The Significance Of State Legislative Assembly Elections Politics Essay

Published Date: 23 Mar 2015

Electoral cycles in India are bifurcated between the national level where Parliamentary polls are held to elect the 545 member lower house, or Lok Sabha, for five year terms in single-seat constituencies using a first-past-the-post system, and within the twenty-eight states and two of the seven union territories to state legislative assemblies, or Vidhan Sabhas. State legislative assembly elections may be held during a Parliamentary term or concurrently with Parliamentary elections in the event that the expiration of a legislative assembly's term is proximate to the date of the Parliamentary poll. In the 2009 Parliamentary elections, for example, the states of Andhra Pradesh, Sikkim and Orissa also held concurrent legislative assembly polls. A wide range of parties contest these polls including national parties such as the Indian National Congress (INC) and Bharatiya Janata Party (BJP) that have a country-wide reach and participate in both Parliamentary and state legislative assembly elections, and regional and state-based parties that may contest only legislative assembly elections. In addition, alliances between parties may dramatically alter an election's outcome and constitute an unpredictable and changeable feature of the political landscape. For example, parties may agree to not contest elections in return for a similar undertaking from their alliance partner with respect to a subsequent poll, or the relationship may be more complex with parties in an alliance allocating constituencies to one another in some parts of a state, while competing vigorously against one another in other constituencies in the same state.

The discernment of trends between legislative and Parliamentary elections in such a complex and unstable landscape is highly problematic. Nonetheless, the popular media is often full of speculation concerning linkages between legislative assembly and Parliamentary elections and the significance of the former as a barometer for predicting a party's performance in the latter. For example, the BJP's defeat in the landmark 2002 legislative assembly elections in Jammu and Kashmir and subsequent 2003 polls in Himachal Pradesh, Meghalaya, Nagaland and Tripura caused considerable speculation in the media of an anti-BJP trend that boded ill for the party's fortunes in the looming Parliamentary election. [1] This feeling of foreboding was, however, replaced by conjecture of a BJP victory following the defeat of the INC in 2003 legislative assembly elections in Rajasthan, Madhya Pradesh, Mizoram and Chhattisgarh. [2] More recently, the INC's shock defeat in legislative assembly elections in the states of Goa, Punjab and Uttar Pradesh in early 2012 was held by some to bode ill for its chances in 2014 Lok Sabha elections. [3] In reality, any interconnectedness between the legislative assembly and Parliamentary elections may be mediated by a range of variables. For example, Yadav [4] (one of the few writers to investigate the linkages between voter choice at the state and national level in India) observes that where two elections are held in quick succession, the party that won the first election tends to be victorious in the second. However, he continues, whereas in the 1970-80s the party that won a Parliamentary election then repeated their victory in subsequent legislative assembly elections, [5] the trend was reversed in the 1990s with legislative assembly election results routinely repeated in the subsequent Parliamentary poll. In contrast, other writers note the decline of the INC as the principal political actor in many states, the rise of strong regional parties in states such as Tamil Nadu and the increasing popularity of smaller parties in both Parliamentary and legislative assembly elections [6] in support of a regionalization thesis where legislative assembly elections are decided at the state and even constituency level independent of national politics. [7] Explanatory factors cited in support of this thesis include the decline of the previously dominant one party model which has been replaced by an alternative where alternation in power is either the norm or a real possibility, and the increase in the number of single-state and multi-state parties as a result of the federalization of the party system. [8] 

The purpose of this study is to examine the statistical evidence for the claim of an enduring and significant linkage between legislative assembly and subsequent Parliamentary election results for major political parties in India. This is achieved through an examination of Parliamentary elections held over the period of 1980-2009, and analysis of the degree to which each party's performance is correlated to its share of the vote in preceding legislative assembly elections held eighteen months (548 days) or less previously in the same state. To the extent that there exists a linkage between the two sets of elections we expect to observe a statistically significant correlation between a party's performance in both polls. In contrast, the absence of any significant correlation counsels against the existence of such a linkage. The study incorporates and builds upon insights from previous analyses that have addressed the influence of local elections on subsequent national polls in the United Kingdom (UK), United States (US) and Europe, as well as the inverse hypothesis that national elections impact party performance in subsequent local elections. In the case of India, however, while there has been some investigation of the links between legislative assembly and Parliamentary elections, [9] these studies have concentrated on shorter-term trends and not included advanced statistical tests. A shorter-term approach may result in a more nuanced understanding given the rapidly changing and sometimes cyclical nature of Indian politics. However, it also raises questions regarding the reliability of any findings given the lower number of observations and the possibility of bias in the selection of the time period covered. The longer-term study conducted here that includes data from a thirty year time period covering nine Parliamentary and forty-five legislative assembly polls with a total of 502 observations avoids these concerns and presents a statistically more rigorous and convincing basis for asserting an enduring connection between legislative assembly and Parliamentary polls.

