{rt-IF! t. V

”ﬁg.

khan «n...

11......)
.€.:..~5c
:4

 

. V
,. MW?

 

....._.;.,.,v
. k .
z :20;

iliiu

a. w 3 y. . . ‘
, . . t on»:
. . n. 3%»

531.. ”.61..

i1.

7!-
rtnma. ,
Egg: . ,
a. V A: I.
» ﬁx
it.
31%;
ﬁn};
is ‘

.mvmmw

4.. .
ﬁrm... sugar)?“
. ”ﬁns".

3:.»

 

any filx al.45Ivt. 9r .sbtl
ll‘.’: 4.... 1.31]... 3....
i...

qua-1.1 v “Etc

 

1 LIBRARY
W Michigan §tate
University

This is to certify that the
thesis entitled

PATHWAYS OUT OF POVERTY IN RURAL MOZAMBIQUE

presented by

BENEDITO ARMANDO CUNGUARA

has been accepted towards fulﬁllment
of the requirements for the

 

 

 

 

Master of degree in Agricultural Economics
Science
Major Professor’s Signature
314% 21/ 22308
J 7
Date

MSU is an afﬁnnative-action, equal-opportunity employer

r.a-.—u-.----u-.-.--u-u----------n— -----c-.-------.--c-o-~o-o-n-o---.-------o-n.

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:IProj/Acc&PrelelRC/DateDueindd

 

PATHWAYS OUT OF POVERTY IN RURAL MOZAMBIQUE
By

Benedito Armando Cunguara

A THESIS
Submitted to
Michigan State University

In partial fulﬁllment of the requirements
For the degree of

MASTER OF SCIENCE
Department of Agricultural Economics

2008

ABSTRACT

PATHWAYS OUT OF POVERTY IN RURAL MOZAMBIQUE
By

Benedito Armando Cunguara

This research addresses three questions: (i) how have the incidence of poverty and the
distribution of income changed? (ii) what factors are associated with rural household
income change? and (iii) what factors are associated with moving out of or into poverty
in the short-run? The analysis presented is based on a nationally representative two-
period panel data set from Mozambique, covering the years 2001/2 and 2004/5. We
cannot unambiguously argue that poverty headcount in 2005 (a year of more widespread
drought) was higher than poverty headcount in 2002, but the poverty gap and squared
poverty gap measures were both higher in the second period. Poverty has spatial,
demographic, and asset holding dimensions. The diversiﬁcation of off-farm income
sources is strongly associated with increases in household income, and thus poverty
reduction. For agriculturalactivities, especially crop production, reduction in
vulnerability to drought is important for poverty reduction. Access to price information is
positively correlated with income and poverty reduction in all areas, and the cultivation
of tobacco is also signiﬁcant in reducing poverty in the central part of the country. The
impact on poverty of growing tobacco or cotton in the north was not signiﬁcant, however.
We also found that increases in landholding size can reduce poverty in rural Mozambique

but will require additional inputs such as labor, fertilizers, and animal traction.

DEDICATION

To my dad, for his encouragement in pursuing graduate studies.
To my son, for his sacriﬁce during the two years of graduate studies.
To my wife, for her love, sacriﬁce, and support in everything.

To my mom and brothers, for their support and love.

ACKNOWLEDGMENTS

The author would like to acknowledge the ﬁnancial support from USAID and the
Ministry of Agriculture in Mozambique for making possible the realization of the
National Agricultural Survey, commonly known as TIA. Ellen Payongayong and the staff
from the Departments of Statistics and Policy Analysis within the Directorate of
Economics of the Ministry of Agriculture have done a great job in cleaning and
organizing the data. The author is also indebted to the households who participated on the

surveys.

For comments on various drafts, the author would like to acknowledge but not implicate
David Mather, Duncan Boughton, Tom Walker, David Tschirley, Eric Crawford, Charles

Ballard, Cynthia Donovan, Gilead Mlay, Deborah Mulcahey, and Peter Ollila.

TABLE OF CONTENTS

 

 

 

 

 

 

 

LIST OF TABLES VII
LIST OF FIGURES IX
ACRONYMS X
CHAPTER I INTRODUCTION 1
1.1. RESEARCH AND POLICY QUESTIONS ................................................................................. 2
1.2. LITERATURE REVIEW ............................................................................................................. 4
1.2.1. WELFARE MEASURE ................................................................................................................. 5

1.2.2. THE CHOICE OF POVERTY LINES .......................................................................................... 6

1.2.3. DISTINCTION BETWEEN TRANSITORY AND CHRONIC POVERTY ..................................... 9

1.2.4. POVERTYANALYSIS METHODS ............................................................................................ 10

1.2.5. ATTRITION BIAS ...................................................................................................................... 12

1.2.6. RELATED STUDIES IN MOZAMBIQUE ................................................................................. I 3
CHAPTER II DATA SOURCES 17
CHAPTER III DESCRIPTIVE ANALYSIS 20
3.1. POVERTY MEASURES ........................................................................................................... 20
3.2. STOCHASTIC DOMINANCE ANALYSIS ............................................................................. 22
3.3. INCOME MOBILITY AND INEQUALITY ............................................................................. 26
3.4. COMPARISON OF AGRICULTURAL SEASONS ................................................................. 32
3.5. POVERTY IN RURAL MOZAMBIQUE .................................................................................. 35
3.5. I SPATIAL DIMENSION OF POVERTY ................................................................................. 36

3.5.2 DEMOGRAPHIC DIMENSION OF POVERTY. ................................................................... 37

3.5.3 OFF-FARM INCOME AND POVERTY ................................................................................ 38

3.5.4 LANDHOLDING SIZE AND POVERTY ............................................................................... 42

3.5.5 AGRICULTURAL PRACTICES AND POVERTY ................................................................. 44
CHAPTER IV REGRESSION ANALYSIS 45
4.1. PREDICTING RURAL HOUSEHOLD INCOME .................................................................... 45
4.2. F lRST-DIF F ERENCE MODEL ................................................................................................ 53
4.3. MULTINOMIAL REGRESSION MODEL ............................................................................... 58

4. 3. I. MOVING OUT OF AND INTO POVERTY IN THE SOUTH ................................................ 61
4.3.2. MOVING OUT OF AND INTO POVERTY IN THE CENTER .............................................. 64

4. 3. 2. MOVING OUT OF AND INTO POVERTY IN THE NORTH ................................................ 66

4.4. POVERTY SIMULATION ........................................................................................................ 69
4.4. I REDUCTION IN DROUGHT VULNERABILI TY ................................................................. 7]

4.4.2 INCREASE IN LANDHOLDING SIZE .................................................................................. 72

4.4.3 ADOPTION OF CASH CROPS ............................................................................................ 75

CHAPTER V SUMMARY RESULTS

APPENDICES

 

 

REFERENCES

 

vi

77

83

93

LIST OF TABLES
Table 1. Food consumption expenditure poverty lines per adult equivalent by province .. 9
Table 2. National Agricultural Survey sample sizes ......................................................... 19
Table 3. Poverty incidence by province and year with different poverty lines ................ 25

Table 4. Mean and median household income per AE by quintiles and head’s gender 27

Table 5. Transition matrix of household income quintiles per AE by year ...................... 28
Table 6. Percentile ratios of household income/AE and Gini coefﬁcient by year ............ 29
Table 7. Distributional summary statistics using reported incomes, 2002 - 2005 ............ 31
Table 8. Shares of different income sources by poverty category and year ..................... 34
Table 9. Shares of each income source by quintile and year ............................................ 35
Table 10. Poverty measures using predicted household income/AB by year ................... 35
Table 11. Poverty transition matrix by province .............................................................. 36
Table 12. Poverty transition matrix by gender of household head ................................... 37

Table 13. Household size and composition in each year by poverty transition category. 38

Table 14. Household farm and non-farm employment by poverty transition category.... 39

Table 15. Mean non-farm income (MTn) by poverty transition category ........................ 41
Table 16. Number of households “escaping” poverty by province between 2002 and 2005

................................................................................................................................... 43
Table 17. Agricultural technology use and access to services by poverty transition

category ..................................................................................................................... 44
Table 18. Results from cross-section models for each year and region ............................ 50
Table 19. First-difference model results by region ........................................................... 55
Table 20. Relative risk ratios from multinomial logit models: southern provinces .......... 62

vii

Table 21. Relative risk ratios from multinomial logit regressions: central provinces ...... 65

Table 22. Relative risk ratios from multinomial logit regressions: northern provinces 67

Table 23. Effect of reducing days of drought on poverty indicators ................................ 71
Table 24. Effect of increasing landholding size (ha) on poverty indicators ..................... 73
Table 25. The effect of adoption of cash crops on poverty indicators .............................. 76
Table A 1. Poverty gap by province and year ................................................................... 91
Table A 2. Squared poverty gap by province and year ..................................................... 91
Table A 3. List of non-farm income generating activities ................................................ 92

viii

LIST OF FIGURES

Figure 1. Distribution of TIA sampled districts in 2003 ................................................... 18
Figure 2. Poverty categories based on predicted incomes ................................................ 21
Figure 3. First-order stochastic dominance ....................................................................... 23
Figure 4. Household income distribution per AB in 2002 and 2005 ................................ 26
Figure 5. Number of household income per AE quintiles changed .................................. 28
Figure 6. Lorenz curves at national level using reported income in 2002 and 2005 ........ 30
Figure 7. Days of drought by province and year .............................................................. 32
Figure 8. Mean days of drought by poverty category in 2002 and 2005 .......................... 33
Figure 9. Mean total landholding size in hectares by poverty category and year ............. 42

ix

ACRONYMS

AE
CBN
CDF
F El

FEWSNET

FGT
GDP
HH
IAF

IF PRI
IIA
INE
IPW
MN L
MN P
MPF
MTn
OLS
PRSP
PSU

SAT
SIMA
TIA
UEM
USAID

Adult Equivalent

Cost of Basic Needs

Cumulative Density Function

Food Energy Intake

Famine Early Warning System Network
Foster-Greer-Thorbecke

Gross Domestic Product

Household(s)

Household Survey of Living Conditions (Inquérito aos Agregados
F amiliares Sobre as Condicées de Vida)

International Food Policy Research Institute

Independence of Irrelevant Alternatives

National Statistics Institute (Instituto Nacional de Estati'stica)
Inverse Probability Weights

Multinomial Logit Regression

Multinomial Probit Regression

Ministry of Finance and Planning

Meticais da Nova F amilia

Ordinary Least Squares

Poverty Reduction Strategy Paper

Primary Sampling Unit

Relative Risk Ratios

Semi-arid Tropics

Agricultural Markets Information System

National Agricultural Survey (Trabalho de Inquerito Agricola)
Eduardo Mondlane University

United States Agency for International Development

CHAPTER I INTRODUCTION

Mozambique’s economy is characterized by limited diversiﬁcation and high dependency
on the agricultural sector, which accounts for a quarter of the Gross Domestic Product
(GDP). Per capita GDP is estimated to be $345, and more than half the population live in
poverty1 (Government of Mozambique, 2006). In addition, 89 percent of the population is
involved in agriculture, including a signiﬁcant segment of the urban population
(Government of Mozambique, 2001). Poverty incidence is higher in rural areas, making
the study of the factors associated with rural income and potential policy interventions

important.

If major pathways out of poverty were identiﬁed empirically, policy interventions could
be designed to facilitate escape from poverty, and to avoid others from falling into
poverty. One of the reasons for the dearth of such studies is the lack of panel data,
particularly from developing countries. This paper exploits a unique nationally
representative two-period panel data set from rural Mozambique covering the years 2002
and 2005, and seeks to analyze the patterns of poverty exits and avoidance of falling into

poverty, by examining the processes of socio-economic mobility among rural households.

The motivation for analysis of poverty dynamics is that some poor households may need
a different type of help through policy than others. Policy research needs to distinguish

between transitory and chronic poverty. Policies that can effectively block the descent of

 

' Poverty incidence is based on ofﬁcial poverty lines from the National Institute of Statistics in

Mozambique, derived from consumption expenditure data.

1

the transient poor into chronic poverty may be different from but complementary to those
to help the chronically poor ﬁnd a way out of poverty. The selection of the right set of
policies depends on an accurate understanding of the underlying factors associated with

rural poverty dynamics.

A major strength of this paper derives from the unique features of the data. Although the
data only cover a relatively short period of time (three years), panel data provide the
opportunity to deepen the understanding of factors associated with transitory and chronic

poverty, and therefore to design more effective policy interventions.

This paper presents one of the ﬁrst empirical studies of poverty dynamics in
Mozambique. So far the available studies on factors associated with rural income in
Mozambique have based their empirical analysis on cross sectional data sets. Moreover,
the poverty literature on Mozambique has not given much attention to the mechanisms to
avoid households from falling into poverty, because poverty dynamics research requires
panel data and the 2004/5 Trabalho de Inquerito Agricola (TIA 05) is the ﬁrst household
panel in Mozambique. The present study addresses these gaps by analyzing the factors

associated with income change and movements out of or into poverty.

1.1. RESEARCH AND POLICY QUESTIONS

The Mozambican government’s ﬁve-year plan for the period 2005-2009, commonly
known as PRSP 11 (Poverty Reduction Strategy Paper 11) sets as its main objective the

reduction of absolute poverty levels (Government of Mozambique, 2006). Government

2

policies and reforms, together with substantial donor support, have contributed to
macroeconomic stability, rapid growth, socio-economic transformation and poverty
reduction. Economic growth has been driven in part by. consistent growth in the
agricultural sector, despite unfavorable climatic conditions in 2000 and 2005
(Government of Mozambique, 2006). This study aims to understand both the factors
associated with income change and the driving factors that enable rural households to

climb out of poverty and to avoid falling into poverty.

Despite the signiﬁcant reduction in poverty incidence in the late 1990’s, more than 60
percent of the rural population still lives in poverty (MPF/UEM/IFPRI, 2004).
Furthermore, one aspect that has not been given much attention in the current poverty
literature on Mozambique is the fact that a considerable number of households may be
moving into poverty, even though overall poverty rates have been decreasing. Reducing
poverty both through lifting poor households out of poverty, especially in rural areas, and
preventing the non-poor from falling into poverty, remains an important development

challenge for Mozambique.

Though poverty reduction depends on the magnitude of economic growth, the
distribution of the beneﬁts of growth also matters. Poverty reduction will be larger if the
beneﬁts of growth go more to the poor than to the non-poor. Thus, when studying
determinants of rural households’ income change, it also becomes important to address

the question of how the income distribution and inequality have changed.

This study will address three questions: (i) how have poverty measures, income
distribution and inequality changed? (ii) what factors are associated with rural household
income change? and (iii) what factors are associated with moving out of or into poverty
in the short-run? Finally, some policy implications will be drawn from the ﬁndings from

the above questions.

The following section reviews the current poverty literature. Chapter two describes the
data used, followed by a descriptive analysis of poverty and income dynamics in chapter
three. Chapter four presents the regression methods and results which provide the basis
for the policy simulation presented at the end of chapter four. Chapter ﬁve summarizes
key ﬁndings from the descriptive and regression analysis, and provides some policy

recommendations.

1.2. LITERATURE REVIEW

An extensive literature on poverty dynamics has been devoted to the sensitivity of results
to important methodological decisions such as the choice of the welfare measure
(Slesnick, 1993; Chaudhuri and Ravallion, 1994; Lipton and Ravallion, 1995),
distinguishing transitory from chronic poverty (Duclos et al. , 2006; Hulme, 2003; Jalan
and Ravallion, 1998), and the choice of the poverty line (Tarp et al. , 2002; Booysen et

al., 2005; Ravallion and Lokshin, 2003; Amdt and Simler, 2007).

This section discusses these methodological issues. It starts with a discussion of choice of

welfare measure, followed by the choice of poverty line. Then, the distinction between

4

transitory and chronic poverty is presented, followed by a discussion of the poverty
analysis methods found in the literature. A discussion of the potential advantages and

limitations of panel data is followed by a brief review of related studies in Mozambique.

1.2.1. WELFARE MEASURE

In a money-metric approach to measuring poverty, various authors argue that there are
theoretical reasons why consumption is believed to be more accurate than income as the
welfare measure. First, in panel data covering shorter periods (or in cross-section data),
the transitory income component becomes very important, making the probability of

misclassifying households as poor (or non-poor) higher (Gradin et al. , 2004).

Second, when using income as the welfare measure, the assumption is that a market
exists for all goods, which is not the case in many developing countries (Thorbecke,
2005). Also, there is no guarantee that households with income at or even above the
poverty line would actually allocate their incomes so as to purchase the minimum basic-

needs bundle.

Third, rural incomes ﬂuctuate in response to the agricultural year, especially in a rural
economy which is dominated by rainfed crop production. In rural Mozambique, crop
income accounted for 73 percent of total household income in 2002, with a 95 percent
conﬁdence interval from 70 to 75 percent. In 2005 crop income accounted for 65 percent
of total household income, with drought being the likely main cause of reduction of crop

income share in the latter period.

Fourth, income is more likely to be underreported than consumption expenditure
(Alderman, 1992). In sum, the general consensus among economists (based on both
theory and observation) is that household consumption varies less over time than does
income, and that consumption is therefore the preferred welfare indicator (since it
represents actual consumption obtained by the household, whereas income represents a

measure of potential consumption).

Nonetheless, as Walker et al. (2004) pointed out, when the research objective is to
analyze the determinants of rural poverty and their implications for agricultural
development, the analysis of data on consumption expenditure may not lead to speciﬁc,
actionable conclusions. Data on the relevant agricultural variables may be missing,
incomplete, or variation in data on consumption expenditure may be relatively small and
more difﬁcult to explain. This paper uses household income per adult equivalent2 as the

welfare measure, given the policy questions addressed.

1.2.2. THE CHOICE OF POVERTY LINES

The choice of poverty line is another important methodological decision for researchers.
A common deﬁnition of the poverty line is the minimum income/consumption level
below which a person is considered to lack adequate subsistence. Poverty rates will be

inﬂuenced by the poverty line that one chooses to use. The literature on poverty lines

 

2 To calculate the adult equivalent units in a given household, adults of either sex receive a weight of 1.0,
children aged 0-4 receive a weight of 0.4, and children 5-14 receive a weight of 0.5 (Deaton 1997).
6

identiﬁes four major methods of setting the poverty lines: the Food Energy Intake (F El)
approach, the Cost of Basic Needs (CBN) approach, the one-dollar-a-day criterion often
used for international comparisons, and a social subjective poverty line (Tarp et al. ,
2002). The main desirable characteristics of a poverty line are its consistency and
speciﬁcity. Consistency is related to making equal classiﬁcation for people with the same
living standard, whereas speciﬁcity is associated with its applicability to the communities
under consideration. Since poverty lines are important in poverty dynamics studies, it is
essential to perform some sensitivity tests to changes in poverty lines in order to assess

the robustness of the poverty measures’.

Amdt and Simler (2007) argue that the CBN approach provides more credible results in
the case of Mozambique, and they use this approach to construct the [AF (Household
Survey of Living Conditions) expenditure-based per capita poverty lines for 1996/97 and
2002/03 (MPF/UEM/IFPRI, 2004). However, per capita poverty lines fail to reﬂect the
needs of household members, as children and adults are given the same weight. An
alternative is to adjust the poverty lines taking into account both the household size and
composition. Fox et al. (2005) presents poverty lines per adult equivalent. These poverty
lines are based on the value of a basket of basic-need goods consumed by the poor in
2002, and they differ by region, reﬂecting regional consumption patterns and price

variations.

