Economic performance and electoral volatility: Testing the economic voting hypothesis on Indian states,1957–2013

Introduction

In most democracies, political parties compete in elections by promising to provide voters policies that will deliver greater economic prosperity. Hence, it is natural to expect that voters will react to economic conditions and attribute to the incumbent party/coalition some responsibility for the performance of the economy. If the incumbent fails to deliver on its electoral promise, voters can be expected to drop their support for the governing party/coalition and switch their vote to either one of the opposition parties or a promising new arrival. Voters switching parties to punish the governing party/coalition for poor economic outcomes arising during its tenure should then produce an increase in vote volatility in the upcoming election.

The impact of changing economic circumstances on electoral outcomes has been widely studied but primarily within developed economies and primarily with respect to whether the incumbent party or party coalition loses the upcoming election or a portion of its vote share (Brender and Drazen, 2008; Gupta and Panagariya, 2014; Pacek and Radcliff, 1995; Uppal, 2009). While the majority of studies find evidence consistent with the hypothesis that economic conditions matter (for India, see Khemani, 2004), there are a sufficient number of ambiguous findings to suggest that winning versus losing and/or the size of the change in an incumbent’s vote share may be too narrow a measure to capture the full impact of economic conditions on election outcomes (Arcelus and Meltzer, 1975; Bengtsson, 2004; Evans and Anderson, 2006; Ravishankar, 2009; Vaishnav and Swanson, 2015; Verma, 2012). Moreover, economic conditions can generate more effects on voting behavior and electoral stability than just those experienced by the incumbent party. Not only can an incumbent be punished for poor economic conditions without necessarily losing the next election, but disappointing performance may prompt the redirection of votes among established parties and/or between established and newly entering parties. This is particularly important in a country like India where the number of political parties competing in elections is often large. ¹ The existence of multiparty competition means that changes in the vote shares of all competing parties matter and a focus on volatility may allow for a more precise measure of the influence of economic conditions on the stability of the party system and electoral outcomes.

While the relationship between economic conditions and political instability has been a topic of growing interest, empirical work using economic variables to explain political instability is still largely confined to cross-country analysis (Aisen and Veiga, 2013; Alesina and Perotti, 1996; Dassonneville and Hooghe, 2017) and country-specific analysis largely confined to those in developed countries (Bischoff, 2013; Dassonneville and Stiers, 2018; Van Der Meer et al., 2012). Studies focusing specifically on developing countries are slowly beginning to appear (see, e.g. Roberts and Wibbels (1999) and Kuenzi et al. (2019) on Latin America and Africa, respectively) but, as yet, only Heath (2005) and Noorudin and Chhibber (2008) have utilized the diversity of India’s states to analyze vote volatility. Heath (2005) studied the impact of the party system and social cleavages on volatility but does not address the impact of economic outcomes on volatility. The study by Noorudin and Chhibber is more directly comparable to ours. Covering 15 Indian states over the 1967–2004 time period, they argue that fiscal space (i.e. the room within the budget that allows higher spending without jeopardizing financial stability) is the key determinant of statewide electoral volatility and find that economic conditions produce no significant effect on volatility. In this article, we use a different model and estimating procedure to extend their study backward to 1957 and forward to 2013. ² In doing so, we find results that modify, to an extent, the conclusions made by Noorudin and Chhibber.

Our work contributes to the growing literature on electoral volatility in a number of interrelated ways. Perhaps our most important contribution relates to how electoral volatility is measured. Most studies of vote or seat volatility measure electoral volatility by using party outcomes at the aggregate legislature/parliament level. However, Katz et al. (1997) have argued that the use of state- or national-level results may be less meaningful than volatility constructed from results at the constituency level. Beginning at the constituency level allows the analysis to pick up more of the individual-level vote shifting that is lost when calculations begin only at a higher level of aggregation. In our work, we calculate and compare volatilities calculated at both the state assembly and the constituency level. To the best of our knowledge, no study has used constituency-based volatility in India to study the economic reasons for vote shifting nor considered empirically what difference the use of constituency-based volatility makes. ³

Second, we test for the relative significance of two decompositions of the relationship between economic outcomes and vote volatility that have received much current attention but have not yet been related to Indian states. The first is whether voter response is symmetric with respect to the direction of economic outcomes. Here the maintained hypothesis is that the punishment voters inflict on incumbents for poor economic performance is stronger than the reward conferred for producing better economic prosperity (Dassonneville and Lewis-Beck, 2014; Nannestad and Paldam, 1997; Park, 2019; Singer, 2011; Stegmaier et al., 2017). The second is whether the volatility response to economic outcomes is symmetric across two distinct party types: volatility arising from the entry and exit of parties (type A volatility) versus vote volatility arising from voters switching among existing parties (type B volatility). Here the maintained hypothesis is that type A volatility is more sensitive to economic outcomes than type B volatility (Birch, 2003; Golosov, 2004; Sikk, 2005; Mainwaring et al., 2017; Powell and Tucker, 2014; Park, 2019; Tavits, 2008).

