2. Motivation
• Specific goal: evaluating the impact of a rural
livelihood project on women empowerment
• General goal: is showing how this method can
be used in measuring empowerment
• Existing approaches are costly and are biased
by subjective perceptions
3. Empowerment is multidimensional
• Empowerment is a multidimensional concept:
Solava and Alkire (2007) found 29 different
definitions in the literature
• Indices have tried to capture
multidimensionality
• This has resulted in long survey questionnaires
4. Example: World Bank Empowerment
Index
• 8 domains:
• Justice
• Politics
• Public services
• Labour
• Goods
• Private services
• Intra-household
• Intra-community
• 164 questions
5. Example: Women’s Empowerment in
Agriculture Index
• 6 domains:
• Production
• Resources
• Income
• Leadership
• Time
• 60 survey questions
6. Subjective perceptions are not
comparable
• People interpret questions in different ways
and meanings vary with the context
• Perceptions can be culturally determined and
being ‘false’
• Perceptions are not comparable across
people, groups or countries
7. Examples of subjective questions
• World Bank Index: “how much influence do
you think you have when the community
selects its leaders?”
• Empowerment in Agriculture index: “To what
extent do you feel you can make your own
personal decisions regarding these aspects of
household life if you wanted to”
8. Andhra Pradesh Rural Livelihood
Project
• Goal of reducing
poverty by building
social capital and
empowering women
• Operates through Self-
Help Groups to
• Channelling funds
• microfinance
9. Women’s empowerment vignette
1. How much freedom/opportunity do you have?
VIGNETTES
1. Neelamma takes a loan from the SHG and starts a grocery
shop. Despite his initial opposition, her husband is won
round to the idea when he sees the good returns from the
business.
2. Manemma takes a loan from the SHG to start a small home
business. But her husband argues with her that the money is
needed for land improvement, and she reluctantly parts
with the loan money.
3. Chandamma wants to take a loan from the SHG of which she
is a member to start a tea stall. She could not take the loan
because her husband and mother-in-law did not agree.
10. Men’s vignette
1. How much influence do you have in your village?
VIGNETTES
1. Davender is in a dispute over some land he claims title to. The
other man claiming this land is a close friend of the Sarpanch,
and, lacking connections himself, Davender has little hope of
winning the dispute.
2. Vinod is in a dispute over some land he claims title to. He
supported the Sarpanch in the last election, bringing him many
votes from his family and families of those who work for him. He
is optimistic that he will win the dispute.
3. Kiran is in a dispute over some land he claims title to. Both he
and the other man in the dispute have some blood relations with
leading men in the village. He has some hope that he may win the
dispute but realizes it may go either way.
11. Goal of vignettes
• Find systematic differences in reporting across
groups
• Comparing self-reported assessments after
purging them of reporting bias
• In the analysis use a two step procedure
(HOPIT model)
12. The Hopit model
• 2-part model (Tandon et al 2003; King et al.
2004)
1.Model reporting behaviour. For example,
women of different caste may rate vignettes
in different ways
2.Model self-assessments using cut-off points
identified in part 1
• Covariates are normally the same in the two
parts
13. Estimation
• 2-step: ordered probit of vignettes and
interval regression using estimated cut-off
points
• Simultaneous estimation by maximum
likelihood. Methods:
Programme in R
Use GLLAMM (generalised linear latent and mixed
model) in STATA
Set the likelihood function in STATA
14. Examples from the literature
• Compare political efficacy in Mexico and China
• Compare reported work disability in US and
the Netherlands
• Compare job-satisfaction across EU countries
• Reported health across demographic groups in
Indonesia, China and India
• First study to compare groups with and
without a project
15. Women’s empowerment and
literacy
0 .1 .2 .3 .4
Women's power by husband's literacy
literate illiterate
no power little power
some power a lot of power
16. Men’s influence and literacy
0 .1 .2 .3 .4 .5
Men's power by literacy
literate illiterate
no power little power
some power a lot of power
17. Women’s empowerment and wage
work
0 .1 .2 .3 .4
Women's power by wage earning status
no-wage wage earner
no power little power
some power a lot of power
18. Men’s empowerment and wage
work
0 .1 .2 .3 .4 .5
Men's power by wage earning status
no-wage wage earner
no power little power
some power a lot of power
19. Women’s empowerment and caste
0 .1 .2 .3 .4
Women's power by caste
SC ST BC OC
no power little power
some power a lot of power
20. Men’s empowerment and caste
0 .2 .4 .6
Men's power by caste
SC ST BC OC
no power little power
some power a lot of power
21. Empowerment and SHG
0 .1 .2 .3 .4
Women's power by SHG membership
no SHG SHG
no power little power
some power a lot of power
22. Ordered probit Hopit
SHG member 0.081* 0.037*
(0.099) (0.712)
Scheduled tribe 0.058 -0.173
(0.499) (0.502)
Other backward caste 0.076* -0.054
(0.225) (0.675)
Other caste 0.041 0.091
(0.622) (0.607)
Age 0.020 0.022
(0.123) (0.354)
Age square -0.001 -0.001
(0.296) (0.321)
Illiterate 0.052 0.202
(0.457) (0.165)
Husband’s education 0.083*** 0.012
(0.000) (0.749)
Female headed household 0.237** 0.140
(0.003) (0.307)
Household size -0.019 -0.052*
(0.134) (0.060)
Land size 0.005 0.001
(0.439) (0.993)
Agricultural labourer 0.157** 0.099
(0.004) (0.387)
Observations 1,431 1,431
23. Adjusted and unadjusted women self-assessments
• Before adjusting empowerment is correlated
with:
• SHG membership
• Husband literacy
• Female headed household
• Earning capacity
• After adjusting none of the variables is correlated
with empowerment except SHG
• Most differences found are perceptual rather
than real
24. Men’s empowerment
• No expected project impact on men’s
empowerment
• Influence correlated with caste, literacy,
household size, wage work
• Correlations hold after adjustments of
perception bias
• Exceptions are: higher caste understate their
influence while OBC overstate their influence
25. Test of homogeneity in responses
Female respondents Male respondents
Homogeneity test Full sample Full sample
All covariates 0.002** 0.379
SHG member 0.096* 0.387
Caste 0.568 0.162
Age 0.514 0.494
Illiterate 0.040** 0.100
Husband’s education 0.041**
Female headed household 0.000***
Household size 0.452 0.343
Land size 0.253 0.826
Agricultural labourer 0.835 0.232
26. Impact of SHG on women’s
empowerment
• The data do not contain a valid control group
• We follow two approaches:
• Instrumented variables
• Difference-in-differences
• In both cases, the impact of SHG on
empowerment vanishes
27. Additional question: non-parametric
approach
5. How would you compare your
freedom/opportunity with that of the
women described above? (less than
Chandamma, same as Chandamma, more
than Chandamma but less than Manemma,
same as Manemma, more than Manemma
but less than Neelamma, same as Neelamma,
more than Neelamma)
28. SHGs and empowerment
Coefficient Standard
error
observations
SHG instrumented hopit
model full sample
0.238 0.844 1,431
SHG instrumented hopit
model restricted sample
0.199 0.794 236
Difference-in-difference
restricted sample
0.202 0.313 232
29. Conclusions
• Much of the differences in empowerment
observed in our Andhra Pradesh sample are
perceptual rather than real
• Empowerment and SHG are correlated but it
appears that causality runs from
empowerment to SHG rather than the other
way
• Vignettes are a simple, unbiased and powerful
tool to assess empowerment programmes