Transcript Slide 1
IW Outes-Leon
Nutrition and Growth in Rural Ethiopia
BREAD Summer School 2008
Ingo W. Outes-Leon, Oxford University BREAD 2008 - 1
Overview A. Introduction B. Data and Empirical Model C. Results D. Conclusions
IW Outes-Leon BREAD 2008 - 2
IW Outes-Leon
A.1 Introduction – Aim and Motivation
Test whether poor nutrition and health has a negative effect on a HHs ability to generate future consumption We estimate effect of adult low BMI on HH consumption growth, after controlling for other HH assets.
Necessary, but not sufficient, condition for nutritional poverty traps to exist.
Combine
Nutritional Poverty Trap (Dasgupta and Ray) - Productivity Effect
Morbidity theories (Deaton (2005) and Fogel (1992)) - Health Effect
With Micro-Growth models
(Jalan and Ravallion (2003), Antman and McKenzie (2005) and Dercon and Shapiro (2007))
And literature on Non-linearity in BMI
(Dasgupta (1993) and Strauss and Thomas (1998))
BREAD 2008 - 3
IW Outes-Leon
A.2 Introduction – Aim and Motivation
Take inspiration from Mankiw, Romer and Weil (1992), estimate:
HH consumption growth (1995 to 2004) on Lagged dependent variable And HH Assets and Other characteristics on baseline (1995) Adult Low BMI enters equation as further component of HH human capital
That is, Conditional convergence type of model.
So no actual test of poverty traps
But able to indicate if low BMI has a drag-down effect on HH growth
BREAD 2008 - 4
A.3 Introduction – Contribution and Challenges Findings and Contribution:
IV methods provide evidence of negative Growth Effect of low BMI
Evidence of Persistence of the 1984 Drought on 1995 adult BMI Application of ‘weak’ IV estimation and inference methods.
Estimation: IV Fuller Estimator Inferences: Conditional Likelihood Ratio (CLR) Moreira p-values IW Outes-Leon
Challenges:
Validity of Instrumentation Methodology
Doubtful, but IV estimates still interesting Disentangle village-specific nutritional effects from individual HH effects
Unobserved Village Effects vs local nutritional poverty traps (e.g. shocks and HH infrastructure)
Quartile regressions might provide further insight.
Treatment of Other Assets, especially Livestock BREAD 2008 - 5
IW Outes-Leon
B.1 ERHS Data – Is low BMI bad for growth?
ERHS: 1470 HHs in 15 villages in rural Ethiopia; 1994 to 2004 period; High poverty: 35% poor HHs in 1995 (Dercon and Krishnan (2003)) Incidence of Adult Underweight: 21% of HH Heads (<18.5) 1995-2004 Growth pa: 4.3% (all HHs) ; 6.0% (Low BMI HH Heads).
.4
.2
0 .3
.1
-.1
Kernel smoothing: No apparent Poverty Trap Quartile Kernel smoothing: ‘low BMI’ effect for poor HHS ?
-.2
15 15.5
16 16.5
17 17.5
18 18.5
19 19.5
20.5
21 21.5
20 Body Mass Index, 1995 22 22.5
23 23.5
24 24.5
25 .15
.125
.1
.075
Quartile 1 Quartile 2 Quartile 3 Quartile 4 .05
.025
0 -.025
-.05
-.075
-.1
15 15.5
16 16.5
17 17.5
18 18.5
19 19.5
20.5
21 21.5
20 Body Mass Index, 1995 22 22.5
23 23.5
24 24.5
25
BREAD 2008 - 6
IW Outes-Leon
B.2 Empirical Model
(1)
(ln
y it
ln
y
) /
p
ln
y
X
1
W
2
S it
3
v it
LHS: Food Consumption Growth (1995 to 2004), per annum RHS: 1995 Baseline HH characteristics:
Lagged Dependent Var: Log Consumption 1995 Life-cycle controls – cohort dummies HH Assets: low BMI, Livestock, Land, Education, HH structure Shocks: Rainfall, price shocks Village Characteristics/ Village Fixed Effects Definition of Variables of Interest:
‘low BMI (Head)’ – dummy for BMI<18.5 of HH head;
‘low BMI (HH Share)’ – share of adults with BMI<18.5; Endogeneity Concerns: low BMI , livestock and lagged dependent variable BREAD 2008 - 7
IW Outes-Leon
B.3 Identification Strategies
Growth (t to t-p) 1984 1994 1995 2004 t-p
Method A: Lagged Endogenous Variables – 1994
Strong but invalid instruments.