In recent work, Kayser and Peress (2012), Fortunato et al. (2018), and Park (2019) have argued that voters use the observed performance arising in other countries as a benchmark for incumbent performance. Aytaç (2018) extends this benchmarking hypothesis from a cross-country comparison of performance to a comparison of current economic outcomes with those arising under earlier periods. ⁴ With data at the state level, we examine the extent to which economic outcomes in adjacent states impact own-state volatility and whether volatility responds to differences in economic performance across adjacent elections.

We also focus on the time interval that matters most for the volatility effect. Broadly speaking, there is a consensus that economic outcomes become increasingly salient as an election-year approaches. There are at least three arguments for this. First, MacKuen et al. (1992) and Erikson et al. (2000) argue that the policies implemented by a government take time to show their effects and voters only become informed of these effects near the election-year. Second, Healy and Lenz (2014) argue that cognitive effort is required to evaluate the performance of a government. As a result, voters simplify the evaluation problem by substituting election-year outcomes for calculating those spanning the entire term. A third perspective relies on different aspects of voter memory, Sarafidis (2007) and Ferris and Dash (2019) argue that voters have an imprecise memory of economic conditions in the early years of a governing term compared to the later years. In an Indian state context, Ferris and Dash find that electorally visible spending in the period closest to the upcoming election matters most for influencing election outcomes. Here we apply the “deteriorating memory” hypothesis by testing whether more recent growth outcomes matter more than the average outcome over the full term.

The remainder of the paper is organized as follows. In the next section, we discuss data sources and the variables used in this study. The third section presents a set of baseline results on the effect of growth on volatility. A number of robustness tests are carried out to check the sensitivity of baseline results. In the fourth section, we analyze the results of the extensions of the economic voting hypothesis. The fifth section presents the concluding remarks.

Data and variable descriptions

The data used in this study come from a wide variety of sources (details and descriptive statistics are given in Online Appendix). The electoral variables on assembly election outcomes are constructed from information collected from the Election Commission of India. ⁵ Data on state populations and gross domestic products (SGDP) come from the Indian Central Statistical Organization, while fiscal data are taken from the Reserve Bank of India Bulletin. The combination of these sources allows us to test our volatility hypotheses on Indian states over the period 1957–2013. More specifically, our data set covers the 179 assembly elections that took place in 14 major Indian states over 56 years. The states and their election periods are listed in Table A2 of the Online Appendix (see Ferris and Dash, 2020).

Following Przeworski and Sprague (1971) and Pedersen (1979), electoral volatility can be defined as

Volatility t ≡ ∑ p = 1 n | v p t − v p , t − 1 | 2

1

where v_pt is the vote/seat share of party p in election t. It measures the net extent of vote/seat shifting among political parties between consecutive elections. ⁶ The volatility index varies between 0 (a stable political system) and 1 (an unstable political system). Using this definition, we calculated vote volatilities at both the assembly and the constituency level (before weighing by each constituency’s share of the total state vote to aggregate upward). Calculating vote or seat volatility at the assembly level is straightforward, however, adjustments must often be made in the constituency-based case because either redistricting or seat expansion generates new constituencies that cannot be matched across time. ⁷ For much of our time period, redistricting was not a concern, but when the calculation of a constituency’s volatility did require a voting outcome from the past election and one was missing, we linked the electoral constituency with its (unchanging) administrative district and used past party averages across that district to construct a proxy past history for each new constituency. ⁸

Figure 1 provides a descriptive overview of the evolution of both constituency and assembly-based volatilities over time for each of our 14 Indian states. Perhaps the most visible features of Figure 1 are that with few exceptions the constituency and assembly volatilities move together in all states and that average volatility at constituency level is usually higher (roughly a quarter higher) than that calculated at the assembly level. While neither shows a strong time trend over the entire period, both volatilities exhibit a steady rise through the 1975–1977 National Emergency followed by a slight downward trend.

OPEN IN VIEWER

Figure 1.

Alternative measures of vote volatility across Indian states, 1957–2013.

A more detailed look at state outcomes indicates that volatility evolves differently across the states. To some extent, this reflects the different rates at which the states reacted to the early dominance of the Congress Party. At the national level, the Congress Party remained undefeated electorally from Independence through the imposition of the National Emergency (between June 25, 1975, and March 21, 1977), but its hold over the electorate was already beginning to break down. However, it was the imposition of the National Emergency that produced a major turning point in Congress’s success. At the national level, the unpopularity of the Emergency led to the first-time defeat of the Congress Party in 1977 by a grand coalition formed under the Janata Party. Similar coalitions were formed at the state level, leading to the Congress Party’s loss of most state elections in 1977 and 1978. The large-scale shift in voting at the state level is reflected in the historically high levels of volatility arising in most states between 1978 and 1980. The elections held after the National Emergency also brought the entry of many new regional and national parties.

Politics in many states is now dominated by competition among state-specific parties with vote fragmentation increasing the frequency of coalition governments. The varying degree of success in forming stable governing coalitions accounts for some portion of the different volatility trends in the post-1977 period.

