian elections and provides empirical estimates of the antiincumbency vote in state and national elections using electoral data from 1977 to 2004. It distinguishes between the “direct effects of ruling”, which relates to how ruling parties perform when they seek re-election to the same office, and the “cross effects of ruling”, which refer to the effect of being in power at the centre on the party’s performance in state elections (and vice versa). The study measures the direct and cross effects of ruling by comparing the probability of re-election for incumbent members of Parliament and legislative assemblies from ruling parties against those from non-ruling parties. The estimation strategy relies on logistic regression analyses using constituencywise electoral data.

The main finding is that while the direct effects of ruling are always negative, the cross effects of ruling are mixed. Being from the national ruling party has a negative effect of 9 percentage points on an incumbent candidate’s probability of re-election in national elections. The direct cost of ruling in state elections is even higher; incumbents from state ruling parties are 14.5 percentage points less likely to win in state elections than incumbents from other parties. The cross effects of ruling – the effect of sharing a party label with the state ruling party in national elections and vice versa – switch from being positive in the first half of the ruling party’s term to being non-existent or negative in subsequent years. In other words, candidates from the state ruling party enjoy an electoral advantage in national elections that are held early in the state government’s term. However, this positive effect of incumbency fades as the state government’s term progresses and the next election rolls around. The cross effects of

march 7, 2009 vol xliv no 10

SPECIAL ARTICLE

national rule in state elections display a similar pattern, though the honeymoon period is shorter.

The paper is presented in the following manner. Section 1 discusses the anti-incumbency phenomenon in Indian elections, drawing upon existing studies of the costs of ruling in democracies worldwide and Indian electoral politics. Section 2 explains the estimation strategies employed in the paper and Section 3 presents the empirical findings. A discussion of the key findings in the context of recent state assembly elections and the upcoming national elections follows in Section 4.

1 The Costs of Ruling and the Indian Case

Anti-incumbency is not a uniquely Indian phenomenon. Early studies of elections in the United States found that voters retrospectively assess the performance of incumbents, and “throw the rascals out” to express their displeasure (Erikson 1988). The fate of ruling parties has since become the focus of two bodies of literature in comparative electoral studies. The first, the cost of ruling literature, is built around the empirical finding that in most western democracies ruling parties suffer a loss of support when they seek re-election (Paldam and Skott 1995). The most widely accepted explanation for this phenomenon relates to electoral arithmetic. In order to win, a party must form a majority coalition of different groups of voters. Once in power, the party is unable to honour all the promises it made to these disparate groups and in the process loses the support of some groups. At the same time, opposition parties are better positioned to attract new groups of voters and form winning coalitions because they can make inconsistent promises (Mueller 1970).

While the cost of ruling literature focuses on the direct effects of ruling on the party’s performance in the next election to the same office, there is a second, considerably larger body of research on the indirect effect of controlling one office on election races for other elected offices. The most famous manifestation of this indirect effect is the midterm phenomenon in US elections (Erikson 1988). The party controlling the While House has historically lost seats in midterm congressional elections that occur during an incumbent president’s term in office. The leading explanation for the midterm effect is the theory of divided government, which hypothesises that voters prefer to moderate national policy by giving control over different government institutions to competing parties (Alesina and Rosenthal 1994).

A more generalised version of this explanation for the midterm effect can be applied to all instances of non-concurrent elections. Voters use staggered elections to different levels of government and different offices within the same level to achieve balanced political representation. Consequently, it hurts to be the party controlling the national executive because voters switch their votes over to the opposition party in other races as a way of keeping the ruling party in check. Scholars have applied this notion of balancing to study staggered elections for different levels of government in federal countries like the US, Canada, and Germany, and have found that national ruling parties typically fare poorly in state elections (Kedar 2006; Erikson and Filippov 2001; Lohmann et al 1997).

