Date post: | 05-Dec-2014 |
Category: |
Technology |
Upload: | essp2 |
View: | 1,392 times |
Download: | 2 times |
Growth, poverty and chronic poverty in rural
Ethiopia:
Evidence from 15 Communities 1994-2004
Stefan DerconJohn Hoddinott
Tassew Woldehanna
There is a growing body of evidence that explores the correlates of poverty in rural Ethiopia
Cross-sectional surveys indicate that, at the national level, consumption poverty (as measured by headcount poverty) has been falling, albeit at an uneven pace over time and space
While informative, this evidence leaves a number of questions unanswered:
To what extent do these trends reflect increased movement out of poverty, or in and out of poverty – there are very different welfare implications associated with theseAre there some households “stuck” in poverty?If so, what are the characteristics of these chronically poor households?Will “standard” interventions (eg infrastructure improvements) help these chronically poor households?Are these households „permanently‟ poor?
INTRODUCTION
INTRODUCTION, cont.
Here we investigate these issues, specifically:What do these poverty dynamics look like
What is the magnitude of chronic poverty in rural Ethiopia
Are the chronically poor „different‟ from other households
The ERHS Data Set
We draw on data collected as part of the Ethiopia Rural Household Survey, a multi-purpose longitudinal household survey covering 15 villages and approximately 1400 households with multiple rounds between 1994 and 2004
Population shares within the sample are broadly consistent with the population shares in the three main sedentary farming systems
The plough based cereals farming systems of the Northern and Central Highlands
Mixed plough/hoe cereals farming systems, and farming systems based around enset that is grown in southern parts of the country.
The sample sizes in each village were chosen so as to approximate a self-weighting sample, when considered in terms of farming system: each person (approximately) represents the same number of persons found in the main farming systems as of 1994.
However, results should not be regarded as nationally representative. The sample does not include pastoral households or urban areas and only covers 15 villages
CONSUMPTION
Consumption is defined as the sum of values of all food items, including purchased meals and non-investment non-food items.
Contributions for durables and spending with some investment connotation, such as health and education expenditures are not included
Although there are good conceptual reasons for including use values for durables or housing , we do not do so; the heterogeneity in terms of age and quality of durables owned by our respondents, together with the near complete absence of a rental market for housing would make the calculation of use values highly arbitrary. Because there has been accumulation of durables over this period, we understate consumption growth.
These values are expressed in monthly per capita terms and deflated using the food price index with base year 1994.
CONSUMPTION
Mean consumption per capita in 1994 was 71.1 birr per capita per month.
By 2004, this had risen to 91.5 birr per capita in real (1994) terms.
Over this ten year period, consumption growth of the mean was on average 2.6 percent per year. This is broadly comparable to the average annual rate of growth of real GDP per capita (2.1 per cent) and the increases reported using nationally representative household consumption data
0.2
.4.6
De
nsity
0 2 4 6 8Log p.c. consumption
Log p.c. consumption, 1994 Log p.c. consumption, 2004
Figure 1
Inequality, assets, school enrollment and perceptions of wellbeing
1994 2004
Inequality Gini coefficient 0.45 0.44
Livestock: Percentage of households owning: Oxen 13 49
Any livestock 81 90
Livestock Real (1994) value of per capita holdings 342 421
Ownership of other assets. Percentage of
households with:
Hoes 59 79
Ploughs 79 87
Beds 49 58
School Enrollment. Percentage of: Boys age 11, enrolled 21 60
Girls age 11, enrolled 13 58
Children, aged 7-14 enrolled 13 45
Public services. Percentage of households
with access to:
Electricity 0 13
Piped water 11 31
Public services. Distance (km) to: Nearest telephone 14.8 9.3
Perceptions of wellbeing. How would you
describe your household circumstances?
(Percentages)
Very rich 0.5 0.4
Rich 5.4 5.9
Comfortable 19.9 30.4
Can manage to get by 25.4 29.2
Never have quite enough 7.9 13.0
Poor 33.4 19.9
Destitute 7.6 1.1
POVERTY
Poverty is based on a cost-of-basic-needs approach.