2. Literature Review

A wide range of studies examining the relationship in party performance between local and national polls have produced varying results across a range of geographical contexts. Early studies in the US, for example, found a limited effect from local campaigning on subsequent national elections with a positive influence on voter turnout [10] and in the nomination and pre-election registration stages. [11] A 1999 study of elections in Canada concluded that a preceding local campaign is an important determinant of a party's vote share in a subsequent Parliamentary election, particularly for opposition parties in seats where the campaign was not successful. [12] This finding was consistent with some earlier studies [13] but conflicted with others. [14] In the UK, a range of studies discovered correlations that varied by party. For example, one study uncovered a modest, but statistically significant, impact in national elections associated with a preceding local campaign for the Labour and Conservative parties, [15] whereas another found a positive relationship for the Labour and Liberal Democrat parties, but not the Conservatives. [16] In contrast, a number of important studies in Europe by Dinkel [17] (in an inverse of the relationship studied here) conjectured the subordination of regional elections to the electoral rhythms of polity-wide politics and concluded that incumbency at the federal level hurts a party in regional level elections. These findings led to a range of studies [18] that were used to explain the underperformance of incumbent parties in the first direct elections to the European Parliament in 1979. [19] Studies on election results in India have been more limited in both number and scope. Using a panel data analysis Webb and Wijeweera [20] find a correlation in party performance between legislative assembly and Parliamentary elections that varies according to party type (national, state and registered), but which fails to account for regional variations. Similarly, Yadav and Palshikar discern a number of different phases in which the correlation between state and federal politics changes and conclude that politics at the state level shapes and filters, rather than pre-determines, the outcome of national elections. [21] This finding is partially endorsed by Chhibber who conjectures some linkage between a party's performance at the Parliamentary and state level. [22] 

While these studies point to a tentative consensus regarding the links between a party's performance at the regional and (supra-)national levels, the correlation varies between electorates and may be influenced by a range of intervening variables. For example, North American studies have posited the influence of religious/ethnic homogeneity, campaign spending and the effect of incumbency as possible explanations for why correlations in a party's performance between local and federal elections may vary. [23] Studies of results from Spain, [24] Greece and Portugal [25] similarly found that the Dinkel effect did not hold in areas where there was a lack of congruence in party systems at the national and regional levels due to the presence of 'historic nationalities' with a strong sense of identity or other intervening variables such as the electoral cycle, socio-economic and cultural factors. In the case of India, Yadav and Palshikar echo previous claims [26] regarding the importance of state politics as a determinant of Parliamentary elections by arguing that politically relevant social cleavages are defined in the arena of state politics where the greater number of actors makes higher voter mobilization more possible, and the building blocks of national collations are formed. [27] Thus, they allege, because voters use Parliamentary elections to pass a verdict on the state government, state elections are an arena of primary choice by voters with the choices made during Parliamentary elections assuming more of a derivative character. [28] Here too, however, we find a range of intervening variables that influence the strength and character of linkages between politics at the state and federal level. For example, Yadav and Palshikar acknowledge the importance of a temporal lag between elections observing that where a Parliamentary election follows a state election in quick succession the state verdict tends to be repeated at the national level. In contrast, they claim, if the Parliamentary election is held in the middle of the term of an elected state government voters are more likely to be critical and an inverse result may obtain. [29] 

In summary, there is a widespread body of evidence of a correlation in political parties' electoral performance between regional and higher-order, (supra-)national elections across a range of electoral systems and geographical areas. However, the correlation is unstable and subject to variation. Moreover, the multiplicity of studies of different variables, time periods, electoral systems and party-types makes any comparison between studies problematic. In addition, a range of intervening variables mediate any relationship in a party's performance between elections with different studies highlighting different variables across different types of electoral systems. In the case of India, several studies have highlighted a possible connection in a party's performance between legislative assembly and subsequent Parliamentary elections, but the exact nature and strength of this connection is unclear. Nonetheless, there is widespread speculation in the mass media and elsewhere of such a connection which, if it did exist, might aid in the prediction of Parliamentary poll outcomes and assist parties to allocate resources where they are most effective in maximizing votes.

3. Method

Our aim is to test the claim that there exists a linkage between the outcomes of legislative assembly and subsequent Parliamentary polls in India by examining a party's share of the vote in both elections held eighteen months (548 days) or less previously in the same state. Our study uses a multivariate model and employs data from the 1980-2009 period obtained from statistical reports produced by the Election Commission of India (ECI) pertaining to the performance of all parties that competed in both elections. The selection of the eighteen month time period between elections is important. A shorter time period would have produced significantly fewer observations and made it more difficult to establish the influence of a time lag on voter support between elections. Conversely, a longer time period would have allowed for more observations, but raised questions regarding the relevance of the preceding legislative assembly result given the presence of other, intervening events during this time that might influence an election's outcome. Accordingly, the time period of eighteen months was chosen because it is sufficiently long to allow enough observations from which to draw reliable conclusions without also raising questions concerning the relevance of the data included. Concerning the related difficulty of including additional variables that may also influence the outcome of an election: Previous studies have generally included only one explanatory variable to account for any statistically significant correlation (i.e. the result from the preceding election) by regressing the percentage of the vote received by a party at the latter election with that from the former. This, however, counter-intuitively suggests that a party's performance in an election result may be accurately modeled by the results of a preceding election. The reality is unlikely to be so simple and experience suggests that a range of variables between elections may influence a party's performance including the length of time between the two elections, incumbency, the death of a candidate, campaign spending, a recent scandal, economic changes and so forth. This, however, creates a dilemma given that the more variables that are included within a statistical model, the greater the difficulty in identifying the effect of each and drawing any reliable conclusions due to possible multi-collinearity problems (i.e. if predictor variables are highly correlated then coefficient estimates may change erratically in response to small changes in the model or the data, producing unreliable results concerning individual variables or the redundancy of certain variables with respect to others).

Consequently, we have chosen to use a multivariate model in which the dependent variable (Y) is the percentage of votes received by a party in a state in a Parliamentary election with the initial four explanatory variables.