 

3 Poverty lines will be varied upward and downward by 20 % when running the Multinomial Logit models

presented in Chapter 4.

The present paper only uses the food component of the poverty lines, calculated by Fox et
al. (2005). For a country as poor as Mozambique, a poverty line that is based on the
intake of food to reach a recommended caloric dietary allowance is likely to be
meaningful marker of rural welfare (Walker et al., 2006), since food represents such a
large part of rural expenditure. For each region, the food poverty line is constructed by
determining the food energy intake requirements for the reference population (the poor),
the caloric content of the typical diet of the poor, and the average cost (at local prices) of
a calorie when consuming that diet. The food poverty line is calculated as the product of
the average daily calorie requirement per adult equivalent and the average price per

calorie.

Poverty lines were inﬂated to 2004/05 prices, using [AF 02 consumption quantities from
the consumption basket for each poverty region. The baskets from 2002 were revalued
using SIAM (Portuguese acronym for Agricultural Market Information Systems) prices
from 2002. Then the cost of an identical consumption basket was computed using SIMA
prices from 2005, and hence, the consumption quantities are the weights for the prices4.
Table 1 presents the poverty lines by province and year. Poverty lines vary considerably
among provinces, with the province of Sofala having the lowest poverty line in 2005. The
inﬂators constructed to update the poverty lines show no price inﬂation in Sofala and
Tete. The reason for this is that the SIAM price of both maize grain and ﬂour fell in rural
markets of these provinces from 2002 to 2005, and maize grain and ﬂour account for 50

percent of the food basket in Sofala and 60 percent in Tete (Mather et al. , forthcoming).

 

4 Details of the construction of the inﬂator by Mather et al. (forthcoming) are found in appendix Al.
8

Table 1. Food consumption expenditure poverty lines per adult equivalent by province

 

 

 

 

 

 

 

 

 

Food poverty line per adult equivalent

Province Meticais (MTN)/year US $Iyear

2002 2005 2005
Niassa 2660 3309 135
Cabo Delgado 2660 3492 143
Nampula 2179 2965 121
Zambézia 2048 2254 92
Tete 2736 2564 105
Manica 2736 2920 119
Sofala 2048 2014 82
Inhambane 3240 4568 186
Gaza 3240 4415 180
Maputo Prov. 5619 7136 291
Sources: Exchange rate is $1 .00 US = 24.50 MTN (Unofﬁcial - July 2005); Ofﬁcial poverty lines from National

Statistic Institute (INE)

1.2.3. DISTINCTION BETWEEN TRANSITORY AND CHRONIC POVERTY

Another methodological issue that researchers have to deal with is how to distinguish
transitory from chronic poverty. The distinguishing feature of chronic poverty is its
extended duration. Though a two-period three-year panel is much better than a single
cross-section, it is still somewhat limited in allowing us to distinguish chronic from

transitory poverty (i.e., a ﬁve- to six-year panel would be preferable).

An alternative approach to identifying the non-stochastic component of income change is
to use the predicted household income per adult equivalent from a regression model that
captures the relationship between a household’s income and its characteristics, in order to
purge the inﬂuence of idiosyncratic (household-speciﬁc) transitory shocks (McKay and
Lawson, 2002). Thus, predicted income more closely approximates longer-term or

expected household income than actual income. This paper uses predicted household

income values to group households into four mutually exclusive poverty status
categories: (i) stayed poor in both periods; (ii) moved out of poverty; (iii) moved into

poverty; and (iv) stayed non-poor in both periods.

One possible way to distinguish between transitory and chronic poverty is as follows.
Households moving out of poverty whose predicted income and asset holdings in the
second period exceed the median predicted income and assets holdings of those staying
non-poor could be considered to have escaped poverty. Landholding size per adult
equivalent was used as a proxy to compare asset holdings of households staying non-poor
and those escaping poverty, because it is the most common asset among rural households,
unlike livestock ownership which varies considerably by province. Table 16 in chapter
three presents the proportion of households who meet these criteria, though it is not

possible to validate it with a two-period panel.

1.2.4. POVERTY ANALYSIS METHODS

A number of different approaches have been adopted to understand the factors associated
with chronic and transitory poverty, or with poverty transitions. Most studies complement
descriptive analysis with an explicitly multivariate approach, generally based on
econometric analysis. Econometric studies generally take two forms: (i) those modeling a
discrete dependent variable measuring dynamic poverty status, and (ii) those modeling
the (generally continuous) underlying variable measuring the welfare. The continuous
approach can be estimated using either the ﬁxed-effect model or the ﬁrst-difference

model, whereas the discrete approach is usually modeled as a multinomial logit model.

10

Various authors argue that the aggregation of a continuous variable into a discrete one
has the disadvantage of loss of information (Ravallion, 1996). Discrete poverty outcome
models such as multinomial logit regressions are also criticized due to the fact that the
choice of poverty line is arbitrary and the assumption regarding the Independence of
Irrelevant Alternatives (IIA) may not always be valid. However, the multinomial logit
model presents three advantages over the continuous regression approach. First, it may be
the case that variables that determine whether one occupies a position in the top, middle,
or bottom of the distribution are likely to differ (Trzcinski and Randolph, 1991). Second,
when studying income mobility, one could argue that small position changes are
negligible, and income changes become important only when the change is large enough
to make the household cross a decile or even quintile threshold (Verpoorten, 2003).
Third, a multinomial logit model can take into account the presence of nonlinearity in a
more ﬂexible way, and estimate differential effects across the distinct categories of losers

and winners compared to a reference category.

In turn, to avoid both the loss of information and the issue of which poverty line to use,
economists usually take a continuous dependent variable approach to understanding the
determinants of household income change. Fixed-effect and ﬁrst-difference models have
been explored by various authors to analyze factors associated with changes in poverty
and inequality (El-laithy et al., 2003; Datt and Jolliffe, 2005; Grootaert et al., 1995).
First-difference models correct for bias in the estimated coefﬁcients due to endogeneity

caused by omitted variable bias. They may also capture the effect of initial household

11

characteristics. Also, the ﬁrst-difference model is often used in an attempt to reduce
multicollinearity problems (Resnick and Bimer, 2006). When the panel data only cover
two-period observations, the ﬁxed-effect and the ﬁrst-difference model estimates and all

test statistics are identical (Wooldridge, 2002a).

In sum, depending on the research question addressed, the researcher can either use the
continuous or the discrete approach. The multinomial logit regression is used in this
paper to address the question regarding the factors associated with moving out of or into
poverty, whereas the ﬁrst-difference model is used to evaluate the determinants of
income change. The next section discusses the problem of attrition bias in panel data,
followed by a brief review of poverty research in Mozambique highlighting some of the

gaps in these studies.

1.2.5. ATTRITION BIAS

Panel data can provide a more accurate inference of model parameters. Hsiao (2005)
argues that panel data usually contain more degrees of freedom and less multicollinearity
than cross-sectional data; hence it improves the efﬁciency of econometric estimates.
Panel data also provide greater capacity for capturing the complexity of human behavior
than a single cross-section or time-series data set (Hsiao, 2005; Maluccio, 2000). For
instance, consider the example provided by Hsiao (2005) regarding the evaluation of the
effectiveness of social programs. Evaluating the effectiveness of certain programs using a
cross-sectional sample typically suffers from the fact that the ex-ante household

characteristics of those receiving treatment are different from those not receiving

12

treatment. Panel data has the advantage that it is possible to observe the before- and after-
effects of receiving the treatment for the same individual unit of observation, as well as
providing the possibility of isolating the effects of treatment from other factors affecting

the outcome.

A potential limitation of panel data is attrition bias in the household sample. Attrition bias
occurs when the probability of an observation not being revisited is non-random with
respect to observed or unobserved household characteristics. One must correct for panel
data attrition in order to obtain unbiased estimates. The use of Inverse Probability
Weights (IPW) has been proposed by Wooldridge (2002b) to address the issue of attrition
bias. In short, IPW will effectively minimize attrition bias if the observed characteristics
of the household are useful in predicting whether a household will be re-interviewed, and
unobserved characteristics are not strong predictors. Mather and Donovan (forthcoming)
provide an IPW correction factor for the data used in this study, based on their analysis of

prime-age adult mortality and poverty in Mozambique.

1.2.6. RELATED STUDIES IN MOZAMBIQUE

Datt and colleagues (2000) analyze the determinants of poverty in Mozambique, using
data from the consumption expenditure survey, [AF 96. Though this was a very detailed
study, it failed to provide actionable recommendations regarding the agricultural sector.
Walker et al. (2004) using rural household income data from the National Agricultural
Survey, TIA 02, have also analyzed the determinants of rural income and poverty in

Mozambique. The latter study was able to propose policy interventions speciﬁcally

13

related to agriculture, which supports the use of income data as the welfare measure when

the research question is to evaluate the determinants of rural poverty in Mozambique.

In their article on the determinants of household income and consumption in rural
Mozambique, Tschirley and Weber (1994) found that incomes and calorie consumption
were highly correlated with landholdings, and argued that landholding size would
continue to be a key determinant of household income and consumption for the
foreseeable future. In line with these ﬁndings, Jayne et al. (2003) also found a positive
association between landholding size and per capita income in several African countries
(including Mozambique), and acknowledged the importance of the initial distribution of

assets to the rate of economic growth and poverty reduction efforts.

Jayne et al. (2003) argue that under existing conditions, the ability of households in the
bottom per capita land quartile to escape from poverty directly through agricultural
productivity growth is limited by their constrained access to land and other resources. An
important recommendation drawn from their study is that the way out of poverty among
the land-constrained households is either to increase landholding size, or to have these
households pulled into productive rural off-farm and non-farm sectors. In Mozambique,
however, the distribution of wage income tend to be concentrated among higher income
households, both in absolute value and as a share of total income, and the reasons for this
pattern are structural in nature and thus not easily addressed in the short-run (Tschirley
and Benﬁca, 2001). The same authors suggested rural formal education as the most

obvious lever to increase the access of the poor to salaried income.

14

The role of landholding size in reducing poverty is also documented in Boughton er al.
(2005) and Boughton et al. (2006). Stagnant crop productivity and limited area expansion
appear to be key constraints to achieving more favorable distributional outcomes. While
various studies have pointed out the importance of landholding size to poverty reduction

efforts, the average farm size in Mozambique is still very lows.

In terms of inequality, James et al. (2006) found that all sections of society enjoyed a
rapid annual increase in consumption between 1996 and 2002, but the rate of growth in
consumption was slightly higher for richer households, which led to a moderate increase
in inequality at national level. Boughton et al. (2005) also found that grth in household
income was concentrated among the wealthier households. There is also evidence on
inequality regarding access to land (Jayne et al., 2001; Marrule 1998) and off-farm

opportunities (Tschirley and Benﬁca, 2001 ).

While the ﬁndings from these studies on the determinants of income and poverty
mentioned above are very informative, they have some limitations. Studies based on
cross-sectional data sets may suffer from a key econometric problem, which is the
omitted variable (or unobserved heterogeneity) bias, which can be reduced when using
panel data. For example, as Maluccio (2000) pointed out, rarely do surveys observe or
measure a family’s preferences and priorities for educating its children. It is quite likely

that families that put a high priority on education will perform additional work to obtain

 

5 The average landholding size in 2002 was 1.7 hectares, with a 95 percent conﬁdence interval between 1.6
and 1.8 hectares.
15

income needed to pay school fees. The same author argues that if we use cross-sectional
data alone to determine the effect of family income on education, we risk making
incorrect inferences, as families with the highest income may also be those that prioritize
education the most. This paper uses panel data, allowing us to control for unobserved

time-invariant household characteristics.

None of the available studies has empirically made the distinction between those moving
into and out of poverty, which is one of the focuses of this paper. Though national
poverty rates have decreased, available studies are unable to identify whether there are
households that suffer from chronic poverty. If appropriate policy action depends on
whether poverty is chronic or transitory, then it becomes crucial to make this distinction.
Little is known about the factors that can be explored to help the non-poor avoid falling
into poverty. The use of panel data provides the opportunity to deepen the understanding
of poverty dynamics in Mozambique and design more effective policy interventions. This
study addresses the gaps from previous studies by controlling for unobserved time-
invariant household characteristics and distinguishing the factors associated with moving
out of and into poverty, using the recently collected panel data set on rural household

income in Mozambique.

l6

CHAPTER II DATA SOURCES

The National Agricultural Survey of 2005, commonly known as TIA 05, was implemented
by the Department of Statistics within the Directorate of Economics of the Ministry of
Agriculture in 2005, for the agricultural season 2004/2005, covering the period from
September 2004 to August 2005. The sampling frame was developed from the Census of
Agriculture and Livestock of 2000. The sample was stratiﬁed by province and agro-
ecological zone. Figure 1 shows the distribution sampled districts in 2003 which are the
same as those for TIA 02. An additional 14 districts were included for TIA 05. In terms of
agricultural potential, the central and northern provinces are more favorable to crop
production, whereas the drier southern provinces are more favorable to livestock

production.

The TIA 05 survey represents the ﬁrst effort to collect a nationally representative panel
data set in rural Mozambique. It covered 94 out of 128 districts (Table 2). Data on small
and medium-sized farms were complemented by group interviews at the community
level, ﬁeld measurements, and a separate questionnaire for all large-sized farms. A total
of 4054 households were re-interviewed in 2005. Of the original TIA 02 sample, about
850 households could not be interviewed in 2005 due to various reasons, migration being
the most common. Only panel households are used in the analysis because poverty

dynamics studies require data from the same households across time.

17

Figure 1. Distribution of TIA sampled districts in 2003

”Tmﬁg- ,k
1'! . u ;

t

 

Table 2. National Agricultural Survey sample sizes

 

 

TIA02 TIA05 Panel
Households interviewed 4908 6149 4054
Number of districts 80 94 80
Primary sampling units 559 656 556

 

 

 

 

Notes: TIA02 covers the agricultural season 2001/02 and TIA05 covers 2004/05
Sources: TIA02 8 TIA05

Consistent with the earlier analysis of Walker et al. (2004), rural incomes are calculated
using TIA data as the value of own production and off-farm earnings, less any paid-out
costs. Income sources include (i) net crop income, (ii) livestock sales, (iii) off-farm self-
employment, net small-business income, (iv) off-farm self employment, resource-
extraction income, (v) off-farm agricultural wage income, (vi) off-farm non-agricultural
wage income, and (vii) remittance income. Rural household incomes from the 2001/02
agricultural season were inﬂated to reﬂect 2004/05 prices. For this purpose, price data
from [AF and SIAM (Agricultural Markets Information System) was used to generate the
inﬂator. Details on each component and the construction of the inﬂator by Mather et a].

(forthcoming) are found in appendix A1.

Data from F EWSNET (Famine Early Warning System Network) was used to estimate
days of drought during the main maize-growing season. The data set is comprised of
rainfall records at the district level. A look at the rainfall data reveals that the agricultural
season in 2005 was worse for rainfed crop production than the agricultural season in
2002, which may have contributed to lower crop income in 2005. Crop production is an

important source of total household income in rural Mozambique.

19

CHAPTER III DESCRIPTIVE ANALYSIS

This chapter covers the descriptive methods and results. Poverty is analyzed by province
and with respect to key household characteristics and other factors that may affect rural
household income in Mozambique. Inequality and income distribution changes are also

assessed. The next chapter will present regression methods and results.

3.1. POVERTY MEASURES

This paper uses the popular Foster-Greer—Thorbecke (F GT) measures of poverty, which
for a given population are deﬁned as

1 q Z-J’i a
(1) Pa =— Z (——)

Ni=1 Z

where a is the poverty aversion parameter,
N is the sample size,
y,- is the predicted (from OLS regressions) household income per adult equivalent,
2 is the food consumption expenditure poverty line per adult equivalent (AE), and

q is the number of households with predicted income/AB below the poverty line.

Three poverty measures are calculated based on three values of or. The headcount index
of poverty measures the proportion of households below the poverty line, i.e., the

incidence of poverty. When a=0 equation 1 is reduced to

(2) P0 =%

20

Poverty incidence from both years is then used to classify the households into four
mutually exclusive categories based on predicted household income per adult equivalent:
(i) moved out of poverty, (ii) moved into poverty, (iii) stayed poor, and (iv) stayed non-
poor (Figure 2). These groups will be used as the dependent variable in the multinomial

logit regression (MNL) model, and they are deﬁned as follows:

Moved out of poverty: if y,-1 < 2 but yiz > z
Moved into poverty: if y,- 1 > 2 but y;; < z
Stayed poor: if y,-, < z for all t
Stayed non-poor: if y,-, > z for all t

Figure 2. Poverty categories based on predicted incomes

 

2001/02 Agricultural Season

[ Poor J | Non-poor ]

2004/05 Agricultural Season

‘1' ‘1' \I' ‘1'
rm Non-poor | ‘ Poor I Non-poor |

Stayed poor Moved out of poverty Moved into poverty Stayed non-poor

 

 

 

 

 

 

 

 

 

 

The headcount index fails to capture the distance that individual income falls below the
poverty line (Jha and Sharma, 2003). Therefore, a second measure of poverty was used,
with (FL The poverty gap index is given by the aggregate income shortfall of the poor as

a proportion of the poverty line and normalized by the population size (equation 3).

21

q _ .
(3) $521623”

1:

 

)

The poverty gap index captures the intensity of poverty. To put it differently, the poverty
gap measures the total amount of income required to remove that poverty. Kakwani
(1980) pointed out two main drawbacks associated with this index: (i) it is completely
insensitive to the number of poor, and (ii) it does not take into account the inequality

among the poor.

A third measure of poverty, the squared poverty gap, is obtained when a=2. It measures

the severity of poverty and is given by:

lqz-y'z
4 P=-—— —-—’—
()2 N=1(Z)

i
By squaring the poverty gap for each household, this measure gives greater weight to

those that fall far below the poverty line than those that are closer to it.

3.2. STOCHASTIC DOMINANCE ANALYSIS

The choice of the poverty line is often discussed in the literature as one of the key
decisions to make when comparing welfare between two periods. If ﬁrst-order dominance
is achieved, the implication is that poverty incidence is unambiguously lower in one
period than in another, regardless of the poverty line that is used. Consequently, it is

appropriate to evaluate the robustness of our conclusions with regard to different poverty

22

lines, which we will do using the concept of stochastic dominance as applied in the

poverty analysis literature (Ravallion, 1992).

Stochastic dominance refers to a set of relations that may hold when comparing a pair of
distributions. For instance, given the distributions of rural household income per AB in
Mozambique in 2002 and 2005, one can evaluate whether the distribution in the latter
period dominates the distribution from the ﬁrst period by plotting both cumulative
distributions in the same graph. Figure 3 shows that the ﬁrst-order stochastic dominance

does not hold since the cumulative distributions intersect".