Using the 1994 data, a food poverty line is constructed using a bundle of food items that would provide 2300Kcal per adult per day.
To this, we add a non-food bundle using the method set out in Ravallion and Bidani (1994) and obtain a poverty line of 50 birr per capita per month in 1994 prices.
TRENDS IN CONSUMPTION AND POVERTY
Mean Consumption
per capita
MedianConsumption
per capitaHead count
poverty Poverty GapSquared
Poverty Gap
1994 71.1 51.6 0.48 0.21 0.12
1995 62.0 45.3 0.55 0.24 0.14
1997 90.9 70.7 0.33 0.12 0.06
1999 88.3 64.5 0.36 0.13 0.06
2004 91.5 65.1 0.35 0.13 0.07
MOVEMENTS IN AND OUT OF POVERTY BETWEEN 1994 (ROUND 1) AND 2004
Household is: Per cent
Poor in 1994
and Poor in 2004 18.8
Not poor in
1994
and
Poor in 2004
14.2
Poor in 1994
and
Not poor in
2004
26.8
Not poor in
1994
and
Not poor in
2004
34.7
Thinking about “chronic poverty”
Being poor “all the time” – counting the number of times a household is observed to be poor
A household is chronically poor if its “average consumption” is below the poverty line (Jalan and Ravallion)
Do such definitions matter? Suppose we wanted to target an intervention to chronically poor households. Consider the following example where poverty line is set at 5
Thinking about “chronic poverty”
• Under the “count” definition, HH1 is “more chronically poor” than HH2, even though it has higher consumption in three of the four periods.
• Under the “average consumption” definition, they are equally chronically poor
Period Household 1 Household 2
Consumption Poverty Status
Consumption Poverty Status
1 4 Poor 1 Poor
2 4 Poor 1 Poor
3 4 Poor 5 Not poor
4 4 Poor 1 Poor
Thinking some more about “chronic poverty”
Given these limitations, an appealing approach is to use allinformation on the magnitude of the gaps between consumption and the poverty line.
The average value of the P2 poverty measure can then be seen as an index of chronic poverty, or more precisely, an index of intertemporal poverty
Chronic Poverty in the ERHS
Percentage of households
(1)
Chronic Poverty Index
Average Squared Poverty Gap per period of poverty
Never Poor 18 0 0
Poor once 22 0.13 0.13
Poor in 2 out of 5 rounds
23 0.32 0.16
Poor in 3 out of 5 rounds
16 0.65 0.22
Poor in 4 out of 5 rounds
14 1.04 0.26
Poor in all rounds 7 1.34 0.27
Correlates of chronic poverty (poor in at least three periods)
Chronic Poor Probit with village dummies
dF/dx
Chronic Poor ProbitdF/dx
Initial (1994) household characteristics
(1) (2)
Education primary or more -0.211 -0.209(3.75)** (3.89)**
Sme educ., below primary -0.022 -0.024(0.46) (0.56)
Ln land holding p.c. in ha -0.048 -0.058(2.19)* (2.84)**
Sex of head -0.017 -0.01(0.34) (0.22)
No. female adults 15-65 0.011 0.007(0.63) (0.46)
No. girls 5-15 0.067 0.056(3.91)** (3.52)**
No. girls below 5 0.112 0.097(4.28)** (3.97)**
No. females >65 0.029 0.014(0.49) (0.25)
No. male adults 15-65 -0.001 0.005(0.05) (0.30)
No. boys 5-15 0.054 0.046(3.20)** (2.91)**
No. boys below 5 0.084 0.07(3.22)** (2.85)**
No. males >65 -0.046 -0.029(0.54) (0.36)
Correlates of chronic poverty (poor in at least three periods)
Chronic Poor Probit with village dummies
dF/dx
Chronic Poor ProbitdF/dx
(1) (2)
Initial village characteristicsDistance to nearest town in km 0.032
(9.64)**Coefficient of variation of rain 0.008
(4.29)**Road accessible to trucks/buses -0.374
(8.24)**Ln mean land holding p.c. 0.152
(3.18)**Mean no. of male adults per hh 0.132
(1.18)Mean no. of female ad. per hh 0.511
(4.91)**
Household Correlates of Chronic Poverty
Education matters: with primary education complete in 1994, the probability of being chronically poor in 1994-2004 is a fifth lower.