X1 Vote percentage received by the concerned party in a prior legislative assembly election at time, t-i

X2 Voter turnout at the Parliamentary election at time t

X3 Percentage share of seats won by the party at the preceding local election

X4 Time lag in days between the legislative assembly election and the subsequent Parliamentary election

This gives us the following model:

To the extent that a party's performance in a Parliamentary election can be explained by the results of a preceding legislative assembly election, then the estimated coefficients of X1 () must be positive and statistically significant. A higher voter turnout at the parliamentary election may affect some party's share of the vote depending upon the number and identity of the parties contesting the election. For example, a lower turnout can magnify the impact of the protest vote in the favor of smaller parties. [30] Therefore, we speculate that the coefficient of X2 will be positive and greater for parties other than the two mainstream parties - the INC and BJP. The number of seats won in the preceding legislative assembly poll may exercise a 'demonstration effect' for some parties, raising their public profile and attractiveness to voters. Accordingly, we hypothesize a positive coefficient for X3. Conversely, for time lag (X4) we anticipate a negative coefficient; as the temporal distance between a legislative assembly and a subsequent Parliamentary election increases the impact of the former on the latter lessens. Voters' memories fade and the events of the preceding election recede in favour of more contemporary events as determinants of for which party electors cast their ballot.

INSERT FIGURE 1 HERE

Our full data set contains 502 observations and covers the period 1980-2009 which includes nine Parliamentary and forty-five legislative assembly elections that were held not less than eighteen months prior to the parliamentary poll. However, because of the high number of parties that contest elections in India, and despite the lengthy thirty year time period covered by the study, only a small number of parties had enough observations to carry out a rigorous analysis resulting in two difficulties: [31] First, in order to carry out hypothesis testing it is necessary to assume that the data are normally distributed which in turn requires a reasonably large sample. Second, a smaller data sample presents problems with respect to degrees of freedom. Every time we estimate a coefficient such as one observation is used up in the process. This means one less observation remains for estimating other coefficients. In large samples this is not a problem. However, in small samples this is a serious issue with the result that where the sample size is smaller than the number of observations it is not possible to estimate the model. These considerations eliminate smaller, state parties from our study as none of them have a sufficient number of observations to permit estimation or valid hypothesis testing. Consequently, similar parties such as the Communist Party of India and its Marxist and Marxist-Leninist offshoots, as well as the Janata Dal and its Secular and United subsidiary parties are grouped together, and only those parties that contain twenty-five or more observations are included in our study. These parties and their sample sizes are:

Indian National Congress (INC): 43 observations

Bharatiya Janata Party (BJP): 37 observations

Communist Party of India (CPI/CPI(M)/CPI(ML)): 53 observations

Bahujan Samaj Party (BSP): 26 observations

Janata Dal (JD/JD(S)/JD(U): 28 observations Independent Candidates: 42 observations

4. Results

4.1 Overall Results

An important feature of our study is the use of statistical tools such as regression analysis to ensure the reliability of our findings. Most empirical analyses of election data employ relatively simple statistical techniques such as the correlation coefficient or covariance test to examine relationships. This has the advantage of providing a straight-forward analysis that does not require a significant degree of specialist knowledge to interpret. However, it fails to provide for a more nuanced analysis and may overlook important weaknesses in the data that negatively affect the reliability of the study's conclusions. There are a number of advanced estimation methods that could be employed to test possible relationships among variables, but they are mostly used in more quantitatively-oriented disciplines such as statistics, economics and econometrics. For these reasons, we opt to use the ordinary least squares (OLS) method to estimate the model which is simple, but very powerful in analyzing relationships [32] and is also popular in the social sciences. Using the OLS method we use sample data to estimate a best-fit regression line achieved by minimizing the sum of squared errors. This assists us to make statistical inferences about the relationship between voting patterns in general elections and preceding local elections and then statistically test the validity of these inferences.

For example, once the estimation is complete, we test whether the coefficient estimates are statistically reliable by examining the relevance of each explanatory variable included in the model to the behavior which the model attempts to explain. To emphasize, the behavior that our model attempts to explain is the variation in the percentage of votes won by major parties in a Parliamentary election and the explanatory variables are the percentage of votes and seats received by a party at a preceding legislative assembly election, voter turnout at the Parliamentary election and the time lag between the legislative assembly and Parliamentary elections. In general, there are two hypotheses; the null hypothesis states that the coefficient is zero, which means the relevant variable should not be in the model while the alternative hypothesis states that the concerned variables should be a part of the model. In this paper, we use the p (probability)-value where, if the p-value is smaller than the level of significance (i.e. 5% or 0.05), we reject the null hypothesis and conclude that that particular variable should be included in the model (the numbers in parenthesis in Figures Two and Three are the respective p-values). We also use the F-test which examines the overall significance of the model. The null hypothesis in this case is that none of the variables are significant in explaining the target behavior (i.e. the percentage of votes obtained by a party in a given state in a Parliamentary election), whereas the alternative hypothesis is that the model is statistically significant because the selected variables explain the behavior. Again, we use p-values in this test with the result that, if the p-value is smaller than the level of significance, we reject the null hypothesis and conclude that that respective model is statistically significant in explaining the Parliamentary voting pattern (F-values and relevant p-values are shown in the last column of Figure Two). Finally, we also report R2. This shows the percentage variation that is explained from the selected model by aggregating the descriptive power of the four explanatory variables. For example, the R2 value for the BJP is 91.2 meaning that the model explains approximately 91% of the total variation of the percentage of votes received by the BJP in the Parliamentary elections included within the study.