Figure 3. F irst-order stochastic dominance

.7 .8 .9
J r 1

.6
A

.4
1

Cumulative percentage of the population
.3 5
l 1

 

 

 

0 5000 10000 15000
Household inoome/AE

2002 — - -- 2005

 

 

 

 

6 Figure 3 excludes the top 5% from each year.
23

The ﬁrst-order stochastic dominance test corresponds to a comparison of poverty
incidence (head count) between the two periods. Thus, we cannot unambiguously say that
poverty incidence in one period is lower (or higher) than in the other period. Since the
ﬁrst-order stochastic dominance was not achieved, we can test for second- and third-
order stochastic dominance, which correspond to the poverty gap and squared poverty
gap measures respectively. Also, if necessary, one can go on to test higher-order
dominance, though the interpretation of these measures becomes less clear (Ravallion,

1992).

A rough approximation to second- and third-order stochastic dominance would be to
calculate poverty gap and squared poverty gap for each province for 2002 and 2005 and
determine whether these poverty measures have declined or risen for all provinces. This
comparison would reﬂect only a single set of provincial poverty lines. Alternatively, one

could repeat the comparison for a range of different provincial poverty lines.

Table 3 shows that poverty incidence at the national level increased by almost 1
percentage point between 2002 and 2005. However, sensitivity analysis indicates that this
ﬁnding is not robust to the poverty line that is used. By increasing poverty lines we ﬁnd
that poverty incidence is higher in the ﬁrst period. In contrast, by decreasing poverty lines
we ﬁnd that poverty incidence is lower in the ﬁrst period. In line with ﬁndings from
Figure 3, ﬁrst-order stochastic dominance was not achieved since using a different

poverty line we rank the distributions differently. In other words, within the range of

24

poverty lines presented in Table 3, poverty incidence at the national level did not change

unambiguously between the two periods.

Table 3. Poverty incidence by province and year with different poverty lines

 

 

 

 

 

 

. Decrease poverty Decrease poverty Increase poverty Increase poverty
Acma' We“ ""9 line by 20% line by 50% line bv 20% line by 50%

Province 2002 2005 2002 £05 2002 2005 2002_ _2205 2002 2005
Niassa 82.6 79.0 73.6 67.9 34.1 38.0 88.7 87.1 94.6 92.0
Cabo Delgado 94.3 82.2 87.9 70.5 56.5 43.8 98.0 88.0 99.5 92.6
Nampula 79.9 88.4 64.2 81.7 27.3 56.2 89.2 93.3 93.1 96.9
Zambézia 81.0 87.7 69.7 82.3 39.7 51.4 86.7 91.6 92.7 95.5
Tete 72.9 81.3 64.9 75.6 41.2 57.1 79.4 84.3 86.4 90.1
Manica 84.1 84.9 78.1 73.6 54.2 54.0 89.6 87.3 91.8 92.4
Sofala 81.9 55.4 70.3 45.8 45.4 18.6 88.4 63.1 94.9 73.9
Inhambane 82.1 80.4 76.1 66.9 43.1 38.8 90.9 88.3 96.2 92.7
Gaza 95.2 92.5 91.6 84.6 71.0 58.3 96.4 94.9 97.5 97.9
Maputo 89.8 93.8 85.7 89.7 63.1 70.7 92.8 94.8 95.9 97.5
National 83.2 84.1 73.4 76.0 43.0 49.6 89.4 88.9 93.8 93.4

 

 

 

 

 

 

 

 

 

 

 

Source: TIA02 & TIA05

At the provincial level, we can unambiguously say that poverty incidence has decreased
(or increased) for some provinces. For instance, regardless of the poverty line that is
used, poverty incidence in the provinces of Cabo Delgado, Sofala, and Inhambane is
lower in the second period, whereas for Nampula and Zambézia provinces poverty

incidence is higher in the second period regardless of which poverty line is used.

Within the range of poverty lines used, the depth and severity of poverty at the national
level were worse in 2005 than in 2002. Poverty gap and squared poverty gap increased
signiﬁcantly in Nampula and Zambézia, regardless of the poverty line that one chooses to
use. Poverty depth and severity have signiﬁcantly decreased in Sofala, regardless of the
poverty line that is used. Results of poverty gap and severity are presented in the

appendix (Tables A1 and A2).

25

3.3. INCOME MOBILITY AND INEQUALITY

Figure 4 shows household income distributions in 2002 and 2005. Ideally we would want
the entire distribution to shift to the right. The distribution in the latter period has become
more spread. The fact that the tails of the distribution are longer in 2005 reveals that the
poorest households have decreased their incomes, and the wealthiest households have
increased their incomes. This pattern is also observed in Table 4, which presents mean

and median household income by quintile and gender of household head.

Figure 4. Household income distribution per AB in 2002 and 2005

 

 

 

 

 

v: _

co ..
.é‘
(D

o .1 _____

I I I I I
-5 0 5 10 15
Log of total HH income/AE
2002 —— 2005 """""

 

 

 

At the national level, mean household income/AB has signiﬁcantly increased from 2002
to 2005 at the 0.05 level. In contrast, median income/AB has decreased, as most of the
increase in household income accrued to those household among the highest income

quintiles. Both the mean and median household incomes have decreased among the two

26

lowest income quintiles. Moreover, male-headed households have signiﬁcantly higher

income levels than female-headed households in both periods.

Table 4. Mean and median household income per AE by quintiles and head’s gender

 

 

 

 

 

 

 

 

Mean Meticais) Median Meticais)

2002 2005 2002 2005

Lowest 494 376 469 351

£233?” 2 1,086 986 1,027 901
quintiles/AE in 3 1,792 1,822 1,736 1,669
each year ' 4 2,941 3,310 2,728 2,960
Highest 8,221 10,243 5,673 6,753

Gender of Female 2,189 2,590 1,359 1,389
household head Male 3,136 3,634 1,877 1,848
National 2,906 3,348 1,751 1,708

 

 

Sources: TIA02 and TIA05

The income transition matrix helps understand the pattern of income change over time. It
shows what fraction of households belonging to a given base-period income group end up
in a given ﬁnal-period income. If all the households had remained in the same income
quintile, then the transition matrix would be an identity matrix and the diagonal elements
would sum to 100 percent, reﬂecting the absence of any income mobility. Table 5 shows
the household transition matrix from 2002 to 2005. From those that were in the lowest
household income quintile/AB in 2002, about 12 percent have moved to the highest
quintile in 2005. This result indicates that the poor are not necessarily doomed to remain

at the bottom.

27

Table 5. Transition matrix of household income quintiles per AE by year

    
   

 
 

  
   
   

    
  

    
   

20.5
21.6
19.6

18.9
22.3
24.7

13.0
16.6
22.5

Household
income quintiles 3
per AE, 2002

100

Figure 5 illustrates quintile movements between the two periods. Households moving

four quintiles backward were located in the highest household income/AB quintile in
2002 but belong to the lowest income/AB quintile in 2005. Likewise, households
switching four quintiles forward are those that were the poorest in 2002, but are now

among the wealthiest. About one quarter of the households remained in the same quintile.

Figure 5. Number of household income per AE quintiles changed

8..

 

 

-2 0 2
Number of household income/AE quintiles changed

28

The fact that the mean income has decreased among the lowest income quintiles suggests
an increase in inequality. One of the common ways of evaluating changes in inequality
between two periods is to use Lorenz curves. The Lorenz curve plots the percentage of
total income earned by various portions of population when the population is ordered by
the size of their incomes. If Lorenz curves intersect, then it is not possible to make an
unambiguous ranking. Inequality increases with increases in the distance between the

Lorenz curve and the 45° line.

The Gini coefﬁcient is bounded between 0 and 1, the upper bound corresponding to the
highest inequality. While not without limitations this coefﬁcient is widely used in poverty
literature for purposes of comparison. Table 6 presents Gini coefﬁcients and percentile
ratios for each year. Gini coefﬁcients are obtained from reported household income/AB.
The Gini coefﬁcient increased from 0.54 to 0.60, which is very high compared to other

countries in Africa7.

Table 6. Percentile ratios of household income/AB and Gini coefﬁcient by year

 

 

 

Percentile ratio Ratio of reported household income/AB
_ 2002 2005
Percentile ratio P90/P10 19.33 22.26
Percentile ratio P90/P50 3.24 4.69
Percentile ratio P10/P50 0.17 0.21
Percentile ratio P75/P25 4.34 5.18
Percentile ratio P75/P50 1.96 2.19
Percentile ratio P25/P50 0.45 0.42
Gini coefﬁcient 0.54 0.60

 

 

 

Sources: TIA02 & TIA05

 

7 Using adult equivalent units, Simkins (2004) estimated the Gini coefﬁcient for South Aﬁ'ica to be 0.64.
South Africa is considered to be among the most unequal in the world (World Bank, 2006). In the late
19905, Gini coefﬁcients for rural Ghana, Zambia and Zimbabwe were 0.37, 0.48 and 0.57, respectively
(Christiaensen er al. 2002).

29

Percentile ratios are also useful in analyzing inequality. They summarize the relative
distance between two points on the income distribution. To illustrate the full spread of the
income distribution, the percentile ratio needs to refer to points near the extremes of the
distribution. The P90/P10 ratio in Table 6, for example, shows that the ratio between the
wealthiest and the poorest households’ income has increased. The P75/P25 ratio better
illustrates the magnitude of the range within which the incomes of the majority of the
population fall. The P75/P50 and P25/P50 ratios focus on comparing the ends of the

income distribution with the midpoint.

Figure 6 shows graphically that inequality has indeed increased between the two periods.
The ratio of the area between the Lorenz curve and the line of absolute equality (the

diagonal) to the area underneath the diagonal is the Gini coefﬁcient in Table 6.

Figure 6. Lorenz curves at national level using reported income in 2002 and 2005

 

 

 

 

-
-
-
-
‘
-

 

0 .4 . .8 1
Cumulative population share, p
2002 -------- 2005
...................... 45°

 

 

 

30

Table 7 provides the distributional summary statistics by income quintile. At the bottom
of the distribution, the maximum household income/AB is about 617 MTn (Meticais) in
2002, and 647 MTn in 2005, which corresponds to 37 percent of the median income/AE
in 2002, and 39 percent in 2005, respectively. Despite the slight increase, the ratio of the
income share between the top income quintile and the poorest quintile is 27 in 2002 but
32 in 2005, indicating that increases in household income were faster among the
wealthiest households. In addition, the maximum household income/AB in each income
quintile is about twice the maximum household income/AB in the immediate lower

quintile.

Table 7. Distributional summary statistics using reported incomes, 2002 - 2005

 

 

 

 

 

 

 

 

 

 

 

Income Maximum HH lncome/AE in each quintile % of the median % Share of All Income
Quintile/AE 2002 2005 2002 2005 2002 2005

1 617 647ﬁ 36.5 39.4 2.2 2.0

2 1290 1169 76.3 71.3 6.7 4.9

3 2189 2207 129.5 134.5 12.1 9.6

4 3968 4409 234.8 268.8 20.7 17.8
_ 5 NA NA NA NA 58.4 65.8
NA: The ﬁrst 4 cells in of the last row n Table 7 are blank because those cells would correspond to the maximum

household income/AE in our sample, and comparing these values with the median is not very useful since the
maximum household income could be an outlier.

Share = quintile group share of total household income per adult equivalent

Sources: TIA02 & TIA05

If the distribution of household income/AE was equal, the share of all household income
for each quintile would be 20 percent. However, the share of all household income is
only about 2 percent for the poorest households (quintile 1). In line with Figure 4, Table
6, and Figure 6, Table 7 suggests that income distribution is more skewed in the latter

period than it was in 2002.

31

3.4. COMPARISON OF AGRICULTURAL SEASONS

Rainfall is a critical factor affecting crop production, which contributes the largest share
of income for most rural households in Mozambique. To compare the quality of the two
agricultural seasons, an estimate of days of drought during the principal maize-growing
season was calculated using rainfall data from both seasons. The estimate is based on a
water-balance model from Mather et al. (forthcoming). Days of drought refer to total
days when cumulative precipitation is inadequate for healthy crop growth. Figure 7
shows that in 2005 there were more days of drought than in 2002. With the exception of
the central provinces of Sofala and Tete, all other provinces experienced a signiﬁcant
increase in days of drought (P-value < 0.01), which contributed to relatively lower crop

production in the 2004/05 agricultural season.

Figure 7. Days of drought by province and year

EL

F

    
 
 

V131 1:»

Nlassa C.Delg

 

   

adoNampuIaZambézia Tete Manlca Sofala Inhambane Gaza Maputo

 

32

Figure 8 shows that there were more days of drought in 2005 than in 2002 for all four
household poverty categories. In addition, households staying poor experienced more
days of drought than the other poverty categories of households. Nevertheless, Figure 8
could be a reﬂection of differences in geographical factors. Days of drought were
therefore computed at the district level. Ideally, days of drought at the village level would
be a better indicator in explaining movements from and into poverty, but data is not

available at this level of disaggregation.

Figure 8. Mean days of drought by poverty category in 2002 and 2005

20 3O 40
l 1

Estimates of days of drought

10
l

 

 

Stayed poor Moved out of poverty Moved into poverty Stayed non-poor

2002 - 2005

 

 

 

Days of drought have a signiﬁcant impact on household income, given that crop income
accounted for about 73 percent of total income in 2002 and 65 percent in 2005 (Table 8).

Because there were more days of drought in 2005, the expectation is for the share of crop

33

income to decrease and the share of non-farm income to increase between periods. This is

indeed the case, except among households falling into poverty. This suggests that

households moving into poverty are those relatively more dependent on rainfed

agriculture as an income source, and hence more vulnerable to weather shocks. However,

the increase in non-farm income should not be interpreted as a structural change since the

two seasons differ in terms of rainfall, and since the data only cover a three year-period.

Table 8. Shares of different income sources by poverty category and year

 

 

 

 

 

 

Income source (%)

Poverty category Livestock Remittance Crog Wa es Self-emoloyment

2002 2005 2002 2005 2002 2005 2002 2005 2002 2005
Stayed poor 2.8 2.6 4.0 5.2 76.0 66.1 6.6 10.8 10.6 15.2
Moved out of poverty 3.0 2.7 3.3 4.2 68.1 54.4 8.5 13.9 16.9 24.7
Moved into poverty 2.2 2.1 2.5 4.8 63.5 68.1 ‘ 11.6 11.0 20.1 13.9
Stayed non-poor 2.1 2.3 1.4 2.8 61.3 54.0 15.7 17.2 19.4 23.7
National 2.7 2.5 3.6 4.9 72.9 64.6 8.0 11.6 12.7 16.3

 

 

 

 

 

 

 

 

 

 

Sources: TIA02 8 TIA05

Crop income is even more important among the poorest households. Table 9 shows that

crop income accounted for about 83 percent of total income among the lowest income

quintile/AB in 2002, and 72 percent in 2005. Thus, the effects of drought will be more

severe among the poorest households. In line with ﬁndings from Tschirley and Benﬁca

(2001), off-farm income is concentrated among high income households.

34

Table 9. Shares of each income source by quintile and year

 

 

 

Income source Livestock Crtm Remittance Salem Self-em 0|oyment
Year 2002 2005 200_2 2005 2002 2905 2002 2005 2002 2005
Lowest 3.3 3.8 82.7 71.8 4.3 7.0 2.1 6.9 7.6 11.2
5.2::de 2 2.9 3.0 81.0 72.4 3.0 4.7 3.6 8.1 9.4 12.2
. 11 WE 3 2.5 2.5 77.2 65.4 3.4 4.5 6.4 11.3 10.5 16.4
q”'” es 4 2.5 2.2 70.3 62.6 3.1 4.4 11.5 13.2 12.5 18.3

 

‘"ea°“Yea' Highest 2.2 1.4 52.9 51.4 4.4 4.0 16.0 19.2 24.4 24.4

National 2.7 2.5 7% 64.6 3.6 4.9 8.0 11.8 127' 16.3
Sources: TlA02 8 TlA05

 

 

 

 

 

 

 

 

 

 

 

 

3.5. POVERTY IN RURAL MOZAMBIQUE

Table 10 presents poverty headcount, poverty gap, and squared poverty gap at the
national level. In line with the results presented in Table 3, poverty headcount did not
increase signiﬁcantly between 2002 and 2005 (P-value = 0.4818). The poverty gap and
squared poverty gap measures have signiﬁcantly increased between the two periods

however.

Table 10. Poverty measures using predicted household income/AB by year

 

 

 

. . Value of poverty indicator at national level
Poverty Indicator 2002 I 2005 I Change
Head count 83.2 84.1 0.9
Poverty gap 40.6 I 43.7 I 3.2
Squared poverty gap 23.5 26.5 3.0

 

 

Notes: Change = poverty indicator in 2005 minus poverty indicator in 2002
Sources: TlA02 8 TIA05

The next sections analyze different dimensions of poverty in rural Mozambique. We start
by looking at spatial dimension of poverty, followed by the demographic dimension. We
then analyze the role of off-farm income in reducing rural poverty, and the relationship
between landholdings and poverty. The last section of this chapter looks at the

relationship between agricultural practices and poverty.

35

3.5.1 SPATIAL DIMENSION OF POVERTY

Table 11 shows that poverty has a geographical dimension. To test whether the

differences in poverty across provinces are signiﬁcant, we compute the )8 statistic, which

has a value of 312. The probability that a random variable drawn from a 78 distribution

with 27 degrees of freedom is greater than 312 is smaller than 0.01. Thus, we reject the

null hypothesis of independence between the province and poverty headcount between

the two periodsg.

Table 11. Poverty transition matrix by province

 

% Head count based on food poverty lines per adult equivalent

 

 

 

 

 

 

 

 

 

 

Poor in 2002 that Non-poor in 2002 that Sa(:2;>le
Province ' Stayed poor Moved out of poverty Moved into pove Stgyed ign-poor
Niassa 70.1 12.2 9.1 8.7 5.6
Cabo Delgado 78.5 15.7 3.6 2.2 11.0
Nampula 74.0 6.2 14.5 5.3 21.6
Zambézia 73.5 7.6 14.2 4.6 23.5
Tete 66.3 7.6 15.4 10.7 8.7
Manica 76.4 7.7 8.5 7.5 6.8
Sofala 49.2 32.2 7.4 11.2 5.3
Inhambane 70.6 11.6 10.7 7.2 8.2
Gaza 90.9 5.0 2.0 2.2 7.0
Maputo prov. 88.9 4.3 5.4 1.4 2.3
National 73.6 9.8 10.8 __ 5.8 100.0
_——Pearson: Uncorrected x m, = 312.0579; Design-based F(12_—2—.46, 293—12): 3.6—1t_P = 0.0000

Sources: TlA02 & TIA05

Non-poor households were least likely to fall into poverty and poor households more

likely to move out of poverty in the central province of Sofala. In contrast, the southern

provinces of Maputo and Gaza have the highest percentage of households remaining

poor. The cases of Nampula and Zambézia provinces are of particular concern since they

constitute about 45 percent of the whole TIA sample, and most of the households moving

into poverty are located in that region. Previous study has also shown that the

 

8 All poverty measures presented in this section are based on predicted household income/AB.

performance of Nampula is below the national average (Boughton et al. , 2006), which

may suggest the existence of poverty traps.

3.5.2 DEMOGRAPHIC DIMENSION OF POVERTY

Besides geographical location, demographic characteristics such as gender of household

head are also related to poverty status. The percentage of households remaining poor is

higher among households headed by a female (Table 12). Furthermore, male-headed

households are more likely to have moved out of poverty than the female-headed ones.

