Migration and Economic Mobility in Tanzania:
Evidence from a Tracking Survey
Kathleen Beegle World Bank
Co-authors Joachim De Weerdt, E.D.I. Tanzania Stefan Dercon, Oxford University
January 2008
1
Background
Much economic analysis of the processes of development and poverty is about the longrun.
Evidence on long-term poverty dynamics remains limited to cross-sectional work, less with panel data:
Few long-term panel data sets; Poor analysis of the evidence, usually only focusing on correlates and descriptives; Panel data sets suffer from high attrition.
2
Background
Attrition strongly related to „rules‟ e.g. LSMS “Blue book” manual suggests interviewing people in same dwelling; most panels go only back to original villages or communities.
BUT
Life-cycle events (death, marriage, etc) make definition of „household‟ not stable over time. „Development‟ usually involves spatial movement (e.g out of agriculture, but also out of village)
....does not sound like random attrition.
3
Overview of this study
Analysis of consumption growth and poverty changes among households from 1991-2004 Households from Kagera, a region near Lake Victoria Drawing on a unique panel data set, involving tracking of all individuals ever interviewed With much attention to finding back everybody wherever they went.
4
Findings
Substantial consumption growth and poverty declines in this period Extent depends on spatial movement involved, justifying ‘tracking’ of movers
Controlling for initial household fixed effects, we find a large impact of physical movement out of the community Results remain surprisingly stable in the 2SLS estimation.
5
KHDS 1991-1994
Kagera Health and Development Survey
households, across Kagera region 4 rounds between 1991/94 Stratified random sample
900
www.worldbank.org/lsms
7
KHDS 2004
Re-interviewing all baseline respondents
Age at baseline 1991-1994 <10 years 10-19 20-39 40-59 60+ Total N (alive end of baseline) 2,081 1,922 1,300 618 434 6,355
8
KHDS 2004
Goal to re-interview all respondents Consistent quantitative survey instruments
www.edi-africa.com
9
KHDS 2004
26 Household members for one panel respondent.
KHDS 2004 results
93% of the baseline households were reinterviewed; 96% of those in 1994. 82% of surviving individuals re-interviewed (above 90 percent for those age 20+ at base).
Individuals found back: 4,432 Individuals death: 962 Individuals not traced:
961 New sample: Living in 2,719 households
11
Tracking households...
912 Original Households 63 Untraced*
832 Recontacted
17 Deceased
2,774 New Households interviewed
12
2,719
households
49% Stayed in the same village
19% Moved to a village nearby the origina l one
20% Moved to another village in Kagera Region, not nearby original village
10% Live in countr y outsid e Kager a Regio n
2% Live outside country : Uganda
13
Location of surviving respondents
re-interviewed not traced Same Village 2797 Nearby Village 626 elsewhere Kagera 636 somewhere Kagera 545 elsewhere Tanzania 314 294 Other Country 59 53 Don't Know 70 TOTAL 4432 961
14
Consumption and Poverty Dynamics
consumption expenditures
Challenge to convert into real (2004) value “narrow” definition to ensure comparability Consumption of household to which individual belongs in each period
Monetary measure of poverty
Poverty line to match poverty levels for those left in Kagera to estimates from HBS for 2001/02 for Kagera (29%)
15
Consumption per capita in KHDS sample (in TSh)
2004 location
within village
mean 1991 155,641
166,565
mean 2004 186,479
230,807
difference means 30,838***
64,242***
N
2611
566
nearby village
elsewhere in Kagera
162,116
262,964
100,848***
571
out of Kagera
Full Sample
169,994
159,217
457,475
225,099
287,480***
65,882***
327
4075
16
Poverty in KHDS sample (in TSh)
2004 location
within village nearby village elsewhere in Kagera
mean 1991 0.36
0.33
mean 2004 0.32
0.22
difference means 0.04***
0.11***
N
2611
566
0.37
0.30
0.24
0.07
0.13***
0.23***
571
327
out of Kagera
Full Sample
0.35
0.27
0.08***
4075
17
Cumulative Density Functions of Consumption per Capita
.8 0
0
.2
.4
.6
1
100000
200000 conspc
300000
400000
500000
1991 2004 moved within Kagera
2004 stayed in village 2004 moved out of Kagera
Consumption growth by move to more/less remote area
Mean Median N 0.13 0.16 2,150 0.52 0.25 0.42 0.86 0.49 0.19 0.34 0.83 1,088 408 417 338
Did not move Move out of community Out of those that moved out of community: Move to more remote area Move to similar area Move to less remote area
19
Consumption growth by move and sectoral change
Stayed in Community 0.18 (1,251) 0.42 (201) 0.12 (88) -0.12 (157) 0.18 (1,697) Moved out of Community 0.28 (477) 1.04 (207) 0.87 (85) -0.00 (88) 0.49 (857) All 0.22 (1,728) 0.67 (408) 0.43 (173) -0.03 (254) 0.27 (2,554)
Stay in Agriculture Move out of Agriculture into NonAgriculture Stay in Non-Agriculture Move into Agriculture from NonAgriculture Total
20
Preliminary conclusions
Moving out of poverty is correlated with moving out of the village. Sampling only those that remain in the village is bound to affect inference. However: is migrating itself a the way out of poverty? Not clear. It could be that a particular characteristic both affects moving out and moving out of poverty…
21
Regression analysis
Δln Cit+1,t = α + βMi + γXit + δih +εit
Explain consumption growth based on initial characteristics (individual, household, community).
