Title: Nested Designs: Field and Natural Experiments and the Role of Qualitative Methods
1Nested Designs Field and Natural Experiments and
the Role of Qualitative Methods
- Thad Dunning
- Department of Political Science
- Yale University
Prepared for the Short Course on Nested Research
Designs sponsored by the Organized Section on
Qualitative and Multi-Method Research, APSA 2009
2Plan for the talk
- Using nested designs to strengthen causal
inference, in a comparative project on cleavage
structures - Two examples
- Cross-Cutting Cleavages and Ethnic Voting in
Mali A Field Experiment - The Salience of Caste Categories in India
Nesting a Field Experiment within a Natural
Experiment - Two of the key points
- Qualitative methods play a central role
- External validity concerns arise in both studies
combining experimental and observational data to
address them - Practical issues how feasible are such studies?
3Example 1 Cross-Cutting Cleavages and Ethnic
Voting
- Do cross-cutting cleavagesthat is, dimensions of
identity or interest along which members of the
same ethnic group have diverse allegianceslimit
ethnic voting? - Classic insights have been extended in the recent
comparative politics and political economy
literature - Yet estimating causal effects is difficult
4 Why no ethnic voting in Mali?
- Mali is ethnically heterogenous
- Yet parties do not form along ethnic lines
- Unlike many sub-Saharan African countries, in
Mali ethnicity is a poor predictor of individual
vote choice - One hypothesis cousinage
- Malians with certain family names are linked
through joking alliances with their fictive
cousins - Two strangers -- say, a Keita and a Coulibaly --
may use cousinage relations to establish rapport
and limit interpersonal conflict
5Cousinage as a cross-cutting cleavage
- The key point cousinage alliances cross-cut
ethnic ties, because they occur across as well as
within ethnic groups - Such cross-cutting alliances may weaken the
correlation between ethnicity and vote choice - Imagine two voters with the same ethnic
relationship to a candidate but different
cousinage relations their preferences over this
candidate may diverge - Dunning and Harrison (2008) use a field
experiment to study the effect of cousinage and
co-ethnicity on candidate preferences - The experimental design can be applied in other
contexts
6Cousinage in brief
- Cousinage relations were codified during the rule
of the emperor Sundiata Keita (c. 1235-1255) and
exist in Mali, Sénégal, Guinea, the Gambia, the
northern Ivory Coast, and Burkina Faso - Though various kinds of cousinage ties exist, we
focus here on cousinage alliances between
patronyms - Standard jokes are used to establish rapport with
cousins (though joking is by no means
automatic) - Our focus is not on explaining the origins or
persistence of cousinage alliances but rather on
estimating their causal effects
7Experimental Design
- Any voter and politician can be placed in this
2x2 table, depending on their relationship with
each other - In Mali, last (family) name conveys information
about both ethnic identity and cousinage
relations, allowing the classification of
voter-politician pairs
Joking cousins
Not joking cousins
Same ethnicity
Different ethnicity
8Experimental manipulation
- We filmed two Malians delivering an identical
political speech (in Bambara) and showed the
speech to Malian participants - We varied the last name of the politician across
different versions of the speech - Each subject was assigned at random, with equal
probability, into one of six treatment
conditions - The four cells in the 2x2 table, plus
- A no name condition
- A same name condition
9Stimulus The political speech
- The content of the speech was intended to be
typical of speeches by candidates to the National
Assembly - Themes need to improve infrastructure,
education, electricity - 56 percent of subjects said the speech reminded
them of a speech they had heard on a previous
occasion - The use of Bambara, Malis lingua franca, does
not imply a particular ethnic identity for the
