Kinship and Consumption: The E¤ect of
Spouses’Outside Options on Household Productivity
Selma Telalagić
January 2014
Abstract
This paper exploits the exogenous variation in land rights in Malawi to estimate the impact
of spouses’outside options on household productivity. In Malawi, some individuals trace their
descent through their mother and others through their father. Descent is exogenously determined by parentage. Where descent is traced through the mother (matriliny), women inherit
household land. This gives them stronger rights to land and better outside options than their
husbands. The opposite is true of patrilineal descent. The estimates show that matrilineal
households consume over 10% more than patrilineal households on average, once a rich set
of control variables is included. Matrilineal husbands allocate their labour more productively
due to the incentives created by land rights. This is a new result to the household literature,
which has focused on intra-household allocation with total resources taken as given. The results
suggest that spouses’ outside options a¤ect household productivity and, thus, total resources
available for sharing.
Keywords: Household productivity, Consumption, Land rights, Matriliny, Malawi
JEL Classi…cation: D12, D13, J12, J16
Department of Economics and Nu¢ eld College, Oxford University, selma.telalagic@economics.ox.ac.uk. The author is grateful to Hamish Low for valuable guidance and to Toke Aidt, Wiji Arulampalam, Ian Crawford, Pramila
Krishnan, Christine Valente, Ansgar Walther and seminar paticipants at the University of Cambridge, University of
Oxford, Université Libre de Bruxelles, CSAE conference and EEA-ESEM Gothenburg conference for helpful comments. Financial support from the ESRC and Faculty of Economics, Cambridge is gratefully acknowledged.
1
1
Introduction
It has been recognised that spouses’ outside options a¤ect how a household allocates its total
resources (Lundberg, Pollak and Wales 1997, Du‡o and Udry 2004, Doss 1996, Hoddinott and
Haddad 1995). This paper asks a new question building on this literature: what is the impact of
spouses’ outside options on the total resources available to the household?1 In other words, do
outside options a¤ect productivity? To answer this question, I examine the case of rural Malawi,
where there is exogenous variation in the inheritance rights of land. Some households are matrilineal, where land belongs to the wife. Other households are patrilineal, where land belongs to the
husband. Descent is predetermined for any individual in Malawi. This provides a useful laboratory
for analysing the e¤ect of outside options on productivity. Using consumption and labour allocation as indicators of productivity and land rights as an indicator of outside option, I …nd that
consumption is signi…cantly higher in households where land belongs to the wife. I show that men
use their endowment of time more productively in these households to generate a larger household
“pie”. The result makes use of the fact that Malawi has one of the highest divorce rates in Africa,
which makes it likely that outside options will be exercised.2
In Malawi, matrilineal and patrilineal tribal groups have co-existed since the mid-19th century
(Phiri 1988).3 Matriliny is considered to a¤ord women greater autonomy than patriliny (Lamphere 1974, Davison 1997, Johnson 1988, Johnson and Hendrix 1982, Dyson and Moore 1983).
Historically, British colonialists, Christian missionaries and other groups believed that matriliny is
detrimental to economic performance (Peters 1997). These beliefs resulted from the observation
that the South is the poorest area in Malawi and is at the same time predominantly matrilineal. In
addition to estimating the impact of outside options on productivity and consumption, this paper
disentangles region and descent in Malawi to assess whether these historical criticisms of matriliny
were justi…ed.
I show that the highest current consumption is observed for matrilineal households, where
women have stronger outside options than their husbands. In the raw statistics (Table 3), I show
that while the North is richer than the South, matrilineal households are richer, on average, in
each region. The regression results demonstrate that matrilineal households consume 10% more on
average in real terms than patrilineal households, once a rich set of control variables is included. I
also show that this higher consumption is not explained by lower savings. The higher consumption
of matrilineal households is con…rmed for per capita expenditure and equivalent expenditure. The
1
In this paper, I de…ne an individual’s outside option as his or her utility when divorced.
Lifetime divorce probabilities in Malawi are between 40 65% (Reniers 2003). Over 40% of women remarry
within the …rst two years after a divorce.
3
A matrilineal descent system is where inheritance passes through the female line. Family land is passed down
from mother to daughter or, more traditionally, from brother to sister’s son. A patrilineal descent system is where
inheritance passes through the male line, from father to son. This a¤ects property division following divorce, where
household land accrues to the wife in matriliny but to the husband in patriliny. Matrilocal residence is where a couple
locates in the wife’s village after marriage. Patrilocal residence is where a couple locates in the husband’s village
after marriage. In this chapter, I use the terms ’kinship’and ’descent’interchangeably to describe the system of land
inheritance that a household follows.
2
2
consumption gap is attributed to di¤erences in labour allocation: men use their time endowment
more productively in matrilineal households.
I develop a theoretical framework of husbands’labour allocation decisions in Malawi that explains why households may su¤er when land rights accrue to men. Wage labour in Malawi is
predominantly carried out by men, who face a decision between allocating time to agriculture and
wage work. Where land belongs to men, husbands are residual claimants of agricultural income
should the couple divorce; as a result, they allocate more time to agriculture at their optimal choice.
As wages are higher than the average product of agricultural labour, this leads to lower household
income and, consequently, lower consumption than for those households where land belongs to
women. There is a mismatch between what is individually optimal and what is optimal for the
household: patrilineal husbands are better o¤ engaging in more agricultural labour, even though
their spouse and children would bene…t if they switched to wage labour. Support for this hypothesis
is obtained in an analysis of labour allocation, which shows that patrilineal husbands allocate more
time to agricultural labour and less to wage labour. They also have lower wage earnings. I also
verify that wages are signi…cantly higher than agricultural productivity, a key assumption of the
framework.
I also analyse how the intra-household allocation of resources to private goods depends on land
rights. I …nd that a signi…cantly higher share of expenditure is devoted to sons’ education and
men’s clothing in patrilineal households, which is consistent with the idea that husbands have
higher outside options in these households.
This chapter makes an important contribution to the household economics literature: it assesses
the e¤ect of spouses’ outside options on the productivity of the household. To the best of the
author’s knowledge, this is a new question to this literature. The chapter also makes a contribution
to the property rights literature by showing that although strong property rights increase individuallevel investment in Malawi, this may not be a desirable outcome at the household level. Where
non-cooperative decision-making about labour allocation cannot be alleviated, weakening men’s
land rights can help achieve a second-best outcome.
The chapter relates to existing empirical work on the e¤ects of asymmetric land rights of spouses.
Udry (1996) examines agricultural production in households in Burkina Faso and demonstrates that
the productivity of husbands’plots is signi…cantly higher than that of their wives’plots, suggesting
that a reallocation of resources could result in a better outcome for the household. One reason
why household outcomes may be ine¢ cient is tenure insecurity: Udry and Goldstein (2008) argue
that women in Ghana do not make long-term investments in their land by leaving it fallow because
fallow land is likely to be expropriated.
Furthermore, my results are related to the literature on property rights. Besley (1995) argues
that individuals may underinvest in their plots due to a risk of expropriation.4 Some authors
argue that men’s weak land rights in matrilineal households in Malawi result in lower long-term
4
Empirical studies supporting this e¤ect include Ali et al. (2011) and Deininger and Jin (2006). In contrast,
Brasselle et al. (2002) …nd no impact of land rights on investment.
3
investment and, as a result, lower income (Place and Otsuka 2001, Kishindo 2010). In this chapter,
I …nd that weak land rights do reduce investment: matrilineal men spend less time on agricultural
work than patrilineal men. However, this is bene…cial to the household because weak land rights
help align private bene…ts more closely with household bene…ts.
The chapter is structured as follows. In Section 2, I present a framework for husbands’incomeearning decisions in Malawi. In Section 3, I describe the empirical strategy and in Section 4 I
estimate the e¤ect of descent on consumption. Section 5 examines labour allocation and Section 6
explores intra-household allocation. Section 7 concludes.
2
Household Decision-Making in Malawi
2.1
Marriage in Malawi
In this Subsection, I describe in more detail the rules governing land rights and other features of
marriage in Malawi. Malawi is a poor country: 57% of rural households are at the poverty line.5
In rural Malawi, individuals belong to tribes, whose rules are particularly important for family life.
Typically, tribes follow either matrilineal or patrilineal descent. Women are considered to be more
autonomous in matrilineal than patrilineal communities because they have access to land. Land
access is especially important in rural, horticultural societies such as Malawi, because it is a crucial
source of livelihood. The other main asset is labour (Takane 2008).
In Malawi, kinship is spatially correlated. Figure 1, a map of Malawi, depicts this dispersion by
district.6 Darker shading represents districts where matriliny is more prevalent, relative to patriliny.
In the Southern region, most districts are predominantly matrilineal. In the Central region, there is
a more equal balance of matrilineal and patrilineal communities. The Northern region has a strong
patrilineal presence.
In matrilineal marriages, the woman receives land from her natal kin. The couple works on this
land as long as they are married. The husband is expected to work for his wife’s family and show
that he is hard-working and useful (Roberts 1964). Should the couple divorce, the wife keeps all of
the family land that she has been given. She continues to work on it and does not have to remarry.
The husband, on the other hand, has to return to his village. He does not have any claims to his
wife’s land. He may be given a temporary plot of land owned by his family to work on, with the
understanding that this is only until he can …nd another woman to marry. Marriage is the crucial
way that a man obtains access to land in a matrilineal setting (Kishindo 2010).
Patrilineal marriages are in many ways the opposite of matrilineal marriages. Marriage is the
primary way that a woman can obtain access to land in a patrilineal setting. The husband receives
5
In 2010. Figure from World Bank (http://data.worldbank.org/country/malawi).
The prevalence of matriliny and patriliny by district is calculated based on the Living Standards Measurement
Study data used in the empirical section of this chapter. For the purposes of this map, in those villages where both
types of descent are practised, half of the households are apportioned matrilineal descent while half are apportioned
patrilineal descent. The …gures are weighted based on the sampling strategy of the data (see footnote 19).
6
4
Figure 1: A map of Malawi depicting the prevalence of matriliny and patriliny by district.