Economic & Political Weekly

EPW
march 7, 2009 vol xliv no 10

While theories about the costs of ruling and electoral balancing find support in a host of industrialised democracies, they have yet to be tested in developing countries. India is the largest democracy in the world and one of the few developing countries with a tradition of elections dating back six decades. It is a federal country with non-concurrent elections for different levels of government. It has a majoritarian electoral system like the US and Canada, and a parliamentary system of government like Canada and Germany. Hence, India offers a good test case for these theories.

The nature of electoral competition in India has undergone dramatic changes in the last 50 years. From 1950 to the early 1970s, the Congress Party dominated national and state politics. Between 1970 and 1990, however, the party began facing greater opposition in both national and state assembly elections. It lost state elections to regional parties in multiple states and its seat share in the Lok Sabha declined dramatically. The final transition to multiparty competition started in the late 1980s and was complete by the mid-1990s. Increased competitiveness over the last few decades has resulted in the frequent defeat of national and state ruling parties, which in turn has contributed to the widelyheld belief that there is an “anti-incumbency factor” at work in Indian elections. The present section analyses this conception of anti-incumbency and formulates hypotheses about the direct and cross effects of ruling in India’s federal system.

In only two out of the nine national elections held since 1977 did the national ruling party win another term in office. With the exception of the national election in 1984, when the assassination of Prime Minister Indira Gandhi led to a huge sympathy vote for the Congress, the ruling party has always seen a decline in its vote share in the next election. The story is similar at the state level; with the exception of the state of West Bengal, where the Communist Party of India (Marxist) has ruled uninterrupted since 1977, ruling parties typically lose their re-election bids and a pattern of alternating ruling parties has emerged in many states.

The term “anti-incumbency factor” was born to describe this regular feature of Indian elections. It has been used primarily in two contexts. First, anti-incumbency is invoked to describe the direct costs of ruling in state assembly elections, specifically the poor performance of state ruling parties. The second context in which anti-incumbency is frequently cited involves the cross effects of state incumbency on national elections. In this application, the performance of state ruling parties in national elections is used to explain the national verdict. Hence, state-level election outcomes, both in state assembly and national elections, are the focus of the popular anti-incumbency explanation.

Despite its widespread usage and popularity, the so-called antiincumbency factor has received surprisingly little scholarly attention. Yadav (2004) provides the only substantive discussion of this now prevalent explanation for election outcomes, which he does in the context of the 2004 Lok Sabha election. The BJP-led National Democratic Alliance ruling coalition lost despite the incumbent prime minister having higher popularity ratings than any of his opponents and the government having ruled during a period of economic expansion. Yadav contends that there was no national swing against the ruling BJP, which relates to his

SPECIAL ARTICLE

argument about the de-nationalisation of politics in the last two decades. Since the 1980s, most states have developed their own party systems and unique political logic. He contends that national outcomes in this era of diverging state party systems are aggregations of several different state-level verdicts. Hence, he analyses the anti-incumbent vote at the state level to explain the outcome in the 2004 national elections. The reason for why the popular anti-incumbency explanation is state-centric becomes clear from this analysis. As the locus of electoral politics has shifted from the national arena down to the states, state-level factors, including the consequences of being the state ruling party, have become increasingly more important for explaining both state and national election verdicts.

The basic anti-incumbency hypothesis, which Yadav calls the simple anti-incumbency explanation, is that state ruling parties perform poorly in state and national elections. Yadav finds this explanation inadequate for explaining the outcomes of the 2004 general election since in only seven or eight of the 18 major states did the ruling party in the state perform poorly. He proposes an alternative incumbency-related explanation, which takes into account the “honeymoon” or the grace period immediately after a state election when the state ruling party enjoys a boost in support. He observes that when the national election follows closely on the heels of a state election, there is a positive effect whereby the state ruling party picks up seats. This amended version of the incumbency explanation captures the election results in a majority of states, though still not all.