LATE interpretation of estimates.
t
Method B: 1984 Drought Shock
IV: self-reported 1984 Drought Shock.
Valid if self-reported shock is unrelated to growth, after controlling for assets.
Some validity tests are passed – But validity remains questionable
Comparison OLS vs IV estimates still interesting:
Analogous to LATE: BMI growth effect of long-term vulnerability BREAD 2008 - 8
IW Outes-Leon
C.1 Naïve and Lagged Endog. IVs
Consumption Growth Regressions Dependent: Change in Log Consumption, 1995-2004 Log Consumption., t-1 Low BMI (Head), t-1 Low BMI (HH share), t-1 Livestock, Cat 1, t-2 Livestock, Cat 2, t-2 Livestock, Cat 3, t-2 Livestock, Cat 4, t-2 Livestock, Cat 5, t-2 Livestock, Cat 6, t-2 Land, t-1 Relative Land, t-1 Schooling, 1994 Chronic h. problem, 1994 Rainfall Shock, t to t-1 Price Shock, t to t-1 Price Index, t-1 _cons Nr Instruments Cragg-Donald F-Stat Cohort, & demo. controls Village Dummies - FE Nr. Observations OLS (1a) -0.1036
-22.76
-0.0112
-1.28
-0.0029
-0.38
0.0374
4.84
0.0017
2.64
0.0013
2.35
0.312
4.37
None Yes No 764
0.0211
1.90
0.0148
1.44
0.0384
3.4
0.054
4.95
0.0568
4.78
0.0814
4.39
-0.0063
-0.48
0.0065
1.3
-0.0003
-0.03
Naïve OLS Model OLS (1b) -0.1021
-21.65
-0.0063
-0.5
0.0196
1.75
0.0153
1.48
0.0347
3.06
0.0496
4.3
0.0557
4.59
0.078
4.2
-0.009
-0.67
0.0076
1.52
0.0009
0.11
-0.0024
-0.31
0.0377
4.83
0.0016
2.36
0.0013
2.48
0.2949
4.1
None Yes No 764
Endog: Low BMI and Consumption, 1995 Lagged Endog IV IV: 1994 IV GMM (2a) IV GMM (2b) -0.074
-3.51
-0.04
-2.17
-0.078
-3.5
0.013
1.11
0.008
0.72
0.035
2.96
0.042
3.48
0.048
3.69
0.073
3.53
-0.02
-1.37
0.007
1.35
-0.003
-0.39
-0.069
-2.49
0.016
1.32
0.012
1.06
0.037
3.16
0.046
3.73
0.051
3.8
0.077
3.64
-0.022
-1.5
0.008
1.61
-0.003
-0.33
0.007
0.86
0.026
2.52
0.001
0.83
0 0.39
0.314
3.75
2 23.11
Yes No 764 0.005
0.72
0.027
2.47
0.001
0.85
0 0.42
0.333
3.84
2 20.88
Yes No 764
Naïve:
No BMI effect Livestock non-linearities Lagged IV:
‘Low BMI (Head)’ reduces growth by 4% (Robustness checks omitted):
BMI effect - robust to changing length of Lag.
But not robust to introduction of Village FE Note: T-statistics reported. Bold coefficients indicate significance at the 10% level. First-Stage estimates not reported. Periods defined as [t=2004], [t-1=1995] and [t-2=1994]. Livestock categories indicate single units of scaled livestock, except for categories 4 to 6 that correspond with ‘4 to 6’, ‘6 to 10’ and ‘more than 10’ scaled units respectively; livestock default category is ‘less than 1’. Cohort and demographic controls include: ‘hh size’, ‘nr male members’, ‘share of female members’, ‘Δ hh size’, ‘Δ nr children members’, ‘Δ nr female members’ and dummies for each decade of age of HH head.
BREAD 2008 - 9
C.2 Quartile IV Regressions – Lagged Endog IVs
IW Outes-Leon Quartile Regressions, Endog: Low BMI (Head), 1995 by Av. Consumption, 1994-1995 Consumption Growth 1995 to 2004 IV Fuller Quartile 1 Panel A - Standard Controls Low BMI (Head), t-1 s.e.
F-Stat (First-Stage) Panel B - Village Fixed Effects Low BMI (Head), t-1 - 0.101
-2.71 *** 49.18
- 0.073
s.e.