politician - Given politicians last name, subjects correctly
inferred ethnicity more than 85 percent of the
time, from 14 ethnic categories - When given no last name, their guesses roughly
mirrored the distribution of ethnic groups in
Bamako - No difference between the two actorsone of whom
was ethnically Peulh, the other of whom was
Bambara
10(No Transcript)
11(No Transcript)
12Subject recruitment
- A door-to-door canvass in all of Bamakos
neighborhoods (quartiers) - A convenience sample -- however, intended to be
as representative of Bamako as possible - After subjects agreed to participate, we obtained
initial subject data, including last name and
self-identified ethnicity - Subjects were then randomized into one of six
treatment conditions, shown the video, and
administered a post-speech questionnaire
13(No Transcript)
14(No Transcript)
15(No Transcript)
16Randomization of Treatment Assignment Creating a
cousinage map
- A challenge linking particular last names in the
cousinage system - Qualitative (open-ended, semi-structured)
interviews played a key role - We used the interviews and drew on the literature
to create a cousinage map, that is, a matrix - the rows of which list potential last names of
subjects - the columns of which give last names associated
with each of the six treatment conditions
17A typical row of the random assignment matrix
(1) (2) (3) (4) (5) (6)
Co-ethnic/ cousin
Co-ethnic/ non-cousin
Non-coethnic/ cousin
Non-coethnic/ non-cousin
Same name
No name
- The final matrix has around 200 last names in
the first column (the names for potential
subjects)
18Randomization creating a cousinage matrix
- We had to revise the matrix after initial field
trials, using field interviews with informants as
well as our initial data to correct errors - An iterative process that very centrally involved
qualitative methods - Challenges
- Expanding the left column to include sufficient
last names (for ease of subject recruitment) - Improving the match between random assignment and
subject perception
19Weaknesses of the experimental design
- Estimated treatment effects may not be large, for
a number of reasons - Stimulus is somewhat artificial?
- May not prime ethnicity
- May not prime cousinage
- Measurement error
- Turning to the analysis
- Intention-to-treat analysis
- Can the experimental evidence explain the
observational relationship?
20Average candidate evaluations, by treatment
assignment
The figure displays average answers to the
question On a scale of 1 to 7, how much does
this speech make you want to vote for (name of
candidate)?
21The salience of ethnicity qualitative evidence
- Comments by experimental subjects as well as
other field interviewers underscore the social as
well as political salience of ethnicity - An ethnic Bamanan subject a politician named
Guindo (an ethnic Dogon patronym) could never do
a good job - An ethnic Bamanan subject the Dogons don't know
how to lead. - An ethnic Songhai suggested that Bobo ethnics
don't know anything about politics, while an
ethnic Malinké subject said the same of Dogons. - An ethnic Soninké subject offered the opinion
that the Malinkés are not intelligent. - Subjects tended to offer more positive comments
about co-ethnics. - Subjects were especially prone to praising
politicians bearing their own patronyms - From a subject named Anne The Anne family is
composed of intellectuals. - From one subject named Sacko A Sacko is a hard
worker. From another The Sackos are very
cultured. - From a Kouyate if a griot (Djely, Kouyate) is a
candidate, it is because he is capable of many
things. - From a Koné, explaining why she paid attention to
the candidates last name The Konés are nobles
(the Konés were members of the caste of nobles
during the Mali Empire) - A subject named Keita, when asked whether she
would be more susceptible of voting for a
candidate sharing her family name said yes, like
uncle IBKa reference to an opposition candidate
during the 2007 presidential elections, whose
patronym is Keita.