At the heart of both these incumbency-based accounts is an electoral logic that is akin to the cross effects of incumbency introduced previously. In the simple incumbency explanation, being in office at one level of government is believed to have an adverse effect on the party’s performance in elections for another level of government. Yadav proposes a more nuanced approach to the cross effects, suggesting instead that the effect is positive during the early part of the state ruling party’s term but turns negative after the “honeymoon” period has passed.

To understand the reasons for the “honeymoon” effect, we need to take a closer look at the nature of electoral politics in India and how it interacts with the federal structure of the Indian state. Scholars of Indian politics have described the country as a patronage democracy (Chandra 2004; Weiner 1989). Patronage rather than policies defines the relationship between voters and parties. Political parties cannot be differentiated spatially along most policy dimensions. Their manifestoes typically make similar official promises about improving governance, controlling inflation, and reducing poverty, and the parties agree on most macroeconomic issues and matters of foreign policy. Instead, they seek voters by offering to use state office to reward supporters with targeted benefits. The Indian state directly and indirectly exercises considerable control over the daily lives and livelihoods of citizens. Politicians use public sector jobs, pork barrel spending, government subsidies, constituency service, and other forms of targeted benefits to attract voters. Even ethnic parties that claim to represent specific social groups draw their support not on ideological grounds but by promising state benefits (Chandra 2004).

The Indian federal system creates different opportunities and capacities for patronage at different levels of government. Since most areas of public administration fall under the purview of state governments, the state ruling party wields considerable influence over government administrators and other public services providers. They also control access to jobs in the public sector and state subsidies. Consequently, incumbents at the state level have a greater potential for providing patronage than their national counterparts.

Intergovernmental transfers are another way in which the federal institutional structure can impact both the capacity for patronage and voter incentives. Research by Khemani shows that the “opportunistic” central government grants more discretionary funds and deficit financing to state governments controlled by the same party (2003; 2001). Most government spending in India is routed via the state government, even for programmes that are funded by the central government. From an electoral perspective, it does not benefit the national ruling party to transfer additional funds to state governments controlled by opposition parties, since voters in the state are likely to credit the state ruling party and not the national ruling party for the extra spending.

These institutional features have several implications for the direct and cross effects of ruling. First, the direct cost of ruling that national and state ruling parties suffer when they run for reelection need not be the same in magnitude. State governments control more sources of patronage than national governments. To the extent that the cost of ruling follows from incumbents failing to live up to the expectations of different groups of voters, the potential for voter backlash is greater in state elections than national elections. Second, the logic of patronage and the federal relations between the centre and the states suggest that Indian voters would prefer to put the same party in charge of different levels of government. In national elections that follow state elections, voters are likely to vote for candidates from the same party that controls the state government in order to maximise their access to the state government. In state elections that follow national elections, voters have the incentive to vote for the national ruling party so as to maximise intergovernmental transfers to the state. Hence, voters are likely to coordinate their votes in both elections, leading to positive cross effects of ruling for national and state ruling parties.

This incentive to coordinate, however, depreciates in the course of the ruling party’s term. Given that ruling parties are likely to lose after a single term in office, the incentive to put the same party in power at another level of government shrinks as the term progresses. This explains the “honeymoon” effect described by Yadav; we expect ruling parties to gain support in elections to other offices during the early part of their term but lose ground the longer they stay in power. The following section describes the empirical methods and data used to measure the direct and cross effects of ruling in India.

2 Measuring the Costs of Ruling

In principle, anti-incumbent votes in an Indian election could be directed at any or all of the following: the incumbent Member of Parliament (MP) or Member of Legislative Assembly (MLA)

march 7, 2009 vol xliv no 10

SPECIAL ARTICLE

seeking re-election, the incumbent party or the party that won in the previous election from the constituency, and the national or state ruling party. Linden (2004) and Uppal (2005) measured the electoral disadvantages faced by incumbent candidates in national and state elections. They both found that incumbents are less likely to get re-elected than non-incumbents. Barooah (2006) analysed the performance of incumbent parties at the constituency level, and found little evidence of an anti-incumbent vote of this kind. The third kind of anti-incumbency – that faced by ruling parties when they field candidates for re-election – is what is being measured here.

