Professional Documents
Culture Documents
Mark Schneider1
May 1, 2008
ABSTRACT
A consensus in public opinion scholarship holds that partisan identity is one of the
most powerful concepts for explaining political behavior, attitudes, and learning. Work
political world, a stable form of identity, and a perceptual screen in attitude formation and
learning (Shivley 1979; Converse 1962; Green et al. 2002; Bartels 2001; Zaller 1992).
and Weldon 2007; Converse 1969; Baker et al. 1981; Brader and Tucker 2001;
Mainwaring 1999; Field 1980). Given the importance of partisanship for understanding
surprising that India been largely ignored. Although Eldersveld (1978) and Field (1980)
found evidence from surveys conducted in the 1960s that over 60% of Indians were
partisan—including people from all classes, castes, religions, and education groups-- we
have seen almost no work on the subject in the past thirty years. Instead, the majority of
work on Indian political behavior addresses vote choice or electoral volatility, implying
1
I would like to thank Prof. Alfred Stepan and Yogendra Yadav for access to the 2004 NES Survey.
2
that partisanship is less relevant than a study of its apparent absence. This paper seeks to
establish whether or not partisanship is relevant among voters in India in effort to begin
The paper proceeds as follows. First, I introduce the case of India and provide a
can be tested with Indian survey data. Second, I operationalize these variables using the
2004 National Election Studies (NES) survey administered by the Center for the Study of
Developing Societies (CSDS). Third, I test hypotheses from the comparative literature
Why India?
India provides a compelling case for the study of partisanship because it is the
most populous and one of the oldest (uninterrupted) democracies outside of the OECD
countries. Unlike Latin America where democracy was interrupted in most countries by
military dictatorships or the former Soviet Union where democracy is quite young, India
provides a rare example of a second wave democracy where partisanship has had nearly
diversity more closely approximates sub-Saharan Africa than the 25 new and established
democracies included in the first wave CSES.2 Thus, a study of partisanship in India
allows us to evaluate the purportedly primary role that ethnic identity plays in partisan
development in one of the few highly diverse countries where procedural democracy is
clearly consolidated (Kohli 2001). Third, decentralized federalism and variation in state
2
According to Alesina et al’s (2003) dataset, India has a linguistic fractionalization score of .81. Fearon’s
(2003) dataset gives sub-Saharan Africa a cultural diversity mean of .71, which is largely based on
linguistic diversity. This is significantly higher than the global average (.48) and the average for Western
countries (.24). Given that India has five dimensions of ethnicity that are politically salient, ethnic diversity
is quite a major issue in Indian politics to a degree beyond that of the OECD countries (See Chandra 2005).
3
party systems and demographics allows for an analysis that takes account of variation in
Why Now?
I have noted that including India in the discourse on partisanship will benefit this
literature. I also argue that this study is essential because rare earlier studies of
partisanship are based on surveys administered when democracy was young and
Congress was an unrivaled power at the state and national level. This is no longer the
case. Yadav (1999) divides Indian politics into three periods. In the first period (1951-67)
Congress held power at the state and national levels. In the second period (1967-1989),
Congress retained power at the national level with rare exception; however its vote share
declined and it lost dominance in a number of states. In the first two periods, politics was
oriented around one’s position on Congress; either you supported Congress or the
opposition. In the third period (1989-present), Congress could no longer hold a majority
in parliament and lost favor in many states. The general party system became much more
competitive, diverse, and for the first time characterized by coalition government at the
center. For example, in the 2004 general elections, the leading Congress won less than 28
percent of total seats in the Lok Sabha and held less than two-thirds of seats in their
ruling coalition, which it shared with 11 other parties. In short, party systems at the state
and national levels hardly resemble those of 1967 when Eldersveld and Bashiruddin
Second, the scope of ethnicity and its salience in electoral politics have changed
since 1967. With the implementation of the Mandal Commission’s recommendations for
reservations for Other Backward Castes (OBCs) in government posts and colleges in
4
1990, party strategies shifted from coded ethnic mobilization to overt ethnic appeals
(Chandra 2000). Lower castes became more politicized and assertive and caste conflict
shaped the party system in ways that it had not under Congress dominance. Moreover, the
birthplace— an event which led to Hindu-Muslim violence in cities and towns across the
country – elevated the Hindu-Muslim cleavage from the domain of extremist parties,
predominantly in urban areas, to that of mainstream politics. Third, media access has
sharply expanded in the past four decades with major expansions of newspaper
most recent studies of comparative partisanship and do not use multiple regression
techniques. Eldersveld and Bashiruddin (1978) ask the following question: ―Generally
speaking which party do you feel closest to?‖ Field asks: ―Is there any political party in
this country that you think would do more good for the people of this country than any
other? Which party do you support‖ (Field 1980)? Eldersveld’s question assumes that
voters identify with a party and provides no negative cues with the likely effect of
evaluation (Achen 1992) than identification (Campbell et al. 1960). Although Field’s
measure is more restrictive than the former measure and may serve as a more reliable
point of comparison, it is based on surveys administered in only four states including two
3
They find 70% of Indians to be partisan based on 1967 data. A measure for 2004 on whether not one likes
a party that require one to name that party--which also should inflate the results—finds 51% to be partisan
(see figure 1).