5
land from his natal kin, which the couple use to earn their livelihoods.7 If the couple divorce, the
wife has no claims to the household land and is forced to return to her natal village, where she
may receive a temporary plot of land. She faces pressure to remarry, however. Her family may
discourage divorce because of the risk that her bridewealth will need to be returned (Schatz 2002).8
Marital residence adds an interesting nuance to the outside options of spouses. Living in the
wife’s village (matrilocality) cements a woman’s power in the household because she is surrounded
by her kin, while her husband is a stranger in the village; the wife’s brother plays a particularly
strong role in the matrilineal-matrilocal household, with the wife often taking orders from him rather
than her husband (Phiri 1983). In contrast, patrilocality can increase the husband’s outside option
because he is surrounded by his own kin. A matrilineal man is only likely to reside patrilocally if
he has no sisters or if the family is particularly wealthy (Peters 1997). However, if a matrilineal
man is the eldest brother to several sisters, he may be expected to reside in his natal village so that
he can look after his sisters and the family land. Although patrilineal couples almost always locate
patrilocally, marital residence is a¤ected by circumstances.9 There may be an endogenous element
to a matrilineal or patrilineal couple’s marital residence. This is in contrast to lineage, which is
exogenous.
These rules for land access imply that matrilineal men tend to own less land then patrilineal
men, while matrilineal women tend to own more land than patrilineal women. Evidence of this
is given in Appendix A. This is important because it determines the outside options of spouses.
Matrilineal husbands have low outside options, as do patrilineal wives. This makes divorce more
accessible for matrilineal women and patrilineal men.
Although land rights follow a clear set of rules based on descent, the rules for other property
are less clear-cut, especially in the case of divorce. Consumption goods tend to be shared equally
on divorce.
There is a strong degree of labour specialisation in rural Malawi. Almost all households derive
a substantial amount of their income from agriculture. Women tend to engage in agricultural
labour, performing many tasks on their own (Hirschmann and Vaughan 1983). Men usually work
agriculturally and for wages. It is rare for women to work for wages, unless they are unmarried
(Spring 1995). This implies that men are predominantly responsible for providing a household’s
consumption goods (Schatz 2002). Domestic labour is predominantly carried out by women (Spring
1995). Matrilineal women have more control over household labour decisions than patrilineal women
7
The question of primogeniture, where inheritance passes to the oldest son at the expense of other sons, is
important to address here. There is no de…nitive evidence on whether this takes place in Malawi. However, the
important distinction is between the land access rights of a husband and his wife, rather than a husband and his
brothers. Even if a man has older brothers and inherits less land than them in a patrilineal setting, he still inherits
some land or at least has access to the family’s land should he need it. This is in contrast to his wife who, by virtue
of being a patrilineal woman, has no land rights at all.
8
Bridewealth is when the husband-to-be or his kin pay an amount in money or kind to the woman’s family. Some
researchers argue that the origins of bridewealth are in compensating the woman’s family for their economic loss due
to the value of the woman’s labour.
9
Matrilocality is not in line with patrilineal kinship traditions and would indicate poor economic circumstances
that forced the couple to move to the wife’s village.
6
(Davison 1997).
Historically, matriliny has been attacked on various economic grounds. In particular, matrilineal
marriages have been said to deter husbands from investing resources to improve household land,
since they have no rights to it should the couple divorce (Phiri 1983). Richard Kettlewell is a popular
example of a colonialist who held a grudge against matriliny because of this hypothesised e¤ect of
tenure insecurity (Peters 2002, Simpson 2000). Despite negative outside in‡uences, matriliny has
remained surprisingly prevalent in Malawi, with around 60% of rural households being matrilineal.
2.2
A Framework for Income-Generating Decisions
In this Subsection I present a framework for analysing the income-generating decisions made by
husbands in Malawi, who choose their labour allocations non-cooperatively as in, for example, Ulph
(1988).10 Husbands decide on how to split their time between wage work and agricultural work; each
type of labour generates household income that is used for consumption. In addition, I assume that
wives supply agricultural labour inelastically. This framework is justi…ed by the labour patterns
observed in rural Malawi, where men work both agriculturally and for wages whereas women tend
to spend most of their time on agricultural and domestic work. There is a social barrier to women
working for wages.11
The husband’s decision is a¤ected by an exogenous probability of divorce.12 The aim of this
framework is to analyse husbands’ investment decisions under uncertainty; this relates to a rich
literature on the impact of property rights on investment (e.g. Besley 1995). Husbands choose
whether to invest in assets that may be expropriated at a later date. Intuitively, the higher the
probability of expropriation, the less investment there is in equilibrium. A similar result is expected
in the present framework: patrilineal men, who have stronger land rights than matrilineal men,
should invest more in their land. This framework demonstrates under which assumptions this is
the case.
A household consists of a husband and wife who each live for two periods, t = 1; 2.13 They
10
This assumption lies in contrast to the collective model of household decision-making (Chiappori 1988), which
assumes that outcomes are Pareto e¢ cient. I assume a non-cooperative framework because of the evidence against
Pareto e¢ cient outcomes in households in developing countries, such as Udry (1996), Udry et al. (1995) and Dercon
and Krishnan (2000).
11
Women may make a decision on a di¤erent dimension: descent may a¤ect their choice between domestic labour
and agricultural labour, if weak property rights mean that they would prefer to invest in their children. As a result,
patrilineal women may spend less time on agriculture and more time on domestic labour than matrilineal women.
However, Section 5 will show that this e¤ect on wives’agricultural labour is not observed.
12
Since Malawi has one of the highest divorce rates on the continent (Reniers 2003), incorporating divorce is
important. I discuss the e¤ect of relaxing this assumption at the end of this Subsection.
13
There are no children in this model. If children consume but do not produce, a simple way of introducing children
would be to include a proportional ’tax’ on all consumption, reducing the amount that both the husband and wife
receive. This would not change the predictions of the model. Assuming that children contribute economically is
unlikely to change the predictions either, as children’s labour is controlled by parents; therefore, their optimal labour
allocation would be similar to their parents’optimal labour allocation, reinforcing the predicted e¤ects. Another way
in which children could a¤ect parents’ behaviour is if they are a vehicle for investment. However, the ancestry of
children is traced in the same way as that of land, so that children belong to the mother in a matrilineal household
and to the father in a patrilineal household. Thus, investment in children should mimic investment in land.
7
marry with an exogenous amount of household land (L1 ) just before the start of the …rst period.
In the …rst period, they derive utility from current consumption, where the husband’s consumption
is denoted by c1 . Spouses also place value on the next period with discount factor
2 [0; 1]. The
husband splits his time, 1, between two types of labour: agricultural and wage (h1 ; 1
h1 ). The
wife supplies 1 unit of agricultural labour inelastically; as a result, all production functions are
written in terms of husband’s labour only. Wage labour allows the couple to enjoy consumption
immediately through the wage rate w1 . The husband receives an exogenous share
2 [0; 1] of the
wage income he earns; the remainder is consumed by his wife. The original stock of land depreciates
at rate D 2 [0; 1]. The husband’s labour increases the value of the land through the function f (h1 ).
Income from agricultural labour is enjoyed in the second period.
In the second period, the couple may be divorced or married, which is determined by an exogenous probability ( ).14 Thus, there are two states of the world in the second period, single and
married: j = S; M . As before, the spouses derive utility from their consumption, with the husband’s
consumption denoted by c2 . Each spouse has a new labour allowance; the husband makes a choice
between agricultural and wage labour (h2 ; 1
h2 ) while the wife continues to supply agricultural
labour inelastically. The wage rate is w2 and wage income is consumed immediately. The spouses
also reap consumption from the agricultural good they invested in in the previous period; this can
be interpreted as consuming crops that have taken one period to grow. Any agricultural labour in
the second period further increases the amount of consumption generated by this land good. The
production function is given by g(Aj2 ; hj2 ), where Aj2 is the amount of land in state j. If the couple
remain married, they each enjoy consumption from the full amount of land, L2 ; in this case, land
is a public good and AM
2 = L2 . If they divorce, the husband receives a share
2 [0; 1] of the
land and the wife receives the rest. This parameter measures the property allocation rule following
divorce and captures the strength of the spouses’relative outside options, with
households and
! 1 for patrilineal
! 0 for matrilineal households. Divorce determines whether income is shared:
wage and agricultural income are split according to the sharing rule
if the couple remain married;
otherwise, they enjoy agricultural income from the amount of land they receive on divorce and only
the husband consumes wage income, as the wife does not engage in wage labour.15
2.2.1
The husband’s optimal labour allocation
The main focus of this analysis is the husband’s labour decision, which he makes optimally by
maximising his lifetime utility. Let u(ct ) denote the husband’s utility function from consumption
in period t. I assume that the functions u( ); f ( ) and g( ) are concave and twice di¤erentiable.
14
In a related paper (Telalagić 2012), I relax this assumption and allow women to divorce their husbands when
they do not generate su¢ cient consumption goods.
15
A crucial assumption here is that
is independent of : the share of consumption received in marriage does
not depend on outside options. This may seem like a strong assumption, but it is justi…ed in the Malawian context
because consumption goods tend to be shared equally on divorce, rather than in line with land allocation. This
implies that if the model were to include savings through durables, men would receive a share
of durables on
divorce. I will examine how intra-household allocation of consumption depends on kinship in Section 6 to test the
assumption of the independence of and .
8
I also assume that the two inputs in g( ) are complementary as in, for example, a Cobb-Douglas
production function. Using U1 (L1 ) to denote the husband’s value function in period 1 and U2S (L2 )
and U2M (L2 ) to denote his value functions in the second period when single and married respectively,
the husband solves the following problem:
U1 (L1 ) = max u(c1 ) + ( U2S (L2 ) + (1
c1 ;h1
s:t:
:
c1 = w1 (1
L2 = (1
)U2M (L2 ))
h1 )
D)L1 + f (h1 );
where
U2S (L2 ) = max u(cS2 )
S
cS
2 ;h2
s:t:
cS2 = w2 (1
:
U2M (L2 ) =
s:t:
hS2 ) + g( L2 ; hS2 );
max u(cM
2 )
M
cM
2 ;h2
cM
2 = (w2 (1
:
M
hM
2 ) + g(L2 ; h2 )):
S
M
After substituting for c1 , cS2 and cM
2 , the husband has three choice variables: h1 , h2 and h2 .
The optimal choice of h1 is given by the following …rst-order condition:
@U1 (L1 )
@h1
=
w1 u0 (c1 ) + f 0 (h1 )(
+(1
) u0 (cM
2 )
u0 (cS2 )
@g( L2 ; hS2 )
@L2
@g(L2 ; hM
2 )
) = 0:
@L2
(1)
This shows that the husband sets his labour supply such that the marginal utility from wage
labour is equal to the marginal utility from agricultural labour. These marginal utilities depend not
only on the wage rate and marginal product of labour, but also on the sharing rule for consumption,
the property rights regime, the divorce rate and the husband’s preferences.
