Migration and Development: A few new facts

Migration, Risk-Sharing and
Subjective Well-being
Some evidence from India 1975-2005
Stefan Dercon, University of Oxford
Pramila Krishnan, Cambridge University
Sonya Krutikova, Oxford University
ICRISAT, India
• 6 villages, semi-arid tropics in Maharasthra
and Andhra Pradesh (3 districts:
Mahbubnagar, Sholapur and Akola)
• Villages extensively studied, longitudinal data
1975-84
• 2005/6 and 2006/07 resurvey of all
households in village plus migrants 2005/06
Purpose
• Briefly report on changes within villages 19752005
• Focus on migration from villages
The data….
Status by 2005
(1)
Full sample of
individuals
included in
1975-1984
(2)
Of which:
Included in
2005
surveys
Numbers
Dead in 2005
Temporarily migrated 2005
Permanently migrated 2005
In village in 2005
No information in 2005
Total
449
111
724
823
23
2,130
NA
40
454
823
NA
1,317
(3)
(4)
Population
Attrition
composition
(by
2005 (based on percentage
points in
(1))
total)
2005
Shares (%)
21%
5%
34%
39%
1%
100%
21%
3%
13%
0%
1%
38%
Overview of Changes in Villages
• All deflated by rural CPIAL
• Quick overview of
– Land and assets
– Consumption
– Income sources
• All suggesting considerable growth VLS1 to
2005
Per capita levels
Implements
(Rps)
Durables
(Rps)
Land Area
Owned (ha)
in 1975 prices
Full sample
vls1
2005
1975- 79
2005
148
754
90
555
Growth (annualized) %
6.5
7.2
1975-83 2005
0.63
0.51
-1.0
Large initial landholding
274
1543
122
815
1.20
0.76
Medium initial landholding
185
526
99
584
0.62
0.59
Small initial landholding
52
304
71
354
0.36
0.32
Initial status: labourers
14
389
46
357
0.07
0.23
772
772
772
772
772
772
Number of Observations
Other changes
• Substantial income and consumption growth per
capita (4% per capita annualised for consumption)
• More than doubling in consumption per capita, with
larger growth in non-food
• Food share down, cereal and pulses share down (69
to 43%), animal protein up (12 to 23%)
• Growth across land distribution groups
• Poverty down from 78% to 18%; landless labourers
down to 28%
Structure of incomes
Shares of Mean Income per Capita
AGGREGATE
LARGE
MEDIUM
SMALL
LABOUR
vls1
0104
vls1
0104
vls1
0104
vls1
0104
vls1
0104
On farm
income (crop 0.64
and livestock)
0.29
0.88
0.33
0.60
0.34
0.48
0.20
0.31
0.27
Labour
income
0.29
0.19
0.10
0.11
0.30
0.11
0.40
0.29
0.68
0.40
0.02
0.05
0.01
0.05
0.01
0.06
0.03
0.03
0.03
0.04
0.05
0.47
0.02
0.51
0.08
0.49
0.10
0.48
0.01
0.30
Transfer
Trade and
Business
Self-Assessed Welfare Positions (2005)
2005
15 years ago
30 years ago
Very Rich (%)
0
0
0
Rich (%)
2
1
2
Comfortable (%)
35
23
15
Manage to get by (%)
45
34
24
Never enough (%)
15
36
40
Poor (%)
2
6
16
Very Poor (%)
0
0
4
Conclusion
• Considerable changes in village living
standards and assets
• Consumption poverty and self-assessed
poverty down
• Big changes in income sources
Conclusion (2)
• Regression consumption growth (recall, doubled =
increased by 100%+ on average)
Strong correlates (with economic significant size)
• those from literate households 30% more growth
• Those educated themselves up to end high school
+17%
• High dependence on crop income in VLS1, doing
worse
• Lower caste groups (SC/ST/some BC) -10 to -20%
So what about Migrants?
• Development correlated with internal migration
– Out of agriculture
– Out of rural areas
“physical mobility, economic mobility, social mobility
all related”
• Scale required is massive:
– E.g. China: last 20 years, from 80% to 55% in
agriculture, much of it involving local or longdistance migration
The data….