5
draw on earlier studies as a reference point and to inform hypotheses for later analyses;
Field (1980) and the recent wave of studies on electoral volatility suggest that
findings on variation in volatility rates across states and understand stability in voting
the aggregate, there is reason to believe that partisanship may have some relevance in
India even if it has more widespread relevance in some states and party systems and not
others.4
In sum, if we think that dominant party systems are more likely to see survey
1967 now that electoral politics is more complex (Huber et al. 2005).5 If we think that
increased ethnic salience makes voters more likely to express partisanship, partisanship
should have been positively affected as well. And if we think that those with greater
access to party messages through the mass media are more likely to be partisan (Brader
and Tucker 2007), we should expect the last 40 years have made a difference. Although
time series analysis is necessary to evaluate these effects, the need for more current
fully developed, partisanship has been found to predict positive attitudes toward one’s
4
See Lupu and Stokes (2007).
5
This remains an open question (Slabad & Slomczynski 1999).
6
party (Campbell et al. 1960), bias information processing (Bartels 2001), and predict vote
choice. Where it is less developed, Brader and Tucker (2001) argue that patterns in vote
choice provide the strongest clues for distinguishing nascent partisans from non-partisans
it is not clear that one should expect partisanship in India to be nascent as in the post-
Soviet world and sub-Saharan Africa. 6 At the same time, it is clear that partisanship has
not achieved the substance that it holds in older single member plurality democracies
falls-- and expect variation on this scale across states where relevant parties vary in age
and party systems vary in diversity and volatility—I use two dependent variables
expected to capture partisanship. First, following from Brader and Tucker (2001), I
include a binary behavioral consistency measure, which codes those who report voting
for the same party in the 1999 and 2004 general elections. Brader and Tucker use a
Indian measure is far more restrictive than their measure due to the five-year spread
strong indicator of being predisposed to vote for that party.7 For my second measure, I
6
It should be noted that India has been a democracy for 60 years, which is 15 years shy of Converse’s 2.5
generation marker for a frozen partisan electorate. Given that most parents have children earlier than age 30
in India we may even say that India has reached the 2.5 generation mark where partisanship levels are
expected to plateau.
7
Note that Brader and Tucker’s measure of core partisans includes a maximum at voting for the same party
within one year. Also, note that I do not claim that partisanship causes vote stability as the reverse causal
argument is also likely. Instead, I use these two measures as corroborating evidence of a party
predisposition.
7
use the NES question: ―Do you like any particular party? Which one?‖8 This measure is
less restrictive and provides a weak form of the self-identification question.9 It attempts
interpreted with caution. First, this question may be less precise than the question used in
past surveys in India and Europe, which asks: ―Do you feel close to any particular party,
which one?‖ The latter question cues a social psychological connection between voters
Finally, the partisanship measure and vote stability measure are significantly
whose political behavior differs from the general trend, those who do not like a party are
approximately equally likely to vote for the same party in two consecutive elections.11 On
the other hand, 61 percent of those coded as liking a particular party report voting for the
same party as compared to 39 percent who do not. This suggests that expressing an
Figure 1
Vote
Stability Total
Party_like 0 1
0 55.60% 44.60% 11,708
1 44.40% 55.40% 11,840
Total 11,010 12,538 23,548
8
I refer to this measure as the party-like measure in the remainder of this paper.
9
Note that the question ―Do you feel close to any particular party‖ was asked in previous NES survey not
available in the U.S. This may be a superior measure to the party-like measure and I will evaluate this in
Delhi in summer 2008.
10
Despite its weaknesses, the NES measure is far from perfectly correlated with vote choice and—unlike
the vote-stability measure-- consistent with a variety of theoretical expectations in the literature on
partisanship in new democracies. This suggests it capture partisanship to a reasonable although not
maximal degree.
11
See table 1 at the end of this paper where I include dummy variables for respondents from scheduled
tribes and states with populations of 30% or more scheduled tribes.
8
questions. (1) Does partisanship grow more pervasive over time (Converse 1969; Lupu
and Stokes 2007; Brader and Tucker 2001; Campbell et al. 1960)? (2) How does political
sophistication influence partisanship? (3) Are those with greater access to information
more likely to be partisan (Brader and Tucker 2002, 2007; Pérez-Liñán 2002)? (4) How
(Ferree 2004; Posner 2005; Mattes and Norris 2003). I present independent variables that
and aggregate forms. At the individual level, the life-cycle argument proposed by
Michigan researchers posits that as one accumulates experience voting for the same party
over the years, partisan stability will increase (Campbell et al. 1960). These findings are
confirmed by a large number of studies based on survey evidence that find partisanship
unlikely to change for older adults (Franklin and Jackson 1983; Campbell et al. 1960;
Leithner 1997; Alwin and Kroskik 1991). Leithner (1991) understands the effect of
cumulative voter participation as a process in which the more times an individual votes
for one party relative to others, the less likely he will vote for a different party with the
ultimate effect of developing partisanship. Unless vote choice is a random walk, which a
consensus of scholarship refutes, it appears that the more one votes, the more likely he
will vote consistently and the more one votes consistently the more likely he will develop
The mechanism driving this trend receives several explanations. Shivley (1979)
argues that voters interested in making an intelligent vote choice but lacking in cognitive
correct vote choice and votes in ten elections, it is plausible that he will be more likely to
adopt an information shortcut than if he has only participated in one election. Second,
Achen (1992) argues that as a voter accumulates evaluations over time, his overall
assessment of a party will change at a decreasing rate. This means that the more one
votes and makes judgments about where she stands on policy relative to political parties,
the less likely one will change her vote. Since party evaluations at time period t depend
on t-i, partisanship grows out of stable positive evaluations of a political party. Applying
this logic to the adoption of partisanship over time, the more a non-partisan makes a
rational vote choice by supporting the same party, the more likely he will defer to that
party label in the future or develop a positive affective attachment to that party.