In the second period, the husband chooses h2 depending on whether he is single or married,
such that
9
@U2S (L2 )
@hS2
@U M (L2 )
@hM
2
= u0 (cS2 )(
@g( L2 ; hS2 )
@hS2
= u0 (cM
2 ) (
w2 ) = 0;
@g(L2 ; hM
2 )
M
@h2
w2 ) = 0:
The husband sets the marginal product of wage labour equal to the marginal product of agricultural labour in each state.
2.2.2
The e¢ cient labour allocation
To assess the extent to which the husband’s optimal labour allocation is ine¢ cient, it is necessary
to derive a benchmark. I de…ne the e¢ cient labour allocation to be that which maximises household welfare. The household’s value function in period 1 is denoted by HW1 (L1 ). I assume that
household utility is a weighted average of the husband’s and wife’s utilities, where the husband has
weight
. The wife’s utility function is denoted v(~
cjt ) with
2 [0; 1] and the wife has weight 1
value function Vtj , where c~jt is the consumption she receives in state j and period t. The wife’s
consumption is …xed in the divorced state, as she does not have a labour allocation decision; it is
de…ned to be c~S2 = s((1
receives (1
)L2 ), where s( ) is a concave function of land input. While married, she
)et , where et is household consumption in period t. Note that et = ct .
There are no externalities in the second period; the husband’s individually optimal choice is also
socially e¢ cient. To see this, consider the husband’s maximisation problem in the second period. In
the married state, the husband chooses labour supply to maximise e2 , his share of second-period
consumption. Since this is proportional to the wife’s share, his choice also maximises her secondperiod consumption, so that it is optimal from the household’s perspective. In the divorced state,
the wife’s consumption is independent of the husband’s choice. Therefore, the husband’s choice is
optimal as long as it maximises his consumption, which is the case by de…nition.
Due to the absence of externalities in the second period, the e¢ cient labour allocation is equivalent to the choice of a ‘household planner’ who dictates the …rst-period labour decision h1 and
leaves the husband to make his own decision in the second period. Therefore, the e¢ cient labour
allocation is de…ned by the solution to the following problem:
HW1 (L1 ) = max u( e1 ) + (1
e1 ;h1
+ ( ( U2S (L2 ) + (1
+(1
s:t:
:
h1 );
D)L1 + f (h1 );
10
)e1 )
)V2S (L2 ))
)( U2M (L2 ) + (1
e1 = w1 (1
L2 = (1
)v((1
)V2M (L2 )))
where
V2S (L2 ) = v(~
cS2 );
c~S2 = s((1
V2M (L2 ) =
s:t:
:
)L2 );
max v((1
)e2 )
e2 ;hM
2
M
hM
2 ) + g(L2 ; h2 ):
e2 = w2 (1
The optimal choice of h1 is given by the following condition:
@HW1 (L1 )
@h1
=
w1 (
u0 ( e1 ) + (1
+ f 0 (h1 )( (
If
+(1
)(1
+(1
)(
)v 0 ((1
)(1
)e1 ))
@g( L2 ; hS2 )
@L2
@s((1
)L2 )
)v 0 (~
cS2 )
)
@L2
u0 (cS2 )
u0 ( e2 ) + (1
)(1
)v 0 ((1
)e2 ))
@g(L2 ; hM
2 )
) = 0:
@L2
= 1, this condition is identical to condition (1). As long as the wife has some weight in the
household welfare function (so
6= 1), the husband’s optimal choice of h1 is not e¢ cient. Letting
h1 denote the husband’s optimal choice of h1 and hE
1 the e¢ cient choice of h1 , the relationship
between the two choices is explained in the following special case.
Special case
= 12 :
Assumption 1
The sharing rule in marriage is equal:
Assumption 2
The utility functions describing the husband’s and wife’s preferences are the
same: u( ) = v( ).
Suppose Assumptions 1 and 2 hold. Then there is overinvestment in land ( h1 > hE
1 ) when
= 1 and underinvestment in land ( h1 < hE
1 ) when
such that h1 =
= 0: Moreover, there exists a
hE
1.
E
2 (0; 1)
Proof. In Appendix B.
The husband overinvests in land when he is the residual claimant of this asset on divorce and
underinvests when he has no claim to it on divorce. His labour decision impacts the utility of the
wife because she receives a share of the consumption he generates; however, the husband does not
internalise this externality. As a result, his labour decision is ine¢ cient.
11
2.2.3
The e¤ect of kinship on the husband’s labour allocation
Although the special case predicts the extent of overinvestment or underinvestment at the boundary
levels of , it does not predict how the husband’s individually optimal agricultural labour changes
with . This is captured by the comparative static
dh1
d ,
whose sign is derived here. It is possible
to derive this sign by totally di¤erentiating the husband’s …rst-order condition for the …rst period
with respect to . First, the following assumption needs to be made:
Assumption 3
L2 (
where g 0 denotes
@g( L2 ;hS
2)
,
@L2
g 00 denotes
g 00
u00
+ g 0 0 ) < 1;
0
g
u
@ 2 g( L2 ;hS
2)
,
@L22
u0 denotes u0 (cS2 ) and u00 denotes u00 (cS2 ).
This condition can be interpreted as a constraint on the curvature of the functions g( ) and u( ). In
particular,
00
L2 gg0 is the production function equivalent to the coe¢ cient of relative risk aversion
for a utility function; the term
00
L2 g 0 uu0 has a similar but less direct interpretation. This assump-
tion implies that both the production function and utility function should not be too concave; in
other words, the rate of diminishing marginal product and diminishing marginal utility should not
be too high. In the case of linear functions, for example, this is always satis…ed as g 00 and u00 are
equal to zero. These assumptions appear reasonable in the context of Malawi. At the levels of
production that households …nd themselves, it is unlikely that strong diseconomies of scale occur.
The following proposition predicts the e¤ect of
on labour allocation.
Proposition 1 If Assumption 3 holds, an increase in the share of land accruing on divorce leads
to an increase in …rst-period agricultural labour at the expense of …rst-period wage labour:
Similarly, second-period agricultural labour is increasing in
in both states:
M
dhS
2 dh2
d ; d
dh1
d
> 0.
> 0.
Proof. In Appendix B.
This proposition states that husbands with a high value of
will have higher agricultural labour
in both periods than husbands with a low value of , assuming that the functions g( ) and u( )
do not exhibit strong diseconomies of scale. Thus, matrilineal men will invest less labour time in
household land than patrilineal men. In addition, if matrilineal households can be accurately de…ned
by the case
= 0 and patrilineal households can be accurately described as having
= 1, then
the special case implies that the e¢ cient …rst-period labour allocation lies somewhere between the
labour allocations of matrilineal and patrilineal men. Since
increases the amount of land available
in the second period due to its positive e¤ect on …rst-period agricultural labour, this also means
that more time is spent on agricultural labour in the second period.
The intuition for why patrilineal men spend more time investing in land than matrilineal men
is that the marginal bene…t of agricultural work is higher for the former than the latter.
12
2.2.4
The E¤ect of kinship on household consumption
For empirical purposes, a household’s consumption is only observed when the couple is still married.
Therefore, the e¤ect of kinship on household consumption can be predicted from the theoretical
framework by analysing the e¤ect of
on e1 and e2 . First, the following assumption is made:
Assumption 4
w2 >
@g(L2 ; hM
1 @g(L2 ; hM
2 )
2 )
+
:
M
M
dh
@L2
@h2
2
dL2
This assumption provides a lower bound on the second-period wage. Increasing
increases
…rst-period agricultural labour, which bene…ts the husband in two ways: it increases the marginal
product of second-period agricultural labour and increases the amount of land available in the
second period. On the other hand, more land means that the husband is less inclined to work for
wages; this results in a loss in wage income. Therefore, intuitively, Assumption 4 says that when
increases and therefore …rst-period agricultural labour increases, the bene…t of this through its
e¤ect on second-period land and agricultural labour is less than the cost through its e¤ect on wage
income. A necessary condition for this to be the case is that the wage is greater than the marginal
product of agricultural labour, which I will test in Section 5.
In the following proposition, I derive predictions on the e¤ect of
on the household’s consump-
tion in the …rst period and in the second-period married state.
Proposition 2 If Assumption 3 holds, then
de2
d
de1
d
< 0. If, in addition, Assumption 4 holds, then
< 0.
Proof. In Appendix B.
Proposition 2 states that a household’s consumption is decreasing in the husband’s share of land
owned. A necessary condition for this to be the case is that the wage is larger than the marginal
agricultural product. Studies have calculated these two productivity measures in developing countries and frequently concluded that wages far exceed the agricultural marginal product, termed a
‘productivity gap’. For example, Gollin et al. (2012) use household surveys to show that even after
considering sectoral di¤erences in hours worked and human capital, an agricultural productivity
gap remains. Vollrath (2009) calculates the ratio of the marginal product of labour in industry to
that in agriculture for many countries; Malawi has the second highest value in the sample, with a
ratio of 13.7. This implies that the marginal product of labour in industry is 13.7 times higher than
the marginal product of labour in agriculture. If this is correct, then two identical households in
Malawi that only di¤er in their labour allocation decision between wage work and agriculture will
have vastly di¤erent incomes. This is linked to early development economics, which argues that
structural change, involving the movement of workers from the primary sector into the secondary
13
and tertiary sectors, is crucial for development (e.g. Rostow 1960). Together, these studies provide
strong support for Assumption 4.
The framework o¤ers predictions on the consumption and labour allocation of matrilineal and
patrilineal households: in particular, households with higher values of
should have lower con-
sumption and higher agricultural labour by men than households with lower values of . A list of
the assumptions and predictions that will be tested in this paper is in Table 1.
Table 1: Predictions and assumptions that will be tested
Prediction/assumption
Section where tested
de1 de2
d ; d
4. Consumption
w>
<0
@g(L2 ;hM
2 )
@hM
2
M
dh1 dh2
;
d
d
5. Labour
>0
5. Labour
corr( ; ) = 0
6. Intra-Household Allocation
is higher for patrilineal HHs
2.2.5
Appendix A
Extensions to the model
There are several possible extensions to the model; I discuss the e¤ect of these extensions on
Proposition 1. First, one can relax the assumption of exogenous probability of divorce. In Malawi,
divorce rates tend to be higher in matrilineal areas where
is low; allowing for this in the model by
assuming a function ( ) would reinforce the e¤ect in Proposition 1 at high levels of
but dampen
it at low levels. This is because a higher divorce rate increases the probability of exercising one’s
outside option. Men with low
would allocate more time to agricultural labour but men with high
would allocate even more time to agriculture. The overall prediction of Proposition 1 would be
maintained.