-1.91 * F-Stat (First-Stage) Panel C – Consumption Quartiles by Village Low BMI (Head), t-1 30.85
- 0.118
s.e.
F-Stat (First-Stage) -2.72 *** 32.11
IV: Low BMI 1994 IV Fuller Quartile 2 IV Fuller Quartile 3
-0.046
-1.50
26.94
-0.0298
-1.17
21.13
-0.0499
-1.88 *
52.61
0.0015
0.06
36.13
-0.0064
-0.25
33.28
-0.0682
-1.78 *
22.03
IV Fuller Quartile 4
-0.0677
-1.74 *
32.49
-0.0294
-0.81
29.99
0.0107
0.36
36.08
BMI effect most likely to exist only among poorest Use village-specific quartiles Effect is largest among the poor.
Robust to Village FE and Village-Specific Quartiles Note: T-statistics reported. Coefficients highlighted in bold indicate significance at the 5% level. F-statistics reported in bold indicate the presence of ‘strong’ IVs. All regressions reported include same controls as in slide C.1
BREAD 2008 - 10
IW Outes-Leon
C.3 Drought IVs – IV GMM
Exogenous IV Regressions IV Set 1: Drought and Food Aid, 1984 IV GMM (1) First Stage - IV Determinants of Low BMI, t 1984 Drought Dummy 1984 Food Aid value 1984 Drought X 1984 Food Aid 0.0724
2.04
-1.15E-05 -2.81
1.18E-05 1.62
Drought Dummy, 1984 to 1994 Endog: Low BMI (Head), 1995 IV Set 2: Drought 1984 IV Set 3: Drought and Food Aid, 1984 to 1994 IV GMM (2) IV GMM (3) 0.0784
2.30
Food Aid value, 1984 to 1994 Drought X Food Aid, 1984 to 1994 0.0603
1.76
-7.15E-06 -3.55
4.48E-06 1.27
Second Stage - Consumption Growth, t to t-1 Aver. Log Cons., 1994 and 1995 Low BMI (Head), t-1 -0.1122
-9.72
-0.1683
-1.43
-0.1149
-8.92
-0.2498
-1.65
-0.1069
-11.10
-0.0981
-1.70
Vill. Consumption, t-1 Vill. Low BMI, t-1 Vill. Livestock, t-1 Vill. Land, t-1 First-Stage Nr Instruments F-Stat Second-Stage Hansen J Overid Test - p-value HH Asstes and Cohort Controls Village Dummies R-Square Nr. Observations
3
5.57
0.31
Yes No 0.051
707 1
5.31
Yes No -0.285
713 3
6.26
0.24
Yes No 0.233
707
Note: T-statistics reported. Coefficients highlighted in bold indicate significance at the 10% level. Controls: HH Assets, Cohort dummies and HH demographic variables IV Set 1, with Village Means IV GMM (4)
0.0125
0.35
-4.60E-06 -1.30
1.12E-05 1.60
-0.0986
-10.30
0.0995
1.09
-0.0180
-0.88
-0.2932
-2.74
0.0123
2.87
-0.1353
-2.37
3
1.23
0.67
Yes No 0.256
707
BREAD 2008 - 11
IW Outes-Leon
C.4 Drought IVs – Weak IVs
IV GMM (First-Stage) F-Stats:
Stock and Yogo (2005) critical values
F-Stats from [5.31] to [6.26] suggest: OLS Bias IV estimates include 30% to 20% of the Weak IV GMM estimates are unreliable – under finite-samples, (Murray (2006)):
IV GMM point estimates can be substantially bias;
Standard Errors tend to be invalid and smaller (!!) Estimation:
IV Fuller and IV LIML Methods are more robust to ‘weak’ IVs
In the literature: IV Fuller method is preferred estimator.
See Stock, Wright and Yogo (2002) and Murray (2006).
Inferences:
CLR Moreira and Anderson-Rubin tests more robust
CLR Moreira methods shown to dominate alternative methods
See Andrews, Moreira and Stock (2005) and Murray (2006).