In order to examine party specific differences, five party groups above are estimated using the Ordinary Least Squares method introduced earlier using the specification given in equation (1) for all five party groups. The results are given in Figure Two below which consists of eight columns. Whereas the first column lists the different parties, the second is the intercept of the linear regression equation which controls for the possibility of other important explanatory variables in the model. Columns three through six are the estimated coefficients of the four explanatory variables. Under each coefficient the probability value corresponding to the estimated coefficient is given in parentheses. Small probability values suggest that the relevant variable is important in explaining the variation of the dependent variable. While there is no universally accepted cut-off point for probability values, a value below 0.05 is generally considered as strong evidence against the null hypothesis of non-relevance. The seventh and eighth columns of the table measure the goodness of the fit of the chosen model in explaining the variation of Parliamentary election results. The seventh column shows R2, which is the percent of the variance of Parliamentary election share explained by the variables in the regression. The eighth and final column shows the F-statistic values which measures the overall significance of the model and suitability for the data set to which it is applied. If the explanatory variables are jointly significant, then the F-statistic should produce a large value and its probability should be less than 0.05. The results overwhelmingly support the appropriateness of the model, except in the case of independent parties which are consequently disregarded in the analysis below.

INSERT FIGURE 2 HERE

4.2 Party Specific Results

Referring to Figure Two we see that the estimated coefficient of the percentage share of voter percentage at the preceding legislative assembly election for the BJP is 1.053 and its probability value is virtually zero. This suggests that the BJP's percentage of the vote in a legislative assembly poll is a statistically significant variable in forecasting the party's percentage of votes cast in a subsequent Parliamentary election in the same state. More specifically, holding all the other variables constant, a one percentage point increase in the BJP's share of the total votes cast in a legislative assembly election increases its share of the vote in a subsequent Parliamentary election by 1.053 percent. A similar positive, but smaller in magnitude, link exists between the vote share percentage received at the parliamentary election and the preceding local government election for the other major party groups. To illustrate; for the BSP (0.626%), CPI (0.955%), JD (0.692%) and INC (0.754%). Interestingly, there is no statistical evidence in support of such a relationship for the independent parties. These findings and possible reasons for them are discussed in greater detail below.

Other interesting findings include that for the CPI there is a statistically significant relationship between the number of seats won in the preceding legislative poll and the party's share of the vote in a Parliamentary election. This same result is not observed with other parties. Voter turnout is not significantly correlated with a party's share of the vote in a Parliamentary election except in the case of the JD and independents. In both cases a one percent increase in voter turnout in a Parliamentary election increases the percentage of votes received by approximately a quarter of a percentage point. Finally, although our tests confirm the suitability of our model for the data set, the model does not work equally well for all parties. To illustrate, in the case of the BJP, CPI, and JD more than 75 percent of the variation in the percentage votes received at the Parliamentary election is explained by the selected variables. The model is moderately good for the BSP and INC where approximately half of the variation is explained by the model. In contrast, the model is less effective in explaining the fortunes of independents that contest Parliamentary polls where only 23 percent of the total variation of their share of votes in Parliamentary elections can be explained by the model. Finally, as far as the time lag is concerned, it is neither statistically significant nor numerically consequential (the coefficient is almost zero in every case). Accordingly, we refrain from drawing any firm conclusions regarding independents and the time lag in our analysis below.

4.3 Inclusion of Regional, Party and Incumbency Variables

A commonly employed tool in studying election results to shed more light on the correlations uncovered is to group the data. For example, similar studies on regional and (supra-)national elections in Europe find a clear variation between different geographical areas leading the authors of these studies to speculate that the distinctive characteristics of some regions affect voting behavior. [33] Accordingly, we similarly group our data into region to examine what effect this exerts on the explanatory power of the model and what conclusions might, therefore, be reasonably drawn from it. Because individual states do not have a sufficient number of observations to permit reliable statistical estimates, we separate the data into the following six commonly-used geographical regions in order to examine whether party performance in Parliamentary elections varies between these. We use the same model as in equation (1), but include dummy variables to capture regional effects. A dummy variable takes a value of 0 or 1 and is represented in Figure Three. In accordance with standard statistical practice, no dummy variable is assigned to one region - Eastern India (EI) - which is the reference region. The intercept value of the equation represents the mean value of the reference region.

Eastern India (EI): Chhattisgarh, Jharkhand, Orissa, Sikkim, West Bengal

Himalayan North (HN): Jammu and Kashmir, Himachal Pradesh, Uttarakhand

North-Eastern India (NE): Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, Tripura

The Plains (PL): Bihar, Chandigarh, Delhi, Haryana, Madhya Pradesh, Punjab, Uttar Pradesh

Southern India (SI): Andaman and Nicobar, Andhra Pradesh, Karnataka, Kerala, Lakshadweep, Pondicherry, Tamil Nadu

Western India (WI): Dadra and Nagar Haveli, Daman and Diu, Goa, Gujarat, Maharashtra, Rajasthan

In addition to the regional dummies we also examine the relevancy of two additional factors in determining the percentage of votes that a party receives in a Parliamentary election: the incumbency factor and the average number of candidates per constituency in the Parliamentary election. A well-established literature covering Western democracies such as the US suggests that incumbency lends a party considerable advantage in an election. [34] In contrast, the evidence for India points to a negative [35] or negligible [36] outcome that may vary between parties, with only one study [37] finding a significantly positive influence. This contrasts with speculation in the mainstream media concerning an anti-incumbency bias that is frequently appealed to as the default explanation for a fall in a ruling party's fortunes at the ballot box. Our study includes only three parties (the INC, BJP and JD) as being incumbent at the Center at the time of a Parliamentary election (the BSP had been a member of the governing United Progressive Alliance but withdrew in June 2008 prior to the 2009 Parliamentary elections). Finally, we include the average number of candidates per constituency in Parliamentary elections, anticipating a negative effect on a party's share of the vote, i.e. the higher the average number of candidates, the greater the choices available to voters and, ceteris paribus, the harder it is for parties and their candidates to distinguish themselves from their competitors in order to win votes. Conversely, elections with a low average number of candidates per constituency may still be fiercely contested, however the difficulty of distinguishing each party from the other, and the risk of a party's support being split between it and a closely aligned rival, are reduced. We label these three additional explanatory variables as follows:

X5 The regional grouping of states. Regional differences are represented via five dummy variables (D).