Including divorce as a choice variable would allow men to choose between the divorced and
married states in the second period. In patrilineal communities, men hold the power to divorce
their wives. Since wives do not make any labour choices, allowing men to choose whether to divorce
would have little e¤ect on their behaviour. Therefore, the sign of
dh1
d
would not change. On the
other hand, in matrilineal settings, women tend to hold the power to divorce their husbands. Women
may use this power to a¤ect men’s behaviour. To allow for this, the divorce probability would have
to depend on the …rst-period labour choice in a pre-de…ned way. One possible assumption is that
women divorce men who do not help them su¢ ciently with their land, which would drive men to
spend more time on agriculture. This e¤ect goes in the opposite direction to the predicted e¤ect
of Proposition 1. A second possible assumption is that men obtain power by working for wages
because they have a source of income that women are unable to earn, which would drive them to
spend less time in agriculture. This would reinforce the e¤ect in Proposition 1. In fact, Kerr (2005a)
14
argues that increasing wage earnings increases men’s power in households in Malawi. In addition,
Telalagić (2012) demonstrates that women in Malawi provide active incentives to encourage men
to earn wage labour. Therefore, it is likely that the second assumption about behaviour is correct,
reinforcing the e¤ect in Proposition 1.
Another possible extension is to introduce savings through assets such as durables, which requires wage income. Such savings would be lower for those men who focus on agricultural labour.
However, the existence of savings would not a¤ect the substitution between agricultural and wage
labour unless the rate of return on assets or utility from assets is particularly high.
A further way to extend the model is to include agricultural income in the …rst period. In this
sense, agricultural labour could provide immediate consumption. The primary force that drives
patrilineal men toward wage labour is that agricultural labour does not generate any …rst-period
consumption; this would no longer be the case, so the predicted sign of
dh1
d
at high levels of
would
be reinforced. For matrilineal men, the existence of …rst-period agricultural income would make
agricultural labour more attractive; however, this is assuming that agriculture is a productive way
of earning income, which is unlikely to be the case. Therefore, the overall predicted sign of
dh1
d
would not be a¤ected.
3
Empirical Strategy
The general relationship this paper aims to shed light on is the e¤ect of spouses’outside options
on household productivity. The speci…c relationship estimated is the e¤ect of land rights on consumption and labour allocation. To measure land rights, I use kinship: whether the household is
matrilineal or patrilineal. This is the best measure of land rights in Malawi, as kinship governs
how land is shared following divorce. However, kinship may capture other factors too, such as the
likelihood of divorce; I address these factors empirically.
In this Section, I explain how the e¤ect of land rights on consumption, labour allocation and
intra-household allocation will be identi…ed in the data. The sequence of the tests is as follows. In
the next Section, I test Proposition 2, namely that matrilineal households have higher consumption
than patrilineal households. For robustness, I examine alternative measures of consumption, sample
restrictions, alternative measures of wealth, savings and tribal …xed e¤ects. In Section 5 I test
Proposition 1, which predicts that matrilineal men allocate more time to wage labour and less time
to agricultural labour than patrilineal men. I also test the necessary component of Assumption
4, namely that the wage exceeds the marginal product of agricultural labour. For robustness, I
examine income and husbands’wage earnings. In Section 6, I examine the e¤ect of kinship on intrahousehold allocation to examine the consistency of the data with existing results on intra-houshold
allocation and to test the assumption of the framework that
Next, I explain each of the tests in more detail.
15
and
are independent.
3.1
Proposition 2: The E¤ect of Kinship on Consumption
In order to test Proposition 2 and analyse the e¤ect of kinship on consumption, I take advantage
of the fact that kinship is predetermined for any individual in Malawi. There is an exogenous
assignation of kinship across individuals. However, due to the way that tribes settled in Malawi,
kinship is not independent of geography. As geography is likely to a¤ect consumption both directly
and through other factors such as prices, covariates that are correlated with geography (G), kinship
and consumption need to be controlled for. These variables capture the exogenous factors that
enter the income generation function. I denote the vector of these covariates by Z. Let Di be a
dummy variable equal to one if household i is patrilineal and zero otherwise. Then, a regression
of consumption on kinship, geography and the covariates Z will give a causal e¤ect of kinship on
consumption as long as
fC1i ; C0i g ? Di j Gi ; Zi 8i;
where C1i is the potential consumption outcome of household i if it were patrilineal and C0i is
its potential consumption outcome if it were matrilineal. In words, conditional on geography and
other regional covariates, the potential consumption outcomes of households across the two kinship
types are independent of their kinship (Angrist and Pischke 2008). If, further, I include in Z all
variables relating to kinship that do not measure land rights, the regression will measure the causal
e¤ect of land rights on consumption. These land rights are then interpreted as capturing spouses’
relative outside options. This framework suggests the following regression equation, which will be
estimated using Ordinary Least Squares:
ln Ci =
+ Di + Gi + Zi + !Hi + ui ;
(2)
where Hi is a vector of household characteristics that are not correlated with descent but
may improve the precision of the estimates. The coe¢ cient of interest is
variable is the log of consumption,
.16 As the dependent
is interpreted as the mean percentage di¤erence between
the consumption of matrilineal and patrilineal households. The framework of the previous Section
suggests that
< 0. However, critics of matriliny would argue that
> 0. The value of
is an
empirical question that is answered in the next Section. Relating to the theoretical framework,
I assume that the sample consists of a mixture of households in their …rst and second periods.
Therefore, the empirical analysis tests for an average of
16
de1
d
and
de2
d .
For robustness, I also examine
The key assumption is that conditional on included covariates, Di is exogenous. If this is not the case, the
estimate of will be biased. Any omitted variables that a¤ect consumption outcomes are likely to be negatively
correlated with Di , implying that economic conditions favour matrilineal households. This is because summary
statistics (not reported) show that matrilineal villages are closer to urban areas and face lower constraints in soil
quality, on average. If this is the case, there will be a downward bias on : the true e¤ect of patriliny on consumption
will be more positive than estimated.
16
savings, which are a measure of future consumption. By analysing both current consumption and
savings, a good picture of lifetime consumption is obtained. I also examine per capita and equivalent
expenditure as measures of Ci for robustness.
3.2
Proposition 1: The E¤ect of Kinship on Labour Allocation
Using the same set of right-hand side variables as in (2), I test Proposition 1 by examining the
e¤ect of kinship on labour allocation. I estimate the following set of equations with Ordinary Least
Squares:17
hw
=
i
hai
hai w
ha+w
i
w
=
a
+
w Di
+
a Di
+
+
w Gi
a Gi
+
+
w Zi
a Zi
+ ! w Hi + uw
i ;
+ ! a Hi +
(3)
uai ;
(4)
=
a w
+
a w Di
+
a w Gi
+
a w Zi
+ !a
=
a+w
+
a+w Di
+
a+w Gi
+
a+w Zi
+ ! a+w Hi +
w Hi
+
uai w ;
ua+w
;
i
(5)
(6)
where hw denotes hours of wage labour by the husband, ha denotes his hours of agricultural
labour, ha
w
denotes the di¤erence between the hours of agricultural and wage labour and ha+w
denotes the sum of the hours of agricultural and wage labour. Proposition 1 implies that
0;
a
> 0;
a w
> 0 and
a+w
w
<
= 0. That is, patrilineal men spend less time on wage labour and
more time on agricultural labour compared to matrilineal men; in addition, the di¤erence between
the two types of labour is higher for patrilineal than matrilineal men, while the sum is no di¤erent
between the two kinship groups, implying a substitution e¤ect. I also test the necessary component
of Assumption 4, namely that wages are higher than the average product of agricultural labour.
The method is explained in more detail in Section 5.1.
3.3
The Independence of
and
In order to test the assumption that
and
are independent, I estimate a series of Working-Leser
expenditure share regressions:
egi =
g
+
g Ei
+
g Di
+
g Gi
+
g Zi
+ ! g Hi +
g Pi
+ ui ;
(7)
where g = 1; :::; n is a set of n categories of goods, egi is the share of total expenditure spent
on good g, Ei is the total expenditure of household i, Pi is a vector of the log of prices of various
goods and the remaining right-hand side variables are as in (2). Total expenditure is instrumented
17
I estimate these equations independently with the same right-hand side variables for all speci…cations.
17
with wealth, measured by house construction material and number of livestock owned, and a Two
Stage Least Squares procedure is used. If
and
are independent, then kinship will not a¤ect the
intra-household allocation of consumption and we will observe
To verify that
g
= 0 for all goods in g.
is higher for patrilineal households, I disaggregate land ownership by spouse
and kinship in Appendix A.
4
Consumption
4.1
The Data and Summary Statistics
The source of the data is the Malawi Living Standards Measurement Study (LSMS), conducted by
the World Bank and the Malawi National Statistical O¢ ce (NSO). Households were interviewed
between March 2010 and March 2011. In total, 12271 households were interviewed, of which 10038
resided in rural areas.18 I restrict the sample to rural households where the household head is
married, which gives a sample of 7350 households. The …nal sample consists of 7136 households,
due to some missing observations. Aggregate real consumption expenditure, both at the household
level and per capita, is provided in the data. I use the household-level measure for most of the
analysis. The consumption aggregate includes food purchased, produced for own consumption
and received as a gift, various household items, the rental value of durables, the rental value of
accommodation and expenditure on health and schooling. Consumption expenditure is de‡ated by
a temporal and spatial price index.19
Summary statistics are presented in Table 2. I disaggregate the data based on descent.20 Details
of variable de…nitions can be found in Appendix C. Matrilineal households own less land and have
fewer members on average, although matrilineal and patrilineal households have approximately
the same number of children on average. Mean education levels are similar across descent types.
However, patrilineal husbands and wives are slightly older on average. On the whole, there are no
major di¤erences in basic characteristics across descent types. The regional dispersion of descent
is clear from the table: while there are close to no matrilineal households in the Northern sample,
18
Villages were selected based on probability proportional to size. Households within these villages were randomly
selected. All summary statistics are weighted based on the probability of being sampled and clustered at the village
level.
19
The price index was calculated by the NSO. It consists of a spatial price index, calculated as a Laspeyres price
index using prices for 29 food items and 13 non-food items with base period February/March 2010, and a temporal
price index, calculated using the monthly CPI for the three regions.
