4
MIT LIBRARIES
HS34
IN MINI
N
II
3 9080 03317 9133
I
|
\f\H'0 $~i
u
'DEWEY
I
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
POWERFUL WOMEN: DOES EXPOSURE REDUCE BIAS?
Lori
Beaman
Raghabendra Chattopadhyay
Esther Duflo
Rohini
Petia
Pande
Topalova
Working Paper 08-1
July 17,2008
RoomE52-251
50 Memorial Drive
Cambridge,
MA 02142
This paper can be downloaded without charge from the
Paper Collection
http://ssrn.com/abstract=1 162358
Social Science Research Network
at
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium
Member
Libraries
http://www.archive.org/details/powerfulwomendoeOObeam
Women: Does Exposure Reduce
Powerful
Lori
Beaman, Raghabendra Chattopadhyay, Esther
Bias:
Duflo,
Rohini Pande and Petia Topalova*
Abstract
We
exploit
random assignment
of gender quotas across Indian village councils to investi-
gate whether having a female chief councillor affects public opinion towards female leaders.
Villagers
who have
never been required to have a female leader prefer male leaders and per-
ceive hypothetical female leaders as less effective than their
performance
is
identical.
erence for male leaders.
Exposure to a female leader does not
However,
it
perceived
likely to
among male
know about
villagers.
alter villagers' taste pref-
weakens stereotypes about gender
and domestic spheres and eliminates the negative bias
is
male counterparts, when stated
Female
in
how female
roles in the public
leaders' effectiveness
villagers exhibit less prior bias,
when exposed
to
them
for the first,
but not second, time.
attitude are electorally meaningful: after 10 years of the quota policy,
to stand for
and win
less
or participate in local politics; as a result, their attitudes are largely
unaffected. Consistent with our experimental findings, villagers rate their
less effective
but are also
free seats in villages that
women
leaders as
These changes
women
are
more
in
likely
have been continuously required to have a
female chief councillor.
RW.J Berkeley and Northwestern, IIM Calcutta, MIT, Harvard and IMF respectively.
paper are those of the authors and do not implicate the International Monetary
Fund, its management, or Executive Board. We thank Ash Center Harvard, Nike Foundation, MIT, YCIAS Yale,
and UNICEF for funding, Alexandra Cirone, Catherine Lee and Kudzai Takavarasha for research assistance, and
Prasid Chakraborly and the SRG team for outstanding fieldwork We also thank Abhijit Banerjee, Tim Besley,
Shawn Cole, Dominic Leggett and numerous seminar participants for comments.
"The authors
are from
The views expressed
in this
Introduction
1
women
In July 2006,
accounted for 17 percent of parliamentarians world-wide, and a
headed the government
in
not reflect legal restrictions -
every country. Rather,
public office
is
many
restricted
(UNICEF,
only seven countries
women can
vote, support candidates
and run
disparities
for office in
do
almost
suggest that in both rich and poor countries, women's access to
by voter and party bias
At the same time, such
These gender
2007).
woman
bias,
even
if
present,
in favor of
may be
male
politicians.
malleable. For this reason, policy-makers
often emphasize the importance of an initial cohort of female politicians and argue that once
voters learn that
women
diminish. This has led to
for
women
in
public
are no less effective than male politicians, gender bias in politics will
more than a hundred
office, either
countries introducing affirmative action policies
by law or through voluntary actions of
1
the last two decades (Krook, 2005; Dahlerup, 2006).
While these
political parties, over
policies
have significantly
increased female representation in politics (Jones, 2004) and often altered subsequent policy-
making (Chattopadhyay and
Duflo, 2004; Powley, 2007),
little is
known about
their
impact
on voter attitudes towards female leaders, despite the obvious importance of this question
for
evaluating the overall effectiveness of quotas.
A
large literature
issues can cause
ller,
1992;
on public opinion formation suggests that low voter awareness on
group stereotypes to play an important
Huddy and
stereotypes
may
A
Terkildsen, 1993).
hinder the entry of
women
associating leadership activities with
men
role in
shaping voter preferences (Za-
widely discussed reason that explains
into politics
is
social
specific
why group
norms which prescribe behavior
(Akerlof and Kranton, 2000; Eagly and Karau, 2002).
Holding leader effectiveness across genders constant, such norms can cause voters to favor, and
elect,
male candidates. Limited exposure to female leaders may,
in turn, fuel biased perceptions
of female leader effectiveness.
While
in
many
extent to which
settings,
exposure to members of another group creates "empathy,"
mandated exposure
to
women
leaders can alter social
norms
the
or perceptions
details of quotas by country are available at http://www.quotaproject.org. Another rationale for affirmative
is gender-specific policy preferences - if male and female leaders have different policy preferences, and
female turnout is relatively low, then affirmative action can help ensure equal political voice for men and women.
2
For example Boisjoly et al. (2006) show that students who were randomly assigned to an African- American
roommate in college are more likely to sympathize with African- Americans and affirmative action.
action
of
women's
ability
is
debated among academics and policymakers. Using international survey
data, Ingelhart and Norris (2003)
across countries.
They argue
document sex-biased attitudes towards
roles in public life
that these attitudes are difficult to change in the short run and
quotas, by violating voters' sense of identity,
may
precipitate a backlash
2004). For instance, a quota-induced increase in female politicians
(Rudman and
may be
Fairchild,
perceived as reducing
the value of a traditionally male activity (Goldin, 1990) and /or as restricting voters' choice
(Thernstrom and Thernstrom, 1997).
However, independent of their impact on voter
leader, can
tastes, quotas,
by exposing voters to a female
improve the precision of voters' information about the expected effectiveness of future
female leaders. Quotas could, therefore, reduce statistical discrimination. Unless
women make
incompetent leaders (causing voters to update negatively), mandated exposure can improve the
perception of
a
women
leaders' effectiveness simply
by reducing the risk associated with electing
woman. 3
This paper provides direct evidence on these
issues.
To motivate our
empirical analysis
we
first
use a simple voting model to show how, in an electoral setting, future male candidates
may
systematically benefit from an initial voter preference for male leaders. Specifically, greater
exposure to male leaders improves voter ability to screen male candidates and as a
result, causes
the expected effectiveness of male candidates to exceed that of their female counterparts
if
voters are risk averse. Statistical discrimination against female leaders will reinforce the initial
and cause the stereotype that women make
taste bias
setting, political reservation,
discrimination even
when
In general, measuring
ficult.
ineffective leaders to persist.
by increasing voter exposure to female leaders, can reduce
voter taste
is
leaders.
To address
statistical
unaffected.
how exposure changes
voters' attitudes vis-a-vis
women
leaders
Less biased voters are presumably also more likely to elect, and be exposed
4
In such a
this concern, our empirical strategy exploits
exposure to female leaders across village councils
in the
random
variation in
to,
is dif-
female
mandated
Indian state of West Bengal.
Since
1998, one third of village council leader positions in this state have been randomly "reserved"
3
One
work less
4
This
reason
is
why quotas
could lead to less competent female leaders
is if,
as in
Coate and Loury (1993), they
anticipation of voter perceptions that quotas lead to less competent politicians being elected.
in
true even with affirmative action, since in most countries even
sentation of
women,
parties can
still
decide where to
field
women
when a law
requires
mandated
repre-
candidates (Frechette, Maniquet, and Morelli).
women
for
in
each election;
in reserved councils
only
women
can run for the position of leader.
This, combined with the passage of a significant period of time since the introduction of political
reservation (two electoral cycles, or ten years)
our empirical analysis.
We
compare
makes West Bengal
villager attitudes
leaders across councils which have been reserved for
villages
an ideal setting
women
towards hypothetical, and actual,
women
Random
once, twice, or never.
for
allo-
cation of reservation implies that a difference in voter attitudes across reserved and unreserved
mandated exposure.
villages captures the causal effect of
An important
innovation of our study
perimental data on voters' taste
female leaders' effectiveness.
plicit tastes are
leaders,
IAT
is
and
We
is
the collection and use of detailed survey
and
ex-
female leaders and their perceptions of gender roles and
for
examine
explicit
and implicit measures of voters'
tastes.
5
Ex-
captured by voters' stated feelings towards the general idea of male and female
implicit tastes by taste Implicit Association Tests (IATs) (Nosek et
an experimental method now widely used
that respondents
who more
associate those concepts.
easily pair
Our
taste
in social
two concepts
IAT
is
in
al.,
The
2007).
psychology, which relies on the idea
a rapid categorization task more strongly
a computer-based double-categorization task which
examines the strength of respondents' association between images of (anonymous) male and
female leaders and normative categories of good and bad. To measure gender-occupation stereotypes we use an
IAT which examines
and leadership and domestic
we asked
the strength of association between male and female
tasks. Finally, to
examine voter perceptions of leader effectiveness
villagers to evaluate the effectiveness of hypothetical female
through vignettes and recorded speeches
Clearly, our
in
which leader gender
is
and male leaders described
experimentally manipulated.
measures of "taste" and "perception of effectiveness" remain imperfect: a
lager's perceptions of a potential leader's effectiveness
may
villager's general distaste for female leaders.
the general "feeling" variables remain
more
That
said,
vil-
influence his feelings towards the
idea of male and female leaders; conversely, the performance rating of a leader
by a
names
may be
our maintained hypothesis
colored
is
that
closely correlated with deep-seated tastes while the
specific "effectiveness" questions reflect voter perceptions of leader
performance.
In villages that have never experienced political reservation, villagers, particularly men,
dislike the idea of female leaders.
5
On
On
a scale of
1
to 10, the average
the use of such measures, see Charles and Guryan (2007); Bertrand et
man
al.
rates his feeling towards
(2005);
Rooth
(2007).
female leaders in general one whole point below those for male leaders in general.
show that men
more
are significantly
IAT measures
male leaders with notions of "good"
likely to associate
relative to female leaders.
They
The average male
ranked the same speech and vignette describing a leader's decision
villager
0.05 standard deviations lower
also perceive female leaders as less effective
when
than male leaders.
the leader's gender was experimentally manipulated to be
female (relative to male). Female villagers' evaluation of hypothetical female leaders, while
negative,
is
not statistically different from male villagers'.
Mandated exposure
to a female leader does not affect villagers' stated taste for male leaders.
IAT show
Neither the "feeling" rating of leaders nor the taste
leaders in villages reserved for a female leader.
and women with domestic
leader effectiveness
among male
activities.
villagers. In the
increased approval of female
However, among male villagers
the stereotype (as measured by the occupation IAT) that
activities
less
men
it
weakens
are associated with leadership
radically alters perceptions of female
It also
speech and vignette experiments, we find that
male villagers who have been required to have a female leader consider hypothetical female and
male leaders equally
effective.
This reduction
in bias is
absent
provide evidence suggesting that a likely reason for this difference
knowledge, and exposure to local
politics,
among female
is
villagers.
We
the lower levels of political
among women.
Consistent with the experimental data, we find that prior exposure improves villagers' evaluation of their actual leader along multiple dimensions. Specifically, in villages where the leadership
position
is
reserved for the
first
time and only time in 2003, leader ratings by male villagers are
where the leader position has never been reserved. Despite
significantly lower relative to villages
their
less
poor evaluation, we show that
first
time female leaders deliver more public goods and take
bribes than their male counterparts.
leaders
in 2003.
is
The
difference in ratings
absent in villages where the leader position
We
do not
find
is
reserved for
prima facie evidence that female leaders
between men and
women
in first
for the
women
second time
time reserved villages
are differentially selected, or undertake different policies, than leaders elected in the second
round of reservation. 6
Our
results suggest that while
The number
of public goods provided by
the second time does not significantly
on average, and
deep preferences and social norms are
all
women
differ.
first
First
difficult
to erode,
time female leaders and women leaders in councils reserved for
and second time female leaders also have similar characteristics
leaders are equally biased in favor of
women
in their policy actions.
affirmative action
and reduce
programs can
significantly
improve perceptions of female leader effectiveness
statistical discrimination against
women
in leadership positions.
Finally,
striking evidence that political reservation leads to significant electoral gains for
medium-run. In the recently concluded
many women stood
for election
chief councillor position
to villages which
The
had
rest of this
and Section
village council elections in
May
for
women
in
women
find
in the
2008, almost twice as
and won unreserved councillor positions
had been reserved
we
in councils
where the
the previous two elections,
compared
either never been reserved or reserved only once.
paper
is
structured as follows. Section 2 provides a conceptual framework,
3 describes the data. Section 4 lays out the institutional context
and our empirical
and Section 6 concludes.
strategy. Section 5 reports the results,
Conceptual Framework
2
In most elections, voters
We
available candidates.
and Cain, 1977) to
make
their choice while uncertain
about the relative competence of the
use a simple model of statistical discrimination (which follows Aigner
may
illustrate that voter uncertainty
distaste for female leaders.
The
result
is
interact with an underlying voter
that female candidates are, on average, not elected
negatively biased beliefs about their effectiveness as leaders persist.
Political reservation
and
can
play an important role in reducing the bias in beliefs.
Consider a single village council leadership election with the outcome determined by plurality
rule.
Candidates
differ in
competence,
resolve disputes. Candidate
We
assume voters are
and
/"(.)
<
0.
i's
competence
risk averse
Risk aversion
is
e.g. ability to
given by
is
and value candidate
with the resources a leader raises
politicians for
The
aversion
of
whom
7?,,
i's
a standard assumption in
the context of village council elections, risk aversion
public resources he or she gets.
extract state resources for the village and
for the village
An important
is
and a
where
77,
=
r\
+
e,
and
effectiveness as f(rji),
many
political
very plausible
if
villager's utility
implication of risk aversion
e<
~
where
iV(0, al).
/'(.)
>
economy models. 7 In
we equate competence
is
is
concave in the total
that voters penalize
they lack information.
typical spatial voting model, which uses a quadratic utility function, implies risk aversion in voting. Risk
is
also
commonly invoked
ambiguous candidates.
to explain electoral
outcomes such
as split-ticket voting
and the punishment
Voter
j's utility
from candidate
i
Uij
where
m
l
is
a male
dummy
defined as:
is
= f(m) +
which equals one
am + Uij,
the candidate
if
can be interpreted as reflecting taste discrimination against
women
implement.
Let Xi €
M, F
u> t j
~
N(Q,
er^) ls
an
represent candidate
i-i-d
random
male.
is
women
~
N{Q,<j* x ).
men, and these
can cause
x and
we only allow the variance
rflx is given
is
rn
+
vlx
sufficient
This
,
where
to cause
are less competent than
However, to focus on the possibility that
risk aversion
(of statistical discrimination reinforcing taste
to suffer a "double penalty"
of the signal to differ across genders.
who maximizes
Voters select the candidate
women
=
77^
across genders
voters have prior beliefs that
beliefs are self-fulfilling).
women
discrimination),
if
0.
voter and politician specific taste shock.
mean
In general, a difference in signal
statistical discrimination (e.g.
assume a >
or a dislike for the policies
each candidate's gender x, and a noisy signal of his/her competence:
v ix
We
Before casting their vote, citizens observe
gender.
i's
(1)
their expected utility.