X6 The average number of parties per constituency in the Parliamentary election

X7 Whether the party was incumbent at the Center at the time of the Parliamentary election

The revised model with these inclusions is given in equation (2) below:

Where δs measure the impact of regional factors, λ measures the impact of the incumbency factor, and γ measures the impact of the number of political parties contesting in the election.

We begin by testing the significance of the model using the R2 and F-tests described above. Comparing the results of these tests between Figures Two and Three, we find that the inclusion of these three additional variables has increased the explanatory power of the model with the R2 values of all parties increasing: BJP (0.912 to 0.932), BSP (0.543 to 0.642), CPI (0.827 to 0.986), JD (0.799 to 0.873) and INC (0.446 to 0.667). Combined with the F-test results we conclude that these three additional variables are important in explaining the voting patterns of the BJP, CP, INC and JD. The low R2 and F values for the BSP indicate that the model is sub-optimal in analyzing voting behavior for this party.

INSERT FIGURE 3 HERE

The key results of Equation 2 may be summarized in the following three points. First, we note the low coefficients of individual regions which vary from -0.002154 for the CPI in the Plains region to 0.115645 for the INC in the Southern region. However, we cannot for this reason, conclude that regional effects are negligible in explaining each party's share of the vote in Parliamentary elections. Not only are some regional coefficients statistically significant (despite being small in magnitude), but the intercept is statistically significant at a 5% level of significance for the CPI. It will be recalled that Eastern India (EI) was omitted from the results as the reference region to prevent perfect multi-collinearity in the model. The statistically significant intercept indicates that EI may be a significant regional factor for the CPI, but not for other parties. 

Second, the number of parties contesting the election has the expected negative sign for all parties and party groups except for the JD. The coefficient is statistically significant for BSP at a 10 percent level of significance. This implies that when more parties contest a Parliamentary election the BSP tends to get a lower percentage of votes. However, the relationship is not particularly strong for other major parties. Third, and finally, the incumbency factor has a positive impact for all parties except the JD. However, none of the coefficients are statistically significant at a 5 percent level of significance. This suggests that incumbency at the Center may have positive impact on gaining more votes in Parliamentary elections for both the BJP and INC, but not for the JD.

In conclusion, there is a close and statistically significant correlation between the percentage of votes won in legislative and Parliamentary elections where, typically, the percentage is higher in the Parliamentary, than legislative assembly elections. The result is similar for both main parties, however, in the case of the INC there are several regions in which the percentage of votes won in the legislative assembly election is higher than in the Parliamentary election. This relationship is clearly visible in Figures Four and Five below where each party's percentage of the vote in both types of elections in each region is shown in chronological order. However, no clear chronological trend is evident for either party. Rather variations between the percentage of votes won in each election occur at various times with the INC experiencing significantly greater variation, which is what we would expect given its lower coefficient (0.754/0.874) versus that of the BJP (1.053/1.102).

INSERT FIGURES 4 & 5 HERE

5. Discussion

5.1 Party Effects & the BJP's Comparatively Strong Showing

Our finding of a strong and statistically significant correlation in a party's share of the vote between legislative assembly and subsequent parliamentary elections endorses the pre-occupation of the mainstream media with regional election results as indicators for Parliamentary polls, and echoes similar research in Canada and the UK which not only uncovered a similar correlation but also a significant degree of variation between parties. [38] It also has important implications for parties and their election strategies. The stronger the correlation in a party's share of the vote between the two elections, the more rational it is for that party to invest valuable resources in a legislative assembly poll in order to maximize the party's share of the vote in a subsequent Parliamentary election. This is because parties with a high correlation value enjoy greater voter loyalty, and therefore experience less volatility in voter support, between the two sets of elections meaning that votes won in the legislative assembly election translate into gains for the subsequent Parliamentary poll. Conversely, for those parties with a weak correlation, investing in a legislative assembly election may be a sub-optimal strategy for boosting the party's chances in the same state in a subsequent Parliamentary election.

The main question raised by our findings concerns the variation between parties in the correlation in voter support from legislative assembly to Parliamentary elections. To briefly review, for both models the link is strong and significant in the case of the BJP (1.053/1.102) but considerably less so for the CPI (0.955/1.062), JD (0.692/0.761) and INC (0.754/0.874). One possible explanation for this variation may be the basis of mobilization for these parties which in turn affects the degree of voter loyalty they enjoy between elections. The BJP, for example, is a member of the Sangh Parivar (a collection of Hindu nationalist organizations) and consequently has strong links with extremist Hindu groups including the Rashtriya Swayamsevak Sangh (RSS) and Vishwa Hindu Parishad (VHP). In contrast, the INC has more of an ideological and class-based appeal predicated on its role in India's independence movement and advocacy for economically and socially disadvantaged sections of society. Consequently, it may be argued that whereas the INC's appeal is predicated upon more transitory factors (the salience of which to voters may vary according to the prevailing socio-economic conditions), the BJP's appeal to the more ascriptive feature of religion lessens the volatility of its support base. Similarly, the JD, which like the INC is a secular-based party that for the period covered by this study also espoused a left-of-center ideology attractive to lower socio-economic classes, also has a comparatively low correlation score.