Table 12. Poverty transition matrix by gender of household head

 

 

 

 

 

% Head count based on rcaplta food povery lines % of

:35“ Of household Poor in 2002 that Non-poor in 2002 that Total
Stayed poor Moved out of poverty Moved into poverty Stayed non-poor HH

Female in both years 89.0 3.4 6.1 1.5 22.3
Now male headed 71.1 21.3 5.9 1.8 1.9
Now female headed 77.5 6.1 14.1 2.3 5.1
Male in both years 68.5 11.8 12.2 7.5 70.7

Total 73.6 9.8 10.8 5.8 100.0

 

 

 

 

 

 

Pearson: Uncorrected 18(9) = 173.3602; Design-based F(6.93,1275.32)= 15.4647 P=0.0000

Sources: TIA02 & TIA05

In general, household size has increased for all households with the exception of those

moving out of poverty (Table 13). The difference in household size and composition

between households moving out of poverty and those staying non-poor was signiﬁcant in

2002, but not signiﬁcant in the latter period. Table 13 also shows that mean household

income/AB in 2002 was signiﬁcantly higher among households moving out of poverty

compared to those who stayed poor. Likewise, households moving into poverty had lower

mean and median incomes than those staying non-poor.

37

Table 13. Household size and composition in each year by poverty transition category

 

 

 

 

 

 

 

Poor in 2002 that... Non-poor in 2002 that... National
Stayed Moved out Moved inIo Stayed non-

poor of poverty poverty poor mean
2002 variables
Mean number of hh members 5.1 5.2 4.5 4.5 5.0
Mean household size (AE) 3.8 3.9 3.5 3.5 3.7
Mean number of members aged 0-4 0.9 0.9 0.7 0.6 0.8
Mean number of members aged 5-14 1.5 1.5 1.1 1.3 1.5
Mean number of members aged 15—59 2.4 2.6 2.4 2.5 2.4
Mean number of male members aged 15-59 1.1 1.3 1.2 1.3 1.1
Mean number of female members aged 15-59 1.3 1.3 1.3 1.3 1.3
Mean predicted HH income/AE 1,558 1,820 4,262 6,678 2,173
Median predicted HH income/AE 1,560 1,817 4,318 6,819 2,191
2005 variables
Mean number of hh members 5.4 5.1 4.8 5.0 5.3
Mean household size (AE) 4.0 3.8 3.7 3.9 3.9
Mean number of members aged 0-4 0.8 0.8 0.8 0.8 0.8
Mean number of members aged 5—14 1.8 1.6 1.3 1.4 1.7
Mean number of members aged 15-59 2.5 2.5 2.5 2.6 2.5
Mean number of male members aged 15-59 1.1 1.3 1.2 1.3 1.2
Mean number of female members aged 15-59 1.4 1.2 1.3 1.3 1.3
Mean predicted HH income/AE 1,478 4,784 1,679 7,846 2,192
Median predicted HH income/AE 1,479 4,776 1,680 7,892 2,199

 

 

 

Sources: TlA02 & TIA05

3.5.3 OFF-FARM INCOME AND POVERTY

There are signiﬁcant differences across provinces in terms of off-farm labor
opportunities. On average, at least one member of each household is self-employed in
Sofala in 2005, compared to one self-employed member for every two households in
Nampula and Zambézia. In general, households dwelling in Northern provinces have

fewer non-farm employment opportunities.

Table 14 presents information on household participation in off-farm income sources in
relation to transition in poverty status between survey periods. Whereas in 2002 there

was no signiﬁcant difference in the number of salaried members between households

38

moving out of poverty and those staying poor, households moving out of poverty by 2005
had signiﬁcantly more salaried members than those staying poor (P-value<0.0001). With
the exception of those moving into poverty, the number of salaried or self-employed
members increased among all other poverty groups between 2002 and 2005 (Table 14).
Likewise, the percentage of households whose head is salaried or self-employed has
signiﬁcantly increased, even among those who stayed poor, but decreased among
households moving into poverty. Though agriculture is very important in rural
Mozambique, especially among the poorest households, Table 14 shows that non-farm
income is crucial in lifting households out of poverty and in preventing the non-poor
from falling into poverty. The loss of non-farm income is associated with falling into

poverty.

Table 14. Household farm and non-farm employment by poverty transition category

 

 

 

 

Poor in 2002 that Non-poor in 2002 that .

Stayed Moved out Moved into Stayed non- N222“
_ poor of poveﬂy ove oor
2002 variables
Number of salaried members (Mean) 0.2 0.2 0.5 0.4 0.2
Number of self-employed members (Mean) 0.4 0.5 0.9 0.8 0.5
Households whose head is salaried (%) 11.7 12.0 30.0 27.4 15.1
Households whose head is self-employed (%) 26.5 37.1 56.1 57.8 32.6
Agriculture is the head's main activity (%) 87.3 80.2 79.1 69.2 84.7
Agriculture is head's secondary activity (%) 9.9 17.1 19.1 27.1 12.6
Agriculture is not head's 3% (%) 2.8 2.7 1.9 3.7 2.7
2005 variables
Number of salaried members (Mean) 0.4 0.8 0.4 0.6 0.4
Number of self-employed members (Mean) 0.6 1.3 0.5 1.2 0.7
Households whose head is salaried (%) 22.1 36.0 25.7 36.1 25.1
Households whose head is self-employed (%) 37.2 70.8 36.8 70.3 42.4
Agriculture is the head's main activity (%) 86.3 77.2 85.8 70.3 84.4
Agriculture is head's secondary activity (%) 10.3 21.5 11.7 27.3 12.6
Agriculture is not head's activity (%) 3.4 1.3 2.6 2.5 3.0

 

 

 

 

 

 

Sources: TlA02 8r TIA05

39

We now look at the contributions of each of the non-farm employment types in lifting
poor households out of poverty and helping the non-poor avoid falling into poverty.
Wage income is divided into three categories: (i) unskilled agricultural wage; (ii)
unskilled non-agricultural wage; and (iii) skilled non-agricultural wage. Self-employment
is divided into four categories: (i) extraction of forestry or fauna products with low
returns; (ii) extraction of forestry or fauna products with higher returns; (iii) other self-
employment activities with low barriers to entry but low returns; and (iv) other self-
employment activities with higher returns. A list of non-farm activities in each of the 7

categories is found in the appendix A3.

Table 15 presents mean values from non-farm income generating activities for all
households in a given poverty category. In general, mean household income from all
types of non-farm activities has increased for all poverty groups except those households
moving into poverty. In terms of wage income, skilled non-agricultural wage income has
the highest mean value in both years, particularly among those who moved out of poverty
or stayed non-poor. Among those households moving out of poverty, mean household
income from self-employment in low return forestry and fauna activities in 2005 was 7.6
times higher than in 2002, while mean household income from other self-employment
activities with high return was 4.5 times. Losing other self-employment income is also
highly associated with moving into poverty. Among households staying non-poor, mean
household income from forestry and fauna activities is at least 7 times higher in 2005
than in 2002. In addition, mean household income from high return other self-

employment activities among those who stayed non-poor is 10.4 times higher in 2005

40

relative to the mean from those moving into poverty, and 14.5 times higher relative to

households staying poor.

Table 15. Mean non-farm income (MTn) by poverty transition category

 

Poor in 2002 that...

 

Non-poor in 2002 that...

 

 

 

 

 

 

 

 

Move ou 0 W110 Stayed non- National

Stayed poor poverty poverty poor mean
Unskilled ag wage income in 2002 163 164 246 536 194
Unskilled ag wage income in 2005 267 501 152 617 297
% change of the mean 63.8 206.1 -38.3 15.2 53.0
Unskilled non-ag wage income in 2002 240 270 302 443 275
Unskilled non-ag wage income in 2005 307 571 273 621 379
% change of the mean 27.9 111.3 -9.8 40.2 37.7
Skilled non-ag wage income in 2002 591 867 1,824 3,489 1,106
Skilled non-ag wage income in 2005 902 2,155 1,550 4,325 1,424
% change of the mean 52.5 148.7 -15.1 24.0 28.8
Self-employment forestry and fauna - low in 2002 79 51 165 28 82
Self-employment forestry and fauna - low in 2005 236 390 93 255 240
% change of the mean 200.7 662.3 43.6 815.5 192.7
Self-employment forestry and fauna - high in 2002 374 542 811 304 426
Self—employment forestry and fauna - high in 2005 386 424 573 2,215 509
% change of the mean 3.2 -21.8 -29.4 627.5 19.6
Other self-employment - low in 2002 329 590 911 870 481
Other self-employment - low in 2005 412 1,063 521 1,093 543
% change of the mean 25.4 80.3 -42.8 25.6 12.9
Other self-employment - high in 2002 706 1,115 2,372 9,904 1,698
Other self-employment - high in 2005 799 5,010 1,112 11,548 1,851
% change of the mean 13.2 349.2 -53.1 16.6 9.0

 

 

 

 

 

 

Sources: TlA02 8 TIA05

Table 15 conﬁrms the hypothesis that the non-poor are less vulnerable to natural

calamities such as the drought that occurred in 2005. Households staying poor were able

to increase their income from all off-farm activities, but the increase was not great

enough to lift them out of poverty. Households moving out of poverty had a large

percentage increase from almost all off-farm activities except high return extraction of

forestry and fauna resource activities. In contrast, households moving into poverty

41

experienced decreases in mean income from all off-farm sources. Thus, diversiﬁcation
into non-farm income opportunities is important for reducing poverty in rural

Mozambique.

3.5.4 LANDHOLDING SIZE AND POVERTY

Figure 9 shows total landholding size in hectares by poverty transition group. With the
exception of those moving into poverty, average landholding size has increased among
all other poverty categories, including those who stayed poor. Landholding size has

signiﬁcantly decreased among households falling into poverty, however.

Figure 9. Mean total landholding size in hectares by poverty category and year

2 3
l l

Landholding size in Ha

1
t

 

   

Stayed poor Moved out of poverty Moved into poverty Stayed non-poor

L 2002 - 2... l

 

 

42

In both years, households staying non-poor had signiﬁcantly more land than those
moving into poverty, and those moving out of poverty had signiﬁcantly more land than

households staying poor.

Returning to the discussion of distinguishing chronic from transitory poverty, Table 16
shows the number of households among those who moved out of poverty that could
meaningfully be said to have escaped poverty. These households meet the criteria that
their predicted income/AB and their total land holding/AB in the latter period exceed the
median income/AB and land holdings of those staying non-poor in each province. About
20 percent of those moving out of poverty can be considered to have escaped poverty on
the basis of these criteria. However, meeting these criteria does not guarantee that these

9“

households escape from poverty” will be permanent, and it would be interesting to

follow these households in future surveys9.

Table 16. Number of households “escaping” poverty by province between 2002 and 2005

 

 

 

Province Moved out of poverty Escaped poverty Percent
Niassa 26 4 15.4
Cabo Delgado 62 17 27.4
Nampula 30 4 13.3
Zambézia 52 6 11.5
Tete 34 4 11.8
Manica 22 2 9.1
Sofala 90 17 18.9
Inhambane 37 17 45.9
Gaza 35 9 25.7
Maputo Prov. 15 2 13.3
National 403 82 20.3

 

 

 

 

Sources: TIA02 8 TlA05

 

9 The results from Table 16 are not population-weighted, because these households represent a small

fraction of the whole population.
43

3.5.5 AGRICULTURAL PRACTICES AND POVERTY

In terms of agricultural practices and access to services, two distinct patterns are
noteworthy. First, the use of agricultural inputs (fertilizer, pesticides, and animal traction)
has increased among households moving out of poverty, even though it declined
nationally and among those moving into poverty (Table 17). Second, the use of fertilizers

also increased among households staying non-poor.

Table 17. Agricultural technology use and access to services by poverty transition category

 

 

 

 

 

 

 

 

 

Poor in 2002 that Non-poor in 2002 that
Moved out of Moved into Stayed non- National
Stayed poor poverty poverty poor

2002 variables

Fertilizer use (%) 2.2 4.8 5.7 17.1 3.8
Pesticide use (%) 5.9 6.8 8.9 13.7 6.7
Animal traction use (%) 10.4 11.3 14.4 19.8 11.1
Hired permanent labor (%) 0.5 0.5 6.8 15.3 2.1
Hired seasonal labor (%) 8.4 18.2 39.8 55.8 15.5
Grow cotton (%) 6.2 9.2 12.1 9.3 7.2
Grow tobacco (%) 2.1 3.5 9.4 16.0 3.8
Received price information (%) 30.7 34.3 55.6 49.9 34.5
Access to extension services (%) 11.4 17.6 21.9 21.9 13.7
Membership to ag. Association (%) 3.0 3.1 7.4 9.7 3.8
2005 variables

Fertilizer use (%) 1.7 6.8 5.0 18.5 3.5
Pesticide use (%) 4.0 9.5 5.5 12.8 5.1
Animal traction use (%) 8.6 11.5 5.3 14.5 8.6
Hired permanent labor (%) 0.7 4.5 2.0 9.3 1.6
Hired seasonal labor (%) 9.5 59.1 15.9 64.0 18.0
Grow cotton (%) 4.6 9.8 7.4 9.6 5.6
Grow tobacco (%) 1.1 7.6 2.0 15.0 2.6
Received price information (%) 35.9 66.2 39.9 62.5 40.3
Access to extension services (%) 14.0 27.7 13.5 26.0 15.7
Membershig to 2. Association (%) 5.6 12.7 7.2 11.7 6.8

 

Sources: TIA02 8 TIA05

44

CHAPTER IV REGRESSION ANALYSIS

This chapter describes the regression methods and results used to obtain predicted
household income/AB in each year, and analyze factors associated with income change
and moving into or out of poverty. It starts with a description of the methods used to
predict rural household income, followed by the description of independent variables.
The ﬁrst-difference model and the multinomial logit regression results are presented later

in this chapter.

4.1. PREDICTING RURAL HOUSEHOLD INCOME

As discussed in Section 1.2.3, one of the difﬁculties when using reported income
measures over time is to distinguish between structural and transitory poverty. This is
particularly true when the data only covers a short period (as in our three-year panel
dataset). Since rural income varies with agricultural season, most of the movement into
and out of poverty between seasons will likely be classiﬁed as transitory. Using the rural
household’s predicted income can potentially purge the inﬂuence of one other source of
transitory income shocks, namely household idiosyncratic shocks (McKay and Lawson,
2002). As a result, predicted income should more closely approximate longer-terrn or
expected household income than actual income in any given period. Therefore, in order
to subsequently classify the households according to their poverty status, we estimate
predicted household income with the following model using the Ordinary Least Squares

method (OLS):

45

(5) 1013011 )=f (X it)

where ya is the household i’s income/AB in year t and X is a vector of the independent
variables, which includes household characteristics, technology, access to services, asset

holdings, and locale-speciﬁc conditions such as differences in agricultural potential.

Demographic characteristics comprise an important subset of independent variables. The
household head’s age is included to capture potential lifecycle effects. Other
demographic characteristics are related to changes in household size and composition, the
head’s years of education, and the gender of the household head. Similar to larger
households, those with a high dependency ratio (ratio of children and seniors to
household size) are expected to have less income per AE than their counterparts. Previous
studies on rural poverty in Mozambique have also shown that female-headed households
have relatively less income than male-headed households (Boughton et al. , 2006; Walker

etaI,2004)

Labor-market participation will also inﬂuence rural income (Tschirley and Benﬁca,
2001). Households with one or more salaried members will have a steadier income than
those relying only on agriculture. Similarly, households with self-employed members are
likely to have a higher and steadier income. Tables 14 and 15 in the previous chapter
highlighted the importance of labor-market participation both in lifting poor households

out of poverty, and to avoid the non-poor from falling into poverty.

46

Since crop income constitutes a large fraction of total household income, landholding
size will have an impact on household income. The expectation is that households that
have access to more land will have higher income levels. Nevertheless, the marginal
effect of increases in farm size is likely to differ across agro—ecological zones or
provinces due to differences in rainfall patterns and soil fertility. Some of the differences
are captured by using regional models rather than a national model. Other assets that can
help explain household income include livestock herd size (cattle), which are expressed

in Tropical Livestock Units (TLU) using conversion ratios from FAO (2007).

Since crop income is an important component of household income and most crop
production is rainfed, we expect differences in rainfall patterns to have a signiﬁcant effect
on household income. The effect of rainfall is controlled for using the number of drought
days at district level. District-level dummy variables capture the inﬂuence of other

district-speciﬁc exogenous shocks to household income.10

Technology use was proxied using variables such as whether the household has used
improved seeds, animal traction, crop rotation, hired labor, and line sowing. The use of
fertilizer was not included as an independent variable because most of the fertilizer used

in Mozambique is applied to tobacco, and we are already controlling for the cultivation of

 

’0 Dummies for agroecological zones were initially included, but then dropped out of the models because
of their potential collinearity with district dummies (dropping agroecological dummies did not change the
results signiﬁcantly). District dummies are preferred in our case because they are more disaggregated than

agroecological zone dummies.

47

this crop. As for any choice variable, technology variables are potentially subject to
endogeneity. An endogenous variable can be deﬁned as one that is partially determined
by other factors within the model (observable explanatory variables that are correlated
with unobservable error terms). One of the key assumptions of Ordinary Least Squares
(OLS) is that explanatory variables are uncorrelated with the error term. When this
assumption is violated, OLS estimation is biased and inconsistent, and hence t-tests and
F -statistics are not reliable. Since testing for the presence of endogeneity when using
survey weights is difficult in many statistical packages, and using an instrumental
variable approach depends on ﬁnding the appropriate instruments, which is usually
difﬁcult, we interpret the results for technology variables with appropriate caution.
Similar caveats apply to the variables on access to rural extension services and price
information, and membership in an agricultural association. The information regarding
the use of improved seeds, crop rotation, and line sowing was only collected in 2005.
Hence, these variables are only included in the 2005 cross-section models and

multinomial logit models.

We suspect that coefﬁcient estimates may differ across regions. Rather than having one
national model, we estimate three regional cross-section models for each year and present
the results in Table 18. The Chow Statistic can be used to evaluate whether the three
regional models are better than one national model for each year and is presented when
discussing the ﬁrst-difference models in the next section. All six models seem to ﬁt the
data reasonably well for both years. Robust standard errors were used in both models to

correct for heteroskedasticity. R-squared statistics for the 2001/02 agricultural season

48

regressions are relatively higher than those for the 2004/05 agricultural season, and in

both years the R-squared is lower in the south.

Growing tobacco is positively and signiﬁcantly correlated with higher income levels in
all three regions except in the south in 2004/05 agricultural season. The effect of growing
either tobacco or cotton in central provinces is higher than the effect of growing the same
crops in northern provinces. Differences between the two regions may stem from the
relatively low performance of cotton and tobacco companies in northern provinces and
differences in rainfall patterns. The result of growing tobacco in the south being
associated with losses in household income per adult equivalent in 2004/05 agricultural
season is probably due to the more severe drought in that year (few households grow

tobacco in the south and they tend to be located in the same area).