BREAD 2008 - 12
IW Outes-Leon
C.5 Estimates and Inferences – Robust to ‘Weak’ IVs
Inference and Estimators, robust to Weak IVs IV Set 1: Drought and Food Aid, 1984 (1) Endog: Low BMI (Head), 1995 IV Set 2: Drought 1984 (2) IV Set 3: Drought and Food Aid, 1984 to 1994 (3) IV GMM IV LIML IV Fuller Wald, p-value CLR, p-value Anderson-Rubin, p-value Wald, Confidence Interval CLR, Confidence Interval Anderson-Rubin, Confidence Interval F Stat (First-Stage) Hansen J Overid Test - p-value R-Square (Second-Stage) Nr. Observations
-0.1683
-1.43
-0.2476
-1.49
-0.1532
-1.80
0.0740
0.0327 *** 0.0583
[-0.519, 0.0239] [-2.1257, -0.0219] [-47.669, 0.0075]
5.57
0.3115
0.0509
707 -0.2498
-1.65
-0.2498
-1.65
-0.1539
-2.03
0.0750
0.0157 *** 0.0157
[-0.5244, 0.0248] [-1.2636, -0.0449] [-1.2636, -0.0449]
5.31
-0.2847
713
-0.0981
-1.70
-0.2065
-2.18
-0.1503
-2.47
0.0390
0.0170 *** 0.0381
[-0.4023, -0.0108] [-0.6793, -0.037] [-0.9795, -0.0096]
6.26
0.2431
0.2327
707
Nr Instruments IV Period Drought Dummy Food Aid value Drought X Food Aid
3 1984 Yes Yes Yes 1 1984 Yes No No 3 1984 to 1994 Yes Yes Yes
Note: T-statistics reported. Coefficients highlighted in bold indicate significance at the 10% level. Wald, Moreira Conditional Likelihood Ratio (CLR) and Anderson Rubin p-values and corresponding confidence intervals reported in square brackets.
BREAD 2008 - 13
IW Outes-Leon
C.6 Further on Drought IVs
Food Aid - Validity
Better connected HHs – obtain more Food Aid and grow faster. No bias appreciable when comparing with ‘IV Set 2: 1984 Drought Only’
Self-Reported Drought likely endogenous
E.g. Better insured HHs in 1984 drought villages – not affected by Drought.
No available HH information prior to 1984 drought.
Use answer ‘Did “food sharing” increase during famine?’ as extra control.
Livestock Endogeneity and Low BMI Bias
Current Implicit Ass: 1984 Drought affects exclusively HH health.
Persistence in Livestock Assets might be behind ‘low BMI’ effects Over-id Drought IV estimates with Endog: Low BMI and Low Livestock
Model is unreliable due to ‘very weak’ IVs.
But ‘Low BMI’ effect remains large [-0.10].
While ‘Low Livestock’ effect is increased substantially; BREAD 2008 - 14
IW Outes-Leon
C.8 Quartile IV Regressions – Drought IVs
Quartile Regressions, Endog: Low BMI (Head), 1995 by Av. Consumption, 1994-1995 Consumption Growth 1995 to 2004 IV Fuller Quartile 1 Panel A - Standard Controls Low BMI (Head), t-1 s.e.
CLR Moreira p-value F-Stat (First-Stage) Panel B - Village Fixed Effects Low BMI (Head), t-1 s.e.
CLR Moreira p-value -0.1756
-2.65
[0.0060] *** 8.35
-0.0448
-1.01
Undefined
F-Stat (First-Stage) Panel C – Consumption Quartiles by Village Low BMI (Head), t-1
1.87
-0.1249
s.e.
CLR Moreira p-value F-Stat (First-Stage)
-1.75
[0.0628]
**
9.85
IV: Drought 1984 IV Fuller Quartile 2 IV Fuller Quartile 3
-0.0373
-0.61
Undefined 3.65
-0.0156
-0.31
Undefined 1.94
-0.0193
-3.90
Undefined 0.01
-0.0112
-0.81
Undefined 0.05
0.0079
0.52
Undefined 0.01
0.0626
2.35
Undefined 0.15
IV Fuller Quartile 4
-0.0091
-0.34
Undefined -0.52
0.017
3.11
Undefined 0.01
0.0248
0.42
Undefined 2.56
BMI effect most likely to exist only among poorest With Village FE – ‘low BMI’ effect disappears Although it is robust to Village-Specific Quartiles Note: T-statistics reported. Coefficients highlighted in bold indicate significance at the 10% level. Moreira Conditional Likelihood Ratio (CLR) p-values reported in square brackets. ‘Undefined’ CLR p-value indicates that confidence intervals were unbounded. F-statistics reported in bold indicate the presence of ‘strong’ IVs. All regressions reported include same controls as in Slide C.4
BREAD 2008 - 15
IW Outes-Leon
D.1 Conclusions
Uncover substantial persistence of 1984 drought on 1995 livestock and BMI
Low BMI has significant negative effect on subsequent HH growth.