To measure the impact of economic circumstances in the economic voting hypothesis, most analysts have used macroeconomic indicators such as economic growth, unemployment, and inflation. Because inflation is not a variable under the control of the state government and state unemployment rates in India are generally unavailable, we have chosen the growth rate of state per capita income as the indicator of economic performance most meaningful to state voters. To assess the time interval over which growth rates matter most for voting behavior, we use two time intervals: the growth rate over the incumbent’s governing tenure and the growth rate arising in the last year of the incumbent’s tenure.

Among the political determinates of electoral volatility, the most commonly used determinant is the effective number of parties (ENPs), a measure of party system fractionalization. ⁹ A larger ENP increase the options available to voters and hence are expected to lead to greater vote switching. In India, one of the most ethnically diverse countries in the world, the party system evolves around ethnic factors such as caste (social class), language, and religion (Dash et al., 2019; Harriss, 1999; Heath, 2005; Jaffrelot, 2012; Yadav and Palshikar, 2003). In the absence of a time-series measure for ethnic divisions in India, ENP also serves as a proxy for the degree of ethnic division across states. A change in ENP between elections is then expected to affect volatility in the same direction.

One consequence of electoral mandates becoming increasingly fractured has been greater reliance on coalitions to govern. In practice, most coalition governments are composed of established electoral partners whose members do not compete with each other. Hence, as the number of parties within a government increases, a decline in volatility is expected.

The Congress Party stands out as the most successful political party in India’s electoral history, winning 46% of elections in our sample period. A dummy variable, Congress, is used to differentiate elections won by the Congress Party from those won by others (1, if the Congress party has won and 0, otherwise). The electoral success of the Congress Party is then a signal of the strength of partisan loyalty and likely to have had a negative impact on vote switching.

Volatility will also be affected by a number of special events, institutional rules, and demographic factors. Because of its traumatic effect on Indian politics, a dummy variable, Emergency, is used to account for the first assembly elections arising after the national emergency. In addition, the presidential rule is sometimes imposed on a state when law and order deteriorates or when no party/coalition is in a position to form the government either in the middle of an electoral term or following an election. When it is imposed, the state legislature is suspended or dissolved and placed under direct central rule. ¹⁰ The imposition of presidential rule may then signal a party/coalition’s inability to govern the state. If so, a presidential rule dummy variable (1, if the rule was imposed during an electoral tenure and 0, otherwise) is expected to be linked with increased vote shifting.

In India, the number of electoral constituencies in each state is determined as a fixed proportion of population size. Because more populous states are likely to have a more diversified/heterogeneous electorate and more assembly constituencies, assembly size is expected to have a positive impact on volatility. ¹¹ In addition, the inability to serve out a full-term will often signal a less successful governing period in a Westminster parliamentary system, so that shorter-lived governments would be expected to be followed by a larger amount of vote shifting in comparison with governments that complete their mandated tenure. The number of years passed since last election would then be expected to be negatively related to volatility. To the extent that new and nonpartisan voters are less politically predictable than established voters with stronger ideological and/or partisan preferences (Hansford and Gomez, 2010), each increase in voter turnout would be expected to increase volatility.

Electoral results and vote shifting may also respond to public policy and provide evidence consistent with a political business (or budget) cycle. Studying the impact of public policy on electoral volatility in Indian states, Nooruddin and Chhibber (2008) have shown that constrained fiscal space hampers a government’s ability to deliver public goods and services, resulting in reduced electoral success and a rise in volatility. Here we define fiscal space as the share of revenue receipts left after meeting nondevelopment expenditures. Nondevelopment expenditures include budget items such as interest payments on outstanding debt, administrative services, and pension payments that governments find difficult to cut in the short run (Chakraborty and Dash, 2017). These are considered to be committed spending items.

Finally, because successive elections are not held in the same year across states, we use the date of each election rather than the election number to pick up any systematic time effects. Further detail on the data and the descriptive statistics of each variable are presented in Online Appendix as Table A1. Fisher’s test for panel unit root, proposed by Maddala and Wu (1999), is the appropriate test for the time-series properties of variables in an unbalanced panel like ours. The results are presented in Table A3 in the Online Appendix and can be seen to confirm the stationarity of all our variables.

The effects of growth on vote volatility

In this section, we present and discuss the results of using an AR(1) dynamic fixed effects estimator to analyze the effects of per capita income growth on volatility. ¹² The panel model estimated is

Y i , t = α 0 + α 1 ( Y ) i , t − 1 + α 2 ( X 1 ) i , t + α 3 ( X 2 ) i , t + γ i + e + ε i , t

2

where Y represents the volatility type (constituency or assembly-based), X₁, the growth rate, and X₂ is a vector of control variables. The γ_i‘s are state fixed effects, e is the year in which the election took place, ε is a random disturbance, and the subscripts i and t represent state and time, respectively. All regressions are estimated with standard errors corrected for heteroscedasticity and clustered at the state level to account for autocorrelation. ¹³