The empirical method used in the paper estimates the costs of ruling at the constituency level, the lowest level of aggregation for which election data is available, while controlling for potential confounders. It measures the effect of being from the national or the state ruling party on an incumbent candidate’s probability of re-election at the constituency level. More specifically, the unit of analysis is the incumbent candidate seeking re-election and the method estimates the effect of being from the ruling party by comparing the probability of re-election between incumbent candidates who are from the ruling party and incumbents from non-ruling parties. By comparing only incumbent candidates who are running on the same party ticket in both time periods and not all candidates, the method controls for local personal votes and local party votes. The assumption is that the two groups

– incumbents from ruling parties and incumbents from all other parties – are similar in all ways that are likely to influence reelection rates other than their affiliation with the ruling party. Hence, any gap between the two in re-election rates is the effect of ruling party affiliation. If being from the ruling party is a source of electoral disadvantage, then the first group should have a lower probability of re-election than the second. However, if there is no cost of ruling, then incumbent parties in the two groups will have the same chances of winning.

Two data sets – one for national elections and one for state assembly elections – were used for the analysis. The data were obtained from the Election Commission of India. It covers the 15 largest states of India and all elections from 1977 to 2004. Data on more recent state elections was not analysed because it was not available for all assembly elections in a format that was comparable to data from previous years. The first election year is treated as the base year to derive incumbency status in the subsequent election year.

The biggest challenge in assembling the data was generating the incumbent candidate variable. This entailed matching the names of the winning candidate from each constituency in a given election to the candidates contesting from that constituency in the next election. The matching procedure involved using a computerised algorithm to first pick out exact name matches and matches when the first and last names are reversed. The second step was “fuzzy” matching using the Levenshtein edit distance, a measure of the “distance” between character strings based on the total number of insertions, deletions, and substitutions required to transform the first character string into the second. All computerised matching was done using the statistical programming language R. All “fuzzy” matches were then manually checked to

Economic & Political Weekly

EPW
march 7, 2009 vol xliv no 10

remove false positives. Since data on by-elections is not incorporated into the data released by the Election Commission, these elections were disregarded. For the purposes of this analysis, all constituencies where the incumbent did not seek re-election on the same party ticket were dropped. The sitting MP contested for the seat in nearly 70% of the seats. Independents were discounted for the obvious reason that ruling party effects do not apply to them.

The dependent variable is binary, indicating whether the incumbent seeking re-election in a constituency won or lost the election. The method relies on a logistic regression model to estimate the effect of being from the ruling party on the incumbent’s probability of re-election. Since coefficients from a logistic regression model are not easily interpretable, the program Zelig (King et al 2006) in R was used to derive the probability of winning for incumbents from ruling parties and incumbents from other parties. The method involves taking repeated draws from the joint distribution of the estimated coefficients and simulating a distribution for the probability of winning. The use of a binary outcome variable instead of other alternative variables like the margin of victory or changes in vote shares is appropriate given that more than two parties are in play in a majority of constituencies. In a multiparty election, an increase (decrease) in the incumbent party’s vote share does not imply success (defeat) in the election. An incumbent party could lose its entire winning margin from the previous election but still not lose the constituency if the votes transfer to a third party instead of going to the incumbent’s main opponent.

The main independent variables are binary indicators of whether the incumbent is from the national ruling party or the state ruling party. For the purposes of this analysis, the national and state ruling parties are defined as the party of the prime minister and chief minister, respectively. A time variable that captures the time difference between the national and state elections is used to test the “honeymoon” hypothesis. For the national elections data, the time variable is coded 0 if the national election took place within a year of the last state election, 1 if the national election took place during the state government’s second year in office, 2 if it took place during the third year, and 3 if the gap is more than three years. Correspondingly, the time variable in the state election data indicates the year of the state government’s term in office during which the national election took place.