Land matters as well, although in percentage terms not as much: doubling land (which implies increasing land by just under one standard deviation) reduces the probability of being chronic poor by about 5 percent.
There are strong effects on having children: they increase the likelihood of being found chronically poor, with children below 5 and especially girls adding most. (This is not due to the use of a poverty definition based on consumption per capita, which would „penalize‟ families with children relatively more)
Village Correlates of Chronic Poverty
The village fixed effects (not reported) are suggestive.A household residing in Gara Godo (an enset growing village in the south) with household characteristics identical to a household in Sirbana Godeti (not far from the large trading town Debre Zeit and the survey site closest to Addis Ababa) is 80 percent more likely to be chronically poor.
Turning to the community characteristics in column (2), the role of road access and distance to towns is striking:
Having a good road reduces the likelihood being found chronic poor by 37 percent
A reduced distance to the nearest small town by about 12 kilometres (which is moving from a distance as in the 75 percentile to the 25th percentile) also brings down the probability of being chronic poor by about 38 percent.
Villages with relatively speaking many female adults are more likely to be chronically poor.
But with only 15 communities, we have to be cautious in attaching strong interpretations to these results.
Taking Stock
In the ERHS data set:
There was considerable growth in consumption between 1994 and 2004.
While this growth was reasonably equitably distributed – with the result that headcount poverty and the P2 poverty index fell – some households moved into poverty while others moved out.
Approximately 37 per cent of households were chronically poor between 1994 and 2004
Being chronically poor is associated with:
Low initial levels of human capital, low land holdings and adverse demographics
Remoteness, poor infrastructure and (possibly) lingering effects of the war
This suggests the following question: Do the determinants of growth differ across households depending on their initial conditions?
A simple household growth model
Consider a standard empirical growth model, allowing for transitional dynamics.
We observe i households (i = 1, …, N) across periods t (t = 1, …, T).
Growth rates for household i (ln yit – ln yit-1) are negatively related to initial levels of income (ln yit-1).
Let δ represent sources of growth common to all households and X reflect fixed characteristics of the household, such as location, that also affect growth. This is captured as a “household fixed effect”
Other sources of growth from t to t-1 are exogenous levels of capital stocks and access to technologies (kit-1) observed at t-1 both of which are time varying.
Lastly, we allow for transitory shocks such as changes in rainfall (ln Rt
– ln Rt-1), to affect growth
Household growth model
Dropping the i subscripts, our basic model is:
ln yt – ln yt-1 = δ + αln yt-1 + βln kt-1 + γ(ln Rt – ln Rt-1) + λX
Household growth model with interaction terms
We focus on three factors that, a priori, we believe may have affected consumption growth in this sample:
The expansion in road infrastructure
The extension programme aiming to increase productivity and
The role played by recurrent droughts, not least the 2002 drought.
We will treat roads and extension as a form of kt-1 with possible subsequent growth effects, and rainfall and other shocks (illness, death, input and output prices) as a source of ln Rt – ln Rt-1.