22The salience of cousinage qualitative evidence
- Comments from experimental subjects and other
field interviewees underscore the salience of
cousinage alliances - Voters anticipate being able to make requests of
as well as sanction their cousins - If (the politician/cousin) is not serious, we
will correct him. (Experimental subject,
explaining why she would be more likely to vote
for a joking cousin) - If the politican does not respect his promises,
we will bring him to heel, because he is a
senanku (cousin). (Experimental subject,
explaining why he would be more likely to vote
for a joking cousin) - One can never hurt (ones) cousin and one must
do what (ones) cousin asks. (Field interview) - Cousinage alliances result in greater
willingness to make voluntary material sacrifices
(of resources, time, willingness to voluntarily
cede in disputes, etc.) (Galvan 2006) - Voters tend to vote for their allies (cousins),
saying that in case of problemsadministrative,
political, or socialthe elected ally would be
more prompt to intervene than he would be even
with a direct member of his own family. (Douyon
2006) - We cannot fully distinguish mechanisms that
explain why cousinage affects political
preferences however, trustworthiness and
credibility appear to play an important role
23The Effect of Cousinage on Perceptions of
Candidate Attributes (Differences of Means,
Cousins Minus Non-Cousins)
The figure reports the estimated effect of
cousinage alliances on subjects evaluations of
the candidates attributes. All variables are
rescaled to run from 0-1, so effect sizes are on
that scale. The darkened circles give point
estimates, while vertical lines show 95
confidence intervals. The analysis pools across
co-ethnicity that is, mean responses of
subjects assigned to the co-ethnic, non-cousin
or non-coethnic, non-cousin conditions are
subtracted from the mean responses of subjects
assigned to the co-ethnic cousin or
not-coethnic cousin conditions.
24Treatment effects, in sum
- Both co-ethnicity and cousinage alliances make
evaluations of the politician more positivebut
cousins from a different ethnic group are just as
attractive as non-cousins from voters own ethnic
group - That is, evaluations of co-ethnic non-cousins and
non-coethnic cousins are statistically
indistinguishable - Estimated treatment effects are similar for
similar questions (such as, on a scale of 1 to
7, how would you rate the global quality of this
speech)? - Trustworthiness and credibility appear to play an
important role - So may social networks voters in both the
cousin and the co-ethnic conditions report having
more friends and acquaintances with the
politicians last name
25 External validity concerns
- External validity concerns are very central here
- One narrow sense of external validity is the
experimental study group representative of the
broader population? - On ethnicity, apparently
- On gender, not at all
- This is a concern our results are valid for the
experimental study group but are they for the
broader population? - On the other hand, treatment effects are very
similar for men and for women (we also use
sampling weights, with caveats, to construct
estimates for the broader population) - But can the experimental results explain the
real-world puzzle? - The issue of variance explained
26Can the Experimental Evidence Explain the
Observational Data? Three Important Points
- Cousinage relations are politically (and not just
culturally or socially) salient - Treatment effects are much stronger for
politically-active subjects - Candidates exploit cousinage relations in
campaigns - Cousinage relations are widespread
- The probability that a politician and a voter
drawn at random are cousins matches or exceeds
the probability that they are co-ethnics - E.g., for a Keita, the probability that a voter
drawn at random is a co-ethnic is about 0.15 the
probability that s/he is a cousin is about 0.52 - Co-ethnic and cousinage ties are negatively