The expectation
of
r/i
given
by
E{r)i\rfix ,x)
=
-
77(1
7)
+
7fo7
2
where 7 = -rf*?-- The error term
e
=
in this
equation
E{r)i\m x ,Xi) -rji
is:
°tx
+ °n
which has variance
2
2
+
°ix
(T„(7„
°f,
The
variance
effectiveness,
is
is
increasing in a„ x
Since /(.)
.
is
concave,
E[f(r)i)\rfi X ,Xi],
declining in the variance of the signal on ability.
Consider the case where the signal on candidate of gender x
relating observed characteristics
to their competence:
8
This
While an
is
the expected leader
rflx
=
jix
and actions Xj
X
x
,
where the
for previously
coefficients are
is
the prediction of a model
observed leaders of gender x
gender
specific.
8
If
the model
a natural assumption, since men and women may differ in both their actions and characteristics.
man may be an effective leader, being assertive may be a liability for a woman.
assertive
is
estimated using
Nx
past leaders' observed actions, then the standard errors of
standard deviation of the resulting prediction error
of the signal for a gender will decrease in the
will
number
decrease with
(5X
and
the
s/Nx Hence the variance
.
of elected candidates (of that gender).
Taste discrimination, or a preference for the policies selected by men, will imply, for equal
expected competence, higher voter
will
be elected only
for her
in
little
relative to the
with a male leader (since
basis to
of the population.
male candidate, the
2
<j
vF
will
This creates a vicious
because voters have a distaste
for
Political reservation increases
Pf and
women
To summarize,
unchanged by
women
if
lead
from a signal about the female
circle
women
is
leaders, risk averse voters will prefer
- a female candidate
leaders,
male
lower (and so their expected effectiveness
is less likely
to
be elected both
and because they are perceived as too
risky.
10
Nf. This, in turn, improves the precision of voters' estimate
competence across genders, reservation
the effectiveness of
women
In other words,
reduces the variance of expected competence of future female leaders.
similar levels of
is
sufficiently favorable
2
be very high, compared to o vM
? This causes statistical
candidates as the variance of their ability signal
of
is
Female candidates
Therefore, voters will see very few
ability to extract information
discrimination: aside from any distaste for
higher).
0).
form a prediction of female politician effectiveness.
candidate will be weak and
is
a >
the few cases where the idiosyncratic taste shock
among a majority
and have
utility
will
Hence,
for
improve the relative evaluation of
leaders.
the pool of female candidates (and their investment in job-specific
skills)
political reservation, then:
Political Reservation
and Taste Discrimination:
Taste discrimination, as captured by a
greater preference for male leaders (relative to female leaders), will be unchanged with political
reservation.
Political Reservation
With
and
Statistical Discrimination: Holding competence constant across
a positive cost of running, presumably very few
women
will
choose to run which
will
accentuate the
signal-extraction problem for female candidates.
10
The voter's decision problem resembles the classic two-armed bandit problem (Bergemann and Valmiki, 2008).
The two arms are the male and female candidate pools, with unknown variance with respect to competence. Voters
get new information on the distribution of competence within a gender only if they select a candidate from that
gender. Since the observations the voter uses to gain information and reduce the variance of the signal are also his
terms of selecting a leader), he must strike a balance between gaining rewards and gaining information.
voter will combine information on the mean with beliefs about the variance of payoffs to maximize
the present discounted value of his payoff. He will favor exploiting information (i.e. choose the male candidate
arm) if he is sufficiently risk averse, faces a finite time horizon or has a sufficiently high discount rate.
reward
(in
The median
genders, political reservation will reduce statistical discrimination as measured by the difference
in
the evaluation of female and male leader effectiveness.
The
literature on affirmative action has identified
our model. First,
leaders, then
political reservation
if
two additional
which
competent female candidates may respond either by under-investing
predicted decline in statistical discrimination. Second,
voting choices constrained then political reservation
in taste discrimination against
The purpose
of our
model
if
may
effect, if present,
Our empirical
(Coate
cause a backlash and lead to an increase
risk averse.
1997).
As mentioned
similar predictions regarding the efficacy of political reservation
is
in skills
would reduce the
is
a natural by-
earlier,
we obtain very
to illustrate that statistical discrimination
product of taste discrimination when voters are
of statistical discrimination
outside
voters react negatively to having their
women (Thernstrom and Thernstrom,
is
lie
worsens voters' prior about the competence of -women
and Loury, 1993), or by simply choosing not to stand. This
if
we assumed
that the source
biased voter priors about average female leader effectiveness. 11
analysis will not try to disentangle different forms of statistical discrimination.
The remainder
of this paper uses detailed data on voter attitudes to test these
about the implications of
3
forces,
political reservation.
two predictions
Below, we start by describing our data.
Data
Our data comes from West Bengal, a middle-income Indian
(Gram Panchayats, from now
on,
GP) have been an
effective
Between June 2006 and November 2007 we surveyed 495
Birbhum
village
district in
we
West Bengal (we randomly
collected data on the quantity
state,
where elected
body
of
governance since 1978.
villages spread across the 165
selected three villages per
and quality
village councils
of public
GPs
in
GP). In each surveyed
good provision, and administered
household and individual surveys to a random sample of 15 households. These surveys, together
with Implicit Association Tests (IATs), form the main sources of our voter attitude measures.
about average competence can exist, for example, if voters conclude that the lower
and wealth among female leaders (relative to male leaders) reflect lower effectiveness.
'initial biased beliefs
of education
levels
Measures of Voter Taste
3.1
We
measure voters' taste by
their explicit
as expressed in survey responses
households per
5
To
village.
and
in
and
IATs.
implicit feelings towards
We
leader, a
"Feeling
of
12
to 10,
1
random sample
collected these data for a
ascertain respondents' explicit taste for male
which has steps from
male and female leaders
how do you
male leader, a female
villager
feel
and female
about a
and a male
Thermometer" which has been widely used
leaders,
X
where
[X]"
we asked "on a ladder
was (separately) a female
This question
villager.
is
adapted from the
in the political science literature to
produce
rank-orderings of parties and candidates (Keller and Mirer, 1974) and to measure partisan
affiliation (Alvarez, 1990;
We
IAT
is
complement the
Weisberg, 1980).
explicit taste
measures with taste IATs (Greenwald
et al., 1998).
a computerized test which aims to measure attitudes of which respondents
explicitly cognizant.
Over the
last ten years
IATs have been widely used (Nosek
et
may
al.,
An
not be
2007) in
various subfields in psychology, neuroscience, market research and, recently, economics. 13
To measure
implicit bias, an
of the association
IAT
uses a double-categorization task to measure the strength
between two concepts. Words from two
different series
Consider the case where the two series are a series of male and female
of
good
or
bad
adjectives.
The respondent completes two
appear on the screen.
first
test blocks.
names and a
series
In the "stereotypical"
block male names and good attributes are clubbed together on one side of the computer screen
(say, left side)
or seeing a prompt, the respondent has to assign
it
example given, a male name should be assigned to the
a
bad
attribute to the right (and a
this association
(i.e.
attributes on the
hearing
to the correct side of the screen.
In the
In each household
good one to the
left
left).
The assumption
left).
easier.
An
is
side (right side).
(and a female
The
males names and bad attributes on the
makes the sorting task
l2
On
and female names and bad attributes on the other
name
to the right)
and
"non-stereotypical" block reverses
right,
and female names and good
that a stronger association between two concepts
automatic association can be detected by comparing the
we conducted IATs with adults aged between 15 and
45.
We targeted at least one respondent
IAT respondents from 1,968 households. We do not use IATs from the
seventy-seven GPs. As our feeling questions were not affected by piloting, we include
per gender in a household, and have 4,378
one pilot village
in
each of
these villages for feelings data giving us 7,182 survey responses across 2,926 households.
13
Rooth (2007) shows that the D-measure
for racial bias of
Swedish employers correlates positively with the
rate at which they call back applicants of different races. Also see Bertrand et
al.
(2005).
response time across the stereotypical and non-stereotypical block.
mean response time between
ineasure: the normalized difference in
"stereotypical" test blocks (Greenwald et
al.,
2003).
IAT
bias
is
given by the D-
the "non-stereotypical" and
A
higher value of the D-measure indicates
to
implement IATs
a stronger implicit stereotype.
To the
we
best of our knowledge,
are the
first
in a field setting in a
low income country. Our IATs used either audio or pictorial prompts and, therefore, did not
assume
literacy or familiarity with
computers (Appendix Figure
a picture of a participant playing the game.
The Appendix
lists
1
shows an IAT screen and
the set of prompts). 14
to exploit variation in respondent exposure to female leaders to
second innovation
is
the malleability of
IAT
bias.
We
Our
examine
are only aware of one other field study which investigates this
question (see Section 3.2).
We
administered two taste IATs; such IATs have been widely used to measure group prejudice
The
(Banaji, 2001).
good
(e.g nice
first
assesses the associational strength
and delicious) and bad
(e.g.
between male and female names and
nasty and horrible) attributes.
The second measures
the association between these attributes and images of male and female politicians (e.g pictures
of either
men
places male
names
3.2
or
women
names
giving speeches or leading crowds). In both cases the stereotypical block
or leader pictures
or leader pictures and
Measure
Our model
and good attributes on one
bad attributes on the other
side.
unlikely to reduce taste discrimination. Instead,
is
that the main channel by which reservation influences voter beliefs
information on prospective
women
would be a greater association
To examine changes
in the
of
leaders.
women
A
is
enhanced
natural consequence of the increase in information
with leadership
activities.
gender stereotyping of occupations we administered an activity-
based IAT. The IAT assesses whether villagers exposed to reservation are
women
and female
of Leadership Stereotype
suggests that political reservation
we hypothesize
side of the screen,
with domestic activities and
men with
leadership activities.
less likely to associate
Specifically,
it
examines
14
Similar IATs have been used for children (Baron. and Banaji, 2006). Wc excluded respondents with an average
response time of over 6 seconds in the first test block or less than 65% average correct responses. We dropped
all
responses
for
which the respondent took more than 10 seconds to categorize any prompt.
ordering of stereotypical and non-stereotypical blocks across respondents.
10
We
randomized the
the association between male and female names and domestic
and leadership
rice)
(e.g. governing
towards associating
women
which were equally
likely to
activities.
sleeping
and eating puffed
To avoid loading the
results
with domestic activity, we chose gender neutral domestic activities
be performed by
Unlike the taste IATs, this
of
and public speaking)
(e.g.
IAT does
not capture a value judgement. Rather,
IAT
to
more strongly with family than with
it is
informative
men and women
as potential leaders.
Rudman and
show that respondents
in the U.S. associate
female names
whether the respondent considers both
Kilianski (2000) use such an
men and women.
career.
There
is
also
some evidence that responses to occu-
pation IATs are influenced by the environment. Dasgupta and Asgari (2004) administered this
IAT
to
American
college students
of female professors were
more
and found that students
likely to associate
women
in colleges
with a higher proportion
with professional
activities.
endogenous selection into college makes a causal interpretation of these results
However,
difficult.
Measures of Leader Effectiveness
3.3
To measure
bias in voter perception of leader effectiveness,
we administered speech and vignette
experiments of the "Goldberg paradigm" type (Goldberg, 1968). Such experiments, which ran-
domly vary the gender
of the protagonist, have been widely used to assess bias in the perception
of the effectiveness of
women
as leaders (Matland, 1994;
and Karau, 2002)). Our experiments were administered
prime-aged male and female respondent
we conducted
in 15
Huddy and
Terkildsen, 1993; Eagly
as part of a longer survey given to a
randomly selected households per
village. Overall,
6,717 male and 6,780 female adult surveys.
In the speech experiment the respondent heard a short tape-recorded leader speech which
was adapted from an actual
village
of six speech recordings (three
village council leader in a village
to a villager complaint
and
meeting speech. Respondents were randomly assigned one
male and three female) and told that
meeting
in
another
this
was a speech by a
district. In the speech,
the leader responds
about a broken tube-well by requesting
effort for local public goods.
15
villagers to contribute
money
After hearing the speech the respondent was asked seven
questions on different aspects of the leader's perceived performance and overall effectiveness,
such as
5
A
"How would you rank
literal translation is
the effectiveness of the Pradhan on a scale of
included in the Appendix.
11
1
to 10?"
and "Do
you think the Pradhan
be successful
will
in
getting resources to the village by lobbying?".
did not include any questions on whether the villager sympathized with the leader; rather,
We
all
questions were about evaluating the leader's action and his or her effectiveness.
Each respondent
also
heard a randomly selected "vignette "in which a situation of resource
was described and the leader decided whether to invest
scarcity
irrigation project.
in a drinking
water or an
Vignettes varied along two dimensions: the leader's choice and the leader's
gender. At the conclusion of the vignette, respondents were asked to evaluate the leader using
four questions, similar to those asked after the speech.
The
leader's gender
was randomly varied across respondents such that a respondent was
exposed to the same leader gender
lagers' responses across the speech
in the
speech and vignette. This allows us to combine
and vignette and ask whether, holding actual action constant,
on average, rank female leaders below male
villagers,
Finally,
we asked
leaders.
villagers to evaluate their actual leader's effectiveness
sions (using a ladder-based scale of
1
to 10).
3.4
on multiple dimen-
Questions included "Do you think the Pradhan
has done a good job looking after the needs of your village?" and
effectiveness of the current
vil-
"How would you rank
the
Pradhan?"
Discussion
Clearly, neither our
"taste"
nor "perception of effectiveness" measure
counterpart to the corresponding parameter in the model. Voters
is
a perfect empirical
may be more
likely to associate
female leaders with bad adjectives in general because they consider them incompetent. Equally,
villagers
may deem
a
woman
leader giving the speech ineffective because they
want to punish
her for violating norms of prescribed behavior. In other words, akin to the utility function
posit, our
measures
may
we
capture both voter taste and perception of effectiveness. However, as
long as the relative weights of taste and statistical discrimination vary across these measures
expect our taste measures to be, on average,
less affected
effectiveness measures.
12
we
by reservation than our perception of
Institutional Context
4
As a background
and Empirical Strategy
for the empirical analysis,
we describe
political reservation in
West Bengal and
our empirical strategy.
The Panchayat System and
4.1
Political Reservation
many prominent
India has had universal franchise since Independence, as well as
including a prime-minister and
now
its first
female leaders,
female President. However, the share of
national and state legislatures averages only 10 percent.
An
women
in
important policy response was the
1993 constitutional amendment which, while mandating a three-tier elected local self-government
institution in every Indian state, the Panchayat, required reservations, or quotas, for
all
three
villages,
tiers.
This policy has led to a dramatic
and the number of
is
now
In each Indian state, the Panchayat consists of a system of village
(Panchayat Samiti), and
Gram
council,
district (Zilla
Panchayat, which
buildings, water, roads, etc.)
The main
Unlike
source of financing for
many Indian
its activities is
The GP
is
in this state
active elected Panchayat since 1978.