20
Descent is measured based on the following question, which was asked to village informants: "Do individuals in
this community trace their descent through their father, their mother, or are both kinds of descent traced?" I label
the category where both kinds of descent are possible as ’dual descent,’even though strictly speaking, each household
will practise one or the other. I report the results for this category but focus on the distinction between matrilineal
and patrilineal households. It needs to be acknowledged that there could be a small element of endogeneity to this
variable, because it measures the descent traced in the village where the couple are resident, which may not be the
descent traced by the couple’s family. As a result, the choice of residence may a¤ect this. However, I assume that
this is not a problem, primarily because individuals are likely to reside in the village of one of the spouses’families.
As inter-marriage between matrilineal and patrilineal individuals is uncommon, the village is likely to have the same
descent pattern as the household itself.
18
53% of the Southern sample is matrilineal. Divorce rates are highest in matrilineal communities,
on average.21
Table 2: Summary statistics by descent
Patrilineal
Matrilineal
Dual descent
P-value P=M=D22
P-value P=M23
2:13
2:06
2:34
0:45
0:42
(0:06)
(0:07)
(0:24)
5:12
5:05
5:34
0:12
0:38
(0:06)
(0:04)
(0:14)
2:97
2:93
3:17
0:13
0:50
(0:05)
(0:04)
(0:11)
41:51
40:13
42:02
0:00
0:00
(0:39)
(0:25)
(1:10)
35:21
34:29
36:34
0:01
0:03
(0:34)
(0:23)
(0:90)
Any schooling (husb)
0:80
0:81
0:80
0:81
0:56
Any schooling (wife)
0:67
0:69
0:65
0:59
0:40
South
0:24
0:53
0:41
0:00
0:00
Centre
0:57
0:47
0:48
0:13
0:05
North
0:19
0:00
0:11
0:00
0:00
Divorce rate
0:08
0:12
0:11
0:00
0:00
(0:00)
(0:00)
(0:01)
Patrilocal
0:41
0:22
0:28
0:00
0:00
Matrilocal
0:08
0:17
0:14
0:00
0:00
N (number of obs.)
2448
4409
279
7136
6857
Land (total, acres)
HH size
# Own children
Age (husb)
Age (wife)
This table reports mean (standard error). Standard errors are not reported for dummy variables.
21
The divorce rate is measured at the district level. It is calculated from the entire LSMS sample and represents
the proportion of household heads who reported being separated or divorced. The …gures are consistent with those
in Reniers (2003), calculated from the 2001 Demographic and Health Survey of Malawi.
19
Further, there is a clear correlation between lineage and marital residence: patrilineal households
tend to be patrilocal, while matrilineal households are almost equally likely to be matrilocal or
patrilocal. Other residence types are possible, such as when both spouses are from the same
village.24
4.2
Expenditure Data
Summary statistics of real expenditure are in Table 3. The raw statistics in this table lie at the
heart of this paper. Much of what has been discussed in the theoretical framework can already be
seen at this level. While the South is indeed the poorest region, as observed by the colonialists and
missionaries, the same cannot be said for matrilineal communities. In fact, matrilineal households
consume more on average than patrilineal households in all regions, with a particularly signi…cant
di¤erence in the Southern region. It appears that patriliny is driving the poverty in the South.
The di¤erence in mean expenditure between matrilineal and patrilineal households is a statistically
signi…cant 11%. In the regression results, I expect to observe lower consumption and more agricultural labour in the South. However, there should be a positive e¤ect of matriliny on consumption
and wage labour over and above this.
Table 3: Real household consumption expenditure (’000s MWK) by descent and region
Patrilineal
Matrilineal
Dual Descent
All
P-value P=M=D
P-value P=M
206:78
209:43
263:17
210:26
0:45
N=A
(8:57)
(N=A)
(42:83)
(8:61)
930
9
79
1018
220:82
252:91
331:89
244:21
0:02
0:01
(8:60)
(8:73)
(81:77)
(7:01)
915
1881
111
2907
141:66
198:77
173:58
188:00
0:00
0:00
(9:79)
(7:81)
(20:88)
(6:57)
N
603
2519
89
3211
All
199:03
224:08
259:85
217:50
0:01
0:00
(5:95)
(5:95)
(45:62)
(4:65)
2448
4409
279
7136
7136
6857
North
N
Centre
N
South
N
This table reports mean (standard error).
22
The column "P-value P=M=D" reports the p-value on the test where the null hypothesis is that the value for
all three groups is the same.
23
The column "P-value P=M" reports the p-value on the test where the null hypothesis is that the value for
matrilineal and patrilineal households is the same. Households in dual descent villages are excluded from this test.
24
Other residence patterns also include neolocality, where a couple sets up their home in a new village, separate
from either spouse’s family. This includes couples from abroad.
20
4.3
Regression Results
In this Subsection, I test whether the di¤erence in mean expenditure observed in the summary
statistics persists when relevant variables are controlled for. This is a test of Proposition 2: do
patrilineal households have lower consumption than matrilineal households? For robustness, I
examine per capita and equivalent expenditure; I also use alternative measures of wealth, restrict
the sample to the Southern and Central regions only, examine savings, include tribal …xed e¤ects
and consider village …xed e¤ects in turn.
I estimate Equation (2), where the primary coe¢ cient of interest is that on the dummy variable
capturing patrilineal descent; I also include a dummy variable for dual descent while matriliny is
the base case.25 The results are presented in Table 4. Each regression includes a vector of basic
characteristics and further controls are added with each speci…cation.26 The key result is that
matrilineal households consume signi…cantly more than patrilineal households, on average, in all
speci…cations. I discuss each speci…cation in turn.
The …rst speci…cation includes basic controls only and no dummy variables for descent. Thus,
households in the South have 7:8% higher consumption than households in the North, while households in the Central region have 20:8% higher consumption than households in the North. Speci…cation (2) adds dummy variables for patriliny and dual descent: matrilineal households consume
19:0% more than patrilineal households on average. The di¤erence is much larger than the initial
di¤erence in means, suggesting that the control variables disadvantage matrilineal households with
respect to their consumption levels. In addition, the dummy for the Southern region loses signi…cance and the coe¢ cient on the Central region falls in magnitude, suggesting that descent explains a
large proportion of the regional variation. Adding geographical variables in speci…cation (3) further
controls for the spatial correlation of descent seen in Figure 1. I include a rich set of geographical
controls, covering temperature, rainfall, soil quality, greenness and agro-ecological zones.27 These
variables explain more than half of the gap between the average consumption levels of matrilineal
and patrilineal households. This suggests that patrilineal households are located within the worst
geographical areas in each region. Indeed, the regional dummy variables lose signi…cance in regression (3), implying that the geographical variables explain the remaining regional e¤ects that were
not explained by descent.
In some ways, including geographical controls accounts for Assumption 3, which addresses
the productivity of the land. The e¤ect of geography on expenditure is likely to work through
agricultural productivity: geography a¤ects the innate productivity of the land, which in turn
a¤ects income and thus expenditure. The coe¢ cients and signi…cance of the geographical variables
25
All standard errors are clustered at the village level; the sample is also weighted based on the sampling strategy,
which selected villages based on probability proportional to size.
26
The basic covariates that are included in all regressions are the amount of rainy and dry season land owned, the
age of the spouses, whether the spouses ever attended school, whether they can read English and/or Chichewa, the
year the consumption expenditure refers to, the year of the rainy and dry season that agriculture is reported on, the
month of the interview, household size and dummy variables for the Southern and Central regions. The Northern
region is taken as the base case.
27
Details of these variables can be found in Appendix C.
21
Patrilineal
Table 4: The e¤ect of descent on consumption
(1)
(2) +descent (3) +geo. (4) +econ.,demog. (5) +gender
Ln(real expenditure)
-0.190
-0.063
-0.094
-0.108
(0.036)
(0.032)
(0.032)
(0.033)
(6) +resid.
-0.104
(0.033)
South
0.078
(0.041)
-0.070
(0.050)
-0.069
(0.082)
0.042
(0.075)
0.078
(0.077)
0.088
(0.077)
Central
0.208
(0.043)
0.097
(0.047)
-0.014
(0.070)
-0.006
(0.067)
-0.013
(0.066)
-0.016
(0.066)
Immigration
0.105
(0.025)
0.103
(0.025)
0.101
(0.025)
Any business empl.
0.071
(0.029)
0.081
(0.029)
0.080
(0.029)
Any wage empl.
0.078
(0.025)
0.070
(0.025)
0.071
(0.025)
Divorce rate
-0.010
(0.006)
-0.009
(0.006)
Women’s group
0.040
(0.029)
0.040
(0.029)
Matrilocal
-0.088
(0.024)
Patrilocal
0.034
(0.019)
Y
Y
Y
Y
7136
0.369
Basic
Geography
Village economy
HH Composition
N
R2
Y
N
N
N
7136
0.262
Y
N
N
N
7136
0.277
Standard errors are reported in parentheses.
Y
Y
N
N
7136
0.333
Y
Y
Y
Y
7136
0.365
denotes signi…cance at 1% level,
22
Y
Y
Y
Y
7136
0.366
at 5% level and
at 10% level.
Table 5: The geographical variables in speci…cation (3)
Category
Variable
Coe¢ cient
Temperature Average daily range
0:016
Temperature seasonality
0:000
Min. temp. of coldest month
0:008
Avg. temp. of wettest quarter
0:007
Rainfall
Avg. 12-month tot. rainfall in 2009, 2010
Avg. rainfall in wettest quarter in 2009, 2010
Avg. start of wettest quarter in 2009, 2010
0:001 ; 0:001
0:001 ; 0:001
0:010; 0:014
Greenness
Total change in greenness in 2009,2010
Onset of greenness increase in 2009,2010
Onset of greenness decrease in 2009,2010
0:005 ; 0:007
0:006 ; 0:003
0:008 ; 0:001
Soil quality
Nutrient availability
Rooting conditions
Excess salts
F = 4:62
F = 22:15
F = 4:67
Agro-ecology
Agro-ecological zones
F = 1:32
Greenness is the emergence of vegetation at the beginning of the growing season.