Baseline results

The coefficient estimates of three different versions of equation (2) for the two different definitions of vote volatility are presented as Table 1 and discussed as baseline results. In columns 1–3, the estimated impacts of real per capita income growth on constituency-based volatility are reported and the corresponding estimates for assembly-based volatility are reported in columns 4 through 6. Columns 1 and 4 show the estimated effects of per capita income growth averaged over the entire tenure of the incumbent governing party, while columns 2 and 5 do the same for the per capita income growth rate in the last year of the incumbent government. Negative signs of all four coefficients indicate that higher income growth overall and in the immediate preelection period is associated with less vote shifting both at the constituency and assembly levels, but the coefficients are statistically significant (at an acceptable level) only for the constituency level volatility. ¹⁴ These results are then consistent with the hypothesis that Indian voters, despite holding strong partisan positions and social preferences, do hold the incumbent political party accountable for the state of the economy at the subnational level. ¹⁵ Finding that the level of vote shifting captured in the constituency measure is more responsive to changing economic conditions argues for the superiority of the constituency measure in picking up more of the vote switching undertaken by individual voters. That is, aggregation at the assembly-level measure averages out some portion of the vote shifting that the constituency-level measure is able to register.

OPEN IN VIEWER

Table 1.

The impact of income growth rates on vote volatility in Indian states, 1957–2013.

	Volatility (constituency)			Volatility (assembly)
	(1)	(2)	(3)	(4)	(5)	(6)
Income per capita growth (entire tenure)	−0.016* (2.07)		−0.002 (0.21)	−0.02 (1.06)		−0.005 (0.27)
Income per capita growth (in last year only)		−0.012** (2.72)	−0.012** (2.36)		−0.014? (1.7)	−0.012 (1.53)
Δ in Effective number of parties (ENP)	0.327** (2.23)	0.339* (2.1)	0.337* (2.12)	0.792*** (5.06)	0.797*** (4.32)	0.787*** (4.56)
Assembly size	−0.045 (0.14)	−0.089 (0.27)	−0.092 (0.28)	1.18** (2.71)	1.09** (2.66)	1.09** (2.61)
Δ in Voter turnout	−0.448* (1.89)	−0.322 (1.5)	−0.331 (1.57)	−0.257 (1.41)	−0.174 (0.88)	−0.192 (1.1)
Years since election	−0.068 (0.93)	−0.066 (0.89)	−0.67 (0.9)	−0.055 (0.43)	−0.039 (0.29)	−0.042 (0.31)
Number of parties in government	−0.093** (2.38)	−0.101** (2.75)	−0.101** (2.73)	−0.22*** (3.75)	−0.23*** (3.74)	−0.228*** (3.81)
Congress government	−0.029 (0.85)	−0.053 (1.27)	−0.051 (1.21)	−0.164** (2.46)	−0.189** (2.55)	−0.185** (2.7)
Election after emergency	0.213*** (4.45)	0.25*** (4.57)	0.25*** (4.44)	0.523*** (6.26)	0.556*** (6.85)	0.553*** (6.55)
Presidential rule	0.01 (0.22)	0.022 (0.49)	0.02 (0.48)	0.061 (0.58)	0.072 (0.7)	0.068 (0.65)
Fiscal space	−0.108 (0.79)	−0.044 (0.34)	−0.047 (0.36)	−0.321 (1.48)	−0.245 (1.17)	−0.255 (1.31)
Election years	−0.005** (2.17)	−0.004* (1.93)	−0.004* (1.96)	−0.014*** (3.72)	−0.014*** (3.25)	−0.014*** (3.77)
Volatility lag	0.17 (1.6)	0.184* (1.87)	0.181* (1.85)	0.134 (1.65)	0.143* (1.91)	0.137* (1.79)
Constant	8.98* (1.86)	8.9* (1.79)	8.71* (1.8)	20.49** (2.81)	21.57** (2.52)	20.8** (2.83)
R 2	0.33	0.36	0.36	0.4	0.4	0.4
F-stat	21.9***	24.1***	62.3***	237.3***	153.1***	604.9***
Elections (states)	150 (14)	150 (14)	150 (14)	157 (14)	157 (14)	157 (14)

Note: All models include state-fixed effects. ? = just misses significance at 10%. Robust t-statistics are given in parentheses. Standard errors are corrected for heteroscedasticity and clustered at the state level. All variables are in logs except Congress government, Emergency, Presidential rule, and Election years.

*, **, and ***, and denotes significance at 10%, 5%, and 1%.

The results presented in columns 3 and 6 include both growth rates, over the entire period of the incumbent’s tenure and in its final year. If the final year effect is found to dominate the effect corresponding to the incumbent’s entire tenure, then data provide additional support for the deteriorating memory hypothesis. The results in columns 3 and 6 show that while all growth rates have their expected negative signs and both suggest a stronger effect arising in the final period, the volatility response is significantly negative only for the final year’s growth rate and only for the constituency-based measure. For the constituency-based measure, then, our findings complement other studies that find that economic outcomes become increasingly salient as the election-year approaches (e.g. Healy and Lenz (2014) and others).