Growing cotton was only signiﬁcant and positively correlated with higher income levels
in 2002/02 agricultural season in the center. This result could be an artifact of the weather
and may also be related to the performance of cotton growers in the north. Days of
drought did not signiﬁcantly change in the center, and cotton companies in the north
seem to have relatively lower performance than their counterparts in the center (Tschirley

etaL,2006)

49

Table 18. Results from cross-section models for each year and region

 

 

WWW
Log of total household income/AE North Center South North lCenter South 2002 2005
Demographics
Numberofmembersaged 04 0.132 '*‘ 0.075 “ 0.063 0.109 *“ 0.010 0.016 0.833 0.799
Numberofmembersaged514 0.094 *“ 0.146 *“ 0.090 “* 0.120 *“ 0.151 *“ 0.111 '“ 1.478 1.689
Numberot members aged 60 orolder 0.174 “ 0.221 ** 0.299 “‘ 0.152 “ 0.072 0.221 0.232 0.267
Numberofmalemembersaged15-59 0.137 *** 0.185 *" 0.193 “* 0.194 m 0.143 *“ 0.166 ** 1.129 1.178
NumberoffemalemembersagedlS-SQ 0.117 ‘“ 0.076 “ 0.158 “* 0.200 *“ 0.163 '“ 0.165 *“ 1.293 1.338

 

 

0.060 ’“ 0.091 *“ 0.070 0.061 m 0.090 m 2.200 1.960

 

Head's years of schooling 0.049

 

 

 

 

 

 

 

Heads age 0.009 0.023 0.016 0.006 0.022 0.010 42.022 44.369

Head's age squared 0.000 0.000 0.000 0.000 0.000 I0.000

Head's gender(1=Male) 0.079 0.249 ' 0.095 0.143 “ 0.068 0.032 0.758 0.725
Labor-market participation

Numberofsalaried members 0.410 ”‘ 0.253 “‘ 0.373 *“ 0.139 *“ 0.121 *‘ 0.357 *" 0.242 0.444

Numberof selfempl. Members 0.286 *“ 0.326 “" 0.303 "‘ 0.259 *“ 0.225 “" 0.171 “‘ 0.477 0.691
Access to public services

Access to extension services (1=Yes) 0.118 ‘ 0.092 0.050 0.083 0.028 0.045 0.137 0.157

Membership to 89. Association (1=Yes) 0.207 *‘ 0.042 0.065 0.070 0.013 0.001 0.038 0.068

Accesslopriceinf0(1=Yes) 0.042 0.236 *** 0.138 0.170 “" 0.313 “ 0.106 0.387 0.403
Agricultural technology

HH used animal traction (1=Yes) NA 0.360 "‘ 0.051 0.166 0.425 m 0.222 0.113 0.088

HH used of permanent |abor(1=Yes) 0.575 **‘ 0.449 " 0.574 *‘ 0.072 0.372 0.258 0.021 0.016

HH used seasonal |abor(1=Yes) 0.328 “’ 0.423 “‘ 0.414 “‘ 0.556 *“ 0.446 *“ 0.380 " 0.157 0.180

HH does line sowing (1=Yes) NA NA NA 0.094 0.015 0.154 NA 0.416

HH doescrop rotation (1=Yes) NA NA NA 0.056 0.200 ' 0.002 NA 0.377

HH used improved seeds (1=Yes) NA NA NA 0.051 0.395 0.518 ‘ NA 0.041
Cultivation of cash crops

HH grows ootton(1=Yes) 0.017 0.235 “ 0.723 0.009 0.062 NA 0.073 0.057

HH grows tobacco (1=Yes) 0.295 "‘ 0.454 ‘*" 0.513 “ 0.382 *“ 0.723 *“ 0.530 *“ 0.039 0.026

Household assets

Landholding size (Hectares) 0.185 “' 0.113 *“ 0.036 ‘ 0.182 *“ 0.115 ‘*‘ 0.123 * 1.762 1.920

Landholding sizesquared 0.002 “‘ 0.001 “‘ 0.000 “ 0.004 *“ 0.001 m 0.002 “

Tropical livestock units 0.012 0.028 *“ 0.026 ”‘ 0.103 '“ 0.027 *" 0.028 “* 0.996 0.985
Agricultural potential

Days ofdrought 0.002 0.006 “" 0.035 “' 0.018 '“ 0.015 "* 0.028 ”' 28.665 45.731

Intercept 7.421 “' 5.956 '“ 9.596 "‘ 7.565 '" 7.187 '“ |9.965 “'

R-squared 0.40 0.42 0.34 0.31 0.29 0.20

Number of observations 1728 1164 1109 1723 1153 1091

 

 

 

 

 

Notes: *“ signiﬁcant at .01 level; “ signiﬁcant at .05 level; ' signiﬁcant at .10 level;
District dummies are included; NA - Not applicable (No observations)
Sources: TlA02 8 TIA05

50

With the exception of household head age, and number of infants in the south, all
demographic characteristics such as household size and composition are signiﬁcantly
correlated with household income/AB. For example, adding one member to the
household decreases the income/AB, regardless of age category, and the effect is usually
larger when the additional member is an elderly person (age 60 or older). Male-headed
households had signiﬁcantly higher income levels than female-headed households in

central provinces in 2002, and in northern provinces in 2005.

The number of years of educational attainment of the household head is signiﬁcantly and
positively correlated with household income/AB, and the coefﬁcients are slightly higher
in 2005. Differences in educational attainment could become more important during a
drought year because non-agricultural wage income can be used to smooth crop income
losses due to bad weather. The number of salaried and self-employed members in the
household is another factor that is correlated with higher household income per adult
equivalent. Households with members engaged in off-farm activities can have a more
steady income source, and make up for losses in crop production during a bad
agricultural season. National means for both salaried and self-employed household

members were higher in 2005 compared to 2002.

As expected, the number of drought days is correlated with household income. In 2005,
days of drought is highly signiﬁcantly and negatively correlated with household
income/AE in all three regions. However, in 2002, days of drought is signiﬁcant and

positively correlated with household income/AB in central provinces. One possible

51

explanation for this unexpected result could be the fact that many household members in
central provinces (particularly in Sofala province) suffering from drought may have

engaged in more lucrative types of non-farm activities.

Turning to agricultural technology variables, the use of animal traction is positively and
signiﬁcantly correlated with household income in both years in central provinces, but not
signiﬁcant in southern provinces. For other agricultural technology variables (2005 only),
the use of crop rotation is positively and signiﬁcantly correlated with household
income/AB in central provinces. The use of improved seeds is positively correlated with
household income/AE only in southern region, while line sowing was not signiﬁcant. The
fact that few households use improved seeds and the more widespread occurrence of
drought in 2005 may be the reason why its use was not statistically correlated with

increases in household income/AB.

In terms of assets, landholding size is positively and signiﬁcantly correlated with
household income/AB in all three regions in both years. The squared term in landholding
size indicates diminishing marginal returns. The impact of landholding size is larger in
northern provinces, followed by central provinces, reﬂecting differences in agricultural
potential between the three regions. With the exception of northern provinces in 2002,
Tropical Livestock Units are positively and signiﬁcantly correlated with household

income/AB in both years.

52

Predicted household income/AE from cross-section regressions presented on Table 18
was used to classify the households into four mutually exclusive groups: (i) stayed poor;
(ii) moved out of poverty; (iii) moved into poverty; and (iv) stayed non-poor. These four
poverty categories are then used as the dependent variable in the multinomial logit

regressions presented in section 4.3.

4.2. FIRST-DIFFERENCE MODEL

This section attempts to model the factors associated with rural income change over time.
The dependent variable is the change in the log of household income/AE between 2002
and 2005. Similar to the cross-section regressions discussed in the previous section, the
ﬁrst-difference model uses reported household income. Predicted income/AB is used in

the multinomial logit regression, presented in the next section.

The ﬁrst-difference model can be represented as

(6) log Ji— =f(AX)
yr(t—1)

where y is the household’s income/AE at time t (or H) and the independent variables
represent changes between the two periods. Variables that do not vary over time drop out
of the model. District dummy variables and household head’s age in the 2002 are

interacted with a time variable.

The two common sources of endogeneity are omitted variable bias and reverse causation.

The ﬁrst-difference model corrects for endogeneity caused by the former case when the

53

omitted variables are constant over time (Wooldridge, 2002a), but does not correct for

endogeneity due to reverse causation.

We estimate four models. As in the case of cross-section models presented earlier, the
ﬁrst-difference model results presented in Table 19 are corrected for heteroskedasticity
using robust standard errors, and they were weighted using the IPW method to correct for
attrition bias (Mather and Donovan, forthcoming). The ﬁrst model uses the entire sample,
whereas the other three models use regional sub-sarnples. The Chow statistic is computed
to test the null hypothesis that the three regions follow the same regression function,
against the alternative that one or more of the slopes differ across the three regions. We
reject the null hypothesis at the .01 level, and thus separate regional models are preferred
to a national model. In addition, some coefﬁcients differ across regions in terms of their

magnitude and sign. The discussion that follows is restricted to regional models.

Differences in rainfall patterns are driving some of the changes in household income/AB.
The increase in days of drought is signiﬁcantly and negatively correlated with reduction
in household income in the northern and southern provinces. The estimated coefﬁcient
for a change in days of drought is not signiﬁcant in central provinces. Because the share
of crop income is high, particularly among the poor, the reduction in crop income due to
increases in days of drought has contributed to a decrease in total household income
among those affected by a drought. The negative coefﬁcient for increase in days of

drought was highest in the southern provinces.

54

Table 19. First-difference model results by region

 

 

 

 

 

 

 

All regions North Center South
Demographics
Change in the # of members aged 04 0.052 ** 0.088 *** 0.007 0.024
Change in the # of members aged 514 0.068 *" 0.068 ** 0.130 * 0.079 *
Change in the # of members aged 60 or older 0.291 *** 0.363 *** 0.087 0.370 **"
Change in the # of male members aged 1559 0.147 "* 0.184 *** 0.146 ** 0.179 **
Change in the # of female members aged 1559 0.152 “* 0.234 *** 0.175 ** 0.068
Head's years of schooling in 2002‘time 0.020 * 0.022 0.043 * 0.012
Head's age in 2002’time 0.001 0.000 0.003 0.002
Change in head's gender 0.262 “* 0.164 0.401 * 0.178
Labor-market participation
Change in the # of salaried members 0.247 m 0.217 *** 0.190 **‘ 0.416 .1...
Change in the #of self-empl. members 0.266 *** 0.278 *** 0.291 *** 0.245 ***
Access to public services
Change in access to extension services 0.002 0.105 0.257 ** 0.100
Change in membership to ag. association 0.063 0.065 0.028 0.043
Change in access to price info 0.152 *‘* 0.114 ** 0.240 ** 0.138
Agricultural technology
Change in the use of animal traction 0.179 NA 0.126 0.236
Change in the use of permanent labor 0.034 0.154 0.025 0.233
Change in the use of seasonal labor 0.281 **" 0.340 *** 0.258 *** 0.046
Assets
Change in landholding size 0.044 *** 0.133 *“ 0.111 *** 0.018 *
Change in landholding size: squared term 0.000 ** 0.002 *‘* 0.001 *** 0.000 *
Change in tropical livestock units 0.041 “* 0.073 *** 0.034 **" 0.025
Cultivation of cash crops
Change on whether the HH grows cotton 0.203 0.106 0.395 0.403
Change on whether the HH grows tobacco 0.284 “* 0.156 0.543 ** 0.273
Agricultural potential

Change in days of drought 0.010 ** 0.008 * 0.002 0.019 ***
Intercept 0.370 ** 0.024 0.016 0.104
R-squared 0.177 0.205 0.218 0.148
Number of observations 4038 1746 1169 1123
Chow Statistic 6.940 ***

 

 

Notes: *“ signiﬁcant at .01 level; *" significant at .05 level; * signiﬁcant at .10 level
District dummies are included; NA - animal traction is not frequent in the north

Sources: TlA02 8 TlA05

55

In general, an increase in household size will result in a decrease in household income per
adult equivalent, other things being equal, regardless of the age group of the new
member. We expect a positive relationship between changes in income and changes in
the number of household members aged 15-59 due to an increase in labor availability
both for agricultural and non-agricultural activities. However, even for this age group, the
additional income generated by new household members does not appear to be large
enough to compensate for an increase in household size, hence resulting in a net decrease

in household income per adult equivalent.

Household head’s years of educational attainment in 2002 and change in head’s gender
were only signiﬁcant in increasing household income/AB in central provinces. Over time
the return to household head educational attainment has appreciated in central provinces,
and households that became male-headed in 2005 but were female-headed in 2002 have
signiﬁcantly increased their incomes. Similar to cross-section regression results, the
effect of educational attainment is particularly prominent during a drought year.
Households with a larger share of crop income that depend on a rainfed agriculture are
more vulnerable to weather shocks, whereas those whose head’s educational attainment
allows them to participate in the non-agricultural labor market can sustain a steadier

income source, and hence be less vulnerable to natural calamities.

In terms of labor-market participation, an increase in the number of salaried or self-
employed members is positively correlated with increases in household income/AB in all

three regions. The impact of a change in the number of household members with salaried

56

employment on household income/AB is higher in the south. Nonetheless, labor market
participation is potentially an endogenous variable. Although the ﬁrst-difference model
corrects for endogeneity caused by omitted time-invariant variables, it does not correct
for endogeneity caused by reverse causation. One way of dealing with such potential
problem is to use an instrumental variable approach. We refrain from using this method

because it is hard in practice to ﬁnd good instruments.

The adoption of tobacco resulted in signiﬁcant increases in household income per adult
equivalent in the central region. Unlike in the south where the coefﬁcient on adoption of
tobacco was not signiﬁcant due to the smaller number of tobacco growers, the absence of
a statistically signiﬁcant coefﬁcient in the north may be an artifact of the weather or

lower performance of tobacco companies in the north, or a combination of both factors.

Increases in landholding size are associated with increases in household income per adult
equivalent. Similar to the cross-section OLS results presented in the previous section, the
effect is higher in the north, followed by central provinces. The squared term on
landholding size show that the marginal effect of increasing landholding size decreases
with the expansion of farm size. Increases in Tropical Livestock Units were also
signiﬁcantly correlated with increases in total household incomes in northern and central

provinces.

57

4.3. MULTINOMIAL REGRESSION MODEL

The multinomial logit regression is used to assess the impact of household characteristics
on the odds of moving from one poverty group to another (relative to the base group).

The multinomial logit model can be represented as:

 

 

£in
(7) P(Yi = j) = J , , for j =1,2,...,J
1+ Ze'BkXi
k=1
and
1 .
(8) P(Y,-=O)= J , , for j=0
1+ ZeﬂkXi
k=1

The X,- variables correspond to the right-hand-side variables that inﬂuence the probability
of a given household belonging to a certain poverty category (Y,). A description of the

independent variables that make up the X,- vector was provided in previous sections.

A household is deemed to have stayed poor if its predicted income is below the poverty
line in both periods, while a household is non-poor if its predicted income is above the
poverty line in both years. Households moving into poverty are those whose predicted
income was above the poverty line in the ﬁrst period but below the poverty line in the
latter period. Lastly, households moving out of poverty have their predicted income

58

below the poverty line in the ﬁrst period but above it in the second period. For this
particular study J=4 (running from 0 to 3), where P( Y,=0) is the probability that the
household has stayed poor, P(Y,=1) is the probability of moving out of poverty, P( Y,=2)
represents the probability of moving into poverty and P( Y i=3) the probability of staying

non-poor.

One of the key assumptions of the multinomial logit model is the Independence of
Irrelevant Alternatives (IIA), which states that adding or removing one of the poverty
categories does not affect the odds among the remaining poverty categories. The
assumption regarding IIA can be tested using either the Hausman or the Small-Hsiao test.
Sometimes the two tests can give inconsistent results (Long and Freese, 2006).
Furthermore, it is not straightforward to perform these two tests when using complex
survey weights. We attempted to use the multinomial probit model instead (as it relaxes
the IIA assumption) but it proved computationally too time intensive. We decided to
continue using a multinomial logit speciﬁcation, and assume that the IIA assumption is

not violated.

The model coefficients presented are relative risk ratios. Standard interpretation of the
relative risk ratio (RRR) is that for a unit change in the predictor variable, the relative risk
ratio of outcome J relative to the benchmark group is expected to change by a factor of
the respective parameter estimate, holding constant the other variables in the model.
Variables with RRR (coefﬁcients) greater than one imply that, given a one unit increase

in the independent variable, the household’s likelihood of belonging to the .1“ group

59

relative to belonging to the benchmark group increases. Therefore, the expectation is that
when the RR is less than one and signiﬁcant in the comparison of moving out of
poverty relative to staying poor, then the same variable will have its RRR greater than
one when comparing households moving into poverty relative to those who managed to

stayed non-poor.

Results from estimation of the four multinomial logit models are presented in Tables 20
through 22. Model 1 is based on actual poverty lines, while in models 2 and 3 the poverty
line has been changed upward and downward by 20 percent, respectively, to evaluate the
robustness of our conclusions in relation to the poverty line being used”. Model 4 uses
actual rather than predicted income, and is included for purposes of comparison. Similar
to the cross-section models and the ﬁrst-difference models, we run regional multinomial

models and discuss the results separately.

The results from the three models that use predicted incomes are consistent (models 1, 2,
and 3). Results from model 4 are slightly different both in terms of the magnitude and the
signiﬁcance of some coefficients. For instance, changes in landholding size, use of hired
labor, tropical livestock units and drought days are signiﬁcant in all three models that use
predicted household incomes but not signiﬁcant in model 4 (where the dependent
variable is based on actual household income). The discussion that follows corresponds

to model 1.

 

” Results from Table 3 showed that poverty incidence can be higher in either period, depending on the
poverty line that is used. Thus, it is important to evaluate whether multinomial logit regression results are

robust to alternative poverty lines.

60

4.3.1. MOVING OUT OF AND INTO POVERTY IN THE SOUTH

Starting with the discussion of demographic factors in the south, we ﬁnd that adding a
new household member (ages 5 and above) decreases the probability of moving out of
poverty (Table 20). The household head’s educational attainment in 2002 is another
demographic characteristic that inﬂuences the odds of moving out of poverty, relative to
staying poor. One additional year of head’s educational attainment is associated with an

increase in the probability of moving out of poverty by 28 percent.

In terms of labor-market participation, an increase in the number of household members
with off-farm income in southern provinces signiﬁcantly increases the probability of
moving out of poverty relative to staying poor. Adding one salaried member increases the
probability of moving out of poverty by 400 percent. Moreover, adding one self-
employed member increases the probability of moving out of poverty by 77 percent. This

conﬁrms the importance of off-farm income for poverty alleviation in rural Mozambique.

Regarding the access to public services, poor households in 2002 that started receiving
price information in 2005 have increased their probability of moving out of poverty by
almost 190 percent. An unexpected result is that access to extension services decreases
the odds of moving out of poverty. One possible explanation for this result is that
households who started receiving extension services may be more dependent on crop

income and hence more vulnerable to drought.