The four agro-ecological zones in Malawi are tropic-warm/semiarid, tropic-warm/subhumid,
tropic-cool/semiarid and tropic-cool/subhumid.
denotes signi…cance at 1% level,
at 5% level and
at 10% level.
are reported in Table 5. Particularly signi…cant e¤ects are observed for temperature variance and
greenness, which is the onset of spring. Soil quality is especially signi…cant, which is intuitive as
this is a crucial factor that a¤ects crop growth. Most of the geographical variables have p-values
less than 1%, suggesting that they explain agricultural productivity well.
Regression (4) includes economic and demographic variables: employment opportunities28 , properties of the village economy29 and household composition.30
28
These characteristics a¤ect the
Employment opportunities are captured by three variables: Immigration, Any business employment and Any
wage employment. Immigration is a dummy variable de…ned by the community-level response to the question "Do
people come to this community during certain times of the year to look for work?" The employment sector variables
are de…ned by the community-level response to the question "Which activities are the three most important sources of
employment for individuals in this community?" Business employment includes beer-brewing, handicraft production,
small-scale industry and any other responses suggesting business. Wage employment includes small-scale trade and
service provision, large-scale commercial industry, professional occupations, civil service, and any other responses
suggesting wage work, such as working on an estate. If any of these activities are listed as one of the three most
important sources of employment for the village, a ‘1’is recorded for the relevant variable. Primary economic activities
(farming, …shing and selling …rewood) are excluded from the analysis because more than 99% of communities report
engaging in those.
29
This category consists of eight variables: one controls for the distance from the household to the nearest road
and one for the household’s distance to the nearest town with a population of at least 20 000, while the remaining
six measure the proportion of surveyed households (excluding the respondent) in the respondent’s village that farm
maize, tobacco, rice, groundnut, cassava and mango.
30
This category consists of variables measuring the number of household members that are male children, female
children, male adults, female adults, male elders, female elders and individuals whose gender or age is missing. It also
includes a dummy variable measuring the gender of the household head and a further dummy variable indicating the
presence of a brother-in-law in the household. When these variables are included, household size is omitted.
23
income-earning opportunity set and the trade-o¤ between di¤erent types of labour. Economic
and demographic characteristics do not explain the consumption gap, which is a statistically signi…cant 9:4%. Employment opportunities are important, however: the presence of immigration,
suggesting a strong village economy, raises average consumption by over 10%. The presence of
business employment raises consumption by 7% on average and the presence of wage employment
raises average consumption by almost 8%.
When variables relating to gender are included in regression (5), the consumption gap does not
change in a signi…cant way. The divorce rate and women’s status do not explain the consumption
gap.31 The summary statistics illustrated that marital residence is highly correlated with kinship. I
add marital residence, measured at the household level, in regression (6). Relative to other residence
patterns, matrilocality is damaging to consumption levels, while patrilocal households fare slightly
better on average. This suggests that, on average across all kinship types, patrilocal households
have the highest consumption. This is not inconsistent with the consumption gap observed between
patrilineal and matrilineal households, as matrilocality is the least common residence choice for all
kinship groups.
A household’s location decision may not be entirely exogenous, even if it is highly correlated
with descent. Patrilineal households have a limited ability to choose their residence. However,
wealthier men in matrilineal communities are more likely to locate patrilocally because they are
more likely to have access to their family’s land. Although I control for the amount of land owned,
there may be a marriage market factor that I am unable to capture.32 Due to this fact, I do not
include marital residence in the remaining regressions of this paper.
I choose regression (5) as the preferred speci…cation. Having controlled for basic characteristics, geography, economic characteristics, household composition and gender, a highly signi…cant
consumption gap of over 10% between matrilineal and patrilineal households persists. This gap is
consistent with the 11% gap observed in means. There is strong evidence to support Proposition
2, namely that matrilineal households consume signi…cantly more than patrilineal households.
4.4
Robustness Checks
In this Subsection, I carry out several robustness checks. First, I verify that the consumption
gap is observed in alternative measures of consumption. I replace real household expenditure in
regression (5) with per capita (pc) real expenditure and equivalent (eq) real expenditure. Equivalent
expenditure is a more accurate measure of per capita expenditure: it gives children a lower weight
than adults because the former consume less.33 These results are speci…cations (7) and (8) in Table
31
Gender includes the proportion of divorced household heads in the district and a dummy variable for the existence
of a women’s group in the village, which acts as a proxy for women’s status.
32
Bargaining power may be an unobservable omitted variable that could bias the results. It will be related to the
marriage market as well as to the choice of residence. However, there is little reason to believe that bargaining power
should a¤ect total consumption except through channels such as fertility and labour choices, which are observable.
In later sections I analyse labour and intra-household allocation, which may be more a¤ected by bargaining power.
33
The weights were chosen by NSO researchers and are as follows: 0.33 for children aged under 1, 0.47 for ages 1-2,
0.55 for ages 2-3, 0.63 for ages 3-5, 0.73 for ages 5-7, 0.79 for ages 7-10, 0.84 for ages 10-12, 0.91 for ages 12-14, 0.97
24
6. It is clear that the e¤ect of kinship on expenditure holds across these alternative measures
of expenditure and is not signi…cantly di¤erent from the gap in regression (5).34 This con…rms
that the consumption gap between matrilineal and patrilineal households is robust to alternative
measures of spending.
Second, I address the issue that land may be an inadequate measure of wealth, insofar as it
is inaccurate or has an endogenous element. In regression (9), I replace land with two alternative
measures of wealth: the number of livestock owned and the type of construction material used
for the house, with the best type (permanent) as the base case. Although the consumption gap
is slightly smaller in this speci…cation, patrilineal households still consume signi…cantly less on
average than matrilineal households.
Third, there is the issue that only nine households in the Northern sample are matrilineal. As
a result, there may be insu¢ cient variation in kinship in the North to provide accurate results. In
regression (10), I restrict the sample to the Southern and Central regions only. The coe¢ cient of
interest is a statistically signi…cant 10:8%, suggesting that the inclusion of Northern households
does not invalidate the results.
Fourth, I discuss savings. It may be that matrilineal households are impatient, in which case
they exhibit higher consumption today at the expense of future consumption. As a result, it would
be misleading to conclude that matrilineal households are more productive. In addition, examining
savings can shed some light on the sign of
de2
d .
Rural households in Malawi have limited savings;
their low income is associated with a high marginal propensity to consume. Direct data on the
amount of savings are not available and less than 1% of households report non-zero values of
interest earned on savings and pension income. Alternative measures of savings are the use value
of durables and the number of livestock, which 85% and 53% of households report having non-zero
values of respectively. Estimating speci…cation (5) but replacing consumption with the use value
of durables or the number of livestock does not yield a signi…cant coe¢ cient on patriliny in either
estimate (results not reported). This implies that these measures of savings are not signi…cantly
di¤erent across matrilineal and patrilineal households. The survey also includes a question on
the household head’s subjective assessment of whether household income is su¢ cient for building
household savings. Matrilineal households report that they are signi…cantly better able to build
their savings than patrilineal households, with a p-value of 0:00. The evidence suggests that either
savings are no di¤erent between matrilineal and patrilineal households or the former have higher
savings. Therefore, matrilineal households are likely to have higher consumption in the future, not
just at present.
Fifth, I consider the possibility of tribal …xed e¤ects. There may be characteristics of tribes that
correlate with descent and consumption but that do not a¤ect outside options, such as work ethic.
for ages 14-16 and 1 for ages 16 and up.
34
There is a possible issue of under-reported household size. There may be several households using the same
cooking facilities for example, even though they refer to themselves as separate households. In this sense it would
appear as though per capita consumption is higher than it actually is. Whether this is more likely to occur in
matrilineal households is ambiguous, however, so this remains a caveat on the results.
25
These need to be taken into account to ensure the validity of interpreting kinship as capturing
relative outside options. In Table 17 in Appendix D, I add a series of dummy variables measuring
the most spoken language in the community to speci…cation (5). As language is highly correlated
with tribe, this is a good measure of tribal …xed e¤ects. The coe¢ cient on patriliny is still signi…cant
and negative, with a slightly smaller magnitude. It is interesting that among those communities
speaking one of the three most common languages (Chewa, Yao and Tumbuka), those speaking
the predominantly matrilineal languages of Chewa and Yao have signi…cantly higher consumption
than those speaking the predominantly patrilineal Tumbuka. In addition, the fact that there is
a negative patrilineal e¤ect over and above these tribal …xed e¤ects further supports Proposition
2; for example, being in a tribe that speaks Chewa is bene…cial but being patrilineal invalidates
some of this positive e¤ect. Comparing the 10:8% coe¢ cient observed in speci…cation (5) to the
8:6% coe¢ cient observed here, it can be argued that there is a 2:2% consumption gap between
matrilineal and patrilineal households due to tribal customs and a 8:6% gap due to the direct e¤ect
of land rights, which are interpreted as capturing spouses’relative outside options.35
(7)
Table 6: Robustness checks
(8)
(9) + other wealth
Ln(pc real exp)
Patrilineal
-0.108
(0.033)
Ln(eq real exp)
-0.106
Ln(real expenditure)
-0.085
(0.033)
Semi-permanent
(10) S&C only
(0.032)
-0.108
(0.034)
-0.273
(0.023)
Traditional
-0.391
(0.021)
# Livestock
0.012
(0.001)
N
7136
7136
7136
6118
R2
0.369
0.351
0.421
0.365
Controls included: Basic (land excluded in (9)), Regions, Geography, Village economy,
Employment, HH Composition and Gender. Standard errors are reported in parentheses.
denotes signi…cance at 1% level,
at 5% level and
at 10% level.
35
An alternative way of capturing geography and other economic variables is to include village-level …xed e¤ects.
However, since kinship is also measured at the village level, there is insu¢ cient variation to estimate the e¤ect of
kinship when village-level …xed e¤ects are included. In fact, when I estimate regression (5), omitting all village-level
variables apart from kinship but including village-level …xed e¤ects, the patriliny variable is forcibly dropped: Stata
does not provide an estimate of the coe¢ cient due to multicollinearity.
26
5
Labour
The driving force of the predicted e¤ect of land rights on consumption in Section 2 is the men’s
labour decision, which I analyse in this Section. Thus far, the results have shown that matrilineal
households have higher consumption. According to the theoretical framework, this is because
matrilineal men allocate a greater share of their labour to wage work. The necessary assumption
for this to be a valid explanation of the results in Section 4 is that wage work is more productive
than agricultural work.36 I test, in the …rst instance, whether this assumption holds. Second, I test
for a di¤erence in labour allocation by kinship.37 For robustness, I also examine income.