The political experience hypothesis in its aggregate form follows from Converse’s
transmission of partisanship. The argument is that as democratic regimes age, voters will
be more likely to have been socialized by their parents toward a party preference. With
respect to new democracies, the expectation would be that we should see an increasing
interrupted-- such as India relative to third wave democracies such as Zambia and Poland.
Dalton and Weldon (2007) apply this model to a dataset on new democracies and find
Hypothesis 2: Those who are older are more likely to develop partisanship.
find that in Russia interest and awareness both drive partisanship while the relationship
between education and partisanship is negative when interest and awareness are
controlled. Political awareness and interest are expected to increase partisanship because
interested and aware respondents are more likely to be exposed to campaign messages,
engaged in the political process, and to vote consistently. Therefore, interested voters are
likely to develop partisanship either in order to reduce the information costs on their vote
decision (Shivley 1979; Dalton and Wattenberg 2000) or because they support the
positions parties take or the social bases they represent (Campbell et al. 1960; Achen
support in earlier studies on India. Eldersveld and Bashiruddin (1978) find that those with
more than a middle school education are the least likely to be partisans to the extent that
illiterates are more partisan than those who completed high school and college.12 Field
(1980) reports a drop in partisanship among the college educated as well. The best
educated in these studies may be non-partisan because they have greater access to
information and do not need to depend on parties for information short-cuts (Dalton and
Wattenberg 2000).
12
College educated people in India are significantly less likely to be strong partisans than illiterates in the
aggregate with 37% and 54% of strong identifiers respectively according to 1967 data.
11
new democracies suggest that as exposure to information on party positions increases that
partisanship will become more likely. For example, survey and experimental data
analyzed by Brader and Tucker (2002, 2007) shows that access to basic information
finds cross-national evidence from Latin America that those who watch television news
more often are more likely to express partisanship. Following Converse (1962), I expect
that those who consume high levels of news media possess large repositories of partisan
information compared to those who consume little or no news. Moreover, I expect that
those with larger repositories of partisan information will be more likely to support a
particular party as new information will be less likely to change their party preference.
Therefore, the expectation from this literature is that low news consumption will translate
Effective Number of Parties. Huber et al. (2005) argue that the more
diverse the party system the more difficult it will be for voters to develop
voters must work harder to learn about their party differential. Following this
logic, the Congress dominant party system in India prior to 1989 should see
higher levels of partisanship than the current diverse party system and states with
12
highly competitive party systems should have lower levels of partisanship than
Party System Volatility. Lupu and Stokes (2007) show that partisanship
Huber et al. (2005) show that respondents (with less than a college education) are
more likely to be partisan where parties have been around for a long time. In India
there is significant variation on party age particularly where regional and ethnic
parties have become most successful. I expect that voters in states where the party
system is highly stable will be more partisan than those in less stable states.
Social Groups. Campbell et al. (1960) hypothesize that the linkage between social
groups and parties will be greatest when groups are salient to politics and parties provide
visible cues of group-party linkages through policies or the social profiles of candidates.
generally considered most politically salient-- suggests that social identity is the
predominant independent variable on vote choice (Birnir 2007; Chandra 2004; Posner
2005; Yadav 2000; Chhibber 1999; Ferree 2004a, 2004b; Norris and Mattes 2003).
Cross-national time series analysis by Birnir (2007) shows that ethnic diversity reduces
electoral volatility in new democracies and Heath (1999) finds that Indian states where
party social bases are more clearly identified by social cleavages saw significantly lower
might drive partisanship. Ferree (2004) tests three types of micro-foundations leading to
voting along racial lines in South Africa. First, ethnic voting can be an expressive act of
considerations do not inform vote choice and identity salience would be most important.
Second, ethnic voting can be conflated with policy interests that are homogenous within
groups. This mechanism means that party policy positions are both known and salient,
which does not apply to the median voter in India (Yadav 1999; Chandra 2004). Third,
ethnic identity may operate as an information short-cut for voters in information poor
supporting the information short-cut hypothesis alone (See Chandra 2004 and Birnir 2007
are likely to support a party along ethnic lines out of strategic reasons, which means that
affective identification with an ethnic group is not necessary. Consistently, using cross-
national Afro-barometer survey data, Miguel and Posner (2006) find that only a minority
identify along ethnic lines rather than along other dimensions although ethnic identity is
highly predictive of vote choice.13 Moreover, Ishiyama and Fox (2006) found the
13
Miguel and Posner asked respondents the following question:
14
in the aggregate and positive and significant only in the ethnic party system of Malawi--
The information short-cut hypothesis implies that where cues signaling the
relationship between parties and social groups are particularly strong, voters from these
groups are more likely to be partisans. A state measure that captures the clarity of group-
party linkages would therefore address the degree to which voters can distinguish parties
by their ethnic composition. For example, if there are four groups {A,B,C,D} in State {S}
and voters understand that parties W,X,Y, and Z are overwhelmingly identified with
groups A,B,C, and D respectively, the party system is ethnically polarized and party-
group linkages are clear to voters (Horowitz 1985). This scenario is the maximal and
simplest form of clarity in group-party linkages. If there are two parties and Party X is
identified with groups A and B and Party Y is identified with groups C and D, group-
party linkages are also quite clear because members of group A realize that they are
party X is identified with A and B but not C and that X is identified with C. The objective
of building a variable that measures the clarity of group-party linkages then should
measure the degree to which voters’ understandings of party social bases approximate the
above examples.