associated - E.g., among eligible subjects who are not
ethnically Malinké, an ethnic Malinké named Keita
is most likely to be a cousin, while among
Malinkés, a Keita is most likely to be a
non-cousin - Voters prefer co-ethnics but an unobserved
omitted variable attenuates the observed
correlation between co-ethnicity and candidate
preferences -
27 Other results (see paper!)
- We also test the observable implications for
party strategy - Parties appear to exploit cousinage networks in
placing candidates on lists - We use Afrobarometer data and surnames of
candidates in the 2007 parliamentary elections to
evaluate this - Note that it is useful to combine observational
and experimental data in this regard
particularly, to link the experimental results to
real-world outcomes and (partially) address
concerns about external validity
28Example 2 Combining Field and Natural
Experiments to Strengthen Causal Inference
- Several recent papers combine field and natural
experiments or merge true experiments at
different levels of analysis - Beaman et al. 2008, Fearon et al. 2009
- Such approaches can be quite useful for measuring
the causal impact of institutions - Here I present research on how electoral quotas
shape the salience of caste in India - I use a field experimental design similar to the
design in Mali but embed the field experiment
in a natural experiment, in which caste quotas
for the presidencies of village councils are
as-if randomly assigned
29The Salience of Ethnic Categories
- Political leadership and electoral rules may
shape the salience of different forms of ethnic
identification - In India, presidencies of some village councils
are reserved, via an electoral quota, for
presidents from lower-caste groups - Such quotas may increase the salience of the
larger caste category on which reservation is
based, at the expense of the individual
sub-castes that comprise the category - Yet estimating causal effects is difficult, if
reservation is related to observed or unobserved
characteristics of council constituencies
30Measuring the Effect of Reservation A
Regression-Discontinuity Design
- Reservation of council presidencies rotates
across village councils on the basis of
population proportions of the targeted groups - I exploited an RD design to construct a study
group of 160 councils, located in the state of
Karnataka, in which reservation is plausibly
assigned as-if at random - The main advantage we can attach causal
interpretations to post-reservation differences
across reserved and unreserved councils - The procedure also produces a study group of
council constituencies in which the population
proportion of the targeted groups varies greatly,
which may help with external validity
31(No Transcript)
32Measuring Caste-Based Preferences A Field
Experiment
- In the 160 selected villages, I then implemented
a field experiment in which I manipulated the
perceived caste relationship between subjects and
a videotaped actor/political candidate (by
changing the surname of the politician) - Subjects were assigned at random to one of three
conditions - Subject and politician share the same sub-caste,
same larger category - Subject and politician from different sub-castes
but the same larger category - Subject and politician are from different larger
categories - The identity of the actor and the text of the
speech were identical across conditions (with a
few caveats) - I can then compare subjects evaluations of the
politician (including vote preference) across
treatments and I can compare treatment effects
across reserved and unreserved council
constituencies
33Caste categories in Karnataka
- The structure of caste in Karnataka raises
interesting questions about the salience of
ethnic categories - Two main jatis comprise the Scheduled Castes
- The Holaya and the Madiga are former Untouchable
sub-castes (Harijans), with some history of
antagonism and competition - There are two dominant jatis among the Backward
Castes - The Vokkaliga and Lingayath sub-castes tend to
dominate politics at the local and state level - So what dimension of caste is most important to
voters? - E.g., Holaya/Madiga or Scheduled Caste?
- Vokkaliga/Lingayath or Backward Caste (anti-SC)?
-
- And what shapes the relative salience of these
categories?
34Reservation The system of rotation
- Reservation for lower-caste politicians of the
council presidencies has been assigned as
follows, starting with the 1994 elections - In each taluk (an administrative unit below the
district), the number N of presidencies to be
reserved for each group (Scheduled Caste or
Scheduled Tribe) is determined by each groups
population proportion in the taluk as a whole - To allocate reservation to particular panchayats,
a bureaucrat lists panchayats in descending order
by the number of council seats reserved each
relevant group the number of reserved seats is
determined by the proportion of each group in the
panchayat population - Then, starting with the Scheduled Caste category,
the bureaucrat moves down the list of panchayats,
reserving the first N presidencies - The same is then done for Scheduled Tribes the
remaining Gram Panchayats are reserved for two
groups of Backward Castes (A and B) or left for
the General category - Within a bin (defined by the number of reserved
seats), if there are more councils than
presidencies to be reserved, reservation is
allocated by lottery - In the next election, the bureaucrat takes up
where he or she left off, rotating reservation to
the next N villages on the list, for each
respective category - If, in any election, a Gram Panchayat is already
reserved for one category (e.g., Scheduled Caste)
but appears in among the GPs that should be
reserved for andother category (e.g., Scheduled
Tribe), the panchayat is skipped and then
reserved for the latter category in the