The argument that religion provides a more secure and predictable base of appeal is intuitively plausible (given that individuals tend to change their religious views and associations less readily than their political values or class-identifications) and might, therefore, explain the variance in voter support between parties. However, explaining the BJP's comparatively high correlation score and apparent voter loyalty by reference to the stable nature of the religious identity that forms the backbone of its appeal to voters begins to sound hollow when we consider that the party with the lowest correlation, the BSP (0.626/0.577), since its founding in 1984 has been regarded as the party of lower caste voters including Scheduled Castes, Scheduled Tribes and Other Backward Castes. In India caste is typically a religiously determined distinction and therefore fixed in the sense that it is an integral component of Hinduism (and other religions in the sub-continent) and individuals are un-free to alter their caste which is assigned from birth by parentage. Consequently, if the non-transitory, fixed nature of religious affiliation is appealed to in order to explain the voter loyalty exhibited by the BJP's high correlation, then a similarly high correlation would also be expected for the BSP. Additionally, there is the high correlation score of the CPI which is second only to that of the BJP's but which, unlike the BJP, is an ideologically, rather than religiously, based party. Once again, if the BJP's comparatively high correlation were explicable by the non-transient nature of its religion-based appeal, then we would expect a much lower correlation for non-religious-based parties such as the CPI.

In making these observations we do not intend to suggest that there is not an important distinction to be drawn between these parties, the basis of their appeal to voters and the possible effect that these may have on the correlations listed above. Rather, our claim is simply that appeal to the comparatively less-transient factor of religion as the basis of the BJP's political mobilization of voters cannot satisfactorily explain our findings. For example, the BSP's low correlation could be attributable to intervening factors such as its comparatively small status and a belief amongst voters that a vote for it in a Parliamentary election is a wasted vote given that it has little chance of leading a coalition at the Center (and zero probability of becoming a single party government). Moreover, it also important to recognize that because our data set includes a significant number of parties, electorates and a long time period its strength in identifying long-term, durable trends is to some degree offset by the difficulty of identifying specific relationships. There is, for example, no reason to suppose that the factors behind a party's correlation in voter support between legislative assembly and Parliamentary elections have remained constant over time or are the same for other, similar parties. Rather, each party's appeal to voters may be different from one another's and have altered over time to reflect the enormous social and economic changes in India that have occurred since 1980.

These difficulties are further compounded by the complexity and size of the Indian electorate making the application of findings from similar studies conducted in comparatively less pluralistic states to the Indian context problematic. For example, European studies that identified regional variations in a party's level of support between federal and regional elections explained this finding by reference to the cultural and national differences exhibited by regions with significantly lower correlations. [39] In contrast, because of India's size and complexity, we can arrive at no similar conclusion. Rather, the cultural, ethnic, linguistic and other ascriptive differences that exist within many of the Indian states that comprise the regional groupings used in our study are greater than those that characterize the European states of these prior studies. Jammu and Kashmir, for example, is geographically and ethnically divided into three regions that are dominated by three different religious and ethnic communities (Hindu-dominated Jammu, Buddhist Ladakh and the Valley of Kashmir which is almost entirely Muslim). Similar levels of diversity are not uncommon in other Indian states. Accordingly, it would be presumptuous to attribute any variation in the regional coefficients identified in our study to differences in 'culture' (however broadly this term were defined). Rather, we surmise that any variations between regions are due to other, political or social factors (e.g. that the Eastern region is statistically significant for the CPI might come as little surprise given that party's dominance in West Bengal politics since 1977).

Similarly while we have no statistically rigorous explanation for the BJP's comparatively high correlation in the percentage of votes the party wins in Parliamentary and preceding legislative assembly polls, we suggest the following as one possible rationalization for this phenomenon. Some authors [40] have noted that the BJP's concept of Hindutva is cultural rather than religious, permitting the BJP to insist that Muslims accept key aspects of Hindu culture while retaining the freedom to practice their faith. This not only allowed the BJP to capture the anti-Muslim constituency, but also to mobilize voters around Hindu religious symbols while retaining a veneer of tolerance towards (religious) minorities that was also an important cornerstone of the INC's secularist identity. By not repudiating the INC's secularist agenda, while at the same time painting the INC as soft on Muslims and other minorities, the BJP has been able to appeal to a broad range of voters from overt Hindu nationalists to secularists and other political moderates who might, or may not, harbor misgivings towards India's Muslim and other minorities. This contrasts with parties such as the INC that relies on more narrowly defined ideological factors to mobilize voters, the BSP's dependence on caste (which might be more important in state-level issues such as land use and the allocation of funds in the local community than at the federal level) and the CPI's reliance on class in their appeal to voters. Thus, we suggest that the multi-dimensional appeal of the BJP not only translates into more votes, but also means that as the party alienates some voters through its policies and statements it captures the support of others and is able to maintain a consistency in its level of support between elections that eludes other parties.

5.2 Other Factors

Turning to the remaining explanatory variables of voter turnout at the Parliamentary election (X2), the percentage share of seats won by the party at the preceding local election (X3) and the time lag in days between the legislative assembly election and the subsequent Parliamentary election (X4) our hypothesizes are mostly confirmed or the results are ambiguous. In Models One and Two voter turnout out is positively related with a party's share of the vote for three parties (BSP, IND and JD) and negatively for the remaining three (BJP, CPI, INC). This suggests that the two main parties fare worse when voter turnout is higher. However, in all cases the magnitude of the effect is small. To illustrate, in the case of the BJP a one percentage increase in voter turnout in a Parliamentary election decreases the party's share of the vote by 0.33 percent (dropping to 0.22 percent in Model 2). This partially confirms our hypothesis that the major parties lose votes as turnout increases as both the INC and BJP exhibit negative coefficients for X2 with the addition of the CPI. Why the CPI should attract fewer votes as turnout increases, but the BSP and JD do not, is unclear. Both the CPI and JD groups of parties are factious, internally competitive, frequently disunited and mobilize voters around ideological factors stressing socialist themes of anti-corruption/exploitation and economic egalitarianism. In the case of the CPI, however, this message seems to find more favor with a core of committed supporters that decreases proportionately as participation in an election increases.