Among the other determinants of vote volatility, ENP stands out as a significant determinant. The data are consistent with the hypothesis that an increase in party fragmentation increases party options and policy platforms for voters and this will be reflected in an increase in vote volatility. The estimated effects are highly significant in all models. Assembly size, used to capture the scale and heterogeneity of the state, is seen to affect volatility positively as expected but significantly only for the assembly volatility measure. The findings for the change in voter turnout are in line with those found by Nooruddin and Chhibber’s (2008) and suggest that voter turnout does not play as important a role in Indian states as it has elsewhere (see, e.g. Hansford and Gomez, 2010). As expected, the smaller the “years since the last election,” the larger is the vote volatility, but the results are weak and never statistically significant. ¹⁶ On the other hand, the expected effect of a larger size of a governing coalition reducing volatility is significant for both volatility measures. The negative coefficient on the dummy variable, Congress government, is consistent with vote consolidation under Congress Party rule.

The data strongly support the hypothesis that the emergency was hugely unpopular among voters throughout the country. In all models, the coefficient estimates are positive and highly significant. Voters also appear to interpret the imposition of presidential rule as a failure of political governance thus encouraging vote switching. In this case, however, the coefficient estimates never become significant. Like Nooruddin and Chhibber (2008), we find some support for the hypothesis that greater fiscal space reduces vote volatility, but its impact in our model is much weaker and its measured effect never becomes significant.

Finally, the coefficient estimates on election years indicate a significant downward trend in volatility as the party system has matured and stabilized over the years. The positive sign on the volatility lag implies some persistence in volatility shocks over elections and its relatively small size indicates that departures from the equilibrium time path revert back relatively quickly. We note here that the estimated effect becomes significant only at the 10% level and only in those cases when the income per capita in the final years of the incumbent’s tenure is included in the model. The small size of the adjustment coefficient also implies that the long-run coefficients will be marginally smaller in absolute size than these short-run estimates and reestimation of unlagged version of the model confirms the same pattern of coefficient significance. ¹⁷

Extensions and further results

In light of more recent concerns and developments in the literature, we extend the analysis in three different ways. Two of these extensions involve the decomposition of performance and volatility in ways that have been studied in other countries but not yet applied to Indian states. The third investigates whether a spillover onto Indian state vote volatility from concurrent economic conditions arising in neighboring states.

Decomposing economic performance: grievance asymmetry

It is often argued that because individuals are typically risk-averse, a higher weight will be attached to negative outcomes than to similar-sized positive ones (Lau, 1985). Applying this reasoning to voters and economic conditions, the grievance asymmetry literature predicts that voters react more strongly to conditions that are bad than that are good (Dassonneville and Lewis-Beck, 2014; Nannestad and Paldam, 1997; Singer, 2011). To assess the importance of the grievance asymmetry hypothesis for Indian states, we test for a different-sized vote volatility response to positive versus negative changes in per capita state income growth rates. Dividing our two period measures of per capita income growth into two separate variables, one with positive and the other with negative growth rates, the models in Table 1 were reestimated and the results presented in Table 2. As in Table 1, columns 1 and 4 refer to income growth rates arising over the entire governing tenure of the previous government, while columns 2 and 5 focus on income growth rates rising in the final governing year. Columns 3 and 6 use both. Note that to make interpretation easier, positive and negative changes are defined so that a negative coefficient sign signals the inverse response of volatility.

The negative signs found for all combinations of income growth periods (entire tenure and last period) and positive versus negative growth rates suggest that in all cases increases in income growth are rewarded and decreases in growth penalized. However, of all the possible combinations, only those negative income changes arising in the final year of governing tenure are found to have produced a significant effect on vote shifting as measured by volatility. Moreover, the significant effect is found at both the constituency and assembly levels. This confirmation of the grievance asymmetry hypothesis within Indian states is reinforced by seeing that the p values for the negative average tenure growth rates are always higher than that for the positive growth ones. Once again, it should be noted that the effects are somewhat stronger at the constituency level than at the assembly level. Like the baseline results of Table 1, the results of Table 2 support the “deteriorating memory” hypothesis irrespective of whether positive or negative growth rates are considered. In terms of the other determinants of volatility, their results are very similar to those found in the baseline case.

OPEN IN VIEWER

Table 2.

The differential impact of positive versus negative income growth rates on vote volatility Indian States: 1957–2013.