61

Table 20. Relative risk ratios from multinomial logit models: southern provinces

 

Moved out of poverty relative to stayed poor

Moved into poverty relative to stayed non-

 

 

 

 

 

 

 

 

 

 

 

poor
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4
Demographics

Change in the if of members aged 04 0.852 0.851 0.851 0.996 0.941 1.480 1.480 0.938
Change in the it of members aged 514 0.728 ’ 0.683 0.683 0.940 1.040 0.872 0.872 0.944
Change in the it of members aged 60 or older 0.224 *" 0.498 0.498 0.393 “‘ 1.833 0.881 0.881 0.787
Change in the # of male members aged 1559 0.343 ‘** 0.582 “ 0.582 *‘ 0.734 * 2.258 *“ 1.812 1.812 0.899
Change in the # of female members aged 1559 0.824 0.772 0.772 1.022 0.388 " 0.269 * 0.269 *“ 1.348
Heads years of schooling in 2002 1.283 “ 1.653 '“ 1.653 “* 1.110 0.861 0.884 0.884 0.908
Heads age in 2002 0.975 0.975 “ 0.975 *' 0.991 1.027 1.029 1.029 0.998
Change in head's gender 2.747 4.508 “ 4.508 “ 1.915 0.372 0.624 0.624 1.261

Labor-market participation
Change in the#0f salaried members 5005 “" 4.426 “* 4.426 “‘ 1.775 ‘“ 0.146 *"‘ 1.156 1.156 0.767
Change in the it of selfempl. members 1.769 ' 1.463 1.463 1.259 0.197 '“ 0.272 *" 0.272 ' 1.120

Access to public services
Change in access to extension services 0303 ** 0.449 '** 0.449 *“ 0.523 ” 0.119 “ 1.270 1.270 0.551
Change in membership to ag. association 3.579 0.860 0.860 1.322 1.108 2.666 2.666 1.451
Change in access to price info 2.923 * 1.790 1.790 1.020 0.781 0.347 0.347 1.288

Agricultural technology
Change in the use of animal traction 0.731 0.545 0.545 1.193 0.718 0.674 0.674 1.018
Change in the use of permanent labor 6.402 *“ 7.889 “’ 7.889 "* 1.091 29.554 " 26.532 26.532 0.824
Change in the use of seasonal labor 6.782 “‘ 5.097 “" 5.097 “* 1.310 0.073 ”‘ 0.034 *' 0.034 “ 1.557
HH did line sowing in 2005 (1=Yes) 0.706 0.439 0.439 0.927 0.752 0.009 " 0.009 “ 0.835
HH did crop rotation in 2005 (1=Yes) 2.013 1.837 " 1.837 ' 1.249 0.784 0.436 0.436 1.266
HH used improved seeds in 2005 5.648 5.502 5.502 0.860 0.148 0.040 0.040 0.251
Household assets

Change in landholding size 1.109 “‘ 1.086 *" 1.086 ** 1.011 1.012 0.979 ‘ 0.979 * 1.044
Change in tropical livestock units 1.232 “ 1.241 ” 1.241 *’ 1.019 0.787 ’ 0.761 0.761 0.960

Cultivation of cash crops
Change on whether the HH grows cotton
Change on whether the HH grows tobacco 0.762 0.320 0.320 0.502 4.124 0.001 “" 0.001 *“ 6.164

Agricultural potential

Change in days ofdrought 0.843 “ 0.844 *' 0.844 ‘* 1.002 1.138 “* 1.106 1.106 1.044 *“
Number of observations 1082

Notes: “‘ significant at .01 level; " signiﬁcant at .05 level; ' signiﬁcant at .10 level; District dummies are included

Sources: TlA02 8 TIA05

62

In terms of agricultural technology, hiring labor for agricultural activities is signiﬁcantly
correlated with a higher probability of moving out of poverty, relative to staying poor.

Other agricultural technology variables were not signiﬁcant.

An additional hectare of land is associated with an increase in the probability of moving
out of poverty, relative to staying poor, of about 10 percent. Relative to other poverty
status categories, households that stayed poor have smaller landholding size. In order for
them to move out of poverty, they have to signiﬁcantly increase their landholding size or
switch to proﬁtable off-farm activities. Tr0pical livestock units were also signiﬁcantly

associated with an increase in the odds of moving out of poverty.

In terms of agricultural potential, an additional drought day signiﬁcantly decreases the
probability of moving out of poverty, relative to staying poor, by 16 percent. Turning to
the discussion of the factors that prevent non-poor households from falling into poverty,
we ﬁnd that most of the factors that can be used to lift households out of poverty can also
be used to prevent the descent path, though the magnitude and signiﬁcance can vary. An
interesting exception is related to household size and composition. While adding a new
female member is not signiﬁcant in changing the odds of moving out of poverty, it
signiﬁcantly decreases the probability of moving into poverty by 61 percent. In contrast,
adding a new male member increases the odds of falling into poverty by 126 percent.

This result indicates that male and female members may have different roles in reducing

poverty.

63

4.3.2. MOVING OUT OF AND INTO POVERTY IN THE CENTER

Table 21 presents the multinomial logit regression results for central provinces. Some of
the factors discussed in the previous section also inﬂuence the probability of moving out
of or into poverty in the center. In this section we focus on the discussion of the factors
that differ from those in the south. In terms of demographic characteristics, households
that were female-headed in 2002 but became male-headed in 2005 have increased their
odds of moving out of poverty, relative to staying poor, by 790 percent. The increase in
the probability of moving out of poverty is high because out of 18 households that ﬁt in
this category (become male-headed), none of them moved into poverty and 4 moved out
of poverty. Becoming male-headed also decreases the likelihood of moving into poverty,

relative to staying poor, by 98 percent.

In terms of change in access to public services, access to price information was highly
signiﬁcantly associated with both increased likelihood of moving out of poverty and
decreased likelihood of moving into poverty. Change in membership of an association or
access to extension was not statistically associated with poverty mobility in the center

however.

The adoption of tobacco signiﬁcantly increases the probabilities of moving out of
poverty, relative to staying poor. Among those adopting tobacco, 62 percent (out of 59
households that ﬁt in this category) have either stayed non-poor or moved out of poverty.

Adoption of such cash crops should be promoted.

64

Table 21. Relative risk ratios from multinomial logit regressions: central provinces

 

 

 

 

 

 

 

 

 

 

 

Moved out of povertyrelative to stayed poor Moved into poverty relative to stayed non-poor
Mode|1 Mode12 Model3 Model4 Modelt Mode12 Model3 Model4
Demographics
Change in the ll of members aged 04 0.905 1.023 1.023 1.099 0.847 1.244 1.244 1.241
Change in the ll of members aged 5-14 0.641 “' 0.792 0.792 1.013 1.166 1.371 1.371 0.936
Change in the it of members aged 60 or older 0.354 ‘ 0.558 0.558 0.870 0.551 3.466 *‘ 3.466 *" 0.813
Change in the#of male members aged 15-59 0.539 “‘ 0.507 *‘" 0.507 *" 0.913 2.211 “ 2.460 ' 2.460 ' 1.092
Change in thellollemale members aged 1559 0.685 *‘ 0.588 *‘ 0.588 “ 0.767 * 1.649 2.784 “' 2.784 “* 1.293
Heads years of schooling in 2002 1.333 “’* 1.186 *‘ 1.186 “ 1.086 0.796 *“ 0.858 0.858 0.863 “
Head's age in 2002 0.996 0.992 0.992 0.991 1.006 1.017 1.017 1.003
Change in heads gender 8.896 *“ 6.294 “‘ 6.294 '“ 1.986 * 0.019 *“ 0.015 “‘ 0.015 *“ 0.423
Labor-market participation
Change in thetlol salaried members 2.033 '“ 1.892 ”* 1.892 *“ 1.215 " 1.091 1.582 ** 1.582 ** 1.198
Change in the#olseltempl. members 2.600 '“ 2.117 “* 2.117 “‘ 1.227 m 0.321 *“ 0.213 "‘ 0.213 “' 0.598 “*
Access to public services
Change in access to extension services 0.866 1.108 1.108 .0810 0.922 1.079 1.079 1.081
Change in membership to ag. association 1.666 1.779 1.779 1.104 1.225 1.328 1.328 0.768
Change in access to price into 3.276 “‘ 2.032 *“ 2.032 *“ 1.443 *' 0.273 '* 0.216 **' 0.216 *“ 0.491 ”‘
Agricultural technology
Change in the use of animal traction 4.604 3.419 " 3.419 " 2.447 *“ 0.326 ‘ 0.221 “‘ 0.221 "‘ 0.882
Change in the use of permanent labor 1.348 1.499 1.499 0.424 0.384 0.311 0.311 2.194
Change in the use of seasonal labor 6.220 '“ 4.669 *“ 4.669 “‘ 1.323 0.172 '“ 0.118 “* 0.118 *“ 0.477 "
HH did line sowing in 2005 (1=Yes) 1.534 1.401 1.401 0.815 1.443 0.981 0.981 0.448 "
HH did crop rotation in 2005 (1=Yes) 2.816 ‘* 3.696 “‘ 3.696 “‘ 2.254 *** 0.778 0.565 0.565 0.710
HH used improved seeds in 2005 13.187 “‘ 8.322 *“ 8.322 *“ 2.290 0.119 *' 0.245 ' 0.245 * 0.722
Household assets
Change in landholding size 0.972 0.995 0.995 0.961 0.751 *" 0.781 “* 0.781 *“ 0.929
Change in tropical livestock units 1.140 “‘ 1.117 '“ 1.117 '“ 1.075 “‘ 0.752 *“ 0.669 “' 0.669 “' 0.927 ‘
Cultivation of cash crops
Change on whether the HH grows cotton 0.784 0.831 0.831 0.941 3.695 4.549 4.549 0.894
Change on whether the HH grows tobacco 34.089 “‘ 65.061 *" 65.061 “" 1.097 0.094 0.102 *“ 0.102 “' 0.613
Agricultural potential
Change in days of drought 1.031 1.053 '“ 1.053 “‘ 1.012 0.714 “‘ 0.706 "' 0.706 *'" 0.979 “
Number of observations 1149

Notes: '" signiﬁcant at .01 level; “ signiﬁcant at .05 level; * signiﬁcant at .10 level; District dummies are included

Sources: TlA02 8: TIA05

65

The use of improved seeds is signiﬁcantly associated with the odds of moving out of

poverty, relative to staying poor. It is also associated with decreases in the probability of
moving into poverty, relative to staying poor. This indicates the importance of improved
agricultural technologies in increasing crop income and ultimately reducing poverty both

lifting the poor out of poverty and to help the non-poor avoid falling into poverty.

The adoption of animal traction decreases the likelihood of moving into poverty, relative
to staying non-poor by 67 percent. Animal traction plays an important role in expanding
cultivated area. Where expansion of cultivated area is possible, adopting animal traction

could have a signiﬁcant impact in avoiding the non-poor from falling into poverty.

4.3.2. MOVING OUT OF AND INTO POVERTY IN THE NORTH

In this section we highlight some of the regional differences regarding the factors
associated with moving out of and into poverty in northern provinces. Table 22 presents

results from multinomial logit regressions for these provinces.

Membership of an agricultural association was also signiﬁcantly correlated with an
increase in the probability of moving out of poverty. Since northern provinces have
relatively limited road infrastructure, belonging to an agricultural association may
facilitate participation in the market by reducing assembly costs for potential buyers.
Likewise, being a member of an agricultural association can improve the access to inputs

and extension messages, which may increase the probability of moving out of poverty.

66

Table 22. Relative risk ratios from multinomial logit regressions: northern provinces

Moved out cl povertyrelative to stayed poor Moved into poverty relative to stayed non-poor

 

 

 

 

 

 

 

 

 

 

Model1 Model2 Mode|3 Model4 Modell Mode12 Model3 Model4
Demographics
Change in thellol members aged 04 0.760 ** 0.666 “' 0.666 *“ 0.918 1.622 *” 1.340 1.340 1.193
Change in thellol members aged 514 0.597 *“ 0.588 *"‘ 0.588 ‘** 0.816 *** 1.401 ‘* 1.359 1.359 1.093
Change in the it of members aged 60 or older 0.492 " 0.942 0.942 0.602 ' 1.181 0.870 0.870 0.953
Change in thettof male members aged 15-59 0.537 *“ 0.459 *‘* 0.459 “* 0.935 1.610 *‘ 1.263 1.263 1.153
Change in the#ollemale members aged 1559 0.477 "" 0.717 0.717 0.689 " 1.138 1.516 1.516 1.190
Heads years of schooling in 2002 1.372 *“ 1.387 *“ 1.387 “* 1.141 *“ 0.812 ‘* 0.808 0.808 0.818 ‘“
Heads age in 2002 0.991 0.983 0.983 0.996 0.977 0.963 0.963 0.989
Change in heads gender 1.643 1.745 1.745 1.585 0.491 * 0.596 0.596 1.105
Labor-marketparticipation
Change in the it of salaried members 1.742 “* 1.704 *" 1.704 “* 1.115 0.258 *‘* 0.228 “* 0.228 **" 0.987
Change in thellotsellempl. members 2.510 “* 2.670 *“ 2.670 “‘ 1.397 “* 0.165 “* 0.224 “" 0.224 **‘ 0.722 ‘*
Access to public services
Change in access to extension services 1.450 * 1.130 1.130 1.079 0.515 * 0.391 ’ 0.391 ' 0.610
Change in membership to ag. association 2.659 “ 3.456 ‘“ 3.456 ‘*‘ 1.233 0.938 0.727 0.727 0.497
Change in access to price into 1.577 ’* 2.278 “* 2.278 “‘ 1.277 ' 0.625 0.698 0.698 1.037
Agricultural technology
Change in the use of permanent labor 2.352 2.232 2.232 0.683 0.301 0.068 ' 0.068 ’ 0.709
Change in the use of seasonal labor 19.913 *‘* 16.440 “’ 16.440 “* 1.861 *“ 0.093 *“ 0.075 *‘* 0.075 *‘* 0.798
HH did line sowing in 2005 (1=Yes) 3.864 “' 5.532 “" 5.532 “* 1.749 “* 0.469 * 0.439 * 0.439 * 0.455 *"
HH did crop rotation in 2005 (1=Yes) 0.859 0.651 0.651 0.841 1.495 1.478 1.478 0.863
HH used improved seeds in 2005 4.614 ‘* 7.071 *" 7.071 “* 1.639 0.562 1.339 1.339 0.877
Household assets
Change in landholding size 1.556 *“ 1.470 *“ 1.470 *“ 0.990 0.497 '“ 0.452 ”* 0.452 *" 0.841 *
Change in tropical livestock units 1.592 *“ 1.530 “‘ 1.530 *" 1.160 m 0.737 * 0.863 0.863 0.987
Cultivation of cash crops
Change on whether the HH grows cotton 1.295 1.161 1.161 1.690 * 0.982 0.822 0.822 0.620
Change on whether the HH grows tobacco 2.996 ' 1.572 1.572 1.018 0.226 0.085 *" 0.085 ‘* 1.830
Agricultural potential
Change in days of drought 0.961 0.981 0.981 0.959 **' 0.457 *** 0.566 '*' 0.566 '*‘ 1.035 '
Number of observations 1710

 

Notes: “* signiﬁcant at .01 level; ** signiﬁcant at .05 level; * signiﬁcant at .10 level; District dummies are included

Sources: TIA02 & TlA05

67

Access to price information was also statistically associated with a higher probability of

moving out of poverty in the north (as it was in the center and south).

Similar to southern provinces, an increase in landholding size is associated with
movements out of poverty. However, the marginal effect of increasing landholding size is
higher in the north. While an addition hectare in the south increases the odds of moving
out of poverty by 28 percent, in northern provinces it increases the probability of moving

out of poverty by 56 percent.

In terms of agricultural practices, line sowing is now signiﬁcantly associated with a
reduction in the probability of non-poor households falling into poverty, and in lifting the
poor out of poverty. Because northern provinces have relatively more rainfall, weeding
practices may have a signiﬁcant impact in crop income, an important source of total
household income in rural Mozambique. Line sowing facilitates the weeding operation,
reducing the competition between the crops and the weeds, and ultimately increasing
crop production. Although line sowing may require more labor, this practice should be

promoted.

Comparing results from the multinomial regressions across the three regions indicates
spatial variation in the response of poverty mobility to agricultural variables. For
example, mobility out of poverty is signiﬁcantly and positively associated with
membership of an association in the north, but not in the center or south. The use of

improved seeds is also signiﬁcantly and positively associated with movement out of

68

poverty in the north and center but not in the south. Access to price information is
positively associated with movement out of poverty in all regions, but the correlation is
only weakly signiﬁcant in the south. In sum, poverty mobility is more responsive to
agricultural variables in the center and (especially) the north compared to the south. The

next section provides the poverty simulation results.

4.4. POVERTY SIMULATION

The purpose of poverty simulation is to evaluate the expected impact on poverty
reduction from relatively small changes in selected variables that are amenable to change
via government interventions. Though poverty simulation results do not tell us directly
what policy interventions the government should consider, they allow us to identify

promising areas for policy intervention based on our empirical ﬁndings.

The simulations presented in this section are based on the estimated ﬁrst-difference
model. By changing the values of selected explanatory variables, we obtain a new
predicted change in income/AB between the two periods (2002 and 2005) for each
household in the sample. The revised predicted estimate of income change is then added
to initial predicted income/AB for 2002 (from equation 5) to obtain the revised predicted
income/AB in 2005 as if changes in the selected explanatory variables had occurred.
Lastly, we compare the poverty measures (poverty headcount, poverty gap, and squared

poverty gap) using the original and modiﬁed predicted incomes in 2005.

Predicted household income/AB is a point estimate, which comes with its own forecast

69

error. Datt (2007) argues that even if predicted household income is above the poverty
line for a given household, there is a nonzero probability that the true value of that
household’s income is below the poverty line, and hence, it is appropriate to treat
predicted income as a stochastic variable. For the purpose of this section, to assess the
relative order of magnitude of possible interventions on poverty indicators, we use point

estimates only.

Our poverty simulation focuses on reducing drought vulnerability, increasing landholding
size, and adoption of cash crops. Though demographic characteristics also play a key role
in poverty reduction, these variables are harder to change in the short-run. For instance,
though it is well documented that male-headed households tend to have higher incomes
than female-headed households, it is not clear how one would induce changes in
household gender. Thus, such variables were not included in the simulation exercise.
Many technology variables were also excluded from the policy simulation because they

were not collected in 2002.

Whereas days of drought and landholding are continuous variables, the adoption of
tobacco was modeled using a dummy variable. To simulate the effect on poverty of an
increase in adoption of cash crops, proxied by the cultivation of tobacco, we increase the
frequency with which the dummy variable takes on the value 1 as opposed to 0. New
adopters in the sample were identiﬁed using a random process to achieve a predetermined

regional adoption rate. Since the tobacco companies that smallholders depend on for

70

access to specialized inputs only operate in selected areas, adoption is restricted to

districts where the crop was already grown in either 2002 or 2005.

4.4.1 REDUCTION IN DROUGHT VULNERABILITY

Table 23 shows the effect of reducing drought vulnerability on poverty incidence in rural
Mozambique. In general there were more days of drought in 2005 than in 2002, but the
increase in days of drought was not uniform across the country. In addition, in central
provinces days of drought did not differ signiﬁcantly between the two periods, and thus

we refrain from simulating the impact of reducing drought vulnerability in that region.