Lastly, we interact kt-1 with a representation of being chronically poor. These are households with below mean levels of per capita land and livestock holdings, no schooling, poor roads and are remote
Annualized Growth in Consumption per capita
Annual Annualized Growth in Consumption
per capita with „low endowments and
remoteness‟ interactions
(1) (3)
Log consumption (IV) -0.365 -0.355(9.03)** (8.53)**
Access to all-weather road 0.159 0.200(5.14)** (4.89)**
Received visit from extension officer
0.071 0.087(1.91)* (1.84)*
Illness shocks (1.14) (1.06)Interaction termsAccess to all-weather road -0.036
(0.63)Received visit from extension officer
-0.015(0.20)
Sample size 4578
RoundHead count
PovertyReceived at least
one extension visit
Access to all-weather road (trucks/buses)
Chronic poor
Non-chronic poor
Chronic poor
Non-chronic poor
Chronic poor
Non-chronic poor
1994 0.83 0.27 0.05 0.07 0.26 0.541995 0.93 0.33 0.02 0.06 0.26 0.541997 0.70 0.12 0.06 0.08 0.23 0.591999 0.68 0.17 0.16 0.11 0.25 0.722004 0.62 0.19 0.15 0.15 0.68 0.70Source: Ethiopian Rural Household Survey
Table 8: Evolution of head count poverty, roads access and for chronically and non-chronically poor households
Evolution of head count poverty, roads access and for chronically and non-chronically poor households
The „puzzle‟ of chronic poverty
The growth regression shows that the „return‟ on public investments is the same for chronically poor and non-chronically poor households (note that such a finding is not consistent with a „poverty trap‟ story
Chronically poor households are exposed to the same changes in public investments.
So why do they remain chronically poor?
The „puzzle‟, cont‟d
The models in Table 6 and 7 are fixed effects models, implying that they control for household heterogeneity in the underlying growth rate. In other words, beyond the factors modeled, each household has its own „unexplained‟, latent part of growth.
The average fixed effects for the chronic poor is -15.4 percent per year, while for the non-chronic poor it is 9.2 percent per year; these means are significantly different from each other at 1 percent and less.
In other words, for the same values of shocks, or roads or extension visits, the growth difference between these two groups is estimated to be almost 25 percent.
The „puzzle‟, cont‟d
The correlation between the household fixed effect and whether one is chronic poor or not, or with the chronic poverty index that allows for the severity of chronic poverty is very high (respectively -0.47 and -0.50).
In other words, the chronic poor face a serious growth deficit, making catching up with the rest very difficult – in terms of time-varying characteristics, they need much „better‟ values to obtain the same level of growth as the non-chronic poor.
What characteristics are associated with these lower levels of latent growth. By retrieving the household fixed effects, we can regress these on initial household characteristics
Fixed „growth‟ effect OLS
(4)
Education primary or more 0.052(2.00)*
Sme educ., below primary 0.004(0.21)
Ln land holding p.c. in ha 0.052(4.08)**
Sex of head -0.017(0.78)
Village characteristicsDistance to nearest town in km -0.005
(3.68)**Coefficient of variation of rain -0.002
(2.63)**Road accessible to trucks/buses -0.01
(0.43)Ln mean land holding p.c. -0.056
(2.57)*Mean no. of male adults per hh -0.025
(0.49)Mean no. of female ad. per hh -0.155
(3.47)**
The „puzzle‟, cont‟d
The most striking result is that the main correlates for lower chronic poverty are the correlates for a higher fixed effect.
For example, moving from the 25th to the 75th percentile in terms of distance to town, would cost 6 percent in latent growth.
Strikingly, unlike remoteness, access to roads is not a significant part the latent growth effect.
Over this ten year period, consumption growth of the mean was on average 2.6 percent per year. Headcount poverty fell from 48 to 35%; poverty severity also fell
While some households moved in and out of poverty over this period, a substantial
fraction, 37%, were chronically poor
Results from the growth model shows that the chronic poor tend to have the same
return from growth stimulating factors, such as improved infrastructure or extension. Also, over this period, they were exposed to the same changes
But the chronically poor start from a serious growth handicap, linked to physical assets, education and remoteness. This contributes to the persistence of their poverty. This fixed growth handicap is the micro-econometric equivalent of showing „club‟ convergence, whereby the initial characteristics matter permanently for long-term outcomes.
CONCLUSIONS