West Bengal Panchayat model.
five years.
council decision-making
veto power, he or she
is
is
These councillors
the only full-time
states,
typical
Gram
is
elect (from
directly elected from each
among
ward
the set of councillors) a
by majority voting. While the Pradhan does not have formal
member
al.,
on Pradhans.
some other Indian
The
and a vice-chief councillor, Upa-Pradhan.
control over the final council decisions (Besley et
In
In fact, the 1993
has an electorate of roughly 10,000 spread across multiple villages.
chief village councillor, Pradhan,
GP
focuses on the village
the State government.
divided into electoral wards, and a ward councillor
by plurality rule every
Our study
states which introduced Panchayats only after the 1993 constitutional
was, in part, based on the
Panchayat (GP)
(Gram Panchayat), block
targeted welfare recipients within the villages.
for identifying
amendment, West Bengal has had an
amendment
close to 40 percent.
responsible for the provision of village infrastructure (public
is
and
Parishad) councils.
in
female leadership across Indian
rise in local
female elected leaders
village-level
women
Pradhan
elections are direct.
L3
of the council
and exercises
significant
2007). In our analysis we, therefore, focus
The
in
amendment
constitutional
each council be reserved
be reserved for women.
1993 required that one third of ward councillor positions
women. And
in
each
district,
one third of the Pradhan positions
In other words, at each election a third of
identified as "reserved" for
Each Indian
for
in
women: only women can be
GPs
in every district are
Pradhan
elected as
in these
GPs. 17
state separately decided the rules for implementing political reservation.
West Bengal the Panchayat Constitution Rule was modified
the positions of ward councillors and Pradhans for
in
women and
In
1998 to introduce reservation for
for
two disadvantaged minorities,
Scheduled Castes (SC) and Scheduled Tribes (ST) (Government of West Bengal, 1998). At the
time our data was collected, two elections had been conducted with reservation:
2003. In Section 6
we
discuss the results of the third election, held in
May
1998 and
in
2008.
In the case of Pradhans, the West Bengal electoral rules state that, prior to an election,
in
a district should be randomly assigned to three
and tables provided
These
lists
GP
GP
elections. In each
starting with
starting with
using
(the
GP
Reserved
in the electoral law):
for
serial
number
number
2
1
GPs
list
on each
on each
numbers and
same was true
reserved for
list
list
was reserved
was reserved.
women. Second,
ST, and Unreserved. 18
features on the
SC
or
ST
list for
for
a
woman, and
We reconstructed
in
2003 every third
the 2003 reservation
list
Chattopadhyay and Duflo, 2004).
there
(since
GP
for
and found that the rule held, with no exception
electoral law tables
in 1998, see
and administrative block
list
SC, Reserved
number rank
are ordered by their serial numbers. In 1998, every
This assignment rule has three consequences.
may be
(on the basis of their serial
are redone at every election to ensure that no
two consecutive
third
lists
GPs
GP
is
First, there is
implicit stratification of
numbers
serial
reserved twice in a row - for instance,
random assignment
women
randomization by
start with a block identifier).
if it
was
first
on a
to being
list in
19
SC/ST
Third, a
GP
1998 but second on a
in 2003.
Figure
1
shows the patterns
in
GP
reservation in our study district of Birbhum.
''Electoral gender quotas usually require that
women
Of the 56
constitute a certain percentage of either party candidate
These quotas are typically constitutionally mandated (as in
Burkina Faso, Nepal, Philippines and Uganda), mandated by electoral law (as in many parts of Latin America,
Belgium, Bosnia and Herzegovina, Serbia and Sudan) or imposed by political parties (as in Argentina, Bolivia,
Ecuador, Germany, Italy, Norway and Sweden). A growing concern that requirements on candidate lists may
not increase actual female representation in politics has led a number of countries (such as Jordan, Uganda and
lists
or the elected body, such as the legislature.
Rwanda) to introduce gender quotas using political reservation, similar to the Indian system.
18
The extent of reservation for SC and ST in a district is proportional to their population share.
19
A
block
is
the administrative unit below the district.
14
GPs
the
reserved in 1998, 20 were also reserved in the 2003 election. Thirty five were reserved for
Pradhans
Birbhum
at the
in
2003 and 74
GPs
reserved for
time
first
in
is
GPs had
women
never been reserved at the time of the survey.
and the
are female,
overall fraction of female
GP
43 percent (due to political disturbances one reserved
Pradhans
did not have a
in
Pradhan
time of our survey).
GPs
In non-reserved
few
women
are elected, but this
number has increased over
The
facie the increase appears to be unrelated to the reservation policy.
GPs
elected in 2003 from
of
All the
women Pradhans
In Table
1
that were only reserved in 1998 (14%)
elected from never reserved
we use 1991 census data
GPs
First Reserved 2003
in 1998,
Prima
fraction of
women
comparable to the fraction
2003 (16%).
to check for systematic differences between villages in
our sample, based on the reservation status of the
Reserved
in
is
time.
GP
and Reserved
they belong
in
to:
Never Reserved, Only
As expected given
1998 and 2003.
the randomization, the village characteristics as of 1991 do not jointly predict the reservation
assignment of
GPs
level or less) differences
6,
having a hand
(6)).
Our
There are
in the local elections.
statistically significant (at the 10 percent
according to reservation status in only three variables (sex ratio under
pump and
having a permanent approach road, see p- values
in
columns
(5)
and
regressions include controls for these three variables (the regression results are not
sensitive to their inclusion).
Empirical Strategy
4.2
As explained above, the sample has
four types of GPs:
First Reserved 2003, Reserved 1998
and 2003, Only Reserved 1998, and Never Reserved. Randomization
allows us to study
GPs
its
reduced form
effect
by comparing the means of outcomes of interest across
with different reservation status. Since
reserved GPs, but relatively few unreserved and
all
previously reserved GPs, have a female Pradhan (Figure
what would be obtained by instrumenting
GP. We,
The
out,
therefore, focus
for
on the reduced form
unit of observation in our regressions
we report separate
results for
of reservation assignment
reduced form
1), this
effect is close to
Pradhan's gender by the reservation status of the
results.
is
villager
male and female
the regression specification differs across these.
15
i
in
villagers.
The
GP
We
first set
g situated
in
block
b.
Through-
have two sets of outcomes, and
of
outcomes are common across
'
respondents, and include the difference in a villager's feelings towards male versus female leadthe
ers,
IAT D-measures, and
The
specifications.
first
we report two
evaluations of the Pradhans. For these outcomes
simply compares outcomes of interest across ever reserved
GPs
and
never reserved GPs,
=
y ig
where
Rg
is
an indicator variable
ever reserved).
The second
Vig
Rg
is
i
an indicator
for the first
GP
the
time
+
GP
for the
+ ab +
Xig-y
e ig
GP
(2)
being currently or previously reserved (now on,
separates reserved and unreserved
= @2Rg2 +
for the
(3R g
fcandl Rg2andl
+
PlRgl
GPs
+ Xigl +
only being reserved in 1998 and
further:
"6
Rg 2
2003 (during the second round of reservation).
in
+
(3)
fyg
for the
Rg 2 an di
GP
'
s
being reserved
an indicator
for
being reserved in both 1998 and 2003. Standard errors are always clustered by GP.
All regressions include block fixed effects
ucation, caste, religion, household size
principal
component
analysis.
We
(aj,)
and proxies
and demographic controls (Xig ): age,
for
ed-
household wealth constructed using a
also include investigator
gender fixed
effects, a
survey period
hand-pumps and presence
of
outcomes consists of the vignette and speech responses. As described
in
indicator and three village controls (under 6 sex ratio, presence of
approach road).
Our second
set of
Section 3.3, each respondent received a vignette about a Pradhan and heard a Pradhan speech
in a
male or female
voice.
are interested in whether
The gender
of the
Pradhan was randomized across respondents.
male and female Pradhans are judged
difference varies with reservation status of the
presented with a "female" stimulus
vignette with a female leader).
y, g
We
(i.e.
differently,
We
and whether the
GP. Let Fi g indicate whether respondent
i
was
heard the speech in a female voice, or was given the
estimate:
= 5Fig + X(Rg
*
Fig + \iRg + X ig j +
)
16
ab +
e ig
(4)
F
and
Vxg
= SFig + A 2 (-Rg 2 * Fig + A 2and {Rg 2and\
)
i
*
ig
)
+ \\ {Rg
\
* Fjg)
+ y^ R k
fj.
k
+ X ig ~j + a b + e lg
,
(5)
k
Rk
the indicator variables
main
control for the
effect of different reservation categories.
In equation (4) the coefficients of interest are S which captures bias towards female leaders
in
unreserved GPs, and A which indicates whether current or past exposure to a female leader
changes the
level of bias. In
equation
we
(5),
are interested in A 2
On many
we ask respondents multiple
issues of interest
outcomes we expect the
on the variables of
coefficients
Ai,
and how they
outcomes within a "family" (Kling
et
al.,
2007).
standard deviation
in the
never reserved sample.
outcomes by estimating the
To summarize, our
the reservation
we
for
We
never reserved
direction.
To
outcome we construct a
GPs and
dividing by the
on the average across these normalized outcomes. 20
effect
from estimating either equation
estimating equation (3) or equation
family" of
obtain an average effect for the "family"
tables report two sets of regressions. Panel
dummy
same
"
report effects which average across
Specifically, for each
normalized transformation by subtracting the mean
Within a
questions.
interest to go in the
avoid drawing inferences based on selected outcomes,
(5).
level of bias for the relevant attitude
5
and
from each other.
differ
of
A2 an di
,
The second
(2) or
to last
row
A
reports the coefficients on
equation
in
measure (usually the mean
(4),
and Panel
B
from
each table reports the baseline
for the
unreserved sample).
Results
Do
5.1
Our model
Villagers
Know
of voter behavior
Their Leader?
is
based on the premise that villagers are aware of local politics
and are able to gather information about leader
providing evidence on this in Table
villagers exhibit relatively high levels of
As
Kling et
al.
(2007),
We
start our empirical analysis
by
2.
Male
in
actions.
we obtain very
knowledge and involvement
similar results
if
seemingly unrelated regression (SUR) system, where we account
we estimate the standardized effects using a
for correlation across outcomes and average
across outcomes to obtain an average effect for the "family" of outcomes.
17
in local politics.
However, we observe a significant gender gap
limited exposure.
row
to last
in
women showing much more
every measure, with
Table 2 provides the means of variables measuring
awareness and activism in the sample of never reserved GPs.
political
male, but only
33%
gender gap exists
13%
The second
in
53%
67%
of
simila.r-sized
of the male,
but only
having ever approached the Pradhan about their needs or village
of female, villagers state
and
A
Pradhan's name.
their current
previous Pradhans (columns (l)-(4)). Further,
for
issues (columns (5)
know
of female, villagers
In these GPs,
A
(6)).
large majority of
men have heard
of the village meeting, even
though a minority has actually attended one (76% and 34%, respectively, see columns (7) and
(9)).
In contrast, only
A
Panel
Panel
we
of the
have heard of the meetings and only
Table 2 shows the coefficient on the "ever reserved" variable
in
the coefficients on the reservation variables in equation
name. For example, men are 11-12.5%
GP
is
Pradhan
GP
the
if
effect of reservation
seriously), or female
male
know
That
villagers
said,
know
equation (2), and
in
(1)
through
(6)
Pradhans and to know
name
the
have attended.
the
of their
name
Pradhan
if
of the previous
was previously reserved.
eral possible interpretations:
thority.
less likely to
7%
columns
(3). In
currently reserved, and between 10-14% less likely to
This negative
5.2
women
see that both genders are less likely to have interacted with female
their
the
B
56%
it
may
on
villagers' interactions
reflect discrimination (villagers
Pradhans may be
their
do not take female leaders
less proactive, or less likely to
an important point from Table 2
know who
with the village leaders has sev-
Pradhan
is,
that even in reserved
is
and have,
be a trusted village au-
at
some
point, seen
GPs
him
a majority of
or her.
Preference for Male Leaders
In Table 3
we examine
differences in villagers' general feelings towards
As a reminder, we measure an
villager's general rating of
explicit preference for
scale).
preference for male leaders by the D-measure in the taste IAT.
IAT
leaders.
male leaders by the difference between a
male and female Pradhans (on a 1-10
ranking of leaders, we also report the rating and
male and female
results for
We
measure an implicit
To benchmark respondents'
respondent preferences for male
versus female villagers.
The second
Villagers in
to last
Birbhum
row of Table
3 presents
summary
statistics for never reserved
GPs.
are not shy about admitting explicit preferences for males. In columns (1)
18
and
(2)
we
exhibit a
bias
is
see that
much
men rank male
villagers 0.69 points higher
smaller, but significant, bias in favor of
male
magnified in the case of leaders (columns (3) and
(4)).
than female
Women
For both genders, this
villagers.
Male
villagers.
male leaders
villagers rate
1.44 points higher than female leaders. For female villagers the difference, while smaller (0.56),
remains
significant.
This distaste
for
female leaders
"ever reserved" (Panel A,
column
is
not ameliorated by exposure.
is
(3))
positive (0.22) and significant for
ing that their relative preference for male leaders
GPs which have
in
Columns
attributes."
These
strengthened
and leader taste IATs. Both demonstrate
among male and female
Neither
is
villagers.
Men
are significantly
results are discouraging of the view that affirmative action can, in the short to
Our evidence
differs
when respondents
possible explanation
that social norms that
rural India.
The
more
affected by reservation.
tends to be muted even
is
finding that
men
are
from that of
quota system which forces them to
women
more biased
elect
rich countries
where
exhibit strong implicit bias (Bertrand et
their implicit preferences are unaffected.
explicit bias
al.,
2005).
A
once reserved GPs, but only as measured
in
women by
medium
should not be leaders are deep-rooted in
That
explicit attitudes, supports the "backlash" hypothesis.
though
is
good with male names while women associate female names with positive
run, alter voter preferences.
by
villagers, suggest-
experienced a gender quota.
a strong same gender preference
1
male
(compared to female leaders)
(5)-(8) report findings for the villager
likely to associate
Rather, the coefficient of
is,
men
protest against the
stating a dislike for female leaders, even
The
fact that this effect
is
strongest in
GPs
that were only reserved once (Panel B, see coefficient on reservation in 1998 or in 2003) leaves
open the
Section
possibility that villagers get used to quotas over time. This possibility
norms which militate against female
leadership,
ences, could also explain consistent results across explicit
combined with same-gender
and implicit attitudes
not female, villagers. Specifically, women's explicit attitudes
ognize that ranking female leaders below male leaders
same sex preference on the part of men
"women are wonderful"
attributes (Eagly and Mladinic, 1989).
Our
discussed in
5.5.
Social
21
is
finding of
countries, where these tests exhibit a
19
is
may
reflect
for
prefer-
male, but
the fact that they rec-
prescribed behavior, even though they
differs
effect
from what is typically found in developed
- and everyone associates women with good
female leaders (and this
like
is
captured by their implicit
two taste IATs
Finally, the very similar results across the
may
villagers
beliefs).
have, in both cases, mostly perceived the gender of the character and did not
associate pictures of female leaders with leadership. In that sense,
IAT may have
both genders suggests that
for
failed to sufficiently
we recognize that the second
capture a "leadership" dimension.