The models include as a control variable margin of victory from the previous election, which is likely to be correlated with the incumbent’s performance in the present election. It is also likely that turnout is correlated with whether the incumbent is from the ruling party and has an effect on the incumbent’s probability of re-election. However, turnout in the current election is not causally prior to the party identification of the incumbent. Since turnout from the previous election is positively correlated with turnout in the current election, it is included as a control variable instead. The number of parties contesting is also likely to have an effect on the incumbent’s probability of re-election. The effective number of parties is measured using each candidate’s vote share, following the formula described in Chhibber and Kollman (1998). The effective number of parties

National RP Other parties

SPECIAL ARTICLE

Table 1: Logistic Regression Results for National Elections

results from the logistic regression model National Direct State Cross State Cross ficant effect on an incumbent’s probability

Over Time for the direct costs of ruling in national of winning.

(Intercept) 0.30 -0.02 -0.48

elections. The independent variable meas- 0.41 0.40 0.42 However, this specification for the in the current election could be endogenous to the incumbent seeking re-election. Hence, the effective number of parties from the previous election, which has a strong positive correlation with the effective number of parties in the current election, is used as a control variable. The last control variable is growth in net state product, which is included to account for standard economic voting.

3 Estimates of Direct and Cross Effects of Ruling

The first set of models measure the costs of rule in national elections. The units of observation are Lok Sabha MPs seeking re-election in the same constituency and on the same party ticket. There were 2,474 such cases, ranging across the 15 largest states of India and the last eight national elections. The number of parliamentary constituencies in each state varies according to the population of the state. Hence, an unequal number of cases are drawn from each state, but they are roughly proportional to the population of the states.

The first column of Table 1 shows the

Figure 1: Direct Costs of Ruling in National Elections (Density)

Effect of National Ruling Party Membership in National Elections

25 – 20 – 15 – 10 – 5 – 0 –

0.0 0.2 0.4 0.6 0.8 1.0

Effect of State Ruling Party Membership in National Elections

25 – 20 – 15 – 10 – 5 – 0 –

0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability

second column of Table 1. The predicted probabilities of victory are shown in the bottom graph in Figure 1. Comparing the two density plots in the graph suggests that state ruling party affiliation has no statistically signi

uring whether the incumbent is from the national ruling party has a negative coefficient, indicating that being from the national ruling party reduces an incumbent’s chances of re-election. The top graph in Figure 1 plots the predicted probability of winning based on this model for incumbents from the national ruling party and incumbents from other parties in national elections. Comparing the two

Previous margin 2.90 2.50 3.29cross effects of state rule on national elec

0.45 0.44 0.46

tions does not discriminate between state

National ruling party -0.36

ruling parties that have recently been

0.10

elected and those that have been in office

State ruling party 0.03 0.81

for several years at the time of the na

0.09 0.16

Previous turnout 1.01 1.32 1.64tional election. The “honeymoon” hypo

0.42 0.41 0.43

thesis suggests that, in the first case,

Effective number of parties -0.37 -0.36 -0.34

incumbents from the state ruling party

0.08 0.08 0.07

derive a benefit from their party affilia-

Growth 0.02 0.02 0.04

tion whereas, in the latter case, sharing a

0.01 0.01 0.01

shows that former have a significantly lower probability of winning than the latter. The difference between the two groups is on average 9 percentage points, with the 95% confidence interval ranging from 4 to 13 percentage points. Moreover, while incumbents from non-ruling parties are more likely to retain their constituencies than not (their mean probability of re-election is greater than 50%), incumbents from the ruling party are equally likely to lose as they are to win when they run for re-election (their mean probability of re-election is 50%).

The second model estimates the cross effects of state rule in national elections

Duration 0.01

0.04

Duration of state ruling party -0.33

0.06

N = 2429 Standard errors are in italics below the point estimates.