subsequent election
35Verifying the assignment procedure
- I obtained reservation data for the entire state
of Karnataka for each presidential term (1994,
2000, 2002, 2005, 2007) - This allows me to verify that the selection
procedure was actually employed - Bureaucrats are required by state regulations to
hold meetings at the taluk level, where the
identity of reserved councils are announced and
the criteria for allocation are explained - This should increase the transparency of the
process and limit the potential for lobbying - Fieldwork confirms that at least some of these
meetings were held
36Village selection a regression-discontinuity
design
- I mimicked the process of reservation as closely
as possible, using census data on group
proportions (the same data used by the
bureaucrats) - By listing panchayats in descending order of
population proportion for each group, and using
reservation data, one can find the threshold
points that is, the cut-point between
panchayats in each category that were reserved
and those that were not - The idea of the regression-discontinuity design
is to select councils on either side of, and
nearest to, the threshold. Observed and
unobserved variables should be very nearly
balancedand assignment to reservation thus may
be plausibly as-if random near the threshold - Local independence is bolstered in Karnataka
because many panchayats with similar group
proportions will have the same number of seats
reserved for membersand then reservation of the
presidency is assigned via a lottery - One hiccup is that I used the underlying
populations proportions (sigh), whereas
bureaucrats used the number of members seats
though the latter are based on the underlying
proportions, larger panchayats will on average be
higher on the list - However, this should not bias inferences, since
councils just above and below my thresholds
should have similar populations, on average
37Reservation as-if randomization checks
Group 1 Reserved for SC or ST (A) Group 2 Unreserved or reserved for OBC (B) Difference of Means (A) - (B) p-value (two-sided)
Mean population (Standard error) 5684.17 (200.44) 6055.3 (180.60) -371.13 (269.80) 0.17
Mean SC population (Standard error) 1119.21 (91.91) 1114.16 (67.84) 5.05 (114.23) 0.96
Mean ST population (Standard error) 505.52 (56.70) 444.85 (43.86) 60.67 (71.69) 0.40
Mean number of literates (Standard error) 3076.63 (111.46) 3315.61 (114.5) -238.98 (159.79) 0.14
Mean number of workers (Standard error) 2860.12 (103.03) 3017.59 (92.41) -157.47 (138.40) 0.26
Mean male population, mean male population age
0-6, mean number of marginal workers, and other
pre-treatment covariates are omitted but also
pass the covariate balance test. P-value for
assignment covariates is 0.97 (SC proportion) and
0.26 (mean ST proportion).
38Design of the field experiment
- I filmed an actor delivering two versions of a
political speech (in Kannada) - During recruitment and prior to random
assignment, subjects revealed their jati and
several other attributes on a screening
questionnaire - Subjects were shown one of the two (identical)
speeches - In the introduction to the video and in every
follow-up question, field investigators varied
the surname of the candidate, according to the
treatment assigned
39Experimental Design
Same caste category
Different caste category
N458
N470 N525
Same sub-caste (jati)
Different sub-caste (jati)
40Politicians surname by treatment condition
(Selected sub-castes)
Subjects sub-caste (jati) Subjects caste category Condition 1 Subject and politician are from same jati and caste category Condition 2 Subject and politician are from different jati, same caste category Condition 3 Subject and politician are from different jati and caste category
Madiga SC Madiga Holaya Gowda Lingayath
Holaya SC Holaya Madiga Gowda Lingayath
Nayaka or other tribe ST Nayaka Madiga Holaya Gowda Lingayath
Lingayath BC Lingayath Gowda Madiga Holaya
Vokkaliga BC Gowda Lingayath Madiga Holaya
Brahmin Forward Deshpande Gowda Lingayath Madiga Holaya
Omitted subject sub-castes include Lambani
(SC), Kumbara (BC), and Bunt (BC)
41Stimulus The political speech
- There were two versions of the political
speech?one programmatic and the other
clientelistic ?with the version to be shown
assigned at random - The programmatic speech focused on general needs
and local public goods - The clientelistic speech focused on jobs,
schemes, income and caste certificates, and other
targeted benefits - The content of the speech appears to have had
relatively little impact on candidate evaluations - In the analysis, I pool across the two versions
of the speech
42Subject recruitment
- A stratified random sample of ten respondents per
village - Four SC residents (ideally, two Holaya and two
Madiga) - One ST resident (Nayaka)
- Five residents from the general category
(including Backward Castes) - Some limited substitution of jatis was allowed
- Residential segregation in villages eased the
sampling - Twenty teams of two field investigators visited
10 villages (on average) - Thus, 200 villages, with 2000 participants
- 40 villages were set aside for a pilot study,
leaving 160 villages and 1600 subjects (in
principle) for this experiment - In each village, surveys were also taken of the
council president, two council members, and the
executive secretary (more later)
43Ethnic distribution of experimental population
Caste category Sub-caste (jati) N Percent
Scheduled Caste Holaya Madiga Lambani 331 228 23 22.87 15.76 1.59
Scheduled Tribe Nayaka 133 9.19
Dominant Backward Caste Lingayath Vokkaliga Bunt 267 246 42 18.45 17.00 2.90
Other Backward Caste Kumbara 77 5.32
Forward Caste Brahmin 97 6.70
44Weaknesses of the experimental design
- As in Mali, estimated treatment effects may not
be large, for a number of reasons - Stimulus is somewhat artificial?