their social group, the next question asks how they base their understanding of
party social bases. On approach would be to simply ask them which groups go
with which parties in a survey although this has not been done. A more feasible
alternative would attempt to map parties to groups based on the mechanisms that
15
voters use to asses group-party linkages. Chandra (2004) argues that voters in
India make this assessment by counting the heads of co-ethnics in prominent party
that party promises are non-credible (Keefer 2006), it should also be the case that
campaign appeals are not credible. Thus, the only reputable information about
group-party linkages comes from observations about the people parties promote
or nominate to top positions. If a scheduled caste voter for example observes that
a disproportionate amount of BSP members of the state assembly (MLAs) are SCs
to a degree beyond all other parties, he then makes the inference that the BSP
that very few Muslims are in positions of power in the Hindu Nationalist BJP, she
makes the inference that the BJP does not represent Muslims. I detail this measure
To test the preceding hypotheses, I present evidence from the 2004 National
Election Survey administered by the Center for the Study of Developing Societies (CSDS)
in New Delhi. This survey includes over 27 thousand observation collected from 28 states
and three union territories. Respondents were drawn through multi-staged stratified
random sampling with respondents drawn from 420 parliamentary consistencies, 932
state assembly constituencies, and 2,380 polling stations. The survey was run in a single
wave that spanned the 48 hours from the time of voting in the 2004 general elections to
the time of vote counting. This marks the largest survey of Indian political behavior to
date with sample sizes above 450 for all states with the exception of Goa. In my analysis,
I include all states with the exception of Jammu and Kashmir, which I exclude due to
potentially confounding factors associated with the three-way conflict between Pakistan,
India, and the Kashmiri Independence Movement. While this state may be included in a
future iteration, it is not clear that partisanship reflects the same process of socialization
in Kasmir as it does in other states. I also drop two union territories with sample sizes
below 50. When accounting for these restrictions and missing data, my analysis include
Political Experience. Political experience includes two variables: age and voting
experience. The age variable is a continuous variable measured in years and includes
17
adults from the ages of 18 to 95.14 If age reflects the degree to which one was socialized
into democracy, the maximum age should include those who were eligible in the
that ranges from 0 to 2.15 Those who report that they did not vote in the 1999 or 2004
general elections are coded . Those who voted in either but not both of these elections are
coded 1 and those who voted I both elections are coded 2. I also test for the interaction
term between age and voting experience. Here, we should expect that older people who
votes consistently will be more partisan. If we can infer that older people who voted in
both elections are also likely to have voted consistently in earlier elections, this captures
interest and education. I measure education as an ordinal variable on a (0,6) scale with
illiterates coded zero and college graduates coded 5. Political engagement is measured
based on responses to the following questions: How interested were you in the election
campaign this year—a great deal, somewhat or not at all? I break this variable into
for media news consumption through television, radio, and newspapers. I code 1 for those
who do not consume a particular type of media. I code 2 for those who report consuming
news once in a while for a given type of media and code 3 for those who report
14
I cut off the years above 95 because I suspect coding error for ages 96-99.
15
Note that it is possible that respondents did not vote in parliamentary elections but did turn out for state
elections, which are particularly important in India. As state assembly election voter turnout was not coded
in the NES survey I am restricted to general election turnout.
18
consuming news media often and code 4 for daily consumption.16 I include separate
variables for each form of media because they are likely to have different effects and
those with at least a high school education.17 Radio is consumed by a wider range of
respondents.
legislative seats at the state level following the formula for the effective number of
1
legislative parties (Taagepera and Shugart 1989; See also Chhibber and Nurruddin
pi 2
2005). I include all parties with at least one percent of seats in this calculation and
calculate from state elections prior to 2004. I only use elections immediately prior to
2004 because party diversity should figure into respondents’ attitudes in 2004. Since
elections for state assembly that year occur concurrently with parliamentary elections this
is required to understand how respondents’ experience with party systems may impact
Party Age. I code party age for an individual party as the number of years from
the first state assembly election a party contested in a states subtracted from the NES
survey year 2004. This number is weighted by vote share—calculated from state
assembly elections in 2004 or immediately prior to 2004-- and summed for up to six top
parties across states.18 For example, in Uttar Pradesh the Congress Party-- formally
founded in 1885-- and the Communist Party (CPI) founded in 1925 both
16
Note that the newspaper question does not mention news consumption explicitly while radio and
television questions do. I consider these parallel since newspapers constitute news by definition.
17
Note that unlike in OECD countries, televisions are primarily owned by those from the middle classes
and above and are rare in the countryside.
18
I restrict parties included in this measure as those who earned at least 1% of the vote.
19
contested in the first state assembly elections in that state in 1951. Therefore I calculate
their party age as 2004 (the year of the NES survey)—1951=53. Finally party age is
weighted by vote share and summed to calculate mean party age for the state party
system as a whole. This last step is also followed by Huber et al (2005) and allows for a
mean party age that captures the parties that most influence the ENPS measure laid out
above.
Social Groups. I argue above that voters in state party systems where group-party
linkages are salient and visible are more likely to be able to discern what party represents
them and eventually develop a psychological attachment to that party than those in states
with less visible cues. If we accept for the moment the consensus among scholars of
Indian voting behavior that caste and religious identities are highly salient in varying
ways across Indian states, the central task is to develop a measure of the visibility of
party-group linkages for state party systems as a whole. I develop two measures in this
section. The counting heads measure cannot be implemented until data on the ethnic
2008. The caste diversity measure captures the diversity of major caste groups across
states.