Gender-Occupation Stereotypes
5.3
Our
results suggest that reservation has not significantly altered voters' taste for female leaders.
In Table 4
lead.
We
we examine whether
it
use an occupation-based
leadership activities with
has
made them more aware
IAT which
men and domestic
tests
whether
activities
of the possibility that
villagers are
more
GPs.
Specifically,
with women.
women
respondents are faster at associating
leadership actions (on average,
deviation faster).
men
IAT
suggests
in leadership activities in never reserved
women
with domestic activities than with
are 0.1 standard deviation faster,
Exposure to a female leader
can
likely to associate
In contrast to the same-sex preference observed in taste IATs, the occupation
that both genders exhibit significant bias against
women
and women 0.15 standard
significantly reduces this association for
men.
This suggests that reservation-induced exposure lowered the strength of the stereotype linking
men with
leadership activities, at least
among male
can lead. To the best of our knowledge, this
the
is
villagers
first
who may have
learned that
study that identifies a causal
women
effect of
a
policy on implicit beliefs, as measured by an IAT.
Overall exposure does not reduce stereotyping
women and
a strengthening of the stereotype associating
exposed to a female Pradhan
for the first
among women.
time in 2003.
If
anything, Panel
B shows
domestic activities among
Section 5.6 provides a
women
somewhat
speculative explanation for this result.
These
results suggest a
more nuanced view
While reservation does not make male
it
villagers
makes them recognize that women can
updated
their perception of the
lead.
competence
.of
of
how
political reservation affects attitudes.
more sympathetic
to the idea of female leaders,
This opens up the possibility that they
women
20
leaders,
which
is
what we turn
may have
to now.
Perceptions of Effectiveness
5.4
Hypothetical Leaders
5.4.1
Table 5 examines
villagers' evaluation of the actions
performed by hypothetical Pradhans as
A
described in the tape-recorded Pradhan speech or in a vignette.
a Pradhan of the
same gender
in
respondent was exposed to
the speech and the vignette and asked multiple questions on
the perceived effectiveness of the Pradhan.
columns
In
variable
(1)
and
we
report the coefficients from regressions in which the dependent
the average of the normalized outcomes across
is
A we
In Panel
vignette.
(2)
see that
men
in
all
questions in the speech
and
the
never reserved villages rate the effectiveness of a
hypothetical female Pradhan 0.055 standard deviation below that of a male Pradhan (with a
standard error of 0.027). While lower and insignificant
of 0.031),
we cannot
reject the hypothesis that
for
women
(0.035, with a standard error
both genders are equally biased.
Reservation alters male evaluations: the coefficient on the interaction between female Prad-
han and ever reserved
Pradhan and
is
0.096 and
is
strongly significant.
Adding the
Panel B, we see a similar impact of reservation across
are positive, with similar-sized,
and
all
is
not strongly significant). In
reservation categories:
some
sufficient to erase statistical discrimination against female leaders.
persists even after the
woman
has
all
statistically indistinguishable, point estimates.
are striking in that they suggest that for male villagers exposure, at
is
on female
interaction with reservation suggests that political reservation completely erases,
its
indeed reverses, this bias (though the resulting pro-female bias
leader
coefficients
left,
office.
coefficients
The
results
point, to a female
Moreover, this
effect
In contrast, reservation has no impact on leadership
evaluation of female villagers.
Columns
(3)-(10) consider different subsets of the effectiveness questions that enter the over-
all
assessment. 22
The
results are
in
never reserved
GPs
evaluate the hypothetical female Pradhan more negatively than her male
counterparts. This bias
the case of
Why
22
They
for all
women
are
are
is
absent in
GPs which have been
villagers, reservation has
women's evaluations
still
remarkably consistent across outcomes.
reserved at
On
each outcome
some point
men
in time.
In
no impact.
of the effectiveness of female leaders unaffected
by reservation?
averaged across the speech and vignette questions. Appendix Table Al reports results separately
speech and vignette questions.
2]
One
possibility
much
the
is
lower exposure of female villagers to local politics (see Ta-ble
2).
For the presence of female leaders to alter an individual's perception of their competence, she
needs to have observed them
it
is
in action.
If
women
unaware
are largely
of local politics then
unsurprising that reservation docs not affect their evaluations. Table 6 provides evidence
which supports
this hypothesis.
We
use the average of the responses to the speech and vignette
questions as the relevant outcome variable.
We expand our set of explanatory variables to include
the interaction of the reservation effect with whether the villager knows the current or previous
Pradhan's name. Columns
that even
among men
OLS
(l)-(3) present the
the effect
results.
driven by those
is
The
who know
estimates, while noisy, suggest
the Pradhan.
concern of reverse causality columns (4)-(6) report IV regressions.
a villager knows the Pradhan's
proximity to the Pradhan
(reads the newspaper,
name with
(lives in
knows the name
woman
of legislator
and has heard
on
this
is
leader in action.
Our
still
women
Section
if
6).
how
social
The IV
noisily estimated).
even
if it
women
least
leaders are less effective
among men. This
updating
suggests that
does not change voter preferences, alter the future electoral
voters place enough weight on leader competence (we provide evidence
And
even
if
electoral
outcomes are unchanged,
enhance voter welfare by improving the information available to
The speech and
of village meetings).
results suggest a relatively rapid
about women's effectiveness as leaders, at
political reservation can,
fortunes of
and
vignette results are consistent with the predictions of a model of statistical
have seen a
of voter beliefs
whether
for
the same village, has the same caste) and political knowledge
discrimination in which villagers update their initial prior that
after they
instrument
a set of variables indicating geographic
estimates are significantly larger for both genders (though
The speech and
We
To address the
political reservation
can
villagers.
vignette experiments allow us to measure bias but
may
not perfectly predict
reservation influences leader evaluation in situations where the information available about
leaders
may
5.4.2
Actual Leaders
differ
with their gender.
Our survey asked each respondent
ferent dimensions using the
experiments.
The
same
We now
directly
examine
this.
to evaluate their actual Pradhan's
1
to 10 ladder
which was used
performance on four
for the vignette
dif-
and speech
questions were purposefully chosen to be similar to those asked about the
22
hypothetical Pradhan in the speech and vignette.
In Table 7
question. In
we report both the average outcome
all
Pradhan
cases villagers evaluate their
reserved in 2003.
Column
row shows that Pradhans
(1)
in
for the four questions
that are
first
reserved for a
for
shows a negative, but smaller and
effective
The
in
male respondents. The
first
is
in
strongly significant.
never reserved
is
GPs
standard
analogous to the speech and vignette results,
considered hypothetical female Pradhans less
coefficients of interest are very similar across individual
in
GPs
first
(BPL) program
outcomes (columns
and
in
preparing the
list
of "below the poverty line"
beneficiaries.
However, consistent with the experimental data, the second row
villagers' evaluation of female
from that of Pradhans
in
Pradhans
level.
in twice reserved
unreserved GPs,
equality between the coefficient of the
percent
(3)-(10)).
reserved in 2003 are ranked lower on general effectiveness, ability
to look after village or respondent needs,
for
"first
women
GPs
is
in
Table 7 shows that
statistically indistinguishable
both men and women. For men, we can reject the
reserved" and "reserved in 1998 and 2003" at the
For women, the coefficients do not
cannot reject that
3
Column
than their male counterparts.
Female Pradhans
1
2003 score 0.2 standard
insignificant, effect for female villagers (-0.07
deviation with a standard error of 0.06). 23 This
where (male) respondents
first
woman
deviations less than Pradhans in unreserved GPs; this coefficient
(2)
GPs
as significantly less effective in
examines the average outcome
GPs
and separately for each
differ; this, in part, reflects
the fact that
we
did not, rate their female leaders lower than male leaders. Finally, row
shows that the current (mostly male) leaders
in
GPs
that were previously reserved in 1998 are
evaluated at par with leaders from never reserved GPs.
These
results are consistent with the
view that exposure to a female leader causes villagers
to update their prior on average female effectiveness.
The
first
cohort of
women
leaders fails to
benefit from this updating, potentially because voters "stick" with their assessment of a specific
leader (Mullainathan and Washington, 2009). In contrast, the second cohort of female leaders
does not appear to suffer from statistical discrimination. Needless to
is
that, unlike the hypothetical leaders in the speech
are
an important caveat
and vignette, male Pradhans and second
These results echo Duflo and Topalova (2004) who, using nation-wide data
more dissatisfied with female leaders in general (also see Duflo, 2005)
23
say,
for India,
showed that
villagers
may be
time female Pradhans
Women
Are First-time
we
In Table 8
is
Pradhans
While
public goods, the average
all
in
GPs
first
reserved in 2003 are
statistically indistinguishable
GPs
of repairs or constructions since
in
column
more
from the
effective
effect in
GPs (column
than Pradhans
GPs
first
(1)). If
from
GPs
last
anything,
in never reserved
GPs. 24
reserved in 2003, the effect in
is
that
women Pradhans
we do not
(2)
any
see
invest in more, but lower quality, public goods.
significant differences in the quality of public
provision across currently reserved, previously reserved or twice reserved GPs.
performance of
the
smaller and insignificant (0.04, with standard error of 0.06).
is
second possibility
However,
number
0.19 standard deviations higher in villages in reserved
twice reserved
A
Leaders Less Effective?
present evidence that, prima facie, male Pradhans do not outperform female
Pradhans. Across
election
better leaders than first-time female Pradhans.
women
elected in first-time reserved
GPs
is
good
Once again, the
indistinguishable from
women
elected
reserved for the second time.
Since public goods are mainly financed by State Government funds, the contrast between
villagers' (especially
Pradhans seem
by
men
in
which
male) negative evaluation of female Pradhans and the fact that female
to invest more, with
no discernible reduction
in quality, is unlikely to
be explained
resenting a "big government" approach. There are, however, (at least) two other ways
villagers
may have
to pay for these goods: voluntary contributions
and bribes. In our
household survey we can identify whether the household paid a bribe to receive a
and whether the household paid bribes and made voluntary contributions
provision. In
likely to
column
(3)
we
see that
on average, individuals
have paid a bribe. 25 Further, in columns
and
(4)
(5)
we observe very
in different reservation categories.
all
card
drinking water
in currently reserved
self-reported villager satisfaction with public goods (we average across
GPs
for
BPL
GPs
are less
similar levels of
public goods) across
26
24
This result is broadly consistent, with Munshi and Rosenzweig (2008) 's finding that in non-caste dominant
GPs, women leaders demonstrate higher competence as measured by public good provision using nationally
representative data.
5
26
This
The
effect
is
driven by
BPL
card; there
is
no difference
in
payments
for
water maintenance,
results are very similar for individual satisfaction measures, except the
unsatisfied with the
way female
BPL list. Men, in particular, are
BPL quota implies that there
leaders allocate them. However, the existence of a
no way to allocate BPL cards to please everyone (our satisfaction results differ from Duflo and Topalova (2004):
an Indian-wide data set they found that, on average, villagers were less likely to be satisfied with public good
provision when women were in charge.
is
in
24
The evidence
women
suggests that
leaders provide
While we may have
a lower effective price.
performance, the weight of the evidence
failed to
more public goods,
measure some
critical
of equal quality, at
aspect of
Pradhan
consistent with our experimental findings.
is
villagers rate female leaders in villages reserved for the first time
That
is,
more negatively even though
they perform at least as well as the average male Pradhan. This bias disappears for the second
cohort of women.
we examine whether female
Finally,
ences. This
men
motivated by the fact that
is
We
Appendix Table A2).
(see
leaders' decisions are biased
women
towards women's prefer-
leaders invest in different types of
goods than
Chattopadhyay and Duflo (2004) and use the
follow
differ-
ence in male and female villagers' complaints in public meetings to construct a measure of female
Column
preferences.
first
(6)
shows that female leaders invest more
pass, this suggests an alternative explanation for
in
women-preferred goods. At
both the apparent taste discrimination
against female leaders and the observed backlash in male preferences for female leaders in Table
3.
However,
Pradhans
this explanation
in first
and second reserved
was particularly strong
To
for
square with the fact that the policy choices of female
difficult to
is
GPs
GPs
are equally pro-woman,
reserved only once (Table
further investigate possible differences in
Pradhan
while the "backlash" effect
3).
selection across first time
and second
time reserved GPs, we compare salient leader characteristics by reservation status
9.
Male and female leaders
come from
in
women
are younger, less educated, and are
landless and less wealthy households.
demographics between
time reserved
reserved
differ:
GPs
GPs
are
women
more
leaders in first
likely to
However, we see
little
for the first
time in 2003 than
(23 percentage points, significant at the 8 percent level).
in experience in serving as
Pradhan, as seen
in
Table
8,
Table
likely to
systematic differences
and second time reserved GPs.
be Pradhan
more
in
That
Women
women
in first
in twice
said, this difference
does not translate into differential
public goods provision. Moreover, their average political experience, which includes experience
in political positions
In
ones.
7
other than council president, does not differ significantly.
summary, the evaluation of the actual Pradhan
The
This
is
first
time
men
are exposed to a female
somewhat, surprising since one
different types of
women) once they
Surprisingly, this
is
may
closely mirrors that of the hypothetical
Pradhan
(either real or hypothetical), they
expect, parties to field different candidates (or voters to select
realize that female leader's decisions reflect the policy preferences of
also true for previously (but not currently) reserved
25
GPs.
women.
women, while
similar, are attenuated
the
first
and
While we can not
insignificant.
performance differences underlie men's
possibility that
The
This difference disappears the second time around.
consider her ineffective.
dislike for
time and their favorable evaluation of Pradhans elected
Pradhans
GPs
in
effects for
fully rule
in
GPs
out the
reserved for
reserved for the second
time, the consistency between these two sets of results strongly points to the possibility that
exposure to female leaders reduced
bias.
Robustness Checks and Alternative Explanations
5.5
we conduct a number
In Table 10
of robustness checks to
examine possible alternative explana-
tions for our results.
A first
possibility
is
that female leaders get worse evaluations than male leaders, not because
of their gender, but because they are
new and inexperienced (roughly 60%
women
have previous experience while the experience of
9).
In the vignette and speech experiments, respondents
leaders
is
who have
I
of Table 10
we
restrict the
evaluation of a Pradhan
elected to any
Another
GP
is
sample to never reserved
Pradhans
significantly lower, see Table
never had a female leader
assume that female Pradhans are new and inexperienced, and therefore
Panel
of male
likely to
GPs and show
uncorrelated with whether the Pradhan
is
may
be worse. In
that a respondent's
new
(i.e.
has not been
women
leaders worse
position prior to 2003).
possibility,
which we alluded
to earlier,
is
that voters give
evaluations simply because they are elected on a reserved seat. This, however, cannot explain
the absence of bias for
idea of reservation.
women
elected from twice-reserved GPs, unless voters get used to the
If correct, this
explanation suggests that voters will rank other reserved
Pradhans below unreserved Pradhans, but
GPs which
were previously reserved
for
women.