	Volatility (constituency)			Volatility (assembly)
	(1)	(2)	(3)	(4)	(5)	(6)
Positive per capita growth rate (Tenure)	−0.013** (2.26)		−0.011* (2.05)	−0.024 (1.66)		−0.025 (1.53)
Negative per capita growth rate (Tenure)	−0.009 (1.59)		−0.02** (2.4)	−0.017 (1.37)		−0.031** (2.23)
Positive per capita growth rate (Last year)		−0.013* (1.81)	−0.013 (1.57)		−0.007 (0.58)	−0.011 (0.92)
Negative per capita growth rate (Last year)		−0.012** (2.22)	−0.015** (2.65)		−0.02* (1.97)	−0.024* (2.09)
Δ in Effective number of parties (ENP)	0.347** (2.36)	0.341* (2.1)	0.342** (2.2)	0.69** (2.77)	0.769*** (4.44)	0.65** (2.59)
Assembly size	0.035 (0.1)	−0.09 (0.27)	−0.181 (0.58)	1.76** (2.66)	1.09** (2.6)	1.55** (2.45)
Δ in Voter turnout	−0.46* (1.95)	−0.321 (1.49)	−0.314 (1.51)	−0.173 (0.93)	−0.117 (0.76)	−0.239 (0.86)
Years since election	−0.084 (1.24)	−0.067 (0.88)	−0.089 (1.21)	0.009 (0.06)	−0.031 (0.23)	0.006 (0.04)
Number of parties in government	−0.089** (2.15)	−0.1** (2.77)	−0.1** (2.8)	−0.251*** (3.64)	−0.237*** (3.93)	−0.264*** (4.08)
Congress government	−0.022 (0.58)	−0.052 (1.33)	−0.039 (0.96)	−0.136 (1.64)	−0.2** (2.6)	−0.154* (1.83)
Election after emergency	0.218*** (5.35)	0.251*** (4.6)	0.264*** (4.93)	0.484*** (4.45)	0.546*** (6.33)	0.535*** (4.69)
President’s rule	0.012 (0.25)	0.022 (0.5)	0.017 (0.35)	0.123 (1.33)	0.071 (0.69)	0.111 (1.15)
Fiscal space	−0.081 (0.58)	−0.145 (0.34)	−0.021 (0.16)	−0.279 (1.33)	−0.228 (1.08)	−0.208 (1.02)
Election years	−0.007** (2.7)	−0.004* (1.93)	−0.005* (2.06)	−0.017*** (3.37)	−0.015*** (3.29)	−0.015** (2.85)
Volatility lag	0.2* (1.87)	0.185* (1.91)	0.185* (1.88)	0.153* (1.92)	0.139* (1.85)	0.133? (1.71)
Constant	12.8** (2.39)	8.83* (1.79)	10.26* (2)	23.44** (2.44)	23.23** (2.59)	20.98* (2.07)
R 2	0.33	0.36	0.38	0.42	0.41	0.43
F-stat	12***	45***	45***	53.5***	927.3***	927.3***
Elections (States)	150 (14)	150 (14)	150 (14)	157 (14)	157 (14)	157 (14)

Note: See notes to Table 1.

Decomposing volatility: Type A and type B volatilities

Following Birch (2003), Golosov (2004), Sikk (2005), Tavits (2008), Powell and Tucker (2014), and Mainwaring et al. (2017), we divide assembly-level vote volatility into vote volatility arising from vote transfers among established parties (type B volatility) and vote volatility arising from votes shifting from parties exiting to parties that are entering the political system (type A volatility). ²¹ The literature that focuses on this separation hypothesizes that the poor economic performance should destabilize existing parties and make the entry of new parties easier. Hence economic circumstances should affect type A volatility more than type B. The results of estimating our three models on type A and then type B volatility are presented in Table 3.

OPEN IN VIEWER

Table 3.

Decomposing volatility: Income growth rates on type A and type B assembly-level vote volatilities, Indian States: 1957–2013.

	Volatility (type A)			Volatility (type B)
	(1)	(2)	(3)	(4)	(5)	(6)
Per capita income growth rate (Tenure)	0.026 (0.54)		0.045 (0.73)	−0.031 (1.52)		−0.025 (0.99)
Per capita income growth rate (Last year)		−0.006 (0.35)	−0.018 (0.75)		−0.011 (1.43)	−0.005 (0.48)
Δ in Effective number of parties (ENP)	0.361 (0.69)	0.234 (0.39)	0.327 (0.6)	0.939** (2.8)	0.995** (2.9)	0.944** (2.81)
Assembly size	2.16*** (4.62)	2.02*** (3.37)	2.12*** (4.31)	0.59 (1.08)	0.556 (0.92)	0.552 (0.96)
Δ in Voter turnout	−1.39** (2.89)	−1.29** (2.66)	−1.29** (2.81)	0.136 (1.17)	0.202 (1.51)	0.157 (1.34)
Years passed since election	0.125 (0.29)	0.124 (0.29)	0.141 (0.34)	−0.181 (1.04)	−0.168 (0.94)	−0.177 (0.99)
Number of parties in government	−0.121 (0.52)	−0.124 (0.53)	−0.137 (0.59)	−0.109? (1.7)	−0.118* (1.87)	−0.113? (1.72)
Congress government	−0.126 (0.5)	−0.103 (0.44)	−0.133 (0.53)	−0.153 (1.69)	−0.179* (1.89)	−0.161* (1.79)
Election after emergency	0.81** (2.6)	0.868** (2.8)	0.873** (2.91)	−0.25 (1.44)	−0.223 (1.24)	−0.237 (1.29)
President’s rule	−0.051 (0.23)	0.015 (0.06)	0.051 (0.26)	0.091 (0.93)	0.116 (1.28)	0.094 (0.95)
Fiscal space	−1.07* (1.98)	−1.06 (1.59)	−0.852 (1.37)	−0.296? (1.7)	−0.235 (1.53)	−0.272* (1.79)
Election years	−0.013 (0.96)	−0.009 (0.67)	−0.012 (0.95)	−0.007* (1.93)	−0.009** (2.26)	−0.007* (1.9)
Volatility lag	−0.206 (1.66)	−0.227* (1.82)	−0.197 (1.54)	0.076 (0.75)	0.107 (1.11)	0.082 (0.78)
Constant	14.5 (0.55)	6.92 (0.28)	12.06 (0.5)	11.15 (1.2)	14.77 (1.6)	11.24 (1.21)
R 2	0.37	0.37	0.38	0.21	0.2	0.21
F-stat	26.9***	45.1***	45.1***	54.5***	36***	114.16***
Elections (States)	84 (14)	84 (14)	84 (14)	157 (14)	157 (14)	157 (14)