Status by 2005
(1)
Full sample of
individuals
included in
1975-1984
(2)
Of which:
Included in
2005
surveys
Numbers
Dead in 2005
Temporarily migrated 2005
Permanently migrated 2005
In village in 2005
No information in 2005
Total
449
111
724
823
23
2,130
NA
40
454
823
NA
1,317
(3)
(4)
Population
Attrition
composition
(by
2005 (based on percentage
points in
(1))
total)
2005
Shares (%)
21%
5%
34%
39%
1%
100%
21%
3%
13%
0%
1%
38%
Destinations of migration
Permanent
migrants
Male
Female
Location
Nearby village
21%
10%
27%
Other village (this district)
16%
11%
19%
Other rural areas
22%
16%
26%
Urban areas
39%
61%
27%
Don’t know/missing
2%
2%
2%
Reasons for migration
Permanent migrants
Male
Female
Work
19%
49%
3%
Looking for work
8%
19%
1%
School/college
2%
4%
1%
Following family
17%
18%
17%
Marriage
49%
4%
74%
Other
2%
0%
2%
Don’t know missing
4%
7%
2%
Views on migration and inequality
On evidence
• Perception of slum living, low wages, high
unemployment paints bleak picture of urban
living
• Evidence from poverty measurement suggests
much higher rural than urban poverty
Views on migration and inequality
On theory:
(a) Labour market theories
• Inequality ‘drives’ migration but outcome is
equilibrium – so why higher rural poverty?
• Inequality drives migration without resolving it (HT)
(b) Household models
• Migration is strategic family decision (NEM)
• with risk-sharing and remittances as one of its
reflections – so strong prediction on intra-household
inequality (not growing) (RS)
The questions
(1) Is there a migration premium?
(2) Is it consistent with standard theory models?
From long-term longitudinal data tracking all
within families, data of up 30 years...
• Evidence:
– of relatively large migration, large “returns” to
migration, including for female migrants
– with a twist on the theory (  or  )
Empirical challenge
• Wages for urban and rural hard to compare
(differentiated labour markets in skills, tasks, etc)
• We need to ensure we have counterfactual: living
standards if migrant had not migrated
– Migrants could be from better families
– M could be those with higher earnings potential
• Setting up via ‘family (risk) sharing model’ as it offers
means of both exploiting data and theory predictions
• Focusing on consumption and subjective well being
(“net of remittances”)
Model
• Suppose we have an extended family group that is in
involved in perfect (risk) sharing. Let us characterize
the outcome and then use this as a basis for testing
deviations from this.
• Let there be (different) (risky) income streams yi for
each household i in a group. (Suppose there is no
savings.)
• Suppose now that these households contract with
each other to get optimal (risk) sharing, and
assuming that the contract is enforceable (binding
sharing rule).
model (2)
“Overidentification” by location: if sharing, location
should not matter, or β=0
Taking to data...
• Model can be used for risk-sharing, but test
nests more general ‘premium’ test
β=0 tests sharing, irrespective of location
But also test for presence of migrant premium, ceteris
paribus, as if in a difference-in-difference framework
Empirical application?
• Following Beegle, Dercon, De Weerdt, RESTAT
2011 on Tanzania
– Initial household fixed effects estimator
– With further IV for time varying individual
heterogeneity
Assessing the impact of migration m
ciJt   .mit   .X it 1   J   iJt
•
Changes in consumption, not levels (in real terms)
= control for time-invariant factors that determine levels
(diff-in-diff)
• Initial household fixed effects, to compare the impact of
migration between family members initially living together
(γj )
= control for all factors that determine changes common to
all those initially living together (“triple difference”)
Specification
•- Individual baseline characteristics (Xt-1 )
= control for all observable individual (time-varying and
time-invariant) factors that determine changes
=individual baseline characteristics: age, sex, education
baseline, caste, family educational and wealth
background, family composition at baseline, nutrition at
baseline.
• One step further: individual level IV
= control for unobservables at individual level
determining changes
Returns to migration….