Lagged Endog IVs and Drought IVs provide results consistent with each other.
Growth effect of Low BMI might not be causal but can be interpreted as growth effect of low BMI persistence (Lagged IV) or HH vulnerability (Drought IV).
This Low BMI burden is overwhelmingly borne by the poor.
Growth effect is large in magnitude
For lowest quartile (with village FE), persistence of low BMI reduces growth by 7% percentage points per annum, for a period of nine years.
Drought IV estimates suggest a very large effect points per annum, for nine-year period.
But partly include village and ‘livestock’ effects.
– up to 15% percentage BREAD 2008 - 16
D.2 Going Forward
Alternative IVs; Alternative tests of validity;
Arellano-Bond Panel Estimation
Can tackle Unobserved HH Heterogeneity Unpack HH FE – seek for 1984 Drought effect
Testing for Mechanism of low BMI:
Income and productivity
Higher morbidity IW Outes-Leon BREAD 2008 - 17
Many Thanks
IW Outes-Leon BREAD 2008 - 18
IW Outes-Leon
C.3 Lagged Endogenous IVs – Robustness
Consumption Growth Regressions Dependent: Change in Log Consumption, 1995-2004 Log Consumption., t-1 Low BMI (Head), t-1 Low BMI (HH share), t-1 Vill. Consumption, t-1 Vill. Low BMI, t-1 Vill. Livestock, t-1 Vill. Land, t-1 Growth: 1995-2004 IV: IV GMM (1a) 1994 Lagged Endog IV IV GMM (1b) IV GMM (2a) 1994 (2b) Endog: Low BMI and Consumption, 1995 Lagged Endog IV, Growth: 1997-2004 IV: IV GMM Village Means Growth: 1995-2004 IV: 1994 Growth: 1997-2004 IV: 1994 IV GMM (3a) IV GMM (3b) IV GMM (4a) IV GMM (4b) -0.074
-3.51
-0.078
-3.5
-0.054
-1.36
-0.054
-1.45
-0.081
-2.36
-0.084
-2.32
-0.034
-0.63
-0.037
-0.74
-0.04
-2.17
-0.042
-1.34
-0.033
-1.77
-0.06
-1.78
-0.069
-2.49
-0.063
-1.36
-0.002
-0.05
-0.06
-1.14
-0.058
-1.86
0.002
0.05
-0.04
-0.71
-0.04
-0.67
0.091
1.31
-0.113
-2.13
-0.033
-0.57
0.119
1.59
0.007
2.26
-0.022
-0.54
0.007
2.22
-0.024
-0.56
0.006
0.98
0.045
0.69
0.006
0.99
0.043
0.67
First-Stage Nr Instruments Cragg-Donald F-Stat Second-Stage Asset, Cohort Controls Village Dummies - FE R-Square Nr. Observations
2 23.11
Yes No 0.459
764 2 20.88
Yes No 0.452
764 2 12.83
Yes No 0.296
735 2 13.47
Yes No 0.285
735 2 10.74
Yes No 0.4822
764 2 10.7
Yes No 0.4734
764 2 8.58
Yes No 0.234
735 2 8.97
Yes No 0.217
735
Note: T-statistics reported. Bold coefficients indicate significance at the 10% level. First-Stage estimates not reported. Periods defined as [t=2004], [t-1=1995] and [t-2=1994]. Controls: HH Assets, Cohort dummies and HH demographic variables Lagged Endog IV, Village FE Growth: 1995-2004 IV: 1994 IV GMM (5a) IV GMM (5b) -0.063
-1.97
-0.065
-1.96
-0.019
-1.09
-0.037
-1.18