- May not prime caste effectively?
- Issues with surnames
- In the field experiment, we may therefore expect
to estimate lower-bounds on the effects of caste - However, the design is likely to give us a good
sense of how reservation shifts the relative
salience of caste categories
45Chart gives average answers to the question, On
a scale of 1 to 7, how much does this speech make
you want to vote for (name of candidate)? Jati
refers to Vokkaliga, Lingayath, Holaya, Madiga,
etc. Caste category refers to SC, ST, OBC, and
so on.
46The relevance of sub-caste
- The effects of sharing a sub-caste on candidate
evaluations are statistically significant,
relative to the other two treatments - In the experimental population as a whole,
politicians from a different sub-caste, but the
same caste category, are statistically
indistinguishable from politicians from a
different caste category - This finding holds for sub-groups, in particular,
for both Scheduled Castes and dominant Backward
Castes - The estimated effect of treatment assignment is
about one-quarter of one standard deviation - This is in the neighborhood of, but somewhat
smaller than, the estimated effects of
co-ethnicity in an experiment in Mali
47Treatment effects, in sum
- Many variables may help explain the preference
for politicians from the same sub-caste, relative
to different caste categories - On the other hand, only the benefits variable
significantly distinguishes among sub-castes,
within the same caste category - Politicians from another sub-caste but the same
category are significantly preferred to
politicians from a different caste category only
on credibility grounds - Does reservation of the council presidency shape
the relative salience of different caste
relationships?
48Chart gives average answers to the question, On
a scale of 1 to 7, how much does this speech make
you want to vote for (name of candidate)?
Reserved panchayat means, reserved for SC or ST
adhyaksha. Unreserved panchayat means,
reserved for OBC or General category.
49The variable measuring credibility combines
survey questions about the post-election
performance of the politician whether he is
trustworthy, has good motives for running for
office, could face the challenges of office,
would do a good job if elected, and would fight
for others and defend his ideals once in office.
The variable is scaled to run from 0 to 1.
50The variable measuring benefits combines
answers to the following two survey questions
If (name of the politician) were elected, people
like you would receive more benefits from the
welfare schemes of the government and If (name
of the politician) were elected, people like you
would have a better chance of getting a job with
the government. The variable is scaled to run
from 0 to 1.
51The variable measuring affection combines
survey questions about the intelligence,
likeability, competence, and impressiveness of
the politician. The variable is scaled to run
from 0 to 1.