The effect of winning seats in the preceding legislative assembly poll (X3) is also ambiguous. Earlier we speculated that parties might benefit from winning more seats in the legislative assembly election through a demonstration effect that raises their public profile amongst voters. However, Figure Two shows a negative effect for the BJP, BSP and JD and a positive effect for the CPI and INC. In all cases the coefficients are small ranging from -0.587 for the BSP to 0.299 for the CPI in Model One and from -0.395 for the BSP to 0.218 for the CPI in Model Two. Whereas for some parties winning seats in the preceding legislative assembly election translates into a higher percentage of votes in a subsequent Parliamentary poll, for others the inverse is true. However, in all cases the effect is small. Conversely, for time lag (X4) we anticipated a negative coefficient. In Model One all parties recorded a co-efficient of 0 which changed to a negative coefficient for three parties in the augmented Model Two but with a very low magnitude (the most significant being 0.0000926 for the INC). Accordingly, we conclude that the period of elapsed time between a legislative assembly and Parliamentary poll does not affect to any significant degree which party voters cast their ballot for.

Concerning the other significant findings, the BSP's declining share of the vote in Parliamentary elections as the number of parties contesting these polls increases suggests that its support is more contingent on the alternative choices available to voters. Again, we speculate that this may be due to problems of transferring caste politics and the other issues used by the BSP to mobilize voter support from the local to national arena. As established, national parties of governance that have held power at the Center and which have also formed state governments, the INC and BJP appear able to bridge the gap between local and national politics more effectively. In contrast, the BSP seems to struggle in transmitting local support in legislative assembly polls to Parliamentary elections, particularly where voters have alternatives available to them. Finally, whereas the INC and BJP are assisted by incumbency at the Parliamentary level, the JD is harmed by it and is less able to translate votes in preceding legislative assembly elections into support at subsequent Parliamentary polls when it is incumbent at the Center. However, some caution is advised in interpreting this finding given that the JD was incumbent at the Center for only brief periods in 1989-90 and 1996-97 as part of broad coalitions characterized by significant bickering and instability. In both cases the JD's incumbency may be regarded as anomalous and atypical in the sense that it occurred at a time of significant transition in Indian politics characterized by the waning power of the previously dominant INC, the rise of the BJP and the development of coalition government as the new norm of government in place of the previously dominant idiom of one party rule.

6. Conclusion

Our study provides compelling evidence of a strong and statistically significant relationship between a party's share of the vote in a legislative assembly election and a subsequent Parliamentary election held within the following eighteen months. This nexus holds for all the parties included within this study but is strongest for the BJP. It is not significantly altered by any of the additional explanatory variables included within this study including the average number of parties per constituency or a party's incumbency at the Center; a party's share of the vote in a preceding legislative assembly election remains the strongest indicator of its ability to attract votes in a Parliamentary poll. In this sense our results endorse the mainstream media's pre-occupation with legislative assembly elections as indicators of a party's likely performance at the Parliamentary level. However, two important caveats are in order. First, the media tends to view legislative assembly results as indicators of a party's nationwide prospects. In contrast, our results indicate that the correlation holds only within the same state as the legislative assembly poll. Second, while this finding indicates that major parties (particularly the BJP) should invest resources in legislative assembly elections as an effective means of increasing their share of the vote in subsequent Parliamentary polls, whether or not doing so results in an increase in seats won will be determined by the distribution of these votes. An increase in the proportion of votes won may result in more Parliamentary seats if these votes are concentrated in key constituencies, or the party's support is sufficiently strong in an evenly balanced race, that only a minor increase in votes is required to 'tip the scales' in its favor across multiple constituencies. Accordingly, parties must be careful to properly analyze the political landscape in a state before committing resources to legislative assembly elections in the hope of later winning more Parliamentary seats.

APPENDICES

Figure 1: Summary of Local Assembly and Parliamentary elections 1980-2009 [41] 

Year of Lok Sabha Election

State

Local Assembly Election Date

Lok Sabha Election Date

Time Lag (Days)