Counting Heads Indicator. The counting heads indicator captures the mechanism
in which voters assess group-party linkages from enumerating the proportion of co-
ethnics in prominent positions in parties. Information about the ethnic identity of party
leaders is the least costly type of information to obtain where ethnic markers are known
and media and politicians alike advertise these identities. When data on this measure
according to six broad caste and religious categories included in the 2004 NES
Tribe, Muslim, and other religious minorities.19 I code the ethnic identities of state
legislators elected in the election prior to 2004 across Indian states. I use ethnic make-up
prior to the 2004 elections because voters are likely to use information about group-party
linkages available to them prior to the election in formulating their answer to the party
identification question. I use state assembly make-up rather than some other measure of
party leadership such as cabinet composition because I expect that voters are more likely
to know the relative share of co-ethnics from elected positions than appointed ones. This
information is likely to be advertised in election campaigns and it is not clear that a large
share of voters pay attention to national or state administrations-- although this remains
When Jaffrelot’s data becomes available in December 2008, I will code this
measure as follows.20 First, I calculate percentages of state legislators from each of the
six NES ethnic categories for each party with at least one percent of state assembly seats
across states.21 Second, I compare percentages to ascertain whether not there are major
differences--20% or more-- across parties. For example, if 50% of the BSP came from
scheduled castes while 10% of MPs from all other parties in the state came from
scheduled castes, it is clear that the BSP represents scheduled castes. If, 40% of MPs
from the BSP are Muslim compared to 50% from SP and 5% from the BJP, no one party
19
The above variables rely on data on the ethnic categories of MLAs collected by Christophe Jaffrelot for
the years 1951-2004 (to be released in December 2008). He codes for the same categories included in the
NES questionnaire.
20
Note that it was expected that these data would become available much sooner. The book in which these
data are used will be released in December 2008 by Rutledge India Press.
21
The six groups include Other Backward Castes (OBC), Scheduled Castes (SC), Scheduled Tribes (ST),
Other castes primarily composed of forward castes, Muslims, and other religions.
21
stands out to represent Muslims. Finally, I create an additive variable for the number of
party’s seats held by a particular group is 20% more than for all other parties are coded 1
if this is true for one group only. They are coded two if two groups are concentrated in
State Caste Diversity. In lieu of data needed for the above measure, I have
developed a measure for the effective number of caste groups and scheduled tribes that
uses the same formula as the effective number of parties outlined above. I choose the
cleavage in Indian politics.22 My caste diversity measure includes data from two sources
due to limitations of the census. I use data on the percentages of STs and SCs from the
census; however, the census does not collect data on OBCs or forward castes. To
calculate these proportions I use the percentages of respondents from OBCs and Other
Castes (Forward Castes) from the 2004 NES survey. These numbers are quite close to
household surveys that measure OBCs and Other Castes for most but not all of the states
I include in my analysis. While the need to draw demographic data from multiple sources
I argue that this measure provides a rough picture of the counting heads
hypothesis. Following Posner (2005), evidence suggests that the ethnic groups that are
For example, in North Indian cities with large concentrations of Muslims, parties actively
mobilize these groups and run candidates from these groups. Similarly, there has been a
22
Note that data compiled by Steven Wilkinson on linguistic diversity did not have a significant effect on
partisanship.
22
trend particularly since 1991 to mobilize scheduled castes and OBCs. For example, the
BSP in Uttar Pradesh and RJD in Bihar have mobilized large segments of the scheduled
castes while the Samajwadi Party has a plurality of OBC support in Uttar Pradesh.
extremely large—that parties will attempt to mobilize these groups, it is also reasonable
to suggest that where there are more effective groups that group-party linkages will be
more visible to voters. This means that although parties may be multi-ethnic in many
cases, when groups are numerically significant, parties have the incentive to mobilize
voters along ethnic lines (Posner 2005). Thus, where group-party linkages are most
obvious, I expect partisanship will be more likely to develop.23 I also interact caste
diversity on the effective number of parties measure as ethnic diversity may only impact
Controls. I include controls for gender, scheduled tribes (STs), and tribal states
where the population includes at least 30% of scheduled tribes. Men are widely known to
be more politically active than women in India and voter turnout is also significantly
lower for women (Chhibber 2008). Second, scheduled tribes are clearly the least partisan
caste group in India and independent variables such as interest that move in the positive
direction for the overall model move in the negative direction for scheduled tribes. This is
likely due to the fact that STs remain poorly integrated into mainstream society in India.
Finally, I create a control for states with population composed of at least 30 percent of
STs. The idea here is that state-level factors such as the effective number of castes or the
effective number of parties will be differently impacted in tribal states. Concerning caste
23
Note that the same analysis may be done on the politically salient dimensions of religion and language.
This data is available off the shelf only for the most populous states in India. I will create these measures
for all states in a future iteration.
23
diversity this is clearly true as STs are considered to be one group. This means that states
with the largest number of STs—which may include many different tribes—are coded as
the most homogenous. Concerning the effective number of parties and other state-level
factors it may be the case that states dominated by STs are different for reasons not
included in the model. For example, many tribal states have undergone some degree of
tribal insurgency in the past and this may explain anticipated variation.
Results from the NES 2004 survey data strongly suggest that partisanship is a
relevant concept in India. Independent variables drawn from all major hypotheses
results.