Table 23. Effect of reducing days of drought on poverty indicators

 

 

 

 

 

, , Percent chan e in poverty indicator

Reduction "1 National lNorth South

days of drought Po P1 P2 Po P1 P2 P0 P1 P2
1 -0.3 -1.1 -1.6 0.3 -1.4 -2.0 -0.3 -1.3 -2.0
2 -0.6 -2.2 -3.1 -0.9 -2.8 -4.0 -0.3 -2.7 -3.9
3 -1.1 -3.4 -4.6 -1.6 -4.2 -5.9 -0.5 -4.1 -5.8
4 -1.4 -4.5 -6.1 -2.1 -5.6 -7.9 -0.9 -5.4 -7.8
5 -1.8 -5.6 -7.6 -2.7 -7.1 -9.8 -1.2 -6.8 -9.7
6 -2.2 -6.7 -9.1 -3.1 85 -11.7 -1.6 -8.3 -11.6

 

 

 

 

 

 

 

 

 

Notes: Po - poverty headcount; P1 - poverty gap; P2 - squared poverty gap

Change in poverty headcount calculated as simulated poverty headcount minus poverty headcount in 2005

P1 and P2 calculated as 100*(simulated poverty indicator - poverty indicator in 2005)/poverty indicator in 2005

Sources: TlA02 8 TlA05

At the national level, poverty headcount, poverty gap and squared poverty gap would
have been lower in 2005 if rural households had not experienced signiﬁcant increases in
drought days. If the increase in drought between 2002 and 2005 had been 6 days shorter,
the national poverty headcount would have been about 2 percentage points lower. The

corresponding reduction in poverty gap and squared poverty gap would have been 6.7

and 9.1 percent, respectively. The reduction in poverty headcount would have been

71

greater in northern region than in the south, which reﬂects the fact that households in the
former region are more dependent on crop production, and hence more vulnerable to
drought than households in the south. Therefore, although the southern region usually
receives less rainfall than the northern region, because the latter relies relatively more on
agricultural income, a reduction in drought vulnerability among households in the

northern region will have a relatively higher impact on poverty headcount.

The government already recognizes the importance of small-scale irrigation as a means to
combat drought, and numerous projects are underway. But irrigation tends to be very
costly and hence the proportion of farmers who have access to irrigation technology
remains very low. Other countries in the region facing similar challenges (e.g., Zambia,
Malawi) are actively testing conservation agriculture practices, but little assessment
appears to have been conducted in Mozambique to date. In the medium-term, genetically
modiﬁed drought-tolerant varieties of maize will become available. It will be very
important for the government to have the necessary policies in place to allow the testing
of such genetically modiﬁed drought tolerant varieties. Thorough study of each of these
options is needed as they could be technically or economically unfeasible in some

regions.

4.4.2 INCREASE IN LANDHOLDING SIZE

In this section we evaluate the expected impact on poverty resulting from simulated
increases in landholding size. Table 24 shows the impact of increases in landholding size

in 2005 by 10 percent. At the national level, poverty headcount would decrease by almost

72

1 percentage point. The effect of increasing landholding size is greatest in the north,
followed by the central provinces. One possible reason for the smaller marginal effect of
increasing landholding size in the south is its lower dependence on crop income, relative
to the other regions. The induced increase in landholding size by 10 percent would also
result in a decline in both poverty gap and the squared poverty gap. Poverty depth would
decrease by 1.8 percent and poverty severity would decrease by 2.2 percent in rural

Mozambique.

Table 24. Effect of increasing landholding size (ha) on poverty indicators

 

Percent change in poverty indicator

 

 

°/ lncre 'n
laondholdaiflzlin National I North I Center I . South
2005 P0 I P1 I P2 I Po I P1 L P2 I Po P1 P2 P0 P1 P2

 

 

10 -0.8 4.8 -2.2 -1.1 -2.2 -2.8 or In 31:] m 41.4
Notes: P0 - poverty headcount; P1 - poverty gap; P2 - squared poverty gap
Change in poverty headcount calculated as simulated poverty headcount minus poverty headcount in 2005
P1 and P2 calculated as 100‘(simulated poverty indicator - poverty indicator in 2005)/poverty indicator in 2005
Sources: TIA02 8: TlA05
The impact of increasing landholding size on poverty headcount seems to be low. The
simulated effect of increasing landholding size may reﬂect drought conditions in 2005. In
a year of better rainfall the effect of increasing landholding size would likely be greater.
In addition, we are simulating for increase in landholding ceteris paribus. Increases in
landholdings are therefore a necessary but not sufﬁcient condition to reduce rural

poverty. Moreover, because poorer households have relatively small landholdings, a 10

percent increase may not be signiﬁcant in lifting them out of poverty.

Several studies have acknowledged the correlation between income and landholdings in
Mozambique at the national level (Tschirley and Weber, 1994; Jayne etal., 2003;

Boughton et al., 2006; Boughton et al., 2007; Walker et al. , 2004). But increases in
73

landholding size are occurring at a slow pace for two possible reasons. First, there are
opportunity costs involved in increasing landholding size (labor, access to basic
infrastructure and services, and others). For example, some households may have to
cultivate more land outside or away from their villages. If new land is available only
outside the village this implies additional time to walk to the ﬁeld for cultivation. In the
case where access to additional land is constrained in the village, leaving may not be
feasible as the new location may not have the basic infrastructure such as wells, schools,

and access to health services.

Second, it may be more difﬁcult to expand landholdings in areas located close to cities.
Land tends to be scarcer in these areas, and the alternative for rural households dwelling
in such locations could be either switching from cultivation of staple crops to higher
value crops or moving into productive non-farm activities. The cultivation of high value
crops is actually occurring in some areas close to cities. For example, in peripheral and
rural areas close to Maputo and Beira (the two largest urban centers in Mozambique)

there are many households growing vegetables for sale.

In areas where additional land is available for cultivation, labor for weeding is well
known to be a major constraint. One technology that could address this constraint is
herbicides. With the exception of cotton (Strasberg, 1994), there are no studies of the
proﬁtability of herbicide use in Mozambique. Boughton et al.(2007) report that the
cotton company in Cabo Delgado province has been piloting the use of low volume

herbicides with cotton growers, using the same application equipment already used for

74

insecticide treatments. The response of farmers has been very enthusiastic despite the
additional cost of the chemical as it enables them to weed both their cotton and food
crops on time. Expansion in the use of herbicide, if proﬁtable, would still be constrained
outside cotton or tobacco contract farming arrangements by the lack of an input

distribution network and lack of access to credit.

4.4.3 ADOPTION OF CASH CROPS

This section presents the effect of adopting cash crops to poverty reduction in rural
Mozambique using the example of tobacco in central provinces. Simulation of cotton was
not included because its adoption was not signiﬁcant in increasing changes in household
income per adult equivalent. Simulation of tobacco in southern provinces was not
included because very few households grow this crop in the south. Although a signiﬁcant
number of households grow tobacco in the north, its impact was not signiﬁcant in
increasing household income/AB in northern provinces, and hence, we do not include the

north.

Table 25 shows the impact of a 10 percent adoption of tobacco in central provinces.
Poverty headcount in the center would have declined by almost 3 percent. The
corresponding reduction in poverty gap and squared poverty gap would have been 4.5

and 4.9 percent, respectively.

75

Table 25. The effect of adoption of cash crops on poverty indicators

  
 

      

 
  
 

in indicator
Center
1

Percent

   
  
  

   

center

Notes: Po - poverty headcount; P1 - poverty gap; P2 - squared poverty gap

Change in poverty headcount calculated as simulated poverty headcount minus poverty headcount in 2005
P1 and P2 calculated as 100'(simulated poverty indicator - poverty indicator in 2005)/poverty indicator in 2005
Sources: TlA02 & TIA05

This results shows that cash crops can potentially have an important impact on poverty
reduction. Therefore, to achieve this level of impact in a broader scale, we need to
understand the reasons why cotton is not contributing to poverty reduction, and why
tobacco is not contributing to poverty reduction in the north, and take counteractive

measures. The next chapter summarizes key results.

76

CHAPTER V SUMMARY RESULTS

The analysis presented in this paper is based on a nationally representative panel data set
from Mozambique, covering the period 2002-05. The results presented were inﬂuenced
by occurrence of a widespread drought in the second period. The ﬁrst question addressed
relates to changes in poverty and income distribution. Though mean household income
per adult equivalent increased between 2002 and 2005, the lowest two income quintiles
experienced a decrease in household income. Consequently, median household income
per adult equivalent has decreased between 2002 and 2005, and the distribution of
income has become more unequal. However, poor households are not necessarily
doomed to stay at the bottom of the distribution as some of the poorest households in
2002 moved to the highest income quintile in 2005. Although at the national level
poverty headcount has increased, we cannot unambiguously say that poverty incidence
was higher in the second period. In terms of poverty gap and squared poverty gap, at the
national level we can unambiguously say that both the depth and severity of poverty was

higher in the second period.

Poverty in rural Mozambique has various dimensions. In terms of a geographical
dimension, Nampula province is an area of major concern. Nampula is one of the most
populated provinces, and despite its agricultural potential, it is one in which the
percentage of households moving out of poverty is among the lowest, and the percentage
of those moving into poverty is among the highest. Overall, poverty incidence has

slightly increased as the percentage of households moving into poverty is greater than the

77

percentage of households moving out of poverty. In Sofala province by contrast, poverty

headcount, poverty gap and squared poverty gap have signiﬁcantly decreased.

The second question addressed in this paper was to understand the factors underlying
household income change. Adoption of improved agricultural technologies and cash
crops is associated with increases in household income, especially in the center and north.
These ﬁndings are consistent with other studies (Walker et al., 2006; Walker et al., 2004;
Boughton et al. , 2006), where the role of agriculture in poverty reduction is
acknowledged. In addition to adoption of improved technologies, diversiﬁcation of
income sources is also associated with changes in household income. For example, in
Sofala province, many household members have engaged in self-employment activities,
the use of animal traction has increased, the average cattle herd size has almost doubled,
more households have access to extension services and price information, and change in
days of drought was not signiﬁcant. The combination of all these factors resulted in
increases in household income, and thus, decreases in poverty incidence, poverty gap,

and severity.

The widespread drought in 2005, particularly in northern and southern Mozambique, has
also contributed to household income change. Households experiencing drought had
relatively lower crop production, which contributed to a decline in total household
income in cases where nonfarm income opportunities were insufﬁcient to compensate.
This result indicates the importance of diversiﬁcation of income sources, especially in

predominantly rainfed agricultural systems, which is the case of rural Mozambique. On

78

the other hand, this result implies that reduction in the vulnerability of agriculture to

drought will lead to increases in household income, and ultimately, poverty reduction.

The third question addressed factors associated with moving out of or into poverty in the
short run. In terms of the demographic dimension of poverty, we ﬁnd that the percentage
of households remaining poor is higher among female-headed households. Male-headed
households are also more likely to have moved out of poverty than female-headed
households. With the exception of those moving out of poverty, on average the household

size has increased for all other poverty status categories.

Given the widespread drought in 2005, it is not surprising that off-farm income played an
important role in changes in household poverty status. Households staying poor were able
to increase their income from all off-farm activities, but their increase was not signiﬁcant
enough to lift them out of poverty. Households moving out of poverty had a large
percentage increase from almost all off-farm activities, except from the extraction of
forestry and fauna resources with a high return (charcoal production and ﬁshing). In
contrast, households moving into poverty had decreases from all off-farm sources. Thus,
diversiﬁcation in non-farm income opportunities is important in reducing poverty in rural

Mozambique.

In terms of assets, we ﬁnd that households staying poor started out with smaller
landholdings. Landholding size has increased for all poverty status categories except

among households moving into poverty. These households started out with more land

79

than the poor in 2002, but their average landholding size was smaller than of those who

stayed non-poor.

In terms of agricultural practices and access to services, two distinct patterns are
noteworthy. First, the use of fertilizers, pesticides, and animal traction has increased
among households moving out of poverty. In contrast, households moving into poverty
have experienced a decrease in the use fertilizers, pesticides, and animal traction. Much
of the change in fertilizer and pesticide use is likely associated with adoption or dis-
adoption of tobacco and cotton. Although the percentage of households using animal
traction has decreased at the national level, households moving out of poverty were able
to maintain (or increase) its use, whereas those moving into poverty signiﬁcantly
disadopted animal traction use. The use of seed and improved cultivation practices is
higher among the non-poor, and reduction in the use of such technologies is correlated
with moving into poverty. Second, there are notable regional differences in the response
of poverty mobility to access to public services and use of improved agricultural

practices.

Most of the results show that factors associated with movements out of poverty are the
same as those associated with movements into poverty. An exception is the use of animal
traction in central provinces, and adoption of tobacco in central and northern provinces.
Animal traction is used to cultivate relatively large areas and households cultivating such
areas are more likely to be non-poor. Thus, adoption of animal traction is likely to be

more effective in helping the non-poor avoid falling into poverty than in lifting the poor

80

out of poverty. The marginal effect of cultivating tobacco is signiﬁcant in increasing
household income among the poor, but not in avoiding non-poor households from falling

into poverty.

For a country like Mozambique, where the agriculture is dependent on erratic rainfall, the
occurrence of drought is one of the major limitations. The simulation results show that
reducing drought vulnerability has a signiﬁcant impact in poverty reduction. Possible
interventions include the adoption of drought-tolerant technologies, such as the adoption
of drought-tolerant seeds or conservation agriculture, or even the development of
irrigation schemes. However, additional studies on the costs and beneﬁts of each of these

opportunities are necessary.

Simulation of the adoption of tobacco in central provinces resulted in signiﬁcant
decreases in poverty headcount. At present, expansion of area under cash crops is limited
to zones where tobacco or cotton companies operate. Similarly, animal traction is
concentrated in the center and south. Broadening the poverty reduction potential of cash
crops will require a more detailed study of the potential for adoption in new areas. The
results also show that tobacco and cotton were not signiﬁcant in reducing poverty in the
north, which has to do with disadoption in 2005 and the relatively low performance of

cotton and tobacco contract farming companies in northern provinces.

In sum, this paper presented empirical evidence regarding the vulnerability of rural

households to occurrence of drought and its impact on poverty. It also showed the role of

81

off-farm income both in lifting the poor out of poverty and in the prevention of a descent
movement into poverty. Reducing drought vulnerability and promoting off-farm
employment opportunities will signiﬁcantly reduce poverty in rural Mozambique. Where
expansion of cultivated area is possible, increasing landholding size can signiﬁcantly

reduce poverty.

82

APPENDICES

83

 

Appendix A-l. Construction of Price Inﬂators for TIA02-TIA05 panel income

analysis (Mather et al., forthcoming)

Summary: We use IAF 2002/03 consumption quantities (ﬂexible adjusted) from the food
consumption basket of each IAF poverty region. These quantities are valued using 2002
SIMA retail prices, then the basket is revalued with 2005 SIAM prices to update the cost
of an identical (ﬁxed) consumption basket. The consumption quantities are therefore the
weights for the commodity prices. The inﬂators which we create are ﬁxed because the
weights are not allowed to change over time (relative to ﬂexible inﬂators, ﬁxed inﬂators

which will tend to overestimate inﬂation, thus underestimating growth).

For commodity l (maize):
Q] j = grams consumed/day of commodity maize from IAF 2002/O3 food basket for
region j

Pl j = SIMA 2001/02 rural retail market price/gram for region j

The share of this commodity’s value in the total value of consumption (at the poverty
line) is = Q1j*Plj / Z; (Qij*Pij) — where i is commodity (from 1 to n) and j is the
province or poverty region. The quantities in a given region’s basket varies by IAF
poverty region (some regions contain two province), though the prices vary by province

in some cases.

Step 1: Convert IA F02 quantities consumption of quantities with SIMA price data covered

84

by SIMA into quantities covered by SIMA (where possible).

Step 2: for [AF 02 commodities for which we have SIMA price data, revalue the
commodity basket using SIMA 2002 prices, computing Pljozt‘Qlj02 for 2002, where Pijo2
is the annual average SIMA retail price/gram for commodity i for that province j (or
region if a price for that province is not available). The sum value of this subset of [AF
basket commodities for a given region j is Z; (Pij°2*Qij°2). This is the ‘cost of a

consumption basket in 2002, for province j’.

Step 3: for that same subset of commodities, compute Pijost‘Qij02 for 2005, where Pij05 is
the annual average SIMA retail price/ gram for commodity i for that province j (or region
if a price for that province is not available). The sum value of these commodities is Z;

(Pij°5*Qij02). This is the ‘cost of a consumption basket in 2005, for province j’.

Step 4: compute the ratio 2; (Pij05*Qij°2) / Z; (Pij02*Qij02); this is our price inﬂator (Plj)
for province j. If we want our panel values to be in 2002 prices, we divide TIA 05 income
by PIj. If we want to the panel values to be in 2005 prices, we multiply TIA 02 income by

Plj.

Computing the annual average SIMA retail price
We compute an annual average SIMA rural retail price (nominal) for 2001/2, and one for
2004/5, for each province (or IAF poverty region). IAF consumption quantities and prices

reﬂect the period of [AF implementation July 2002 —June 2003. The ﬁrst wave of the TIA

85

panel was implemented in the fall of 2002, and collected recall data for the period
October 2001 — September 2002 (the 2001/02 agricultural year). The second wave was
implemented in the fall of 2005, and collected recall data for the period October 2004 —
September 2005 (the 2004/05 agricultural year). Since we want to control for price
inﬂation between the TIA panel surveys, we compute an annual price where the year is
deﬁned as October-September (Oct 2001- Sept 2002; Oct 2004 - Sept 2005) to

correspond with the TIA recall periods.

SIAM collects weekly price data from 3 traders in a given market. For the given
commodity and rural retail market, we ﬁrst take the mean of the 3 trader prices for a
given week at that market. Second, we compute the median (nominal) price for each
quarter of the year for each market (the ‘ﬁrst quarter’ of TIA recall is October-
December). Third, if there is more than one market in a given region, we compute the
mean quarterly price across the markets in that region. Fourth, we compute the annual
price as the mean of the quarterly prices for that commodity in the given region.
Computing quarterly prices (step 2) before the annual price ensures that we aren’t
weighting too heavily quarters of the year in which there happen to be more observations.
If a province does not have price observations for a given commodity, we use the price
from the neighboring ‘IAF poverty region’ province (for example, Cabo Delgado and

Niassa are in the same IAF poverty region).

There are several aspects of this method which are less than ideal, but which we still ﬁnd

preferable to using the INE urban CPI from Maputo, Beira, and Nampula. First, the SIAM

86

prices we use do not correspond temporally to the reported quantities consumed in IAF
given that the [AF 2002/O3 survey covers a different period of time relative to TIA
2001/02. If we were to use prices from the IAF period then we might underestimate price
inﬂation from 2002 to 2005, since IAF prices represent a period which occurred nearly a
year after the TIA 02 recall period. Second, while the SIAM prices are from the same
region as the IAF quantities, they technically do not correspond spatially to the IAF
quantities. We choose not to update the IAF food basket values using SIAM prices only
for 2005, since the SIMA and [AF prices (even if they were for the same year, which they
are not) would likely be somewhat different given differences in the location and method
of price reporting between IAF and SIAM. IAF prices are derived from the reported ‘retail
value’ of retail-purchased or home consumption of commodities. MN F analysts bounded
these household prices by price observations from the nearest village market taken at the
same time as the household survey. By contrast, while there are many fewer observations
of SIAM prices per province (only 1-3 rural markets per province), the SIAM price data
itself is likely to have less measurement error given that it is measured directly by SIAM
staff as well as recollection collected using SIAM-standardized methods (i.e. standard
containers) to ensure accurate quantity conversion (to a kilogram price). This rationale is
not meant to criticize IAF price data, but rather simply to demonstrate why we chose not

to value the [AF 2002 quantity using an IAF price for 2002 and a SIAM price for 2005.