In Panel
II
we show
Pradhans lower than non SC Pradhans (potentially because they are
previous reservation for
time female leaders
is
does not affect
we show
the speech and vignette.
It is
GP
women
reservation in general.
26
SC
in
SC
also discriminated against),
was previously reserved
that previous reservation for
exposure to
that while voters rank
Similarly the negative evaluation of
this.
unaffected by whether the
Finally, in Panel IV,
in
women
Pradhan would improve
their ranking of reserved
for
SC
(Panel
first
III).
does not reduce voter bias
that makes a difference, not exposure to
Women
Are
5.6
Leaders Hindered by Voter Bias?
women
Reservations ensure that
difficult.
we report
In Table 11,
experience on the job
We
first
men
is
on
for this question
Pradhans
their lives than
for
how
affected by
GPs
in
first
well they are received by voters.
(on a scale from
life
reported happiness). 28 In column
self
women
reserved for
This difference
is
coefficient
absent for Pradhans elected in
GPs
is
GPs
In column (2)
in either
We
sample to
first
more on-the-job
women who
Pradhans elected
(for
see that, relative to
leaders elected from
creating
restrict the
1998 or 2003
Duflo, 2004).
Table 11 we
with
and the mean
-0.47,
reserved for the second
and second
first
strongly significant.
is
we
using the
5,
less satisfied
time; moreover, the difference in stated satisfaction levels of female Pradhans in
time reserved
to
1
(1) of
2003 are significantly
in
Pradhans elected from unreserved seats (the
3.7).
job more
their
based on surveying Pradhans which suggest that their
results
consider Pradhans' stated satisfaction with
standard phrasing
see that
is
But discrimination may make
get elected.
in
elected from second-time reserved GPs.
GPs
shown
that their gender created problems. While not
in
women
1998 or 2003) report their gender as
(in either
Thirty-four percent of the
difficulties.
GPs
1998 we use survey data from Chattopadhyay and
women
time reserved
were elected Pradhan from reserved
first
time female Pradhans state
the table, only 10 percent of the second
time Pradhans did.
column
Finally, in
(3)
we expand
the sample to
and examine whether the Pradhan plans
the
first
time
in either
We
run
Pradhans elected
for re-election.
1998 or 2003 are 12% more
unreserved GPs. This difference
in
to
all
likely to
Both the
in
GPs
reserved for
absent for Pradhans in twice reserved GPs.
is
and lower
bias
Incumbents
plan to not rerun than incumbents
interpret these findings as suggesting that bias lowers
time female leaders.
in either 1998 or 2003,
life
and job satisfaction
absent for second time
level of satisfaction are
leaders.
Although somewhat speculative (especially since we do not see any other
women),
this
may
also explain
why,
significantly less likely to associate
in
Table
4,
women with
after
for first
one round of reservation,
leadership activities, except in
effects for
women were
GPs
that had
been reserved twice.
"
face
The question
is
asked,
"Which
picture best describes
very satisfied and the frowning-crying face
is
how
satisfied
you are with your
life
right
now
very dissatisfied? (Show picture with 5 faces)."
27
if
the smiling
Conclusion and Epilogue: Electoral Outcomes
6
Indian villagers prefer male leaders and have negatively biased priors on the effectiveness of
female leaders. Male villagers rate their feeling towards female leaders more than one point below
that towards male leaders; moreover, in never reserved villages hypothetical leaders performing
the
same
We
action are evaluated as less effective
have argued that,
in large part,
if
they are female.
the overall "feeling" for male leaders reflects taste-
based discrimination, or a deeply rooted social norm that
leadership positions.
Such
social
norms appear hard to
it is
inappropriate for
alter,
women to
fulfill
even after ten years of political
reservation. In contrast, the effectiveness rating of a leader in a speech or a vignette,
which we
suggest reflects statistical discrimination, appears to be easier to change.
We
flects
do not
find
any evidence that negative feelings towards the idea of female leaders
worse performance by female Pradhans.
Pradhans on many dimensions
If
of performance:
anything, female Pradhans outperform male
for instance,
at a lower price. Moreover, villagers' rating of the
they provide more public goods
performance of their actual female Pradhan
closely tracks their ranking of hypothetical leaders:
below men (possibly reflecting
the
statistical discrimination),
first
cohort of female leaders
but the second cohort
with men. Another alternative, and potentially complimentary, explanation
is
rated
is
rated at par
is
that
men
dislike
female Pradhans because they implement pro-woman policies.
This explanation
inconsistent with the fact that first-time and second-time female
Pradhans are equally as
to implement
pro-woman
policies but only first-time
Taken together, the weight
reservations, or
leaders
makes
more generally quota systems, may play an important
it
translate into a
Our
results
difficult to predict
priori
permanent improvement
do suggest that
if
However,
villagers'
however,
likely
and that
political
role in reducing bias in
continued preference for male
whether a temporary program of reservation can
in the electoral fortunes of
women.
reservation improves electoral outcomes for
should be stronger after two cycles of reservations than after one cycle.
May
is,
Pradhans receive negative evaluations.
of the evidence suggests that exposure matters,
perceptions of female leader effectiveness.
re-
2008 Panchayat elections provide a unique opportunity to
was marked by a strong anti-incumbent voting -
The
women, the
effect
recently concluded
test this prediction.
This election
after 30 years of uninterrupted electoral success,
28
Communist Party
the
election,
it is
lost in
a majority of GPs.
very likely that political parties would not
seat unless they expected her to
As we mentioned
elect the
in Section 4.1,
we analyze the
GPs) that were not reserved
Columns
women who
ran
ward
is
a female candidate in an unreserved
indirect: directly elected
results,
which
May
women
in 2008.
show the impact
ward councillors
2008 election have not yet occurred.
reflect villager preferences, are available.
electoral results for the 1185
for
wards (spread across the 165 Birbhum
There were, on average, three candidates per
GP
(4)
and
for,
and won, ward councillor positions. In non-reserved wards belonging
(5)
ward
councillors were
of previous
reservation on the fraction of
Pradhan was reserved both
we
in
women (33%
of the
wards are reserved, so roughly
see a striking difference for
wards
for
wards
in
in
in unreserved
wards
in
GPs
GPs where
1998 and 2003. Specifically, relative to wards
GPs, the proportion of female candidates
to
Pradhan before 5.2% of the candidates, and 4.8%
women). These numbers are unchanged
councillors are
reserved only once. However,
of
election
that had never been reserved for a female
of the elected
all
Pradhan
Pradhan. The Pradhan elections following the
In Table 11
GPs
field
this
win the position.
However, ward councillor election
ward.
Given the highly contested nature of
in
double reserved
38%
of
that were
the position
never reserved
GPs
increased
by 3.3 percentage points and the proportion of women elected more than doubled (10.1% versus
4.8%).
29
The increased entry
to field
stand
women
in
of
women
into politics
may
reflect either a
change
in parties'
response to altered citizens' attitudes, or a greater willingness of
for election, after
they see that the second cohort of female Pradhans
is
willingness
women
to
well liked. 30
Regardless of the mechanism at work, these results provide striking evidence that while ten
years of exposure to
leaders,
women
by giving voters
a
women
may
not have changed voters' stated preference for male
chance to learn about the effectiveness of
improved women's access to
Beyond
leaders
women
leaders, they have
political office.
their implications for the possible role of quotas in inducing durable changes in
representation in politics, these results provide some of the
first
evidence on the potential
29
Bhavnani (2008) reports similar findings for urban wards of Mumbai, where previous reservation for women
improved future representation of women on unreserved seats.
30
Even though we do not see impact of reservations on the attitude of the average woman, wc have argued that
the most likely explanation is that many of them arc not exposed to politics. Women who are capable of standing
for election must, however, be the most politically aware.
29
role for public policy in mitigating voter discrimination.
of
women
is
widely documented, there
is
While the
political under-representation
almost no credible evidence on whether public policy
Much of the evidence from
outside the political
sphere suggests that policies which limit awareness of a candidate's gender
may be an optimal
can influence voters' belief systems and prejudice.
short run response to discrimination (Goldin and Rouse, 2000).
political settings
where such strategies are
important medium-run
figures
may
role.
While the
Our
results suggest that, in
infeasible, political affirmative action
first
generation of
women who
may play an
are powerful political
encounter significant prejudice, their experience can pave the way
for
others to go
further.
References
Aigner, D. and G. Cain (1977). Statistical Theories of Discrimination in Labor Markets. Industrial and Labor Relations Review 30, 175-187.
and R. Kranton (2000). Identity and Economics.
nomics 115(3), 715-154.
The Quarterly Journal of Eco-
Akerlof, G.
M. (1990). The Puzzle of Party Identification
American Politics Quarterly 18, 476-91.
Alvarez,
-
Dimensionality of an Important Concept.
M. (2001). Implicit Attitudes can be Measured. In H. Roediger and J. Nairne (Eds.), The
Nature of Remembering: Essays in Honor of Robert G. Crowder. Washington, DC: American
Banaji,
Psychological Association.
Baron, A. and M. Banaji (2006). The Development of Implicit Attitudes: Evidence of Race
Evaluations from Ages 6 and 10 and Adulthood. Psychological Science 17(1), 53-58.
Bergemann, D. and J. Valmiki (2008). Bandit Problems. In
The New Palgrave Dictionary of Economics.
Bertrand, M., D. Chugh, and S. Mullainathan (2005).
Economic Review 95(2), 94-98.
S.
Durlauf and L. Blume (Eds.),
Implicit Discrimination.
American
and V. Rao (2007). Just Rewards? Local Politics and Public Resource
Allocation in South India, mimeo, Harvard.
Besley, T., R. Pande,
Bhavnani, R. (2008). Can Governments Remedy Political Inequality? Evidence from Randomized Quotas in India, mimeo, Stanford.
Duncan, M. Kremer, D. Levy, and J. Eccles (2006). Empathy or Antipathy? The
Consequences of Racially and Socially Diverse Peers on Attitudes and Behaviors. American
Economic Review 96, 1890-1906.
Boisjoly, J., G.
J. Guryan
Working Paper 13661.
Charles, K. and
(2007).
Prejudice and the Economics of Discrimination.
NBER
Chattopadhyay, R. and E. Duflo (2004). Women as Policy Makers: Evidence from a Randomized
Policy Experiment in India. Econometrica 72(5), 1409-1443.
30
S. and G. Loury (1993). Will Affirmative-Action
American Economic Review 83(5), 1220-1240.
Coate,
Dahlerup, D. (2006). Women, Quotas and Politics.
Policies Eliminate Negative Stereotypes?
New
York: Routledge.
Dasgupta, N. and S. Asgari (2004). Seeing is believing: Exposure to counterstereotypic women
leaders and its effect on the malleability of automatic gender stereotyping. Journal of Experimental Social Psychology 40, 642-658.
Why
Duflo, E. (2005).
Political Reservations.
Journal of the European Economic Associa-
tion 3(2-3), 668-678.
Duflo, E. and P. Topalova (2004). Unappreciated Service: Performance, Perceptions, and
Leaders
Women
Mimeo.
in india.
Eagly, A. and S. Karau (2002). Role Congruity Theory of Prejudice toward Female Leaders.
Psychological Review 109, 573-598.
Eagly, A. and A. Mladinic (1989). Gender Stereotypes and attitudes toward
Personality and Social Psychology Bulletin 15(4), 543-558.
Frechette, G., F. Maniquet,
and M.
Morelli.
Incumbents
Interests,
women and men.
Voters Bias and Gender
Quotas. American Journal of Political Science forthcoming.
Goldberg,
P. (1968).
Goldin, C. (1990).
A
Are
Women
Prejudiced Against
Women?
Transaction
5,
28-30.
Pollution Theory of Discrimination. Mimeo, Harvard.
Goldin, C. and C. Rouse (2000). Orchestrating Impartiality: the Impact of "Blind" Auditions
on Female Musicians. American Economic Review 90(4), 715-741.
Government of West Bengal
Department of Panchayats
The West Bengal Panchayat (Constitution) Rules, 1975.
Rural Development.
(1998).
&
Greenwald, A., M. Banaji, and B. Nosek (2003). Understanding and Using the Implicit Association Test: I. An Improved Scoring Algorithm. Journal of Personality and Social Psychology 85(2), 197-216.
Greenwald, A., D. McGhee, and
Cognition:
1464-1480.
The
J.
Schwartz (1998). Measuring Individual Differences in Implicit
and Social Psychology 75(6),
Implicit Association Test. Journal of Personality
L. and N. Terkildsen (1993). Gender Stereotypes and the Perception of Male and Female
Candidates. American Journal of Political Science 37(1), 119-147.
Huddy,
Ingelhart, R. and P. Norris (2003). Rising Tide:
the World. Cambridge: Cambridge University.
Gender Equality and Cultural Change Around
Jones, M. (2004). Quota Legislation and the Election of
Experience. Journal of Politics 66(4), 1203-1223.
Keller, S.
Women: Learning from
and T. Mirer (1974). The Simple Act of Voting. American
the Costa Rican
Political Science
Review
68,
572-91.
J., J. Liebman, and L. Katz (2007).
Econometrica.
Kling,
Experimental Analysis of Neighborhood
31
Effects.
Krook, M. (2005). Politicizing Representation: Campaigns for Candidate Gender Quotas Worldwide. New York: Ph.D. diss Columbia University.
Matland, R. E. (1994). Putting Scandinavian Equality to the Test: An Experimental Evaluation
of Gender Stereotyping of Political Candidates in a Sample of Norwegian Voters.
British
Journal of Political Science 24, 273-292.
S. and E. Washington (2009). Sticking With your Vote: Cognitive
Political Attitudes. American Economic Journal: Applied Economics.
Mullainathan,
and
Munshi, K. and M. Rosenzweig (2008). The Efficacy of Parochial
and Competence in Indian Local Governments. Mimeo.
Politics: Caste,
Dissonance
Commitment,
Nosek, B., A. Greenwald, and M. Banaji (2007). The Implicit Association Test at Age 7: A
Methodological and Conceptual review. In J. A. Bargh (Ed.), Automatic Processes in Social
Thinking and Behavior.
Powley, E. (2007). Rwanda: The Impact of Women Legislators on Policy Outcomes Affecting
Children and Families. Background Paper, State of the World's Children.
Rooth, D. O. (2007). Implicit Discrimination
in Hiring:
Real World Evidence. IZA Discussion
Paper 2764.
Rudman,
L. and K. Fairchild (2004). Reactions to Counterstereotypic Behavior: The Role of
Backlash in Cultural Stereotype Maintenance. Journal of Personality and Social Psychology 87(2), 157-176.
Rudman,
L. and S. Kilianski (2000). Implicit and Explicit Attitudes
Personality and Social Psychology Bulletin 26(11), 1315-1328.
Thernstrom,
ible.
New
S. and A. Thernstrom (1997). America In Black
York, NY: Simon and Schuster.
UNICEF
(2007). The State of The World's Children:
dend of Gender Equality. New York: UN.
Weisberg, H. (1980).
Behavior 2, 33-96.
Zaller, J. (1992).
A
And
Women
Toward Female Authority.