Table 2: Logistic Regression Results for State Elections

State Direct National National Cross Cross Over Time

(Intercept) 0.24 -1.30 -1.63

0.04 0.18 0.19

Previous margin 0.72 2.69 2.62

0.05 0.20 0.21

party label with the state ruling party becomes a liability. To test this notion, I introduce the time variable, which measures how long the state ruling party has been in office when the national election takes place. The time variable is interacted with the state ruling party membership variable to evaluate if the effect of being from the state ruling party changes over the course of the state party’s term in office. The results from the model are

National ruling party -0.26 0.31 reported in the third column of Table 1.

0.05 0.08

Figure 2 (p 97) shows how the probability

State ruling party -0.14

of winning for incumbents from state

0.01

ruling parties compares with other in-

Previous turnout 0.48 1.95 2.41

0.04 0.19 0.21 cumbents in national elections over the

by including a binary variable indicating whether the incumbent is from the state ruling party. The model disregards the timing of elections. Hence, it tests the simple “anti-incumbency” explanation described

Effective number of parties -0.02 -0.05 -0.05course of the state ruling party’s term in

0.01 0.03 0.03

office. The top-most graph plots the pre-

Growth 0.01 0.02 0.01

dicted probabilities when the national

0.01 0.01 0.01

election happens within a year of state

Duration -0.01

0.02 elections. It shows that the probability

by Yadav (2004), which holds that state Duration of national ruling party -0.27 of re-election is significantly greater for ruling parties fare poorly in national elec-0.03 incumbents from state ruling parties than

N = 7089 tions. The results are presented in the Standard errors are given in italics below the point estimates. other incumbents. The mean “honeymoon”

march 7, 2009 vol xliv no 10

State RP Other parties

SPECIAL ARTICLE

effect in the first year is 18 percentage points, with a 95% confidence interval ranging from 11 percentage points to 25 percentage points. However, this positive effect diminishes with time. Moving down the rows, the graphs in Figure 2 show that the

Figure 2: Changes in the Cross Effect of State Rule in National Elections

Effect of State Party Membership in National Elections Over Time

20 10 0

Other parties State RP

0.0 0.2 0.4 0.6 0.8 1.0 Year 1

20 Other parties State RP
10 0

0.0 0.2 0.4 0.6 0.8 1.0 Year 2

20 10 0

Other parties State RP

0.0 0.2 0.4 0.6 0.8 1.0 Year 3

20 10 0

State RP Other parties

0.0 0.2 0.4 0.6 0.8 1.0 After Year 3

Figure 3: Direct Costs of Ruling in State Elections (Density)

Effect of State Ruling Party Membership in State Elections

50

40

30

20

10

0

0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability

Effect of National Ruling Party Membership in State Elections

50

40

30

20

10

0

0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability

Figure 4: Changes in the Cross Effect of National Rule in State Elections

Effect of National Ruling Party Membership in State Elections Over Time

40 20 0

Other parties National RP

0.0 0.2 0.4 0.6 0.8 1.0 Year 1

40 20 0

Other parties National RP

0.0 0.2 0.4 0.6 0.8 1.0 Year 2

40 20 0

National RP Other parties

0.0 0.2 0.4 0.6 0.8 1.0 Year 3

40 20 0

National RP Other parties

0.0 0.2 0.4 0.6 0.8 1.0 After Year 3

difference between incumbents from the state ruling party and others in terms of their chances of winning switches from being large and positive in the first two years to zero in the third year and negative after that. In other words, the longer the gap between the national election and the state election,

Economic & Political Weekly

EPW
march 7, 2009 vol xliv no 10

State RP Other parties
Other parties National RP

the worse incumbents from the state ruling party perform in national elections.

A second set of models perform similar analyses using results from state assembly elections in the same 15 states. The sample includes 7,258 cases of incumbents who sought re-election on the same party ticket. Since I define the state ruling party as the party of the chief minister, I exclude elections where there was no chief minister immediately before the election was held. This typically occurs when the national government declares president’s rule in a state. The number of elections per state included in the sample varies for this reason and because states are on different election cycles.