1980

Mizoram

24 Apr 1979

3-6 Jan 1980

255

Sikkim

12 Oct 1979

113

1984

Jammu & Kashmir

5 Jun 1983

24-28 Dec 1984

570

1989

Tamil Nadu

21 Jan 1989

22-26 Nov 1989

307

1991

Manipur

12 Feb 1990

20 May 1991 - 15 Jun 1991

475

Himachal Pradesh

27 Feb 1990

460

Bihar

27 Feb 1990

460

Gujarat

27 Feb 1990

460

Maharashtra

27 Feb 1990

460

Madhya Pradesh

27 Feb 1990

460

Orissa

27 Feb 1990

460

Pondicherry

27 Feb 1990

460

Rajasthan

27 Feb 1990

460

1996

Maharashtra

9 Feb 1995

27 Apr - 30 May 1996

459

Manipur

16 Feb 1995

452

Gujarat

20 Feb 1995

448

Orissa

7 Mar 1995

433

Arunachal Pradesh

11 Mar 1995

429

Bihar

11 Mar 1995

429

1998

Punjab

7 Feb 1997

16 Feb 1998

374

1999

Mizoram

25 Nov 1998

25 Sep 1999

303

Madhya Pradesh

25 Nov 1998

11-25 Sep 1999

296

Delhi

25 Nov 1998

5 Sep 1999

284

Goa

4 Jun 1999

5 Sep 1999

93

2004

Himachal Pradesh

26 Feb 2003

10 May 2004

439

Meghalaya

26 Feb 2003

20 Apr 2004

419

Tripura

26 Feb 2003

30 Apr 2004

419

Nagaland

26 Feb 2003

5 Apr 2004

404

Mizoram

20 Nov 2003

20 Apr 2004

152

Delhi

1 Dec 2003

10 May 2004

161

Madhya Pradesh

1 Dec 2003

5-10 May 2004

156

Rajasthan

1 Dec 2003

5 May 2004

156

Chhatisgarh

1 Dec 2003

20 Apr 2004

141

2009

Himachal Pradesh

14 Nov to 19 Dec 2007

13 May 2009

528

Gujarat

11-16 Dec 2007

30 Apr 2009

503

Tripura

23 Feb 2008

23 Apr 2009

425

Meghalaya

3 Mar 2008

16 Apr 2009

409

Nagaland

5 Mar 2008

16 Apr 2009

407

Karnataka

10-22 May 2008

23-30 Apr 2009

345

Chhatisgarh

14-20 Nov 2008

16 Apr 2009

150

Jammu & Kashmir

17 Nov - 24 Dec 2009

16 Apr - 13 May 2009

145

Delhi

29 Nov 2008

7 May 2009

159

Madhya Pradesh

27 Nov 2008

23-30 Apr 2009

150

Mizoram

2 Dec 2008

16 Apr 2009

135

Rajasthan

4 Dec 2008

7 May 2009

154

Figure 2: Results of Party Differences

Party

Linear Intercept

% of Votes Received (X1)

Voter Turnout (X2)

% of Seats Won (X3)

Time Lag (X4)

R-squared

F-stat

BJP

0.273

1.053

-0.330

-0.101

0.000

0.912

83.020

(0.001)

(0.000)

(0.008)

(0.323)

(0.804)

(0.000)

BSP

-0.022

0.626

0.073

-0.587

0.000

0.543

6.240

(0.513)

(0.005)

(0.249)

(0.362)

(0.464)

(0.001)

CPI

0.032

0.955

-0.055

0.299

0.000

0.827

696.532

(0.063)

(0.000)

(0.080)

(0.017)

(0.394)

(0.000)

IND

-0.295

1.271

0.690

0.229

0.000

0.233

2.812

(0.181)

(0.194)

(0.040)

(0.843)

(0.088)

(0.039)

JD

-0.138

0.692

0.276

-0.002

0.000

0.799

22.879

0.120

0.000

0.084

0.982

0.916

0.000

INC

0.208

0.754

-0.183

0.073

0.000

0.446

7.635

0.035

0.003

0.179

0.501

0.952

(0.000)

Figure 3: Results of Regional Differences with Incumbency & Number of Parties

Linear Intercept

Explanatory Variables

Regional Variables (X5)

R-squared

F-stat

% of Votes Received (X1)

Voter Turnout (X2)

% of Seats Won (X3)

Time Lag (X4)

Number of Parties (X6)

Incum-bency (X7)

HN

NE

PL

SI

WI

BJP

0.185509

(0.0682)

1.102752

(0)

-0.201731

(0.1842)

-0.122818

(0.2707)

-0.0000408

(0.6386)

-0.000438

(0.8488)

0.035858

(0.0887)

-0.012054

(0.7699)

0.007483

(0.8615)

-0.010873

(0.766)

-0.007854

(0.8605)

0.039774

(0.2839)

0.932746

31.520

(0)

BSP

0.018838

(0.6812)

0.577369

(0.0343)

0.027406

(0.7397)

-0.395484

(0.6253)

-0.0000015

(0.9718)

-0.001536

(0.0863)

N/A

-0.018364

(0.3549)

N/A

0.002678

(0.8739)

-0.010692

(0.6923)

-0.005877

(0.7325)

0.642683

3.197567

(0.020602)

CPI

0.069142

(0.0067)

1.062897

(0)

-0.115371

(0.0112)

0.218994

(0.0881)

0.00000235

(0.9084)

-0.000661

(0.1697)

N/A

-0.027308

(0.0236)

0.012721

(0.3001)

-0.002154

(0.757)

0.02049

(0.2544)

-0.004823

(0.5498)

0.986766

305.7176

(0)

IND

-0.038293

(0.8507)

1.779886

(0.0507)

-0.016342

(0.9619)

-1.264766

(0.2495)

-0.000112

(0.6028)

-0.001707

(0.7698)

N/A

0.030499

(0.7337)

0.315573

(0.0016)

0.002479

(0.9824)

0.034431

(0.7477)

0.008487

(0.9402)

0.54073

3.649845

(0.019)

JD

-0.212903

(0.0848)

0.761307

(0)

0.288166

(0.1661)

-0.013959

(0.8664)

0.0000198

(0.8802)

0.002288

(0.3833)

-0.027193

(0.4443)

0.027858

(0.7001)

0.060128

(0.164)

N/A

-0.004316

(0.9337)

0.004326

(0.9323)

0.873425

11.73081

(0.000009)

INC

0.206012

(0.0665)

0.874143

(0.0004)

-0.315678

(0.0548)

0.1688

(0.1075)

0.0000926

(0.4476)

-0.00063

(0.8473)

0.019887

(0.4889)

-0.01969

(0.7031)

-0.03772

(0.0017)

-0.044913

(0.3611)

0.115645

(0