White:
One Nation,
Indivis-
and Children: The Double Divi-
Multidimensional Conceptualization of Party Identification.
Political
The Nature and Origins of Mass Opinion. Cambridge: Cambridge University
Press.
32
A
Appendix
IAT
A.l
We
administered three IATs.
Taste
IAT
1:
The prompts used
IAT
in each
are listed below.
Villagers Male names: Badal, Shyamol, Saurabh, Gopal, Anup, Ashok, Tapan, Raju
Female names: Geeta, Minoti, Anjali, Rekha, Jharna, Minu, Rupa, Basanti
Good
Bad
attributes: good, nice, fun, happiness, love, clean, sweet, heaven
attributes: bad,
Taste
IAT
2:
Good and Bad
mad, sorrow, inauspicious,
sad, dirty, spoiled, hell
Leaders Male and Female Leaders: Pictures
Attributes:
Same
as Taste
IAT
(available from authors)
1.
Gender Occupation Stereotype IAT: Domestic
activities:
Eating puffed
rice,
Listening to the radio,
Sleeping, Taking rest, Clay modelling, Harvesting, Caring for animals, Threshing rice
Leadership words:
ing,
Gram Pradhan,
Public speaking, Govern, Chairperson, Lobbying, Leader, Campaign-
Taking bribes Male and female names: same as
A. 2
in Taste
IAT: Villagers
Vignette and Speech
Note Respondents were
given one of four vignettes, where we randomly varied the gender of the Prad-
han (Male (Tapan Das) or Female (Sandhya Das)) and the investment decision
Respondents received one
recordings.
Vignette
We
of six speech versions
randomized the gender
READ OUT: We
South 24 Parganas.
We
of
-
Pradhan
three male voice recordings and three female voice
at the respondent level
Pradhan of
will read a short description of the
will ask
(Irrigation or Water).
CHANDI
village
you some questions about what you think the Pradhan should have done.
There are no right or wrong answers. Please answer each in terms of your own reactions.
Pradhan Tapan Das [Pradhan Sandhya Das] has been serving
the end of the year approaches, there
villagers
water.
have been pressing him
in district
is
[her] to
only a limited
amount
make improvements
There was enough money to make investments
still
of
Panchayat
money remaining
one area.
Prior to
villagers at the
no safe drinking water available
:;::
for ten
months. As
in the
budget. Yet,
two major areas: irrigation and drinking
in only
Pradhan Tapan Das [Pradhan Sandhya Das] consulted with
expressed frustration that there was
in
his [her]
.
Gram
in
making a
Sabha.
the village.
decision,
Many
Many
people
people,
especially children, were getting sick. Others were upset about the quality of the irrigation system. Poor
irrigation
system meant that,
in
dry years,
Sandhya Das] considered the demands
carefully,
Das [Shrimati Sandhya Das] knew the health
off
many
people lost their crops.
and wondered what to
cost of
bad water
do.
Shri
On
quality. Yet,
Tapan Das [Shrimati
the one hand, Shri Tapan
wouldn't everyone
be
better
with better irrigation. After careful reflection Pradhan Tapan Das [Pradhan Sandhya Das] decided to
improvement [drinking
invest in irrigation
Speech
READ OUT:
Labhpur
in district
You
will
water].
hear a tape-recorded speech from the village meeting of
West Dinajpur.
We
will ask
Gram Panchayat
you some questions on the effectiveness of the
leader.
There are no right or wrong answers. Please answer each in terms of your own reactions.
VILLAGER: The
tube well of our Kumarpara
your locality has been done
PRADHAN:
As a
partially,
not functioning.
The
repairing job of the tubewell in
but the same work at Nutangram has been completed.
For repairing of tubewells
result of which, other
is
maximum amount
Panchayat
of funds of the
is
being drained out.
works can't be done. From the next stage you, the people, should take mental
preparations that the minor repairing jobs of the tubewells won't be done by the Panchayat.
if
the work involves a large amount of money,
e.g.
if
a pipe
is
needed then
Rs.250, Rs.300, this type of works will be done by the Panchayat.
people have to take initiative to collect subscriptions to do
will
this.
But
it
involves the
for the
I
mean
money above
minor repairing jobs the
In the future, the plan of the
Panchayat
be "plans with equal sharings" ("Samobhagi Parikalpana"). The Government won't provide
money.
The Government
provide some amount of
will
money and
people either by giving labor or helping financially. In this
done.
Suppose a
village road
work and the Panchayat
will
the Panchayat did mostly.
A. 3
Public
Good
all
the
the rest have to be borne by the
the work of the Panchayat have to be
has to be constructed, then the people of the village will do the earthen
supply the morram. Therefore the people
Then
preparation have to be taken.
way
that
I
the total work can be
would now
made with
like all villagers to
will
now
share the jobs, which
a success. So in the next stage that
approve the village budget.
Variables
Quantity and Quality Variables: Water and
sanitation quantity variable: a
dummy
for
whether a
tubewell was built, a tubewell was repaired, a sanitation pit was built, a sanitation pit repaired. Water
and Sanitation quality
variable:
handpumps
are perennial, provide clean water, no stagnant water, have
drainage and sanitation - no stagnant water, drainage
34
facility.
Irrigation quantity variable:
dummy
for
whether an irrigation
pump was
built or repaired.
road was built or repaired since 2003.
Roads quantity
Roads quality
variable: a
dummy
for
whether a metal
number
condition of road (1-5) and
variable:
of
potholes in 100m. Transport quantity variable:
number
of transportation related infrastructure (bus stop,
bus service, taxi). Transport quality variable:
dummy
for
Schools and other education
shelter.
was
facility
built,
and an indicator
a
dummy
for a
CE
for
whether such
Center/CE
primary schools, middle schools,
quality variable: whether
math
was
number
for
all
quantity variable: a
facility
was repaired, a
libraries
and secondary
built,
a
dummy
of doctors.
whether a health
Health Quality variable:
having a labor room.
Fair Price
facility
facility
bus stand has
for
whether any educational
dummy
for
whether there
is
a
creche
SSK, Anganwadi,
Schools and other education facilities
schools.
primary schools have drinking water,
for
if
dummy
Library. Educational facility quantity variable:
Health Quantity variable: the number of health
test scores.
facility
facilities
presence of a bus stand and
latrines,
a
facilities,
was repaired
(0
if
blackboards and reading and
dummy
no health
for
whether a health
facility existed),
and
having tap or hand-pump water and an indicator
Shop Quality measure
includes:
whether prices displayed, no bad
behavior of shop keeper, and no complaint against shop.
Satisfaction Variables:
Average satisfaction
is
the satisfaction across five different kinds of public
goods: water provision, public transport, schools, health care and
computed
public good satisfaction measure
is
variables take the value of one
the question was answered
if
fair price
shops.
Each
of the individual
as the average of the responses of a set of questions.
provision: satisfied with the quantity of the water supply
in
All
the affirmative. Satisfaction with water
and the quality of the water supply respectively.
Satisfaction with public transport: satisfied with the frequency, reliability, cost of public bus system, the
quality of the buses and the behavior of the driver/conductor.
(if
Satisfaction with schools: satisfied with
available) the school's building, playground, recreational facilities, classrooms, toilets, drinking water,
meals, quality of the teaching, quality of study material and behavior of teacher. Satisfaction with
price shops:
satisfied
fair
with the quality of items, the quantity of items, the fairness of shopkeeper and
the availability of items in the shop.
Satisfaction with health care: satisfied with treatment, behavior
of doctors, behavior or paramedical staff quality of medicine, cleanliness inside the facility in outpatient
health
facilities
and
satisfied with treatment, behavior of doctors, behavior or
paramedical
staff quality
bathrooms
in inpatient
of medicine, quality of food, cleanliness inside the wards, cleanliness inside the
health
facilities.
35
—
OS
©
£
-a
w Io
<2
<u
O.
§ s
1*1
B
u.
o
c
a
oo
en
K
T
D
00
o\
co
C5
o
o
CM
£
L>
©
^3
o
e
»s
os
o
O c
u
o
G
C
a
id
o
-G
ri
t
'J
c/i
Of
OS
flj
04
c
3
O
§
u
r;
s
u
-a
B
fa
£
I
art
e
0J
o
o
3
6D
i-i:-_i
0.
CO
o
o
o
to
o
in
o
CO
o
o
suuqpcij 0|Biuaj uoijdkjj
CM
o
o
Table
1.
Villaee-level Randomization
Census Variable
Total Population
Fraction
SC / ST
Population
Average Household Size
,
Sex Ratio Under 6
Literacy
Fraction
Women
Literate
Fraction Irrigated Land
Village has a Bus or Train Stop
Village has Permanent Approach
Village has
Road
Tube Well
Village has
Hand Pump
Village has Well
Village has
Community Tap
Number of Schools
Number of Health
Check
First
Reserved
Only
Reserved
1998 and
Reserved
Never
2003
2003
1998
Reserved
(1)
(2)
(3)
(1)
Diff:(l)
and
Diff: (3)
and
(2)
(4)
(6)
(5)
1.318
1,197
1,267
1,362
-102.279
-78.373
(1318)
(1295)
(1418)
(1555)
(178.537)
(189.410)
0.432
0.502
0.450
0.471
0.004
-O.001
(0.281)
(0.256)
(0.235)
(0.265)
(0.056)
(0.033)
5.436
5.222
5.362
5.497
0.137
-O.059
(0.612)
(0.449)
(0.571)
(2.454)
(0.093)
(0.135)
1.093
1.034
1.046
1.036
0.206
0.015
(0.590)
(0.276)
(0.218)
(0.226)
(0.113)
(0.025)
0.387
0.351
0.391
0.373
0.014
0.021
(0.119)
(0.129)
(0.121)
(0.137)
(0.032)
(0.017)
0.287
0.265
0.296
0.275
-0.003
0.023
(0.125)
(0.123)
(0.113)
(0.134)
(0.033)
(0.016)
0.599
0.485
0.607
0.500
-0.038
0.054
(0.321)
(0.349)
(0.323)
(0.352)
(0.054)
(0.049)
0.280
0.433
0.290
0.259
-0.130
0.061
(0.451)
(0.500)
(0.456)
(0.439)
(0.097)
(0.054)
0.110
0.300
0.176
0.194
-0.225
-0.006
(0.314)
(0.462)
(0.383)
(0.397)
(0.082)
(0.052)
0.910
0.933
0.912
0.977
-0.022
-0.018
(0.288)
(0.252)
(0.285)
(0.151)
(0.020)
(0.031)
0.124
0.000
0.100
0.057
0.000
0.013
(0.331)
(0.000)
(0.302)
(0.233)
(0.000)
(0.012)
0.440
0.533
0.402
0.528
0.043
-0.066
(0.499)
(0.503)
(0.493)
(0.500)
(0.082)
(0.074)
0.065
0.050
0.042
0.010
0.022
-0.018
(0.248)
(0.220)
(0.202)
(0.098)
(0.020)
(0.017)
1.190
1.233
1.265
1.160
-0.167
0.149
(0.895)
(0.810)
(0.889)
(0.910)
(0.160)
(0.111)
0.170
0.217
0.118
0.218
-0.036
-0.015
(0.403)
(0.490)
(0.324)
(0.809)
(0.090)
(0.069)
Overall Effect: F stat
0.300
0.600
Overall Effect:
m
0.586
0.438
Facilities
p value
N
60
105
108
Notes:
1
The census
variables are from the 1991 Census of India.
variables have between
454 and 478
total
N
refers to the
maximum number of observations.
Infrastructure
observations across GPs.
2 "First Reserved 2003," "Reserved 1998 and 2003," "Only Reserved 1998," and "Never Reserved" are indicator variables for
GPs reserved
for a
female Pradhan for the
first
time
in
2003,
in
both 1998 and 2003, only
in
1998, and not reserved
in either
election, respectively.
3
Columns
(l)-(4) report
means with standard deviations
in
parenthesis
Columns
(5)-(6) report tests of differences of
means
across columns (1) and (2) and columns (3) and (4), respectively. Standard errors are in parentheses. Tests are based on
regressions with block fixed effects and standard errors are clustered by
GP
«
u
r3
OjO
0)
wi — v> — t— ©
—
©_©„
° —
o ° o o o
9 p, 9 p. ° p.
JS >>
>
~
<3
<^>
<
vo Tf
vo
© o
o
© o
o
CO
J
~3
u
—
'
o
"^
-
—
"Q
-z
o 2
P o
9 o.
-1-
© o
d d
MS
d
I
c
On
d d
VO
JO
u
•a
J3
u
s
CN
3 vO
S
M On
in *
S o
o o P o
P
9 d © © P d
r»l
r->
o
•a
L,
CO
•j3
0)
<-^
o
vo
o
d
©
ON
no o
in
d
d p
g
o
c
13
o a
o
CO
NO
(L>
2
O
4)
00
!3
—
^
o m
o
P
d o d
<N
S 2
o
9
o p
("-
r-i
ON rO
m
o
P O
d d d
ON 00
O
d
in ©
r- o
d d
tVO
o
Xo
CS
O
a.
o
ft.
D.
<
JP
-a
K
n
—
en - * J m
_~ ° ~
© o P o
P o o
o _
P o
9 P,
-r vo
m
©
p
d d
JJ
o
j=
On
r--
r-
vo
<s
L,
ft,
c
u
X u
**•
c
VO
— CM ©
©
q
— 3 d q
— p
©
d ©
-T
O ©
9 P.
:
©
d
On
o
d q
d
CO
In
vo
vo
c
„
o
P
©
no
C3
D.
"3
rn
f;
m N n
p
© P (N
P ©
©
P
P © d o d o
**"
o
C3
r«i
r--
*^"
in*
m
©
cn o
© d
p
c
^a
CJ
-p
.a j§
u
2
5
S
S «
ft.
o
P
o
"3r^
M
c-
^
-
m
iN
r~ oo 00 r- -r
©
ro ^- Ov
© CN
©
q c;
© cq © d
d O
©
p
©
n x r»
^ S
^ ©
©
© ^
© © s
o
ov
On
r^i
*~*
©
ON
— On
©
in ©
On
2
r-
P o
9 o
9ddo9o
CN ON
m ©
©
© ©
r<->
c
<N en
P O
9 o
—
-
©
S "*
© © ©
d 9 ©
^>
a~
©
©
vo
vo
m
c
©
©
(N
-O
c;
C
0O
On
rs
ra
*—
"O
HJ
e
Q>
W
4>
-
CO
ON
On
u
E;
r.
ft,
E
—
u
£
t/5
1J
ft<
i
00
ON
ON
t
4J
to
<u
D
U
'./
^
^
~B
O
a
=
O
<a
B
>
X 5
O"
C
«
C3
<
IG t3
U
a
u
c
J^
—
2C
X
-—
u JD
3
O
c
O
T3
E
"
JC
a
j
c
00"
EH
>^
5
h
a
<U
eb
<U
j3
^
C
O
1
ra
>
<
CU
K JO
G
u 3
n. O
—
O
.—;
en
C
J5
t/1
SO
'«, "O
g
fc^
t-
2
CO
o
"3
Oi
--
II
—
15
p
CO
y
1)
J3
K
-rr
T3
-C
1i
o
o
>
rd
a>
-a
j-
L~
y
8
VO
00
ON
ON
c
5
on
-=
©
©
CN
«
S
u
ESO
-
-J
u
3
a
u
"z:
u
W
XJ
^
62
^
p
V
'—
>
O
-II-
z
fc
u
<
c/j
"c3
E
cj
Bj
^°
2 5~
© S
9 ©
o O 00
o
o O
o
r-
2s
_
T>
a.