Results from the first model that measures the direct effect of ruling in state assembly elections are included in the first column of Table 2. The corresponding graphs of predicted probabilities are in the top row of Figure 3. The probability density plots show that the cost of ruling in state elections is greater than the cost of ruling in national elections. Incumbents from the state ruling party are 14.5 percentage points less likely to win than other incumbents; the 95% confidence interval ranges from 12 percentage points to 17 percentage points. Much like the direct costs of ruling in national elections, the means of the two probability densities straddle the 50% cut-off, which implies that ruling party incumbents are more likely to lose than win, while the opposite is true for other incumbents.

The next model tests for a simple cross effect of national party membership in state elections. The results are presented in the second column of Table 2. The corresponding graph, which is in the bottom row of Figure 3, shows that incumbents from the national ruling party membership do worse than other incumbents during state elections. These calculations disregard the timing of the state elections in relation to the previous national election. The final model takes this timing into account, in order to test for a “honeymoon” effect in state elections. The regression results and corresponding graphs are in the third column of Table 2 and Figure 4, respectively. The graphs show that there is a “honeymoon” period for national ruling parties in state elections, during which incumbents from the national ruling party fare well in state elections. The mean positive effect of 6.5 percentage points is much smaller that the corresponding honeymoon effect in national effects. More significantly, the advantage dissipates by the second year and turns into a liability by the third year.

In sum, Figures 2 and 4 reveal that ruling parties perform well in other elections in the start of their term in office. Contrary to the simple anti-incumbency account, they benefit from a “midterm” gain in support during this period. These results support the hypothesis that voters have the incentive to elect the same party into office at both levels of government, but this incentive depreciates as the term of the incumbent party approaches its end.

4 Conclusions

This paper measures the anti-incumbency factor in Indian elections. The estimates presented show that incumbents from the national ruling party are more likely to lose than win when they come up for re-election and are 9 percentage points less likely to

SPECIAL ARTICLE

win than incumbents from other parties. Similarly, incumbents more generally. On the one hand, it reveals that the effects of from state ruling parties are 14.5 percentage points less likely to ruling are not as simple as the popular discourse surrounding the win than incumbents from other parties in state elections. “anti-incumbency factor” would have us believe. On the other it Hence, at both levels of government, the effects of ruling are shows that theories about balanced government developed to exoverwhelming negative, confirming that there are direct costs to plain electoral behaviour in western democracies may not carry ruling in Indian elections. The fact that state ruling parties suffer over easily to other contexts especially in developing countries a higher cost of ruling than national ruling parties is not sur-if the underlying assumptions about what motivates voters are prising given the interaction between patronage-based demo-either not applicable or inaccurate. cratic politics, party competition and India’s federal structure. This research also raises important questions about the Indian voters care more about state politics than national politics of development. The extent to which politicians expect a politics. They are more vigilant about what state governments second term shapes their incentives in their first term in office. do and, according to these results, harsher when it comes to Incumbency advantage in US elections has been a source of voting out incumbents. concern because it implies low rates of turnover and political

The cross effects of ruling, which capture the effect of ruling at competition and it is argued that this makes politicians less reone level of government on the party’s performance at another sponsive. However, incumbency disadvantage turns out to be level, are not consistently negative as suggested by the divided equally problematic. If most politicians do not expect to be government hypothesis. It is not the case that voters attempt to around for very long, they have the incentive to get as much temper the national ruling party by putting an opposition party personal gain out of their current position as possible. Hence, in charge of the state government or vice versa. Instead, they both incumbency advantage and incumbency disadvantage attempt to coordinate their votes, at least initially. During the could adversely affect accountability. early part of ruling party’s term in office, there is a positive There is some evidence from recent elections that the pendu“midterm effect” whereby incumbents sharing the party label lum is swinging away from strong anti-incumbency towards perform well in other elections. However, as this positive effect greater parity. Out of the 24 state elections that have been held diminishes, ruling party membership becomes a liability. Hence, since 2005 in which there was a clear winner (i e, discounting the analysis presented in this paper confirms the “honeymoon” hung verdicts), the state ruling party won 12 and lost 13. The hypothesis and provides empirical estimates for the length upcoming national election in 2009 will be a timely test of of this period and the size of the effect. State ruling parties enjoy whether the anti-incumbency factor and the honeymoon effect a “honeymoon” period of two years after they are elected analysed in this paper were temporary phenomena or are perinto office during which incumbents from their party are more manent features of Indian elections. According to the evidence likely to win than other candidates in national races. Beyond presented in this paper, the UPA government is likely to face a this two year window, incumbents from the state ruling party strong anti-incumbency vote in the next general election. The no longer have an edge over others. Incumbents from the extent to which it can ride out this trend, however, will depend national ruling party have a much smaller honeymoon window crucially on how much it benefits from the positive honeymoon of one year in state elections. effect in states where it recently won office and whether it can