2 11.21
Yes Yes 0.5
764 2 11.1
Yes Yes 0.498
764
BREAD 2008 - 19
IW Outes-Leon Variables
Av. Food Cons. Growth, (t to t-1)/p Log Food Cons., t-1 Height (HH head) BMI (HH head), t-1 Low BMI (HH head), t-1 * Low BMI (HH share), t-1 Livestock (Index), t-1 Low Livestock, t-1 * Land (Hect. per AEq), t-1 Relative Land, t-1 Schooling (HH head), 1994 * Chronic health problem, 1994 * Rainfall Shock, t to t-1 Food Price Index (FPI), t-1 Price Shock (ΔFPI), t to t-1 Adult Illness Shock, t Livestock Shock, t ** Agricultural Shock, t ** Δ HH size, t to t-1 Δ Nr. Female members, t to t-1 Δ Nr. Children, t to t-1 HH size, t-1 Nr. Males Adult, t-1 Female share in HH, t-1 HH head Age, t-1
Nr Observations Al HHs (1)
0.0430
3.8006
167.3367
19.9525
0.2147
0.2155
2.6872
0.6505
0.3868
1.1116
0.4398
0.2160
-0.0432
120.4468
-5.6561
0.3704
-0.2497
-0.1770
-0.9908
-0.5131
-0.3639
6.6361
1.6950
0.4958
46.8261
764 High Cons. HHs (2)
-0.0139
4.4676
167.6845
20.1294
0.1414
0.1575
2.8906
0.5838
0.4985
1.1926
0.4869
0.1885
-0.0022
122.4427
-6.5490
0.3377
-0.2452
-0.1586
-0.5209
-0.2906
-0.0942
5.7827
1.4764
0.4996
44.8808
382 Low Cons. HHs (3)
0.0999
3.1335
166.9889
19.7757
0.2880
0.2735
2.4838
0.7173
0.2752
1.0307
0.3927
0.2435
-0.0842
118.4509
-4.7633
0.4031
-0.2542
-0.1954
-1.4607
-0.7356
-0.6335
7.4895
1.9136
0.4920
48.7663
382 Normal BMI (Head) (4)
0.0382
3.8865
167.1656
20.6267
0.0000
0.1031
2.7796
0.5872
0.4097
1.1319
0.4733
0.1800
-0.0296
121.0354
-6.1511
0.3367
-0.2588
-0.1692
-0.7633
-0.3983
-0.3383
6.4700
1.6000
0.4951
45.6156
600 Note: (*) Indicates dummy variables. (**) Index between (0) and ( 1), negative indicates a more serious shock. ‘High’ and ‘Low’ consumption HHs correspond with the top and bottom halves of the 1995 food consumption distribution. Variables used in the regressions models not reported here include: village asset means and decade-age dummies. Consumption growth variable is defined as the average growth between t and t-1, where p takes a value of nine years Low BMI (Head) (5)
0.0605
3.4863
167.9628
17.4860
1.0000
0.6268
2.3491
0.7166
0.3031
1.0377
0.3171
0.3476
-0.0929
118.2934
-3.8453
0.4939
-0.2168
-0.2056
-1.8232
-0.9329
-0.4573
7.2439
2.0427
0.4984
51.2472
164 BREAD 2008 - 20
Normal BMI - 1995 Low BMI - 1995 Nr Observations By Villages Haresaw Geblen Dinki Yetemen Shumsha Sirbana Godeti Adele Keke Korodegaga Trirufe Ketchema Imdibir Aze Deboa Adado Gara Godo Doma Milki Kormargefia Karafino Bokafia All villages IW Outes-Leon Low BMI (Head) 1995
No Yes
713
12.50% 8.33% 26.79% 10.26% 8.62% 15.91% 14.29% 34.55% 22.73% 81.82% 25.00% 10.00% 35.59% 22.22% 0.00% 16.00% 14.29% 16.67%
