Indian Politics & Policy Volume 3, Number 1, Spring 2020 | Page 20
Indian Politics & Policy
terest to students of Indian politics. The
NES data has been used extensively to
study voting decisions and other substantive
political questions in India. 16
Since the data in this study are
measured using nominal or ordinal
scales, I convert all variables into a set
of dichotomous dummy variables that
take on values of either one or zero. The
dependent variable in this study—the
incumbent vote—is taken from a question
in the survey that asks “who did
you vote for?” and measures whether
the respondent voted for an incumbent
or not (1 = vote for the incumbent, 0 =
otherwise). In this analysis, all constituent
members of the Congress-led United
Progressive Alliance (UPA) in 2014
and the BJP-led NDA in 2019 are considered
incumbents. I drop all responses
that are either undisclosed and those
who did not vote. The total number of
observations in the sample used in this
study is 36967. Note that in 2014, the
UPA contested 540 seats and the NDA
contested 543 seats in 2019.
The independent variable of interest
is a voter’s retrospective evaluation
of household economic conditions
over the five-year period of incumbent
rule. This variable is constructed using
a question asked in both surveys:
“As compared to five years ago, how is
the economic condition of your household
today—would you say it has become
much better, better, remained
same, worse or much worse?” In addition
to the above five response categories,
respondents also had the option of
choosing “no opinion.” I created three
variables that represent those who said
that their household economic conditions
had (a) improved, i.e., respondents
choosing “much better” or “better”
(Econ.Impr), (b) remained the same
(Econ.Same), and (c) deteriorated, i.e.,
respondents choosing “worse” or “much
worse” (Econ.Wors). In the models I estimate,
the status quo condition Econ.
Same serves as the reference category.
Incumbent party attachment (or
partisanship) is measured using the
questions, “Is there any political party
you particularly feel close to?” and “(If
yes) Which party?” Respondents who
identify UPA coalition members (in
2014) and NDA coalition members (in
2019) as the party they feel closest to is
coded as 1 and others are coded as 0.
The other independent variables
function primarily as control variables
and include respondent characteristics,
such as caste-community identity, economic
class, gender, and rural, that are
likely to have an impact on vote choice.
For the caste-community identity variable,
I combine caste (Dalit, Adivasi,
Other Backward Classes, and upper
caste) and religion (Hindu, Muslim,
Christian, and Other, including Jain,
Buddhist, and Sikh) categories resulting
in Hindu Dalit, Hindu Adivasi, Hindu
OBC, Hindu upper caste, Muslim minorities
(Dalits and Adivasi), Muslim
OBC, Muslim others (i.e., Muslim respondents
who do not identify as Dalit,
Adivasi, or OBC), Christian minorities
(Dalits and Adivasi), Christian others,
Other minorities (for instance, Sikhs
who identify as Dalit), and Others who
do not identify as Dalit, Adivasi, or
OBC (for instance, Jain upper caste). All
female voters are coded as 1 and others
as 0 for the gender variable. Similarly,
16