Note: See notes to Table 1.

The results in Table 3 show that per capita income growth rates across Indian states do impact type A and type B volatilities differently, to an extent. However, only in column 4 is a higher income growth rate associated with a significant reduction in volatility and this is with respect to type B volatility rather than type A. Hence the data are not consistent with a fall in growth rates over the incumbent’s governing tenure generating greater volatility by destabilizing existing parties. The data are somewhat more favorable to the destabilizing hypothesis if the per capita growth rate in the period before the election is considered. However, even here the data cannot distinguish the size of the effect produced on type A volatility from that arising in type B. The results do indicate, however, that volatility arising from the entry and exit of parties did increase dramatically following the emergency declaration and fall with the existence of greater fiscal space in state budgets. Greater fiscal space has a similar effect on type B volatility. However, in the case of voter turnout, the results are quite different. An increase in voter turnout is associated with a significant decrease in type A volatility and, while type B volatility does not decrease, it comes close, in the case of last period income growth, to becoming significantly larger. That is, the data suggest that voter turnout is associated with the vote consolidation among the established parties with the desire for greater participation satisfied through internal voting. Finally, while the data suggest that both volatility types have fallen through time, the overall fall as indicated in Tables 1 and 2 is not due to the decrease in volatility arising from the entry of new parties. Rather, it is through a reduction in vote switching among established parties, type B volatility, that overall volatility has decreased significantly.

Neighborhood effects

Studies by Kayser and Peress (2012), Aytaç (2018), and Park (2019) argue that what matters is not so much how well the domestic economy is doing, but how it performs relative to an international and historical reference point. We adapt this theory to the Indian federation by comparing own state outcomes to those arising in neighboring states and those arising in the recent past.

We adapt Aytaç’s (2018) extension of Kayser and Peress’s international reference point model to estimate the following model:

Y i , t = α 0 + α 1 Y i , t − 1 + α 2 X 1 ​ i , t + α 3 X 2 ​ i , t + α 4 Z 1 ​ i , t + α 5 Z 2 ​ i , t + α 6 Z 3 ​ i , t + γ i + e + ε i , t

3

where equation (3) adds to (2) only the Z variables. Z₁ is the difference between own states’ average SGDP growth rate and the average of its adjacent neighbors over the time period of the government’s tenure and measures the neighborhood effect. ²² Z₂ is the average SGDP growth rate of adjacent neighbors over the time period of own government’s tenure. Z₃ is the difference in own average SGDP growth rates between election tenures. ²³ The results are presented in Table 4.

OPEN IN VIEWER

Table 4.

The impacts of income growth on volatility with neighborhood effects Indian states: 1957–2013.

	Volatility (constituency)			Volatility (assembly)
	(1)	(2)	(3)	(4)	(5)	(6)
SGDP growth rate differences with neighbors (Tenure)	0.056 (0.54)		0.065 (0.53)	0.209 (0.93)		0.309 (1.38)
SGDP growth rate of neighbor states (Tenure)	0.066 (0.65)		0.077 (0.63)	0.194 (0.86)		0.302 (1.38)
SGDP growth rate difference between elections (Tenure)		0.004 (0.48)	0.003 (0.29)		0.009 (0.49)	0.001 (0.06)
Per capita growth rate (Tenure)	−0.061 (0.6)	−0.004 (0.29)	−0.071 (0.62)	−0.204 (0.93)	−0.026 (0.69)	−0.314 (1.54)
Per capita growth rate (Last year)	−0.013** (2.45)	−0.012** (2.25)	−0.013** (2.32)	−0.015 (1.65)	−0.011 (1.33)	−0.017 (1.64)
Δ in Effective number of parties (ENP)	0.328** (2.14)	0.344** (2.13)	0.338** (2.16)	0.757*** (3.98)	0.652** (2.48)	0.66** (2.45)
Assembly size	0.022 (0.05)	−0.123 (0.39)	0.008 (0.02)	1.45** (2.21)	1.68** (2.71)	2.27** (2.42)
Δ in Voter turnout	−0.319 (1.5)	−0.318 (1.57)	−0.314 (1.45)	−0.305 (1.64)	−0.221 (0.82)	−0.251 (0.86)
Years passed since election	−0.021 (0.83)	−0.021 (0.89)	−0.025 (1.03)	0.007 (0.16)	0.018 (0.36)	0.014 (0.28)
Number of parties in government	−0.096** (2.32)	−0.103** (2.73)	−0.097** (2.3)	−0.213*** (3.22)	−0.267*** (3.79)	−0.244** (2.81)
Congress government	−0.043 (0.98)	−0.051 (1.18)	−0.042 (0.93)	−0.148* (1.94)	−0.167** (2.17)	−0.123 (1.37)
Election after emergency	0.255*** (3.96)	0.253*** (4.14)	0.257*** (4.07)	0.503*** (4.88)	0.484*** (3.67)	0.468*** (3.68)
President’s rule	0.017 (0.43)	0.022 (0.54)	0.019 (0.47)	0.069 (0.59)	0.136 (1.31)	0.127 (1.12)
Fiscal space	−0.03 (0.25)	−0.038 (0.29)	−0.017 (0.14)	−0.169 (0.89)	−0.271 (1.48)	−0.143 (0.87)
Election years	−0.004 (1.53)	−0.004** (2.3)	−0.004? (1.72)	−0.012* (2)	−0.013*** (3.67)	−0.008 (1.62)
Volatility lag	0.17 (1.61)	0.174 (1.69)	0.162 (1.49)	0.118 (1.6)	0.118 (1.34)	0.095 (1.16)
Constant	7.49 (1.03)	8.94* (2.09)	7.91 (1.1)	11.6 (0.94)	15.56** (2.41)	2.75 (0.21)
R 2	0.37	0.36	0.37	0.41	0.43	0.43
F-stat	140.62***	739.36***	362.94***	90.02***	63.56***	48.97***
Elections (States)	150 (14)	150 (14)	150 (14)	157 (14)	157 (14)	157 (14)