21.60% 1984 Drought Shock No Yes
286 52
338
273 102
375 1984-1994 Drought Shock No
244 44
288 Yes
315 110
425 Nr Observ.
559 154
713 Av. Food Aid Value (1984) Av. Food Aid Value (1984-1994)
677 Birr 446 Birr
707
2001 Birr 771 Birr
707 37.5% 83.3% 8.9% 66.7% 34.5% 95.5% 42.9% 25.5% 47.7% 15.2% 0.0% 100.0% 32.2% 42.2% 73.0% 84.0% 85.7% 91.7% 47.4%
62.5% 16.7% 91.1% 33.3% 65.5% 4.6% 57.1% 74.6% 52.3% 84.9% 100.0% 0.0% 67.8% 57.8% 27.0% 16.0% 14.3% 8.3%
52.6%
31.3% 83.3% 5.4% 33.3% 22.4% 88.6% 37.5% 18.2% 45.5% 15.2% 0.0% 100.0% 27.1% 20.0% 70.3% 80.0% 85.7% 91.7%
40.4%
68.8% 16.7% 94.6% 66.7% 77.6% 11.4% 62.5% 81.8% 54.6% 84.9% 100.0% 0.0% 72.9% 80.0% 29.7% 20.0% 14.3% 8.3%
59.6%
45 37 25 14 12
713
32 24 56 39 58 44 56 55 44 33 40 40 59 938 Birr 49 Birr 406 Birr 6,430 Birr 71 Birr 6 Birr 3 Birr 122 Birr 26 Birr 70 Birr 63 Birr -
627 Birr
4532 Birr 17,055 Birr 406 Birr 9,960 Birr 118 Birr 236 Birr 3 Birr 206 Birr 30 Birr 515 Birr 314 Birr -
1734 Birr
-
Note: Change in number of households from 713 to 707 is due to missing food aid information for 6 households. Pairs of columns (3)-(4) and (5)-(6) add to unity by village BREAD 2008 - 21
IW Outes-Leon Year 1995 Round Nr of Household Worse Drought – Vulnerability Module, 20 year recall period – Nr of Household – 2 nd Worse Drought 1994 Round – Vulnerability Module,10 year recall period Nr of Household – 3 rd Worse Drought Total Food Aid Value (in Birr) Nr of Households Total
1973 1974 1975 1976 1978 1979 1980 1981 1982 1983
1984
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 6 5 1 5 2 4 7 6 16 37
375 1 2 3 3 1 5 2 4 17 12 514
3 1 1 2 2 22 62 22 11 31 13 15 11 5 15 79 8
303
1 1 19 6 8 8 4 2 13 2 6 68 7
145
-
67,342 142,418 37,579 46,787 4,025 2,494 7,286 3,015 8,246 12,667 331,859
22 24 28 106
475 (*)
201 51 14 9 12 8
Note: Highlighted area indicates information used in the construction of the different sets of instruments. (*) Food aid totals reported include double-counting of households. That is, 201 households received food aid in 1984, but only a total of 371 different households received food aid at least once during the 1984 to 1994 period.
BREAD 2008 - 22
IW Outes-Leon Quartile Regressions, Endog: Low BMI (Head), 1995 Endog: Low BMI (Head), 1995 by Av. Consumption, 1994-1995 Consumption Growth 1995 to 2004 IV Fuller Quartile 1 Panel A - Standard Controls Low BMI (Head), t-1 -0.101
-2.71
[0.0023] 49.18
F-Stat (First-Stage) Panel B - Village Fixed Effects Low BMI (Head), t-1 -0.073
-1.91
[0.0556] F-Stat (First-Stage) Panel C – Consumption Quartiles by Village Low BMI (Head), t-1 30.85
-0.118
F-Stat (First-Stage) -2.72
[0.0013] 32.11
IV: Low BMI 1994 IV Fuller Quartile 2 IV Fuller Quartile 3
-0.046
-1.50
[0.1797]
26.94
-0.0298
-1.17
[0.4368]
21.13
-0.0499
-1.88
[0.0733] 52.61
0.0015
0.06
[0.9993]
36.13
-0.0064
-0.25
[0.8180]
33.28
-0.0682
-1.78
[0.0972] 22.03
IV Fuller Quartile 4 -0.0677
-1.74
[0.0848] 32.49
-0.0294
-0.81
[0.4603]
29.99
0.0107
0.36
[0.7445]
36.08
IV Fuller Quartile 1 -0.1756
-2.65
[0.0060] 8.35
-0.0448
-1.01
Undefined 1.87
-0.1249
-1.75
[0.0628]
9.85
IV: Drought 1984 IV Fuller Quartile 2 IV Fuller Quartile 3
-0.0373
-0.61
Undefined 3.65
-0.0156
-0.31
Undefined 1.94
-0.0193
-3.90
Undefined 0.01
-0.0112
-0.81
Undefined 0.05
0.0079
0.52
Undefined 0.01
0.0626
2.35
Undefined 0.15
IV Fuller Quartile 4
-0.0091
-0.34
Undefined -0.52
0.017
3.11
Undefined 0.01
0.0248
0.42
Undefined 2.56
Note: T-statistics reported. Coefficients highlighted in bold indicate significance at the 10% level. Moreira Conditional Likelihood Ratio (CLR) p-values reported in square brackets. ‘Undefined’ CLR p-value indicates that confidence intervals were unbounded. F-statistics reported in bold indicate the presence of ‘strong’ IVs. All regressions reported include same controls as in Slide D.2
BREAD 2008 - 23