Moved
outside
community
(1)
OLS
all
(2)
IHHFE
all
(3)
OLS
men
(4)
IHHFE
men
(5)
OLS
Women
(6)
IHHFE
Women
0.217
0.205
0.268
0.287
0.183
0.164
Always significant at 1%
Controls for sex, caste, age, schooling, shocks 1984-2005, living conditions at baseline
Specification IV
-Instruments
= control for unobservables at individual level
determining changes
= predictors of migration, not directly determining ‘incomes’
= predictors explaining why member x went and not member y
= relational variables (birth order) plus push factor
interacted with age window at baseline: rainfall at the age
of 16
First stage, strongly significant, Cragg-Donald 9.42
Results: 0.67 for men, 0.65 for women (sign 1%)
Answers
• Is there a premium to migration? (HT): YES
• Is this premium fully exploited? NO
• Are families smoothing over space? (RS): NO
But not a simple story of educational investment
(life-cycle), sectoral, urban-rural shift...
Intra-Family Inequality after migration
High premium ‘unexploited’
• So Why Undermigration? Theory just wrong?
Are we getting the point?
• They are not ‘sharing’ in space? But what if ‘location’
matters per se? Location as a taste shifter?
Are we getting the point?
•
•
•
•
For example: “urban needs”
As in “keeping up with the Jones’ consumption ”
Are they ‘sharing’ in this space?
If θ(location), then finding migration effect could be
consistent with risk-sharing
• Can we test?
– Do we have data closer to bist cist γ, and not just cist?
– Possibly via subjective wellbeing data!
– We would expect that this ‘controls’ for taste shifter better,
so no more migration effect.
Assessing the impact of migration m
• we have data on changes in perceived wealth
• we also have data on levels of happiness, life
evaluation, etc.
Subjective assessment of wealth
Household living in now
(2005)
All
Migrants
Nonmigrants
Household lived in 30
years ago
All
Destitute/Poor/Never
enough
19.9
51.5
Can manage to get by
41.0
25.3
Comfortable
35.4
18.5
Rich/very rich
3.7
4.8
100
100
1,158
1,158
Total Observations
Nonmigrants
Migrants
Subjective assessment of wealth
(percentages)
Household living in now
(2005)
All
Migrants
Nonmigrants
Household lived in 30
years ago
All
Nonmigrants
Migrants
Destitute/Poor/ never
enough
19.9
17.2
24.9
51.5
58.7
37.5
Can manage to get by
41.0
46.1
31.2
25.3
24.3
27.2
Comfortable
35.4
35.0
36.3
18.5
15.1
24.9
Rich/very rich
3.7
1.7
7.6
4.8
1.8
10.3
100
100
100
100
100
100
1,158
761
397
1,158
761
397
Total Observations
Nostalgia bias?
• Results may be affected by recall.
• Can we use cross-section? Needs strong assumption
on observability of pareto weight
Nostalgia bias?
• Alternatively: when living together, no compensation
for subjective well-being. We treat is as if we were all
in initial household at similar subjective wellbeing
(and so in fixed effect)
Perceived wealth and happiness IHHFE
Migrant IHHFE
Migrant IHHFE
With IV
(1)
(2)
(3)
(4)
(5)
Consumption
growth
Levels of
perceived
wealth
Changes in
perceived
wealth
Happiness
Ladder of
life
0.205***
0.078
-0.188
-0.051
-0.235
0.785**
0.012
-0.059
-0.222
-0.295
Interpretation
• OVERALL consistent with sharing!!!
• Migration lowers subjective well being (how one assess
own wealth)
=Consistent with subjective well-being =relative concept
=Could reflect more difficult conditions (being outsider,...)
=could reflect ‘relative’ comparison but also huge
nostalgia effect
• As a migrant, your initial family ‘allows’ you to have a
huge consumption premium, to compensate you for your
miserable existence (taste shifter)
• Consistent with literature on subjective wellbeing as
relative experience
Overall conclusion
• Families may allow inequality to emerge as
part of ‘sharing’ strategy
• HERE: with higher material wellbeing to
compensate for otherwise lower overall or
subjective wellbeing
• Still: UNDERmigration in terms of material
wellbeing (given seemingly high returns)
• Policy?