Note: See notes to Table 1.

The results indicate that none of the neighborhood and/or historical effects in any of the six models are significant or even approach significance. It follows that there is little evidence that over our time period growth rate differences from those in neighboring states made any significant difference to how voters evaluated the performance of the party running their own state government and produced any effect on their voting. Similarly, growth differences between tenures do not play a significant role in influencing vote volatility, further reinforcing the “deteriorating memory” hypothesis. ²⁴

Concluding remarks

This article tests the economic voting hypothesis on Indian states by examining how economic conditions impact electoral outcomes as represented by vote shifting and measured by changes in vote volatility. Information from 179 assembly elections held over the 1957–2013 time period across 14 major Indian states was used to calculate vote volatilities both at the state assembly level and at the individual constituency level (before aggregating to form a state average). This electoral data were then combined with political, economic, and budget information from a variety of sources to form a cross-state panel based on state assembly elections. Using per capita income growth as our measure of the economic condition most relevant to state voters, we investigate the effects of income growth on both vote volatilities. The results suggest first that higher-income growth significantly increases electoral stability by reducing the degree of vote volatility. Second, constituency-based vote volatility is found to be more sensitive to income growth rates than is volatility measured at the assembly level, suggesting that the use of state-level measures may hide volatility present but observed only at the constituency level. By finding that vote volatility responds more strongly to the growth rate arising in the final period of the incumbent government than to the average growth arising over the full life of the incumbent government the data also give support for the presence of a “deteriorating memory” hypothesis in the context of vote switching.

As with Nooruddin and Chhibber (2008), we find that increases in fiscal space reduce volatility, but with results that are not as singular or significant as those found earlier. We find other major determinants of volatility to include changes in the expected number of political parties, ENP, the size of the state assembly, and the number of coalition parties. At a purely political level, the period immediately following the declaration of the emergency stands out for its contribution to increasing volatility. Finally, the level of vote volatility has slowly declined over the years, suggesting that within each state the political party system is stabilizing as the country matures. All these results pass a variety of robustness checks.

From the benchmark case, we consider a number of recent extensions. Questioning the symmetrical impact of good and bad economic conditions on vote shifting, we find, in common with others, that Indian state voters punish the negative growth outcomes more severely than they reward similar sized positive outcomes. Once again, it is in the final year of governing tenure where the strongest effect is felt. When we decompose total vote volatility, we find that the growth rate over the entire governing tenure influences significantly only vote switching among established parties (type B volatility). There is a suggestion that the growth rate in the last year of the tenure influences both types of volatility, however, neither of these indicated effects are significant at standard levels of significance. Finally, the data give no support to the hypothesis that vote volatility is influenced by whether or not own state growth performance exceeds or falls short of that arising among immediate neighbors.

It is often argued that voters in developing countries pay less attention to economic conditions than to political factors such as political ideology and/or ethnic considerations involving caste or social class, religion, language, and culture. Such considerations form the basis for writers such as Bratton and Van de Walle (1997), Lindberg and Morrison (2008), Ferree (2010) in Africa, and both Chandra (2004) and Wilkinson (2004) in India. While not minimizing the importance of any of these considerations, our findings imply that economic conditions in India do matter and play an important role in determining voting behavior. Our work suggests not only that theories of economic voting have an important role to play in understanding electoral outcomes but also that the effects of economic outcomes on vote volatility may provide insight into the political business cycles and the stability of the political system.