© o
9 d
OJ
nj
E
Cfl
S * 2 "~
— o
3; © m * m
© o © o
o o
© d © d © ©
fi
o
o o
d p
oo
« a
— m
S
s
\p
V
c
U.
cd
jiS
ap
13
o
a
T3
H
r--
o\ cn
V~l
©do
n vi
in
5
3
33
u
o O
m
o ©
CN
=
T3
E
o
r^
o
^
If ~pop©
d d d d
i
CA
TO
c
K^
fc
00
cn
O
cn
d
£
(N
m "* t
© o m
o o
© © © d
Lr
5!
2 ci
©
© ©
©.
"3-
—
— vo
o
©
d d
•>*
Mt~-
CN
3-
s
*r\
o
ft<
m
— CN
o
s
5
£1
E
_u
—
o O
C3
Tf
£
u
s
I
>
ft.
•* r- •*
— p
©
©
3
90
0-
odd
—
n
—
en in
cn
cn
U1
c
ri
tJ-
<r^
o
J2
o
* ©
© d d d d d
d
O
g
T3
Oi
.22
u
3
(U
X
LS
—
c
o
1-
t»
O
© ©
ft.
o
o
2 © © p
d d © d d d
CN
©.
•a
00
-
cn
d
oo
CN
<u
CO
s
d d
U
u
3
U
bO
>
flj
OJ
in
b
TO
'5.
p,
to
O
<L)
00
yj
E
a.
_
to
-:
a
a>
•o
E
T3
(^1
J4
to
E
c
TO
>
oo
§s
r—
© ©
CS
oo
cn
—
0\
v£>
d
CN
i/-i
O
c
u
Q
c
<U
TO
5
o
U
to
<
to
HJ
C
u
rsi
0>
<
aj
0>
5
U
i
TO
<u
c
II
en
o
o
CN
a
-
cn
o
o
cn
u
5,
E
a
C/3
-^
u
oc
CJ
3-
U
w
O
T3
C
a
O
CN
"Q
1—4
s
p4
oo
T3
-C
u
.-
M
&
~
>
II
o
-
oi
tt
a
Oi
p.
E
5
u
CN
'—
>p
TO
=;
<u
,«3
t«—
5u
-£E
T3
cr
'~
o
(L)
to
-a
=
'J
QJ
~
U
to
U
o
c:
u
u
-
o
s
QJ
.5
e
s.
-o
QJ
C
4J
-:
c
a
c _o
TO
3
>
c
JH
—
—
<
<u
u
1
-->
U
IN
o
0)
T3
U
u
to
5
c
c2
Z —
<
c
o TO
i o
to
i
to
H
CO
H
"_i
TO
"5
a
r_
is
>
C^
X
"to
c
<u
TO
—
TO
cz
TO
U
>
CO
a!
a
3
to
o
:r
V
t/1
g
—
—
—
-C
~ H
-1-
—
3
U
3
eg
'
QJ
-3
"o
~D
JS
ti-i
1
~
TO
oCJ
"C
~ ~
>
Table
4.
Gender and Leadership: Implicit Associations
IAT (D-measure of
bias against females)
Leadership/Domestic and
Male/Female
Male
Female
111
(2)
A
Panel
Ever Reserved
-0.077
0.021
(0.030)
(0.040)
Panel B
First
Reserved 2003
Reserved 1998 and 2003
Only Reserved 1998
Test:
2003 = both 1998 and 2003 = 1998
Mean of Never Reserved Sample
N
-0.086
0.122
(0.040)
(0.053)
-0.036
-0.106
(0.047)
(0.072)
-0.094
-0.027
(0.040)
(0.049)
0.507
0.010
0.110
0.150
(0.021)
(0.027)
477
357
Notes:
1
Dependent variables
measure indicates
are the
IAT D-measure,
a stereotype associating
defined in notes to Table
women
leadership activities. Reservation variables are as defined in Tables
2
The regressions include
clustered by GP.
3
the controls defined in notes to Table 2,
The sample
The p-value from
a
Wald
test
size reflects the
3.
A
positive
with domestic activities and
1
men
and 2 notes.
and standard errors are
sampling strategy (see Table 3 notes).
of the equality of the coefficients on First Reserved
2003, Reserved 1998 and 2003 and Only Reserved 1998
is
Dwith
reported.
in
;
,
1
—
;
I s
<u
»J
c
t-
Os
so OS
tl
M O
S M
O O
o o o o
ea
E
m *£ —
^ t
'O 'O I* >n
p © p o p o
© © © © ° ©
oo c^ in
°. o
9 o
<-?
<u
u-
P
.E
in
c
o
TJ
03
o
o
© n
o
q
o q
d d d
Os
<D
03
2
3
a>
tD
|
53
O
m
c
c
<n
m
o ro
d =
-3-
1/1
rsi
©
o o
d d
i
<u
a!
-rr
iri
i
r oo
m
c o
~ ©
o q
d d
E
— r- oo
\o
in h in
t
o o oo
o o o
o
o © o © © ©
SO
&
u
m
o ©
d =
i/">
so
in
»n
o
.=
«3
£
2
S
H,
9
V a
OS
ITS
— C
— = ~
=
^
~
ra
73
~[fl
t/n
U
U
u ~
re
>
S
U
o
«
CO
rsi
-3-
Os
c3
rsi
m
-f
o
—
o m
E
Csl
o o
d d
:
<u
oo
CM
en
rn c
©
~ c
— c o
d
03
2
d
ro
— 5sO
— cr »n
o
d d —
so
so
— —
S
o
o rM
9 d
©
ro
o c
d d
— in oo
O
o so i- \o
»n
—< p -^
p o
o o o p
o o o
OO os
)p.
SO
fcu
c
V*s
on
.5
oo
^
so rn
r-1
3-
d
o C »n
o
•n
»n
—
S?
-
3 S -
-E
-
<
i-
ra
^L)
U-
_
tn
rsi
4J
"3-
rsi
m
o oo
o ^t
o
03
2
^3-
r-si
rsi
v-i
M
-_<
^
—
o >n
o
CJ
SO
3
u
in m *n oo
-o in h ^ >n
S
P o
p o o O
^T O O O O O
tN
^
°. o
rs)
-rr
TD j:
c
o
5
o
"-p
3
Q
p
u
,o
—
03
— ^
m — OS
O
o
o o o
SO
*T)
S
i
-_
E
m iQ ro
o
- i
— so
o O oo
c o —
o o o o o o
rn
rj
* m
o e
rsj
t-~
rsi
—
«
u
03
<U
s
—<0
£
z
V
M
a
u.
u
JU
f
.
>
<
2
o
q
03
c_?
53
vn
— o os
m
o o m
o
d © ©
r*j
r~rsi
so
•£}
S
-3-
Ml 13
—
- O O
O O h
O O
d d d ci d ci
oo r- r~
r- oo
3<*
SO
r-
o os
o ro
o
H
° o d o
03
>
q o o o o
^ d o d d
P d
9 d
rsi
o o
<u
5
u
^
+
cy
2
2
i-
t c
S
>
=
°
^ >
a
5
2
—
ii
II
m
J3
a>
r^
°°
os
os
ra
JH
C/l
CD
£
CU
£1
u
CO
03
03
•^i
E
E
o
3
£
d
PL,
03
H £
Pm
„U
<L>
o
u-
Ph
ii
^
O
£
"O
^
C3
c
SJ
<l>
— —
rs"
u
>
ij
oo
—<U
'
m
O
o
(N
O
s
D
<u
O —
'-'.
C
u
f
o
ii
T3
3
m
o
o
d
O
*N <N
"^T
oo
X
c>
O"
_=
—
r
ra
3"
"5
ca
(4
(J
B #
t
-n "S
-C
Oh
J=
i
QJ
a <
—
tL,
t.
1
S
<S)
c
'
a
>
< ^
„
—
oZ
U
13
3>
- o
ur 2"
J5
8 ^
JT-
e
>,
—
(NinS^O^O^-MSOCN^
— © O O ~
—
P
p
p p
3 on
in C^
oopooooooopooo
—
-":
--;
;
p
1 O
,
o
~-
P — r-_ o — P
~-
—
— — —
u
r
C
—— 3
^
In
<s>
.
doPodoooPoPcJdo
:
a!
2
E
-a
_3
^
ac
>
_ oc
—^
i/i
S3
O
o-
E -g
n
rf
-'
i
">
H
£o
.2?
3
re
" o> n *
n
GO CN VO oo
O ~ O~ — o
00P0— O O O
ddpo'ddddpdpd
'
a
T3
'
r*^
11
o
m
Q.
X
PJ
uj
v-it^inrninfit^i^i^'n —
PopoPopoooopop
^ddd^ddddddddd
<1'
— ©\voovo^5r^om
r<-i
«
> u u
e
P. •£
re
S -a
td C
2 is
u v
'
I
iC r^ (N
in in
*
O
o o d
PoooooPooo
pdddddpddd
c ^
""-"'
o>
<D
<
o
5
~o
<
^
b
re
C J2
J5
>o vo IN
sy m
m
— * ^ f 2
o o q p p p
p p p P o o o
d d d d d d d ^ d 9 d d d
C-
=t
PJ
<*
r^-
'i
,,
I o
Cu -S
D.
<U
OJ
O
u
:
a)
c
o
o
—
c
j
"
re
5
oo
-o
ca
1 1
^
o
<u
C
*-
<U
fli
re
—
->
re
QJ
^^
^
.2
oi
u
>
</5
tu
*
c
a
-3
—
-
-a
q
«
JC3
V
»—
CL<L>
o
c
>
tt)
*
c
$
<u
-a
L.
a.
B
P
<u
U
u.
T3
t
f
C/l
t/J
<U
u
>
u
aj
tu
W
>
ti
-
i-S
c 3
-^
£ 2
-2
I -SI
U
U.
s
re
*
-a
>
4J
§£
S
m
td
u.
o
§
^
.2
5
o
I ^3
c
Z —
rsi
t;
"
—
*r.
—
tfl
_0>
«
E
-J
<D
ft,
U*
m
m
c
^
cd
_CJ
03
2
i
— —
f- rn co Tt
so
«n ^f *n
o p
P o p p
o
d
o d
o
9
!>
S2
^1 fN »
^
2
^ o — o t
d
o
d
9
<
VI ar- o -*
*
so
3
o "i
o o fo
P o <n
ci
6 d d d
9
"3-
03
o
Mc
(9
C3
M
g
^
^
s
o
>.
z
CD
-J
<L>
C/l
-
U
d
u
^2
bfl
03
c
co
B
u
Cl>
o
_
""!
—
m
3 o
O
p
3 d
>
>
73
m r> £; O
~ o
SO O SO S so
f^ o o o P o
9 d d d 9 d
^r
O
^ ST
o P o
9 d 9 d
2; rn* £3
V£>-
sc
o
CN
2
(LI
o
— o
m
rco
..
•*j~>
"»
~
^
P o
9 d
"~ CN
so
oo
o SO
° h
O "
CD c O
(N
-O
o
o o
o
.
?? o t; ST a «
Os K. CN
&
o 9 o
9
d d 9 d 9 d
o o
m
cn
CN »n
C o
<L>
>
u
u.
CO
—
t!>
rt
sS
•—
(D
ft.
o
Os
O
-
co
VI
u
SO
M"
CN
SO
ZZ
.2
o
u.
O
9
d d
SO
0O Tf
so
-T
d o
oi
to
CO
2
cn
so
9
N so
o o
o 9 ro o
>o
r
C O
© o
l
o
3
O
—g
_=
j
T3
eg
o
X
-
c"
53
UJ
O
odd
cn
Ov
^r
co
o o o
E
Cl>
O *
(N ^>
d
ft.
I~-
Os
CN
Vi
m
d d
«
a
c
ca
5
>
<
cu
to
p
t
(0
CO
2
n
c Os
P in — r- c
in
H o o o c p
9 d d d d d
T3
CO
Os
o o
p
d
d p
r-i
u
SO
^5
SO
<4-l
o
ca
C
T3
ca
OO
Os
Os
E;
<L»
W
(U
i—
II
B
CD
s
m
c
c
CN
ac
-
>*
O
O
o
o
CN (N
W
1—
-O T3
CO
CU
t/3
3
a
-OO
Q>
ON
1
'
T3
:o
O
0>
-a
d>
>
a>
T3
C/J
eg
!L*
oc!
>i
irt
U
G
Cci
c
••*
(U
H
^5
T?
=
—
ra
cu
a.
CI
TO
C3
i>
£
u
w
cu
-W
VI
cu
f-
oi
CU
cn
u
_3
_-
C
<u
o
C/J
"3
.c
cn
o c
'1
4J
-CJ
o
£
m
~
T3
G
^
U
u
5d
5
ca
c
o
tfc
cu
1_
(U
U
3
-5
H
Z O
O
E
o
o
t
<+-i
C
C/l
u
u
6
cn
c
II
rn
m
o
o
o
o
CN CN
•5
CO
~
X)
J3 J3
1/5
on
a
U-i
<U
o o
-i-*
X
CO
H
c
"—
sS •£
II
rN
U
CO CO
OS ON
OS OS
tu
-s
5
3 i
o m
CO
c
a
c
C E
c
C/l
1
U
>
2
III
=
CO
cu
0)
CO
>
u
M
V
o-
cn
U
3
3
>
c
~3
tU
D.
U
< £
tN
ro
<u
~
;
'
1
w
_c
<D
(J
c
y t>
s
g .0
—
n
O
0>
u
£
B
a.
.£P
re
u
On ON co
r- in »n
no rn
m
NO
en
in
O
O
O
<* en
ON
O O
1
< E
LJ_
On m
3 * 2 *n
*
"*
9
00
9 O N
9 O X
OO
o
o
9
9
e<->
in
r>
-^r
On
r-
<"
O
in
no
<3-
r-
U
M
rz
u.
U
^ (N °°
£ m £
9 o 9
© 9
L/-,
O
_n
>
< £
<?'
•a
s s
9 o
n t^
O
m
m
o J3
9 o
o 9 o
NO - (N On PT
2
S
£3 >o
Q o °
o 9 o
9 O O o 9 o
o
o
"
-
r-
o
o o
—
(N NO
OO ON
^r in
'
"=*
a
<D
NO ^t"
r- co
1
Z 2
00 a,
<
— On
O O no
— q m
O
o o O 00
O O
(N
on
>
o
bS
t~~
0000
CO
ON
1
—
II
en en
a O
cs
-3
CO
b
3
a
e
3
ON
~0
>
CO
a
00
en
O
O
CN
a
C4
OO
c
cn
CO
*~~
00
ON
On
-0
O
tt5
5
_
-D
a)
II
II
en
en
r
<N
1
4_;
JJ
Crt
VI
S
s "-'
es
B
5
-
O
u
- Hu
<u
a
O
^
_
H
(N
Cj no
ii3
—
o
u T3
> OJ
to
cd
c3
^o on r^ *t (N on
in
o
o o
o ^
o o n
o N
o
o o o o o o
ca
o
C3
-J_
t£
S
<
«
-i,
a
ca
CJ
Ph
O
c
-e
—
i<u
<
Oh
—
o
B
0>
E
—
r- oo
g;
<N oo On
u
£3
o.