5.1
Wages and agricultural productivity
In order to give an indicative idea of whether there is support for Assumption 4, I calculate the rate
of return to wage work and agricultural work for husbands in the sample. In line with the theoretical
framework, I assume that while married, agricultural income can be treated as a public good. Also
in line with the framework, I assume that wages are kept by husbands on divorce but agricultural
income is allocated based on kinship: matrilineal husbands do not receive any agricultural income
on divorce, while patrilineal men receive all of their agricultural income. In order to take these
expectations into account, I calculate the annual probability of divorce by assuming a model where
individuals move between the two states of marriage and divorce according to a Markov chain (see
Appendix E for full calculations). The probability of divorce in each period can be calculated if the
stationary proportions of married and divorce individuals are known, along with the probability
of moving from the divorced to the married state (the remarriage probability). I assume that the
current distribution of married and divorced individuals in the nationally representative LSMS
sample is the stationary solution to this model. I calculate the remarriage probability using the
Malawi Longitudinal Study of Families and Health (2010), which asks husbands about when their
marriages started and ended. Using this information, I calculate the average proportion of divorced
husbands who remarry within a year. This is the annual remarriage probability. I calculate this
separately for patrilineal and matrilineal husbands. I feed this information into the model to yield
the annual divorce probability. These …gures are in Table 7.
The return to wage work is calculated as the average hourly wage in the sample. Respondents
engaged in wage work reported the amount of their last salary and the period of time it covered;
from this, I calculate the hourly wage of each individual. I also calculate the average hourly wage
paid by the Malawi Social Action Fund (MASAF) public works programme for comparison; this is
a lower bound on wages as the programme o¤ers a social safety net in particularly poor villages.38
36
A caveat on this is that if labour and consumption are codetermined, the former will not explain the latter.
However, in the theoretical framework, the assumption is that the husband treats the two decisions separately.
dhM
1
The theoretical framework predicts that dh
> 0 and d 2 > 0. Since I have assumed that the sample consists of
d
a combination of households in their …rst and second periods, I test for a weighted average of these two e¤ects.
38
The average hourly MASAF wage is calculated as the average of the reported male and female MASAF wages,
divided by four. This is because the daily MASAF tasks have been estimated to take four hours (Chirwa, Mvula
37
27
The wage is purposefully set below the market wage (Dzimadzi and Chinsinga 2004); in this sense
it operates di¤erently to employment guarantee schemes elsewhere (such as the National Rural
Employment Guarantee of India, which sets wages above the market clearing wage).
The agricultural product is calculated as the estimated value of consumption from own production in the last year divided by the annualised number of hours of own-farm agricultural labour by
all household members.39 However, this is not a completely accurate representation of the return
to agricultural labour for matrilineal men, as their expected return depends on the probability that
they will be divorced next year. Assuming they don’t receive any agricultural income on divorce,
their expected average product of agricultural labour is, in its simplest form, the probability that the
marriage will not end next year times the average product of agricultural labour. Since I assume
that patrilineal men keep all their land on divorce, their average product of agricultural labour
(APAL) is the same as their expected APAL.
I take a weighted average of these three calculations for matrilineal and patrilineal households;
this is in Table 7. There are four key points in this table. First, the estimated annual divorce
probability is almost twice as high among matrilineal households than patrilineal households. This
is consistent with the stylised fact that matrilineal communities tend to have more marriages and
divorces than patrilineal communities, with a shorter average duration of marriage. Indeed, the remarriage rate is higher among matrilineal communities, suggesting that their turnover of marriages
is higher. Second, the estimated wage is signi…cantly higher than the estimated APAL for both
groups. This provides strong evidence for Assumption 4. The di¤erence is particularly pronounced
in matrilineal communities when the divorce probability is taken into account in estimating the
expected rate of return to agricultural work. The expected APAL is signi…cantly lower than the
wage for matrilineal husbands, which rationalises the decision of matrilineal husbands to engage in
more wage work than patrilineal husbands. Third, the MASAF wage is not signi…cantly di¤erent
from the APAL. As the MASAF wage is set below market wages, this is consistent with the idea
that wages are substantially higher than the APAL. Fourth, the estimated wage is not signi…cantly
di¤erent across matrilineal and patrilineal communities. This suggests that di¤erences in consumption and labour allocation are not attributable to di¤erences in returns. Therefore, these results
provide convincing evidence that the rate of return to wage work is signi…cantly higher than the
rate of return to agricultural work.
and Dulani 2004). Note that in the survey, the MASAF wage is only reported by those villages that have a MASAF
program, which are likely to be particularly poor.
39
Assuming that production exhibits diminishing marginal returns, if the average agricultural product is lower than
the wage, then the marginal agricultural product is also lower than the wage. I assume that the average agricultural
product is the same for all household members. This is a simplifying assumption that ensures the identi…cation of
the agricultural product, because it is not possible to identify how much of consumption from own production came
from the labour of each individual household member. A further important assumption for these calculations to have
a valid interpretation is that the wage and agricultural product are constant. If they are not and people only accept
wage work when the wage is high enough or only work for agriculture if the average product is high enough, then the
calculated average wage and agricultural product will over-estimate their true values. Although the over-estimation
itself is not a problem for testing Assumption 4, an issue arises if they are over-estimated by di¤erent amounts.
28
Table 7: Estimates of the wage and agricultural product
Parameter
Matrilineal
Proportion married (pm )
72:4%
Proportion divorced (pd )
12:3%
Annual remarriage probability ( )
63:3%
Annual divorce probability ( )
10:7%
Average Wage
177:36
(21:16)
N
829
MASAF Wage
N
Average Product of Agricultural Labour
N
Expected Average Product of Agricultural Labour
H0 : Average Wage=APAL
H0 : MASAF Wage=APAL
Patrilineal
76:4%
8:7%
45:5%
5:2%
185:44
(22:93)
425
111:27
(24:31)
828
161:56
(59:95)
343
115:74
(10:72)
3506
103:31
0:01
0:87
122:51
(20:17)
1881
122:51
0:04
0:48
Units are MWK. This table reports mean (standard error). The expected average product
of agricultural labour is calculated as ((1
probability and
)+
) AP AL, where
is the divorce
is the share of land kept on divorce (which equals one in patriliny and zero
in matriliny). Tests of null hypotheses report p-values and signi…cance: * indicates
signi…cance at 10%, ** at 5% and *** at 1%.
29
5.2
Labour Allocation
In order to analyse the impact of kinship on labour allocation, I …rst present summary statistics
of labour allocation in Table 8. This table shows the number of hours each spouse spends per
week on each type of activity, disaggregated by lineage.40 Women tend to work harder when
they are patrilineal, while men tend to work harder when they are matrilineal. This is consistent
with the idea that women have more autonomy in matrilineal communities, while men have more
autonomy in patrilineal communities. Men tend to split their time between di¤erent types of
economic labour, while women tend to engage in agricultural labour and domestic work, devoting
less than an hour per week on wage labour. This is in line with the division of labour assumed in
the theoretical framework. Both spouses tend to allocate more labour to agricultural work when
they are patrilineal, while both spouses tend to allocate more labour to wage work when they are
matrilineal. Patrilineal spouses allocate more time to ganyu labour than matrilineal spouses. This
makes sense, as ganyu labour is typically carried out by the poorest households (see footnote 39).
I verify whether these labour allocation patterns are still present in a regression analysis when
relevant variables are controlled for. I estimate Equations (3)-(6) speci…ed in Section 3. The
results for husbands’and wives’labour are in Table 9. Husbands’labour allocation is in line with
the theoretical predictions. While there is no signi…cant di¤erence between the total labour time
of patrilineal and matrilineal men, the former spend approximately 1 hour and 30 minutes more on
agriculture and a similar amount of time less on wage work per week, on average. This substitution
e¤ect is con…rmed in regressions (IV) and (V), as patrilineal men spend over three hours more on
agriculture than wage work, while the sum of these two labour types is not signi…cantly di¤erent
between matrilineal and patrilineal men.41
The results for women’s labour show that patrilineal women work harder than matrilineal women
on average. Most of this additional work takes the form of agricultural labour. They also appear to
substitute a small amount of agricultural labour for wage labour. There is no a priori reason why
this should be the case. One possible reason is the omission of an important unobservable variable:
40
Total labour includes own-farm agricultural labour, wage labour, business activities, ganyu labour and unpaid
labour. Ganyu labour is agricultural labour on other people’s land, which is paid a very low wage either in cash or
in kind. Engaging in ganyu labour is a sign of food insecurity as households are choosing to work on others’ land
for a low but immediate wage instead of working on their own land. Wage labour captures any work, excluding
ganyu, carried out for a wage, salary or commission. Domestic labour is time spent fetching water and …rewood. The
questionnaire did not ask about more typical domestic tasks like cooking and cleaning. In addition, there is no data
on leisure, which is why the total number of hours is not equal to the number of hours in a week.
41
Although the regressions control for employment opportunities, a more careful analysis of employers can verify
that the observed di¤erence in wage work hours is an active choice made by households rather than a result of
employment opportunities. When examining the share of individuals working for di¤erent types of employers in the
whole sample, it is observed that signi…cantly more matrilineal than patrilineal men work for private individuals
and companies and state-owned organisations. No di¤erence is observed for government and religious employers.
However, when the same analysis is conducted for the restricted sample of those husbands who engage in wage work,
no di¤erence in shares of matrilineal and patrilineal husbands working for di¤erent employers is observed apart from
government employers. However, this is signi…cant at the 9% level and the government only employs 4% of the whole
sample. These …ndings suggest that any di¤erence in employers observed for the whole sample is driven by the fact
that matrilineal men are more likely to work for wages in the …rst place. Conditional on having decided to work for
wages, almost no di¤erence in employer is observed. This suggests that there is little di¤erence in the employment
opportunities available to matrilineal and patrilineal households.
30
Labour
Total
Agricultural (own-farm)
Wage
Ganyu
Domestic
N
Table 8: Labour hours per week
Matrilineal Patrilineal Dual descent
Husb.