I begin this analysis by looking at the effects of the variables outlined above on
the development of partisanship in India. Partisanship and vote stability measures are
estimation using a logit model. Coefficients and standard errors are presented in table 2 at
the end of this paper. I decompose interaction effects and main effects among the
proportion partisans and voters who supported the same party in 1999 and 2004 in table 1.
Partisanship and Volatility. These data make it clear that partisanship and vote
stability measure relatively distinct concepts. Beginning with raw state figures on both
dependent variables displayed in table 1, we can see that the two dependent variables
often do not move together. For example, Karnataka has the lowest level of partisanship
24
outside of the tribal states with below 40 percent expressing a preference; however, the
percentage of voters who support the same party falls above the mean at 55 percent.
Moreover, although the partisanship measure is less restrictive than the vote stability
measure, there is not a clear pattern in these numbers. Ten states have more partisans than
stable voters and 19 have more stable voters than partisans. This suggests that there is not
Second, results from the logit model show that independent variables linked to
partisanship have different effects on vote stability. For example, both measures of
interest in politics have positive and significant effects on partisanship; however, the only
interest variable with (weakly) significant effect on vote stability takes a negative sign.
Similarly, radio news consumption has a highly significant and positive effect on
partisanship and an equally significant negative effect on vote stability. Finally, while
males are very significantly more likely to express partisanship than women, there is no
significant difference across genders when it comes to party loyalty at the polls.24
The marginal effects of voter turnout interacted with age also show different
trends across dependent variables. The effect of voter turnout on the probability that an
individual is a partisan across age cohorts pales in comparison to that for vote stability.
An increase of one unit (voting in one election) on the voter turnout measure leads to an
increase in the probability of party loyalty in vote choice many times the size of
partisanship across all age cohorts. Large differences also exist in interpretation of the
interaction term between effective number of parties and effective number of caste
groups where the effects are much larger for vote stability than partisanship. Finally, it is
24
Note that this may be explained by women’s reluctance to offer political opinions in Indian surveys and
the possibility that women vote the preference of the head of the household during elections. This is an
empirical question outside the purview of this paper.
25
interesting to note that while men are highly significantly more partisan than women
(99.9% significance level), there is no significant difference between men and women
concerning vote stability. Approximately 53% of both groups voted for the same party in
1999 and 2004. While analysis of voting behavior falls outside the purview of this paper,
it is clear that while partisanship predicts vote choice in India, they are not similar
concepts of varying strength. For this reason the remainder of this analysis emphasizes
findings on the partisanship measure with some discussion of how both measures
compare.
These data also show that partisanship is well predicted by major hypotheses in
the literature on partisanship. All variables not included in interaction terms and all
partisanship works in a similar way in India as it does in other new and older democracies.
That I find such consistency in terms of significance suggests that the party-like measure
captures partisanship to a reasonable degree although it will not be clear whether or not
or how much it is inflated relative to other measures until comparative survey analysis
can be completed.
acquisition of partisanship posit that as voters get socialized into democracy over time,
partisanship becomes more likely (Brader and Tucker 2002;Converse 1969; Lupu &
Stokes 2007). I test this hypothesis by analyzing the effect of age on the probability that
one is a partisan across levels of voter turnout (table 2) and subsequently the marginal
effect of voter turnout on partisanship across age cohorts (table 3). These data show that
25
I established that interaction terms are significant with likelihood ratio test. In all cases chi squared
values were significant at the 95% confidence level or above.
26
age has a modest negative and significant effect on partisanship when voter turnout is 0.26
The minimal effect of age on partisanship is not surprising although it suggests that those
who simply lived in democratic India for a large number of years—even if they did not
The more interesting finding concerns the marginal effect of voter turnout on
(cumulative probabilities of 25% to 90% on the age variable respectively) the percent of
marginal increase in the probability of partisanship rises from 13.8 percent to 29.8
percent respectively. A similar but much larger trend exists for vote stability. The
probability that a 60 year-old voted for the same party twice is nearly two times as large
as that for a 27 year old. From these data we see that voter turnout and its interaction with
age has a powerful impact on the development of affective ties to parties. If one votes
often she is more likely to be partisan and if one votes often and is older—meaning they
are likely to have voted often for extended period of time-- the probability of partisanship
significant (90% level) effect on partisanship. The relationship is twice as large and
significant (99% confidence level) for the vote stability regression. These findings are
consistent with Shivley’s information shortcut hypothesis and consistent with findings on
partisanship in new democracies. This hypothesis posits that when individuals have the
capacity to process information without the aid of information shortcuts from party labels
they will not become partisans. In the Indian case, it is plausible that the better educated
26
Note that when the range on the age variable is reduced from 18-95 to 18-75, this size of the coefficient
becomes several times smaller and loses significance in the logit output although the negative sign remains.
When age is restricted to >65 it changes signs although this is not particularly meaningful
27
not only have the capacity to make their own electoral decisions but are also disaffected
with party politics. Future research should distinguish between whether or not the most
educated are less partisan because they are disillusioned with party politics that often
ignores policy and emphasizes patronage distribution that they do not need and ethnic
Both interest measures have positive and highly significant effects on partisanship.