X
-*
^3
oo
J3 -* Jg
o>
2
— 2 —
^
O
o
""!
'
i
«
~
— m
- O
O.
Si-sa
«
o
-3
E
5
DO
<u
en
r^
1
oo 25 oo
— 2 o 2
9 o
r-
e
00
2
S
<N
9 o
o 6
p
C o
ca
'
o
=
d
V
~
X
75
0)
<u
TJ
^
:z
x:
~
(D
x
£
m
o oo
O
— rN on
ro o
o o — m
o o o o o
i
c
"3
u
a
C
CO
£
=j
O
r_
o
—
O On
oo
o
_ oo
o — UT
9 o
o o o 9 o
£">
oo
oo
-*
o O
O O
(N
-c
r-j
O o
<
T] oo in on
—
— O o
o — H
9 m
o o 9 o o o
r-
U
r-
l
-
fN>
-sr
on
o
o o
IT)
i3
O
C
c
en
u
u
to
en
O
o
M
<
tn
—
i
en
5
g
U
ON OO -*
oo
oo £N no
o —
o o
o o o
©
o
(N O —
— g
O- ^> -*
S
"^
no
^ r- S
-7
O
O O O
^
°^ ON
r-~
u r
> —
W
*Q
D.
i £
5
x
cd
-—
u
C/l
"O
T3
en
u
> s
—
C
v.
u
-C
o
u
— t
C
u
rN
P>
~
P
— — in
On
O )£
O o — o
9 o
"9 o o o 9 o
2
a>
>
T3
"
T3
_e
i
9 o
EJ
^ ON
^ ss «
£9
OO
w-i
c3
Cu
a.
QJ
CJ
Eg
£
2 o
00
u
~
-
5
=
o
^
o
73 73
C-
7j
u
•=
(n oo rg
rn
rJ
r-~
o
<N
-
c
'5
cu
u
J=
>
B
c
a ae
cd
^^
>
c
z
aD
en
C4
JD
en
V
as
c
u
i:>
t
s
c
tl
c
u
c i
o n*
3
a
O £
en
u ~
u
-3
a.
-J
M
jS
z
-3
tl
_f
-^
en
^_,
C^
>i
C
CO
1>
=
U
*S
i
Table
Robustness
10.
Average
I.
Evaluation of
First
Time
Evaluation of
II.
in
2003 Term
in
2003
*
Reserved
Reserved 2003
(for
Only Reserved 1998
IV.
Reserved 2003
SC
Panel
-0.100
-0.030
(0.067)
(0.064)
-0.082
-0.052
(0.058)
(0.053)
(for
*
Woman
for
SC
in
1998
0.125
0.006
(0.132)
(0.119)
Reservation
Woman)
Reserved 1998 and 2003 (for
First
0.007
(0.073)
1998
Evaluation of Female Pradhans and
First
0.016
(0.072)
2003
in
SC
Reserved for
III.
in
SC
Reserved for
-0.212
-0.070
(0.074)
(0.075)
Woman)
0.042
0.003
(0.087)
(0.063)
Woman)
Previously Reserved for
0.007
-0.028
(0.086)
(0.076)
SC
0.040
-0.007
(0.132)
(0.131)
Reservation on Female Leadership: Speech and Vignettes Results
A
Female Pradhan
Female Pradhan
Panel
(21
SC Pradhans
SC
Reserved for
Female
(D
New Pradhans
Pradhan
as
effect
Male
*
Ever Reserved
-0.046
-0.040
(0.035)
(0.034)
SC
-0.014
0.020
(0.039)
(0.039)
0.004
0.017
(0.042)
(0.045)
B
Female Pradhan
*
Reserved for
SC
in
2003
Female Pradhan
*
Reserved
SC
in
1998
for
-0.028
0.030
(0.047)
(0.048)
Notes:
1
The regressions
in
Panel
I
—I II
use the average evaluation coefficient for Pradhan (explained in notes to Table 7) as the
dependent variable.
2
In Panel
sample
3
In Panel
the explanatory variable of interest
I
is
II
In Panel
whether the Pradhan was elected for the
the explanatory variables of interest are whether the
and the interaction of being reserved
4
is
first
time
in
2003 and the
restricted to the set of never reserved Pradhans.
III.
for a
woman Pradhan
the explanatory variables of interest are the
in
GP was
reserved for
SC
in the
1998 and being reserved for an
women
2003 and 1998 elections
SC Pradhan
in
2003.
reservation status indicators and the interaction of
"Previously Reserved for SC" indicator with the "First Reserved 2003" indicator. The regressions also include the
interaction of
5
The
in
"Only Reserved 1998" and "Reserved 1998 and 2003" with "Previously Reserved
SC"
indicator.
regressions in Panel IV use the average perception of Pradhan effectiveness in the speech and vignettes (explained
notes to Table 5) as the dependent variable. The explanatory variables of interest are the indicators for female
Pradhan
speech/vignette and the interaction of this indicator with the reservation for
in the
addition, the regressions include the indicator variables for ever reserved for
2003 and 1998
6
for
in
in the
in
SC
status
of the GP.
Panel A, and reserved for
In
SC
in
Panel B.
All regressions include the controls defined in Table
defined
SC
notes of Table
1.
2,
and standard errors are clustered by GP. Reservation types are
1
1
x
©
c
<N
_c
!/.
C
c
o
o
o
©
'-C
u
<u
V
c
£
ta
»
(N ^ \o oo n
—
«n m o — —
o ©'
o o o o —
o
o o © o
oo »n ir>
^oo
(N
O — _
0-
O
5
o
-
c
o
c
g
f2
£
-H:
i
2
£
C/l
aj
xi
CO
cfl
^.
>
u
u
4>
o
5
u
OJ
Os
C/3
a
c O
l/J
B
c
L>
*z!
•->
«J
w
)0 - \o
S
— ~— ^ vo
—
~ ~ —
~ —~ —
-2 ~
P O © O ° o o
r*1
-_
cu
v_
—
oo
N n
2
O IN
<N
"1
o
c
iC
u o
.5
—
CL
->
o
a
-6
3 Z
a
o X
<L>
«!
4J
(U
1/1
§
i-
3
=
— £
13
.—
-n
O
ra
rj
1)
&
v?
u
p
lc
4>
£X
a
3
p
^J-
tS
—
5
,
g
3
o
s
o
c
o
&c
CN
V)
<D
i>
oi
O
"
c
a.
o
r-
T3
c S
"j
*"'
ra
——
a
u
1J
6
u
JS
a
=a
"3
o
o
>
ao
in
ti-.
(fi
^
a.
en
£
M ~
—
(S o 2
j3
-a
u
c
u
,_;
U
7
O
<
=§
a
_o
Z <
2
(S
a
Cl.
o o
o
°l
5
"31
1)
=
B
CL)
>
<§
5
— O
3
CD
r- on
22 ^> no
g
r- mc [n \o 5- o
~
CN m (N — (N o
p © o o P © ©'
rsi
;
^^
oe
c
r^
r*-i
fn
~
o
__
r^
o
ND
no
-
—
«
<
~
s
o
acu
>
CL>
1)
-^
X.
i
S
>
CD
o
£
=
<|
3 9
S
u
Cu
CU
c
S
a
(A
B
~5
U
S
u
c
Q
cu
0)
ab
CQ
&
XI
5
o
u
-5
>^
U
r-
•5^5=
c
s
'C
S3
C/l
o
u
o a
3
fS
^
a
O
DMo
c
ra
on
on
•«=
^:l
u
iu
ra
o ^
€
g
,
»
2
h
,u
S3
B
'5
w
6
o
-
>. .3
S -a
m
o
o
(N
p*"l
3
5
£
2
o
o
(N
^
T3
V3
CS
a
00
a)
on
fj
&
3C
ON
1
g
:"
OJ
Ol
C
'""
'=
°
«
cu
5
>
— w
-°
cq
—
-=
S
-O
~
or
c
^
-a
I
59
tl^
1 i
*—
6i
1 1
"i
4-.
u
<N
i/i
+3
aj
—
n
c
u
>-
o
Z -
U
Bi
i
r^i
"O
S
tti
«
O
QJ
S R
a €
S « 2
c c
_ -—
o
O
<N u u u
.
ro
^ t;
°
u w
!9
•a
S3
=;
5 <
i
TO
q
«
5
q
b
>
CL
6
J3
-;
feh
no m m ro o o o
© © © ©
-
oo
m
T ©
— o ~- o On
o
o
o © © © ©' o
no
r**i
p-)
iri
p\i
ia
r- 00 M
no
m
o t
o ©
o o
© © o ©
VO ^ "i " O
c. 00 — o p to o — — P o
—
m —
if © no
©
O
O O O
©'©©'©
t
*T O >0
o o
o o
© O © O
NO
(N
m
m On
— —
t
o wn
o o r*
o
o © o o
:
NO
CO
iTi
TT
=
o o o o
l S
<
O
U
^
~
O
*
V
r-
cu
«
o o
©
© © o ©
rvl
OC
rn
Op
O ©
— — — —
O O © O
o o
o ©
— — — —
© © o ©*
vn
O m
On
nO
NO
On
o o ©
*
rs
°0
—
5
©
Ov nT
N
—
~
— — ©
on m
5 © —
n in
P
©
tj-
q
=
^
— ©
© ©
1«I
Cd
<*1
vo"
£
a s
^O
ON
CO
r-j
£^
C
_Z
M
(N
m (N
no
T m
in
NO
in
©
•*r
On
r*
rJ
On
no
^
c«
s
-^ '5
S
S
o
(*-
©©©'©©'
© o p p © ©
o o © o ©' o
m cc
r- co
© o r© r^
© co
© no
©
© © o o © o
•—
r»
oo
r-
Ji
=
o
O
-s
»n
© © On
© © o
o © © ©
r-
o
r- no r- r- no
©
© © © ©
VT
-3-
On
On
feb.
CO
On
O
© o m
o o
O O © ©
© © © ©
©
r*-,
p © p — p ©
m
© ©
m
m r— >n
o no
© no
© o
© r©
© © ©'©©'©
=S-2§5 g
r^T
_
-^r
— O © ©
© ©' o o
^"
P On On
™ On ^ qq
©
—
©
P © P
° © P ©' P ©
n*>
*n
© © ©
£J rs
£
o
*-.
r- On no ©
_ o — —
o o © o
>
fT
o o © © o ©
© ©' ©' ©' © o
o
P in
© P rP © P ©'
;
a
—
rj
©
*d-
— r- no o
m
© t
o © «
o
o o o o
r~
o
o
© o o © © o
2?
TO
f O
— NO
co
o © on
o
o
©©'©'©
On
Nf>
© © © © ©
o o o © ©
© ©
E
-a
-o
"
—
TT
TT —
O —
— © © © © ©
© © ©'©©'©'
© © © © o
m no no
on — t—
—
p © o — ©
o ©' o © o o
© © © © ©
© © p
p ©
©'©'©' o
© © © © ©
o © © o ©
^ v ~
§ 5 S
©©'©©©
©©'©©' ©
<
—
o o
;
-a
_ _
ST
C 5
E "S
«
u u
5
Is
e9
I a |
i |-|
cu
3
>"
«
E
Z
S
Its J
§ £ «
E
a
Q
1!
>
6b
"
u
.
Table A2. Effect of Female Leadership on Public Goods Quantity
Coefficients on
GP
GP
Reserved
Previously
only
N
At Least One
New
Tubewell was Built
495
At Least One Tubewell was Repaired
At Least One Drainage/Sanitation
At Least One Drainage/Sanitation
At Least One
At Least One
Irrigation
Irrigation
Number of Metal Roads
Pump was
Pump was
482
Facility
Facility
was
Built
was Repaired
396
495
Built
319
Repaired
Built or Repaired Since
Number of transportation
495
2003
related infrastructure
495
495
(bus stop, bus service, taxi)
At Least One Educational
At Least One Educational
Facility
Facility
was
Built
was Repaired
At Least One Community Education Center
There
is
a
NGO
(PHC, Health Sub Center)
Facilities
At Least One Health
At Least One Health
Number of Trained
Facility
Facility
was
(0 if no fac)
Dais, Untrained Dais and
Private Doctors
Average Effect
495
495
495
Built
was Repaired
465
495
Child Center/Creche
Number of Health
495
495
495
in
Reserved
2003
GP
in
1998
Reserved
in
Mean of
2003 and
Never
1998
Reserved
0.152
0.073
0.160
0.365
(0.066)
(0.063)
(0.088)
(0.482)
0.208
0.130
0.080
0.628
(0.067)
(0.064)
(0.089)
(0.484)
0.053
-0.113
0.052
0.428
(0.067)
(0.059)
(0.091)
(0.496)
0.150
-0.017
0.032
0.178
(0.067)
(0.062)
(0.071)
(0.384)
0.137
0.005
-0.013
0.180
(0.053)
(0.051)
(0.050)
(0.385)
0.110
-0.078
-0.005
0.417
(0.092)
(0.086)
(0.123)
(0.495)
0.274
0.046
0.079
0.118
(0.117)
(0.070)
(0.065)
(0.448)
0.074
0.250
0.303
1.302
(0.175)
(0.160)
(0.225)
(1.20!)
0.053
-0.030
0.026
0.117
(0.042)
(0.036)
(0.055)
(0.322)
0.165
0.039
0.001
0.296
(0.072)
(0.069)
(0.097)
(0.458)
-0.007
0.030
-0.001
0.009
(0.010)
(0.023)
(0.009)
(0.095)
-0.045
-0.039
-0.027
0.045
(0.016)
(0.021)
(0.023)
(0.208)
-0.025
0.027
-0.005
0.257
(0.049)
(0.052)
(0.084)
(0.468)
0.011
-0.004
-0.018
0.014
(0.015)
(0.014)
(0.009)
(0.116)
0.061
0.016
0.047
0.009
(0.023)
(0.016)
(0.024)
(0.095)
-0.069
-0.158
0.384
1.014
(0.232)
(0.226)
(0.423)
(2.012)
0.192
0.039
0.097
(0.070)
(0.061)
(0.082)
Notes:
1
All regressions include block fixed effects, and standard errors are clustered by
2 Reservation status defined in Table
3
Average
effect coefficients are
GP.
1
from a regression on the average of all the dependent variables
normalized by the mean and standard deviation of the never reserved sample.
listed in the
rows above
Figure Al: Female/Male Politician Good/Bad Stereotypic Block Example
Figure A2: Participant Performing
IAT