To ensure that we are able to capture price seasonality throughout the year, we only use
data from SIAM markets which have recorded at least one price for each quarter of the

year (this is the case of the majority of SIAM markets). If a market is missing just one

87

quarter out of the four for a given year, we use price inﬂation by quarter for that product
in that region to impute the missing price for the product/market in that quarter. For
example, if a Niassa rural market is missing a maize price for the 3rd quarter, we use the
average price inﬂation in the north for maize between the 2nd and 3rd quarter to inﬂate the
observed Niassa rural market maize price in the 2nd quarter and thus impute a 3rd quarter
price. In all but one or two cases, this was the only price missing for the 8 quarters of
price data (i.e. the two years of the TIA panel), and a majority of markets had complete
price data for all 4 quarters for both years. As follows are the number of SIAM rural
markets used relative to those with incomplete price data: (number of markets with one
quarter missing/number of markets used) = maize in grain (1/7), maize ﬂour (2/6),
cowpea (2/6), common beans (2/9), rice (2/8), cooking oil (1/6), peanuts — small seeds

(2/5), peanuts — large seeds (3/5), and sugar (2/6).

IAF basket commodities used in our baskets

We use the 2002/03 adjusted ﬂexible quantities of each commodity in the IAF food
baskets for each IAF poverty region (obtained from MPF, 2004 via Charming Amdt), and
use only those commodities for which we can match a SIAM price (or an imputed price,

in the case of sweet potatoes).

IAF commodity name = SIAM commodity name
Milho branca em grao = Grao de milho branco nacional
Farinha de milho branco com farelo, nacional = Farinha de milho branco nacional corn

farelo

88

 

Farinha de mandioca = (converted to fresca)
Mandioca fresca = mandioca fresca

Feijao nhemba = feijao nhemba

F eijao manteiga = feijao manteiga nacional
Arroz corrente = arroz corrente

Oleo = Oleo alg/avul

Amendoim pequena = amendoim pequena
Acucar nacional = acucar castanho nacional

Batata doce = imputed retail price

IAF consumption baskets often include consumption of the same commodity in different
states/forms (i.e., ﬂour, with husks, grain, dried/fresh, etc). Since we tend to only have
SIAM price data for one of the product forms, we convert the IAF quantities of some
commodities to the form for which we have SIAM price data, using TIA commodity

conversion factors as constructed for TIA 02.

SIAM '5 coverage of cassava ﬂour and fresh cassava is geographically very sparse (we
only have 4 quarters of price data for Mocuba). However, we have a considerable number
of farmgate cassava prices from TIA across all regions. We compute the marketing
margin (%) between the SIAM ﬂour price and the TIA farmgate cassava price for
Mocuba. We then apply this marketing margin to the TIA farmgate cassava price data by

region to impute retail cassava prices for each region.

89

SIAM does not collect price data for every commodity in the IAF consmnption baskets.
Most of these commodities we simply must drop from the IAF consumption basket.
Column F in Table A-2 shows how much of the total IAF expenditure basket for a given
region is covered by the commodities in our adjusted baskets. However, for sweet potato,
we impute a 2002 retail price by applying the cassava marketing margin to the TIA

farmgate price data on sweet potato.

9O

Table A 1. Poverty gap by province and year

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. Decrease poverty Decrease poverty Increase poverty Increase poverty

Actual poverty Irne line by 20% line by 50% line b1r20% line by 50%
Province 2002 2005 2002 2005 2002 2005 2002 2005 2002 2005
Niassa 36.8 36.4 26.5 26.8 7T 9.4 45.0 44.1 54.4 53.1
Cabo Delgado 48.8 39.4 38.2 30.2 16.0 12.8 56.8 47.0 65.2 55.8
Nampula 31.5 47.1 21.2 37.5 6.2 18.1 40.5 54.4 50.7 62.7
Zambézia 38.3 45.8 28.9 35.9 12.5 16.0 45.9 53.2 54.6 61.3
Tete 38.3 47.9 30.9 40.3 16.1 23.5 44.5 53.8 52.4 60.4
Manica 47.5 47.7 38.9 39.8 21.0 24.4 54.1 54.1 61.3 61.2
Sofala 41.2 23.0 32.0 16.1 15.1 5.2 48.6 29.0 57.2 37.0
Inhambane 41.3 37.2 31.7 28.0 13.3 12.3 48.9 45.2 57.7 54.2
Gaza 57.2 49.3 48.0 39.6 26.5 18.9 63.6 56.8 70.3 64.7
Maputo 51.6 56.6 42.8 47.7 22.5 28.1 58.3 62.9 65.7 69.5
National 40.6 43.7 31.1 34.6 13.6 16.6 48.2 50.9 56.9 59.0
Source: TlA02 & TIA05
Table A 2. Squared poverty gap by province and year

. Decrease poverty Decrease poverty Increase poverty Increase poverty

Acma' ”New ""9 line by 20% line b1r50% line by 20% line by 50%
Province 2002 2005 2002 2005 2002 2005 2002 2005 2002 2005
Niassa 19.1 19.8 11.9 13.0 2.6 3.6 25.8 26.1 34.6 34.4
Cabo Delgado 28.4 22.8 19.7 15.9 6.1 5.4 36.0 29.1 45.1 37.3
Nampula 15.5 28.6 9.3 20.7 2.1 7.8 21.9 35.5 30.6 43.9
Zambézia 21.9 27.1 15.2 19.1 5.3 6.7 28.1 34.0 36.2 42.5
Tete 24.2 31.9 18.2 24.9 8.6 12.1 29.5 37.8 36.4 44.8
Manica 30.5 32.2 23.1 25.5 10.5 13.8 36.7 38.0 44.4 45.2
Sofala 24.6 11.9 17.6 7.5 6.8 2.4 30.9 16.3 38.9 22.4
Inhambane 23.8 21.4 16.4 14.9 5.7 5.4 30.4 27.5 38.7 35.7
Gaza 37.4 30.3 28.6 22.1 12.5 8.5 44.5 37.3 52.7 45.9
Maputo 33.1 37.9 24.9 29.7 10.3 14.3 39.8 44.7 48.0 52.5
National 23.5 26.5 16.4 19.2 5.9 7.5 30.0 32.9 38.3 41.0

 

 

 

 

 

 

 

 

 

 

 

Source: TIA02 8. TIA05

91

Table A 3. List of non-farm income generating activities

 

 

 

 

Category I Activities

Unskilled agricultural wage jUnskiIIed agricultural wage from small and large farms
Unskilled non-agricultural wage IUnskilled non-agricultural wage

Skilled non-agricultural wage [Skilled or specialized wage labor and mining:

 

. . Cut ﬁrewood, sticks, palm tree leaves, and grass, collect
Extraction of forestry or fauna products wrth low return Ihoney and bush fruits, and hunting

 

Extraction of forestry or fauna products with high return Charcoal production and ﬁshing

 

andrcraft, carpentry, cloth makinngcycIe andradﬁ)

Other self-employment activities wrth low return repairing, blacksmith, brick or concrete block production

 

Home-made beverages, purchase and sale of beverages,
Other self-employment activities with high return food and non-food products, purchase and sale of livestock.

agro-grocessing and milling activities
Source: TlA02 & TIA05

92

REFERENCES

93

Alderman, Harold. (1992). Incomes and Food Security in Ghana. CFNPP Working Paper
N° 26. Ithaca, NY.

Amdt, Charming and Kenneth Simler. (2007). Estimating Utility — Consistency Poverty
Lines with Applications to Egypt and Mozambique. Paper presented at the conference on
Economic Development in Africa, Centre for the Study of Aﬁican Economics, University
of Oxford.

Benﬁca, Rui. (2006). An Analysis Of Income Poverty Effects in Cash Cropping
Economies in Rural Mozambique: Blending Econometric and Economy-Wide Models.
PhD Dissertation, Michigan State University, East Lansing, Michigan

Booysen, F ., van der Berg, 8., Burger, R., von Maltitz, M., and Gideon du Rand. (2005).
Using an Asset Index to Assess Trends in Poverty in Seven Sub-Saharan Aﬁican
Countries. Presented at the International Conference in The many dimensions of poverty.
Brasilia, Brazil 29-31 August 2005.

Boughton, Duncan, David Mather, David L Tschirley, Thomas Walker, and Ellen
Payongayong. (2005). Pro-Poor Rural Economic Growth for Post-Civil War Recovery:
Myth or Reality in Mozambique? Accepted as a poster paper for the 26th International
Association of Agricultural Economics Meeting, Brisbane, August 2006.

Boughton, D., Mather, D., Tschirley, D., Walker, T., Cunguara, B., and Ellen
Payongayong. (2006). Changes in Rural Household Income Patterns in Mozambique
1 996-2002 and Implications for Agriculture’s Contribution to Poverty Reduction.
MINAG Research Report No.61E. Maputo: Ministry of Agriculture

Boughton, Duncan, David Mather, Christopher B. Barrett, Rui Benﬁca, Danilo Abdula,
David Tschirley and Benedito Cunguara. (2007). Market Participation by Rural
Households in a Low-Income Country: An Asset-Based Approach Applied to
Mozambique. Faith and Economics Vol 50:64-10].

Chaudhuri, S. and Martin Ravallion. (1994). How Well do Static Welfare Indicators
Identify the Chronically Poor. Journal of Public Economics 53(3): 367-394.

Christiaensen, Luc , Lionel Demery and Stefano Patemostro. (2002). Growth,
Distribution and Poverty in Aﬁica: Messages from the 19905. The World Bank,
Washington DC.

94

Datt, Gaurav. (2007). Simple Formulae for Predicting Poverty Measures from Regression
Models of Household Living Standards. World Bank

Datt, G., Simler, K., Mukherjee, S., and Gabriel Dava. (2000). Determinants of Poverty
in Mozambique: 1996 — 97. IFPRI Food Consumption and Nutrition Division Working
Paper No.78, Washington DC.

Datt, G. and Dean Jolliffe. (2005). Poverty in Egypt: Modeling and Policy Simulations.
Economic Development and C ultural Change 53(2): 327—346.

Davidson, Russell and Jean-Yves Duclos. (2006). Testing for Restricted Stochastic
Dominance. Departmental Working Papers 2006-20, McGill University, Department of
Economics.

Davidson, Russell and Jean-Yves Duclos. (2000). Statistical Inference for Stochastic

Dominance and for the Measurement of Poverty and Inequality. Econometrica 68(6):
1435-1464.

Deaton, A. (1997). The Analysis of Household Surveys. Washington, DC: World Bank.

Dercon, S. (2002). Income Risk, Coping Strategies and Safety Nets. World Bank
Research Observer 17(2): 141-166.

Duclos, Jean-Yves, Abdelkrim Araaryand, and John Giles. (2006). Chronic and
Transient Poverty: Measurement and Estimation, with Evidence ﬁom China. ECINEQ
Working Paper 2006-35.

El—laithy, H., M. Lokshin, and Arup Banerji. (2003). Poverty and Economic Growth in
Egypt 1995-2000. World Bank Policy Research Working Paper 3068.

Fox, Louise, Elena Bardasi, and Katleen Van den Broeck. (2005). Poverty in
Mozambique: Unraveling Changes and Determinants. Africa Region Working Paper
Series No.87.

95

Government of Mozambique. (2006). Council of Ministers. Poverty Reduction Strategy
Paper 11. Maputo.

Government of Mozambique. (2001). Council of Ministers. Poverty Reduction Strategy
Paper I. Maputo.

Gradin, C., Olga Cantc'), and Coral del Rio. (2004). Inequality, Poverty and Mobility:
Choosing Income or Consumption as Welfare Indicators. Paper Prepared for the 28th
General Conference of The International Association for Research in Income and Wealth,
Cork, Ireland, August 22 — 28, 2004.

Grootaert, C., Ravi Kanbur, and Gi-Taik Oh. (1995). Why Some People Escape from
Poverty and Others Don ’t: An A frican Case Study. The World Bank Policy Research
Working Paper 1499.

Hsiao, Cheng. (2005). Why Panel Data. Institute for Economic Policy Research Working
Paper No. 05.33. University of Southern California.

Hulme, David. (2003). Conceptualizing Chronic Poverty. World Development 31(3):
403-423.

Jalan, Jyotsna and Martin Ravallion. (1998). Transient Poverty in Post-Reforrn Rural
China. Journal of Comparative Economics 26(2): 338-357.

James, R., Amdt, C., and Kenneth Simler. (2005). Has Economic Growth in Mozambique
Been Pro-Poor? IF PR1 Food Consumption and Nutrition Division Working Paper
No.202, Washington DC.

Jayne, T.S, Takashi Yamano, Michael Weber, David Tschirley, Rui Benﬁca, David
Neven, Anthony Chapoto, and Ballard Zulu. (2003). Smallholder Income and Land
Distribution in Africa: Implications for Poverty Reduction Strategies. MSU International
Deve10pment Paper N°24.

Jha, Raghbendra and Anurag Sharma. (2003). The Spatial Distribution of Rural Poverty
in the Last Three Quinquennial Rounds of NSS. Australian National University Working
Paper. Canberra, Australia

96

Kakwani, Nanak. (1980). On a Class of Poverty Measures. Econometrica 48(2): 437-446.

Lipton, M. and M. Ravallion. (1995). “Poverty and Policy”, Chapter 41 in Behrman, J.
and TN. Srinivasan, Handbook of Development Economics, vol. 3A, Elsevier Science
B.V., pp. 2551-2657.

Long, J. S., and J. F reese. (2006). Regression Models for Categorical Dependent
Variables Using Stata. Stata Press.

MADER. (2002). Relatorio Geral do TIA. Maputo, Mozambique: Ministry of Agriculture
and Rural Development.

Maluccio, John. (2000). Attrition in the Kwazulu Natal Income Dynamics Study 1993 —
I 998. IFPRI Food Consumption and Nutrition Division Discussion Paper No.85,
Washington DC.

Marrule, Higino. (1998). Land-Poor in a “Land-Abundant” Setting: Unraveling a Paradox
in Mozambique. Masters Plan B Paper. East Lansing: Michigan State University

Mather, David and Cynthia Donovan. (Forthcoming). The Impacts of Prime-age Adult
Mortality on Rural Household Income, Assets, and Poverty in Mozambique.

Mather, David, Ellen Payongayong, and Benedito Cunguara . (Forthcoming). Household
Income and Asset Dynamics in Rural Mozambique, 2002-2005.

McCulloch, Neil, and Bob Baulch. (1997). Distinguishing the Chronically from the
Transitory Poor: Evidence from Rural Pakistan. IDS Working Paper 97.

McKay, Andrew and David Lawson. (2002). Chronic Poverty: A Review of Current
Quantitative Evidence Chronic Poverty. Research Centre Working Paper No 15.

97

Ministério da Agricultura e Desenvolvimento Rural (MADER). (2002). Trabalho de
Inquerito Agricola 2002. Departamento de Estatistica, Direccao de Economia, MADER,
Repr’rblica de Mocambique, Maputo, Mocambique.

Ministério da Agricultura e Desenvolvimento Rural (MADER). (2003). Trabalho de
Inquerito Agricola 2003. Departamento de Estatistica, Direccao de Economia, MADER,
Republica de Mocambique, Maputo, Mocambique.

Ministério da Agricultura (MINAG). (2005). Trabalho de Inque’rito Agricola 2005.
Departamento de Estatistica, Direccao de Economia, MADER, Republica de
Mocambique, Maputo, Mocambique.

MPF [U EM/IF PR1 (Mozambique Ministry of Planning and Finance/Eduardo Mondlane
University/Intemational Food Policy Research Institute). (2004). Poverty and well-being
in Mozambique: The Second National Assessment (2002-2003). Maputo.

Ravallion, Martin. (1992). Poverty Comparisons: A Guide to Concepts and Methods.
LSMS Working Paper 88. The World Bank.

Ravallion, Martin. (1996). Issues in Measuring and Modeling Poverty. Economic Journal
106 (438): 1328-1343.

Ravallion, Martin and M. Lokshin. (2003). On the Utility Consistency of Poverty Lines.
World Bank Policy Research Working Paper No. 3157. Washington, DC.

Resnick, Danielle and Regina Bimer. (2006). Does Good Governance Contribute to Pro-
Poor Growth: A Review of the Evidence ﬁ-om Cross-Country Studies. IFPRI
Development Strategy and Governance Division Discussion Paper No. 30.

Simkins, Charles. (2004). What Happened to the Distribution of Income in South Africa
Between 1 995 And 2001 ? University of the Witwatersrand.

Slesnick, D. (1993). Gaining Ground: Poverty in the Postwar United States. Journal of
Political Economics 101(1): 1-3 8.

98

Tarp, F ., Simler, K., Matusse, C., Heltberg, R., and Gabriel Dava. (2002). The Robustness
of Poverty Proﬁles Reconsidered. Food Consumption and Nutrition Division of the
International Food Policy Research Institute Discussion Paper No. 124.

Thorbecke, Erik. (2005). Multi-Dimensional Poverty: Conceptual and Measurement
Issues. Paper prepared for The Many Dimensions of Poverty International Conference,
UNDP International Poverty Centre, Brasilia, August 29-31, 2005.

Trzcinski, E. and Randolph, S. (1991). Human Capital Investment and Relative Earnings
Mobility: The Role of Education, Training, Migration, and Job Search. Economic
Development and Cultural Change 40(1): 153-168.

Tschirley, David, Colin Poulton and Duncan Boughton. (2006). The Many Paths of
Cotton Reform in Eastern and Southern Africa: Lessons from a Decade of Experience.
International Development Working Paper 88. East Lansing: Michigan State University.

Tschirley, David and Rui Benﬁca. (2001). Smallholder Agriculture, Wage Labour, and
Rural Poverty Alleviation in Mozambique: What does the Evidence Tell Us?
MINAG Research Report No. 41. Maputo: Ministry of Agriculture.

Tschirley, David and Michael Weber. (1994). Food Security Strategies under Extremely
Adverse Conditions: The Determinants of Household Income and Consumption in Rural
Mozambique. World Development 42(2): 159-173.

Verpoorten, Marijke. (2003) The Determinants of Income Mobility in Rwanda, 1 990 —
2002. KULeuven Working Paper.

Walker, T., D. Tschirley, J. Low, M.Tanque, D. Boughton, E. Payongayong, and M.
Weber. (2004). Determinants of Rural Income, Poverty and Perceived Well-Being in
Mozambique in 2001-2002. MINAG Research Report No. 57E. Maputo: Ministry of
Agriculture

Walker, T., D. Boughton, D. Tschirley, R. Pitoro, and Alda Tomo. (2006). Using Rural
Household Income Survey Data to Inform Poverty Analysis: An Example from
Mozambique. Contributed paper prepared for presentation at the International
Association of Agricultural Economists Conference, Gold Coast, Australia, August 12-
18, 2006.

99

Wooldridge, J. (2002a). Introductory Econometrics: A Modern Approach. South Western
College Publishing. '

Wooldridge, J. W. (2002b). Econometric Analysis of Cross Section and Panel Data.
Cambridge, Massachusetts: The MIT Press.

100

ltilitrgtllrtltlylﬁgr