Those who report a great deal of interest in the election were over twice as likely to be
partisans as those said they were somewhat interested. Brader and Tucker (2002) suggest
that those who are engaged in politics are most likely to be partisan regardless of
education because they are motivated to vote effectively. Those who are interested and
less educated-- following the Shivley hypothesis-- should be more partisan than those
interested and highly educated because information shortcuts will be more necessary for
confirmed in the data. Newspaper readership is significant at the highest level across both
dependent variables and has the largest coefficient. Radio news consumption has
significant effects in opposite directions across the dependent variables. Television news
holds the weakest effect, which is only significant on the partisanship measure. These
findings confirm expectations by Brader and Tucker (2002) that exposure to campaign
propaganda through the mass media increases the probability that one has a party
preference. It should be noted here that exposure to party messages through the media is
quite distinct from the effect of education. Education makes respondents more critical of
28
electoral politics while the mass media-- with the exception of the English language
expect that respondents in states where party systems are highly stable are more likely to
identify with parties. Respondents in states where the same party has contested elections
for many decades had had a longer time to develop partisan ties relative to respondents in
more volatile party systems. Moreover, in states where effective parties entered the party
system very recently it may be more difficult for respondents’ partisan preferences to
stabilize. In such contexts we may expect a particularly low level of party loyalty at the
polls and low levels of affective ties to parties as well. The 2004 data confirm this
from literature on ethnic voting and the social psychological literature on partisanship, I
hypothesized that where group-party linkages are most obvious to voters that partisanship
should be more common than where these linkages are less obvious. The compromise
measure I proposed looks at whether the effective number of caste groups at the state
level translates to higher levels of partisanship. Clearly, at issue concerning the degree to
which castes map to parties is the effective number of parties. I present data on the
interaction between caste diversity and the effective number of parties in tables 5 and 6.
These data show that the probability of partisanship increases with both ethnic diversity
and the number of parties in a state. It is also clear that ethnic diversity has a much larger
effect across party systems than the effective number of parties has across caste
29
demographies. I find that the effect of caste diversity is at least twice as large as ENPS on
the partisanship variable. This finding supports the group-party linkages hypothesis.
counting mechanism of partisanship outlined earlier in this paper is not clearly confirmed
with the compromise measure of caste diversity. The expectation with this finding is that
parties are more likely to mobilize along ethnic lines where groups along a politically
salient cleavage dimension are more numerous. Nonetheless, this measure does not
capture the degree of ethnic homogeneity of party leadership. Although the caste
diversity effect grows as the number of viable parties increases-- which suggests that
states with more ethnically polarized party systems have more partisans—it does not
provide information on the degree to which parties are ethnically homogenous. For
example, in Uttar Pradesh three of four parties court Muslims, three of four parties court
Brahmins, and at least two parties court the scheduled castes.27 This means that states
with high caste diversity are likely to have a variation on the degree to which parties map
to caste groups when caste diversity is held constant. It is not possible to know the degree
to which voters develop partisan tendencies following the counting heads hypothesis until
data on the ethnic makeup of party leadership compiled by Jaffrelot becomes available
later in 2008.
Moreover, the caste diversity measure does not provide information needed to
evaluate causal mechanisms through which counting the heads of co-ethnics among party
leaders translates into partisanship. If voters base their party preferences on expectations
about future distribution of benefits, we should see that parties identified with particular
27
The Congress Party, BSP, and SP court Muslims. The Congress Party, BSP, and BJP all court Brahmins.
The BSP and Congress court scheduled castes (See Chandra 2001 for a broader discussion).
30
groups maintain disproportionate levels of patronage distribution for the groups they
represent. Recent research on ethnic patronage suggests that this pattern is not inevitable
in clientelistic democracies (Kasara 2007). Further research is also needed to evaluate the
(Chandra 2004).28
Conclusion
This paper presents two primary contributions. First, it provides strong evidence
to support the relevance of partisanship in India and adds that country to the dialogue on
partisanship in new democracies. This evidence is rooted in the finding that major
predictors of partisanship in Russia (Brader and Tucker 2002), Latin America (Lupu and
Stokes (2007), and western democracies (Green et al. 2002; Huber et al. 2005)
scholarship on partisanship (Chandra 2004; Birnir 2007; Green et al. 2002). Preliminary
evidence also suggests that caste mobilization translates to partisanship. A new measure
on the ethnic composition of parties across state legislatures formulated but not included
in this paper will provide a more explicit test of Chandra’s ethnic head counting
28
This topic will be addressed in a research project recently initiated with Simon Chauchards.
31
Bibliography
Alwin, Duane and Jon Krosnik. ―Aging, Cohorts, and the Stability of Sociopolitical
Orientations Over the Lifespan.‖ The American Journal of Sociology, Vol. 97, No. 1
(July 1991), pp. 169-195.
Bannon, Alicia, Edward Miguel, and Daniel N. Posner. "Sources of Ethnic Identification
in Africa." Afro-barometer Working Paper, 2004.
Barnes, Samuel, Peter McDonough, and Antonio Lopez Pina. ―The Development of
Partisanship in New Democracies: the Case of Spain.‖
Brader, Ted and Joshua Tucker. ―The Emergence of Mass Partisanship in Russia, 1993-
1996.‖ American Journal of Political Science, Vol. 45, No. 1. (Jan., 2001), pp. 69-83.
Brader, Ted and Joshua Tucker. “Reflective and Unreflective Partisans? Experimental
Evidence on the Links between Information, Opinion, and Party Identification.‖ Working
Paper (2007).
Calvo, Ernesto and M. Victoria Murillo. ―Who Delivers? Partisan Clients in the
Argentine Electoral Market.‖. American Journal of Political Science, Vol. 48, No.4 (Oct.,
2004) 742-757.
Campbell, Angus et al. The American Voter. Chicago: University of Chicago Press,1960.