Resolves time-invariant sources of endogeneity (risk aversion?, ability) Further Address household effects (δih) using “initial household
Consider moving as endogenous.. The search of IVs
22
FE” (832 to 2719 households) Controlling for individual level factors for ( Xit)
Instrumenting strategy
Migration pull factors
Being a male, age 5-15 at baseline interacted with distance to regional capital Being age 5-15 at baseline * rainfall deviation between rounds Relational and positional variables in the HH
Migration push factors
Social relationships within the household
Age rank * age 5-15, male/female child of head, spouse or head
23
Table 10: Consumption Change & Mobility
Moved outside community Kms moved (log of distance)
(1) IHHFE 0.363*** (0.025)
(2) IHHFE
(3) 2SLS 0.372** (0.151)
(4) 2SLS
0.121*** (0.006)
0.104** (0.043)
24
Instrumenting strategy
tests validity of instruments F-stat of instruments
11.70 for movement 9.07 for distance of move weak instrument problem once we try finer distinctions in moving out.
CDF of baseline PCE for movers and non-movers overlap: suggesting either that omitted variable bias is small or biases “balance out” (highly able leave, less able leave)
25
Table 11: 1st Stage results
(1) Moved Baseline covariates: excluded instruments Head or spouse Child of head Male child of head Age rank in HH * age 5-15 Km from reg. capital * male * age 5-15 Average rainfall deviation * age 5-15 -0.218*** (0.038) -0.097*** (0.032) -0.115*** (0.037) 12.383 (8.008) -0.001*** (0.000) 0.000** (0.000) (2) Distance moved -0.635*** (0.147) -0.417*** (0.123) -0.338** (0.144) 58.015* (30.894) -0.002** (0.001) 0.001** (0.000)
26
Table 12: Consumption Change & Characteristics of the Move
(1) IHHFE Characteristics of the move Move to more remote area Move to similar area Move to more connected area Km moved Distance moved if to similar area Distance moved if to more connected area 0.177*** (0.036) 0.097** (0.044) 0.485*** (0.047) 0.073*** (0.011) 0.033** (0.015) 0.070*** (0.013)
27
(2) IHHFE
Other findings
Moving out of agriculture associated with higher growth Strong additional effect from migration along with this sectoral move Table 10 consistent with adult equivalent consumption (v. per cap)
28
Conclusions
Strong consumption growth and poverty declines overall Moving out of the village is strongly correlated with consumption growth Education and individual characteristics matter for moving out and for growth
29
Conclusions
IHHFE results show large gains to consumption for movers. Migration is linked with a 37 percent higher growth compared to those that stayed in the same community 2SLS results are similar suggesting that relevant sources of heterogeneity are controlled for using the initial household fixed effects and individual controls from baseline.
30
Conclusions
Gains are highest for movers to more connected areas, but also higher for those moving to more-remote areas. Without tracking We could never have identified this. Consumption growth would have been understated.
31
Reasons for moving from original homestead, by location in 2004
Found work To look for work Posted on a job Looking for land Schooling Marriage Divorce Parents died To care for a sick person To seek medical treatment Following inheritance Other family problems Follow parents Follow spouse Follow relatives New house Other (specify) Unanswered Missing Total Same village 5 10 0 76 8 159 11 15 3 0 52 62 29 5 3 20 49 5 1,638 2,150 Nearby village 10 16 4 29 16 180 11 13 1 2 21 29 17 1 2 5 22 2 19 400 Within Kagera 27 59 3 43 23 140 11 12 3 4 16 26 31 1 12 0 21 2 3 437 Outside Kagera 15 64 9 10 36 37 3 5 0 3 1 12 22 4 11 0 15 3 1 251 Total 57 149 16 158 83 516 36 45 7 9 90 129 99 11 28 25 107 12 1,661 3,238 Same Nearby village village 1.0 2.6 2.0 4.2 0.0 1.1 14.8 7.6 1.6 4.2 31.1 47.2 2.2 2.9 2.9 3.4 0.6 0.3 0 0.5 10.2 5.5 12.1 7.6 5.7 4.5 1.0 0.3 0.6 0.5 3.9 1.3 9.6 5.8 1.0 0.5 --100 100 Within Kagera 6.2 13.6 0.7 9.9 5.3 32.3 2.5 2.8 0.7 0.9 3.7 6.0 7.1 0.2 2.8 0.0 4.8 0.5 -100 Outside Kagera 6.0 25.6 3.6 4.0 14.4 14.8 1.2 2.0 0.0 1.2 0.4 4.8 8.8 1.6 4.4 0.0 6.0 1.2 -100 Total 3.6 9.5 1.0 10.0 5.3 32.7 2.3 2.9 0.4 0.6 5.7 8.2 6.3 0.7 1.8 1.6 6.8 0.8 -100