52The causal effects of reservation
(1 same jati, same caste category 2
different jatis, same caste category 3
different caste categories)
Estimated effect, reserved panchayats (A) (t-statistic) Estimated effect, unreserved panchayats (B) (t-statistic) The causal effect of reservation (A-B) (t-statistic)
Vote preference (1-2) 0.20 (1.45) 0.23 (1.77) -0.02 (-0.11)
Vote preference (1-3) 0.31 (2.30) 0.12 (0.91) 0.19 (1.03)
Vote preference (2-3) 0.10 (0.77) -0.11 (-0.88) 0.21 (1.18)
Affection (1-2) 0.03 (1.50) -0.00 (-0.06) 0.03 (1.15)
Affection (1-3) 0.06 (4.07) -0.00 (-0.06) 0.06 (2.93)
Affection (2-3) 0.04 (2.48) -0.00 (-0.00) 0.04 (1.79)
Credibility (1-2) 0.03 (1.82) 0.01 (0.62) 0.02 (0.88)
Credibility (1-3) 0.06 (3.63) 0.03 (1.93) 0.029 (1.28)
Credibility (2-3) 0.03 (1.76) 0.02 (1.30) 0.01 (0.05)
Estimated effect, reserved panchayats (A) (t-statistic) Estimated effect, unreserved panchayats (B) (t-statistic) The causal effect of reservation (A-B) (t-statistic)
Monitoring (1-2) 0.02 (1.13) 0.01 (0.46) 0.01 (0.43)
Monitoring (1-3) 0.03 (1.30) 0.01 (0.25) 0.02 (0.70)
Monitoring (2-3) 0.00 (0.14) -0.01 (-0.23) 0.01 (0.27)
Preferences (1-2) 0.01 (1.12) 0.02 (0.85) 0.01 (0.23)
Preferences (1-3) 0.07 (3.28) 0.02 (0.72) 0.06 (1.94)
Preferences (2-3) 0.05 (2.29) -0.00 (-0.18) 0.05 (1.79)
Benefits (1-2) 0.05 (2.17) 0.02 (1.11) 0.03 (0.79)
Benefits (1-3) 0.08 (3.72) 0.03 (1.49) 0.05 (1.63)
Benefits (2-3) 0.01 (0.32) 0.03 (1.48) 0.02 (0.81)
53The effects of reservation, in sum
- Reservation of the council presidency seems to
make caste matter more - In the sub-group analysis, all (save one) of the
statistically significant effects occur only in
reserved panchayats - Reservation shifts the relative importance of
caste categories, making the larger caste
grouping relatively more important - Reservation intensifies distinctions between
politicians from ones own larger caste category
and politicians from a different caste category. - It sharpens distinctions between politicians who
are from a different sub-caste, but from the same
caste category, and politicians from a different
caste category. - Reservation also blurs distinctions between
politicians from the same larger group category
(whether or not they come from the subjects own
jati), with one exception - In general, reservation most strongly shapes
measures of affection, as well as perceptions
in-group preferences
54Interpreting the results
- Evidence on the special importance of affective
factors contrasts with many studies. - The politics of dignity?
- Another reason reservations distributive
effects may in fact be more limited than previous
evidence suggests (Dunning and Nilekani 2009) - In other work, I try to explain the limited
distributive effects patterns of party
competition at the local level seem important - What about external validity?
55External Validity Comparing Means in the Study
Group and the State of Karnataka
Study Group State of Karnataka
Mean population (Standard Deviation) 5869.7 (1912.03) 6132.1 (2287.1)
Mean SC population (Standard Deviation) 1116.7 (805.7) 1129.7 (760.2)
Mean ST population (Standard Deviation) 475.2 (506.5) 512.5 (715.8)
Mean number of literates (Standard Deviation) 3196.1 (1133.4) 3122.7 (1326.7)
Mean number of workers (Standard Deviation) 2938.9 (979.3) 3005.9 (1092.5)
Number of Panchayats 200 5760
56Practical considerations
- Are studies like these feasible for doctoral
dissertations? - Yes!
- Researchers spending extended time in the field
have a logistical advantage - Conducting interviews oneself or with RAs, over
an extended period of time, reduces costs - Great opportunities to merge qualitative
fieldwork with field or natural experimental
designs - Other considerations