These findings have implications both for the study of electoral build alliances that inoculate it from cross effects of incumbency politics in India and our understanding of retrospective voting in other states.

references – (2003): “Partisan Politics and Intergovernmental (http://www.democ.uci.edu/research/conferences/ Transfers in India”, World Bank Policy Research pluralitypapers/uppal2.pdf).

Alesina, A and H Rosenthal (1994): Partisan Politics, Divided Government, and the Economy (New York:

Working Paper, No 3016. Weiner, M (1989): The Indian Paradox (New Delhi: Cambridge University Press).

King, Gary, Kosuke Imai and Olivia Lau (2006): “Zelig: Sage Publications). Everyone’s Statistical Software”, URL: http:// Yadav Y (2004): “The Elusive Mandate of 2004”, Barooah, V (2006): “Incumbency and Parliamentary

Elections in India”, Economic & Political Weekly, gking.harvard.edu/zelig. Economic & Political Weekly, 39 (51): 5383-95.

41 (8): 739:746.

Linden, L (2004): “Are Incumbents Really Advantaged?

Chandra, K (2004): Why Ethnic Parties Succeed (Cambridge: Cambridge University Press).

Chhibber, P and K Kollman (1998): “Party Aggregation and the Number of Parties in India and the United States”, The American Political Science Review, 92: 329-42.

Erikson, R S (1988): “The Puzzle of Midterm Loss”, The Journal of Politics, 50: 1011-29.

Erikson, R and M G Filippov (2001): “Electoral Balancing in Federal and Sub-national Elections: The Case of Canada”, Constitutional Political Economy,

12: 313-31. Kedar, O (2006): “How Voters Work Around Institutions: Policy Balancing in Staggered Elections”, Electoral Studies, 25: 509-27.

Khemani, S (2001): “Federal Politics and Budget Deficits: Evidence from the States of India”, World Bank Policy Research Working Paper, No 2915.

The Preference for Non-Incumbents in Indian National Elections”, Viewed on 10 December 2008 (http://www.columbia.edu/~ll2240/Incumbency% 20Disad.pdf).

Lohmann, S, D W Brady and D Rivers (1997): “Party Identification, Retrospective Voting, and Moderating Elections in a Federal System: West Germany, 1961-1989”, Comparative Political Studies, 20: 420-49.

Mueller, J E (1970): “Presidential Popularity from Truman to Johnson”, The American Political Science Review, 64: 18-34.

Paldam, M and P Skott (1995): “A Rational-Voter Explanation of the Cost of Ruling”, Public Choice,

83: 159-72. Uppal, Y (2005): “The Disadvantaged Incumbents:

Estimating Incumbency Elects in Indian State Legislatures”, Viewed on 10 December 2008

available at

Alter Media Bookshop

Brahmaswom Madom Bldg, M G Road Trissur 680 001 Kerala Email: info@altermediaindia.com

march 7, 2009 vol xliv no 10

Dear Reader,

To continue reading, become a subscriber.

Explore our attractive subscription offers.

Click here

Back to Top