Chandra, Kanchan. ―Elite Incorporation in Multiethnic Societies.‖ Asian Survey, Vol. 40,
No. 5, Modernizing Tradition in India. (Sep. - Oct., 2000), pp. 836-865.
Chandra, Kanchan. Why Ethnic Parties Succeed: Patronage and Ethnic Headcounts in
India. Cambridge ; New York : Cambridge University Press, 2004.
Chhibber, Pradeep, Sandeep Shastri, and Richard Sisson. ―Federal Arrangementsand the
Provision of Public Goods in India.‖ Asian Survey. Vol. 44., No.3 (June 2004).
Chhibber, Pradeep and John Petrocik. ―The Puzzle of Indian Politics: Social Cleavages
and the Indian Party System.‖ British Journal of Political Science, Vol. 19, No. 2. (Apr.,
1989), pp. 191-210.
Chhibber, Pradeep and Irfan Nooruddin. ―Unstable Politics: Fiscal Space and Electoral
Volatility in the Indian States.‖ (To be published in Comparative Political Studies in
August 2008.
Converse, Phillip. ―Of Time and Partisan Stability.‖ Comparative Political Studies.
Vol. 2 (1969).
Dash, Shrinkant. ―Transmission of Ideology and Partisanship in India, Canada, and the
United States.‖ Political Psychology, Vol. 13, No. 4 (Dec. 1992).
Dalton, Russell and Martin Wattenberg. Parties Without Partisans: Political Change in
Advanced Industrial Democracies.” Oxford: Oxford University Press, 2000.
Dalton, Russell and Steven Weldon. ―Partisanship and Party System Institutionalization.‖
Party Politics, Vol. 13, No.2 (March 2007).
Eiffert Ben, Edward Miguel and Daniel Posner. ―Political Sources of Ethnic
Identification in Africa.‖ Afro-barometer Working Paper, 2007.
Eldersveld, Samuel. Citizens and Politics: Mass Political Behavior in India. Chicago:
University of Chicago Press, 1978.
Franklin, Charles and John E. Jackson. ―The Dynamics of Party Identification.‖ APSR,
Vol. 77, No. 4. (Dec., 1983), pp 957-973.
Green, Donald, Bradley Palmquist, and Eric Schickler. Partisan Hearts and Minds:
Political Parties and the Social Identities of Voters. New Haven : Yale University Press,
2002.
Heath, Oliver. ―Party systems, political cleavages and electoral volatility in India A state-
wise analysis, 1998–1999.‖ Electoral Studies., Vol. 24, No. 2 (June 2005) pp177-199.
33
Huber, John, Georgia Kernell and Eduardo L. Leoni. Political Analysis. Vol. 13, No.4
(2005).
Iyengar, Shanto. ―Children's Partisan Loyalties in a New Nation: A Research Note Using
Indian Data.‖ The Public Opinion Quarterly, Vol. 42, No. 1. (Spring, 1978), pp. 115-125.
Kasara, Kimuli. ―Tax Me If You Can: Ethnic Geography, Democracy, and the Taxation
of Agriculture in Africa.‖ American Political Science Review (2007), 101: 159-172.
Keefer, Phillip and Razvan Vlaiku. ―Policy, Research Working Paper; No. WPS 3472.‖
World Bank (2002).
Kitschelt, Henry and Steven Wilkinson. ―Patrons, Clients, and Policies: Patterns of
Democratic Accountability and Political Competition.‖ New York: Cambridge University
Press, 2007.
Leithner, Christian. ―Of Time and Partisan Stability Revisited: Australia and New
Zealand 1905-90.‖ American Journal of Political Science, Vol. 41, No. 4. (Oct., 1997),
pp. 1104-1127.
Mattes, Robert. ―Public Opinion Research in Emerging Democracies: Are the Processes
Different?‖ Afro-barometer Working Paper ,2007.
Miller, Arthur and Thomas Klobucar. 2000. ―The Development of Party Identification in
Post-Soviet Societies.‖ AJPS, Vol. 44, No. 4 (Oct., 2000), pp 667-686.
Mullainathan, Sendhil and Ebonya Washington. ―Sticking With Your Vote: Cognitive
Dissonance and Voting. NBER Working Paper 11910.
Nooruddin, Irfan and Pradeep Chhibber. ―Unstable Politics: Fiscal Space and Electoral
Volatility in the Indian States.‖ Comparative Political Studies Vol. 41, No. 8 (August
2008).
Norris, Pippa, and Robert Mattes. "Does Ethnicity Determine Support for the Governing
Party?" Afro-barometer Working Paper, 2003.
Roberts, Kenneth and Erik Wibbels. ―Party Systems and Electoral Volatility in Latin
America: A Test of Economic, Institutional, and Structural Explanations.‖ APSR (1999).
34
Sanders, David., Jonathan Burton, and Jack Kneeshaw. ―Identifying the True Party
Identifiers: A Question Wording Experiment.‖ Party Politics, Vol. 8, No. 2 (2002).
Table 2
Vote
Party-Like Stability
Political Experience
Age -0.009* -0.016***
0.003 (.003)
Voter Turnout 0.021 0.633***
(0.066) (.073)
Age*Voter Turnout 0.004** 0.013***
0.002 (.002)
Political Sophistication
State-Level Variables
Marginal Effect
Turnout Marginal Effect Party-like Vote_stability)
0 -.009 -.016
(.2%) (1.6%)
1 -.006 -.003
(.6%) (.3%)
2 -.003 .01
(.3%) (1%)