22:61
21:23
18:98
(0:62)
(0:77)
(2:32)
P=M=D
0:16
P=M
0:16
Wife
13:70
(0:44)
15:39
(0:69)
11:74
(2:01)
0:06
0:04
H
11:73
(0:42)
12:26
(0:57)
10:51
(1:68)
0:55
0:46
W
10:94
(0:40)
11:80
(0:59)
9:55
(1:81)
0:32
0:22
H
5:28
(0:49)
3:75
(0:36)
5:02
(2:08)
0:04
0:01
W
0:47
(0:09)
0:26
(0:07)
0:40
(0:32)
0:17
0:06
H
2:66
(0:17)
2:88
(0:26)
1:66
(0:38)
0:02
0:48
W
0:94
(0:08)
1:57
(0:16)
0:55
(0:16)
0:00
0:00
H
0:84
(0:07)
0:62
(0:08)
0:60
(0:23)
0:12
0:05
W
8:33
(0:18)
8:55
(0:20)
7:51
(0:59)
0:22
0:41
4409
2448
279
7136
6857
This table reports mean (standard error). Columns 4 and 5 report p-values for the rejection of the null hypothesis.
31
Table 9: The e¤ect of descent on labour allocation
(I)
(II)
(III)
(IV)
(V)
Husband’s Labour Total
Agric
Wage
Agric - Wage Agric + Wage
Patrilineal
-0.139 1.509
-1.644
3.153
-0.136
(1.018) (0.701) (0.843)
(1.199)
(0.983)
N
7136
7136
7136
7136
7136
R2
0.149
0.181
0.147
0.170
0.152
Wife’s Labour
Patrilineal
N
R2
(VI)
Total
1.788
(0.798)
7136
0.162
(VII)
Agric
1.433
(0.664)
7136
0.208
(VIII)
Wage
-0.332
(0.167)
7136
0.048
(IX)
Agric - Wage
1.765
(0.691)
7136
0.193
(X)
Agric + Wage
1.101
(0.679)
7136
0.192
Controls included: Basic, Region, Geography, Village economy, HH Composition
Gender and the price index. Standard errors are reported in parentheses.
denotes signi…cance at 1% level,
at 5% level and
at 10% level.
bargaining power. If bargaining power is omitted and if husbands with higher bargaining power
do not allow their wives to engage in wage labour to safeguard their power, then a downward bias
on the coe¢ cient of patriliny will result: a negative e¤ect which is mistakenly attributed to land
rights, rather than low bargaining power.
The results suggest that labour allocation may explain the consumption gap observed in Section
4. Matrilineal households consume more than patrilineal households. Simultaneously, matrilineal
men spend more time on wage labour and less time on agricultural labour. Therefore, there is
support for both propositions of the theoretical framework.
5.3
Robustness: Income
To verify the robustness of the labour allocation results, I examine the e¤ect of kinship on income:
patrilineal households should earn less income and, in particular, less wage income by the husband
than matrilineal households. Table 10 supports this prediction.42 Patrilineal households earn less
income than matrilineal households and patrilineal husbands earn less wage income than matrilineal
husbands, on average. The income gap represents approximately 40% of mean income for the entire
42
The construction of the income aggregate follows the method of Hoddinott and Haddad (1995). Income was
calculated as the sum of crop sales, wages from employment, earnings from ganyu, pro…t from business activities,
remittances and other gifts, for all members of the household. I do not include income from livestock sales. Wage
earnings represent the ’wages from employment’component of income earned by the husband and thus exclude any
earnings from ganyu. The relative importance of these components in income is as follows: ganyu forms the largest
share of income on average (29:9%), followed by crop sales (29:0%), wage employment (16:1%), pro…t from business
(13:6%) and remittances (11:4%). The share of wage earnings in income is lower for patrilineal than matrilineal
households, while the share of ganyu earnings in income is higher for patrilineal than matrilineal households. This is
consistent with the labour patterns observed in the previous Subsection.
32
Patrilineal
N
R2
Table 10: Income
(XI)
(XII)
Income
Wage earnings (H)
-32.474
-15.817
(15.534)
(6.111)
7136
7136
0.082
0.088
Controls included: Basic, Region, Geography, Village economy,
HH Composition and Gender. Standard errors are reported
in parentheses.
at 5% level and
denotes signi…cance at 1% level,
at 10% level.
sample. The wage earnings gap is 19:2% of mean income, which is not inconsistent with the gap
observed for total expenditure.43
The evidence shows that matrilineal households consume signi…cantly more than patrilineal
households; at the same time, patrilineal men dedicate more of their labour to agriculture than
wage work and earn less wage income. This suggests that while matrilineal men are deterred from
investing in land, this is bene…cial to the household. This is an example of ‘positive’hold-up. The
husband’s weaker outside option increases the productivity of the household.
6
Intra-Household Allocation
The analysis in Section 4 focused on consumption at the household level and per capita. The
analysis of per capita consumption may not re‡ect consumption at the individual level, as the
bene…ts and costs of di¤erences in household productivity may accrue asymmetrically. If this is the
case, a welfare comparison of matriliny and patriliny is not clear-cut. To explore this issue, I examine
the intra-household allocation of expenditure. Inferring individual consumption from household
consumption data is di¢ cult; however, one can look at goods that are private by de…nition. Two
such goods are clothing and education. The former is important for adults whereas the latter is
important for children. I examine spending on men’s and women’s clothing and sons’and daughters’
education, as well as total household education spending, total household clothing spending and
food spending. First, I present mean expenditure shares of these goods, disaggregated by kinship
(Table 11). Patrilineal households spend a signi…cantly higher expenditure share on food, education
43
The gap in wage earnings is even closer to the gap in purchased consumption, which is obtained by excluding
consumption from production and gifts from the consumption aggregate and estimating regression (5) with the log
of purchased consumption as the dependent variable. The coe¢ cient on patriliny is 16:3% with a p-value of 0:000
(N = 7135). This is a much larger gap than the gap in total expenditure and is almost the same as the wage earnings
gap. This is consistent with the idea that matrilineal households consume more because of husbands’ preference
of wage labour over agricultural labour, which would make the gap particularly pronounced for that component of
consumption that requires cash payment. The regressions are not reported fully because I calculate the purchased
consumption aggregate from the raw consumption data using the LSMS/NSO guidelines, which do not give full
details of unit conversions and prices. Therefore, the aggregate may not match the LSMS/NSO method completely.
However, the correlation between the two measures, when comparing total expenditure, is high.
33
and sons’education. However, from summary statistics alone, it is not possible to ascertain whether
this is due to the di¤erence in total expenditure by kinship or whether kinship has an e¤ect on
intra-household allocation over and above its e¤ect through total expenditure.
In order to explore the direct e¤ect of kinship on intra-household allocation, I estimate a series
of Working-Leser expenditure functions as in Equation (7) in Section 3, examining the e¤ect of
kinship on the share of expenditure devoted to the various categories discussed.44 Although the
literature tends to instrument expenditure with income (e.g. Attanasio and Lechene 2010), I do
not do this because of the typically low savings of households in Malawi discussed in Section 4,
which imply that income and expenditure are highly correlated. As a result, income will be just as
endogenous as expenditure. Instead, I instrument expenditure with the number of livestock owned
and the construction material of the dwelling, which together capture wealth. The right-hand side
variables are as in speci…cation (5); I also include the log of household size and the log of prices of
various goods as additional controls.45 The results of these Two Stage Least Squares regressions
are in Table 12.
The key result is that descent has a weak positive e¤ect on the expenditure share devoted to
sons’education and men’s clothing and no e¤ect on the remaining variables in Table 12. Patrilineal
households allocate an additional 0:1% of their expenditure to sons’education and men’s clothing,
compared to matrilineal households. This suggests the presence of gender bias and is consistent with
the idea that husbands are more empowered in patrilineal households. However, the magnitude of
the e¤ect is small, which suggests that the assumption of the theoretical framework regarding the
independence of
and
is a reasonable one.
The coe¢ cients on other variables of interest indicate that the regressions are well-speci…ed. For
example, the share of expenditure allocated to food in regression (a) declines signi…cantly with per
capita expenditure, which is in line with Engel’s law. In contrast, the share of expenditure accruing
to clothing is signi…cantly increasing with per capita real expenditure. The e¤ect of household
composition also indicates that the regressions are well-speci…ed. While all demographic groups
but one increase food expenditure, there are di¤erential e¤ects for other expenditure categories.
For example, regression (b) shows that the share of expenditure allocated to education is increasing
in the number of adults in the family but decreasing with the number of children. An additional
adult is associated with an approximately 0:5% higher share of expenditure allocated to education,
whereas an additional child is associated with an approximately 0:3% lower share of expenditure
allocated to education.
44
This analysis could be improved on by estimating a full demand system, such as an Almost Ideal Demand System.
However, the lack of su¢ ciently detailed price data prevents this (see footnote 45).
45
The regressions include the log of the cost of milling maize and the cost of milling rice and the log of the prices
of maize grain, maize ‡our, rice, bread, scones, beans, cabbage, tomatoes, banana, milk, egg, chicken, …sh, beef, tea,
salt, sugar, oil, chips, soap, a toothbrush, toothpaste, clothes soap, vaseline, chitenje cloth, trousers, coca cola, beer,
cigarettes, a watch, …rewood, charcoal, para¢ n, a bicycle, a mattress and a mosquito net. On average, the goods for
which prices are available represent 55:2% of spending.
46
One observation was not included in the expenditure share analysis because the disaggregated expenditure calculated from the data did not match the value calculated by the NSO and the reason for the di¤erence was not
clear.
34
Food (%)
Table 11: Summary statistics of expenditure shares
Patrilineal Matrilineal Dual Descent P=M=D
66:65
63:64
62:31
0:00
(0:46)
(0:36)
(1:57)
P=M
0:00
Education (%)
1:12
(0:07)
0:94
(0:05)
1:45
(0:38)
0:05
0:03
Daughters’education (%)
0:47
(0:04)
0:43
(0:03)
0:59
(0:16)
0:55
0:50
Sons’education (%)
0:59
(0:04)
0:46
(0:03)
0:55
(0:11)
0:04
0:07
Clothing (%)
2:60
(0:13)
2:86
(0:10)
2:61
(0:27)
0:24
0:11
Women’s clothing (%)
0:93
(0:05)
1:02
(0:04)
0:88
(0:11)
0:22
0:13
Men’s clothing (%)
0:62
(0:04)
0:60
(0:03)
0:56
(0:10)
0:85
0:73
N 46
2448
4408
279
7136
6857
This table reports mean (standard error). Columns 4 and 5 report p-values
for the rejection of the null hypothesis.
Regressions (c) and (d) show that there is preferential treatment in educating daughters and
sons, depending on household composition. An additional female adult reduces the percentage share
of expenditure devoted to sons’education by 0:3% and increases the share devoted to daughters’
education by 0:7%. Similarly, an additional male adult increases the share spent on sons’education
by 0:6% and reduces the share spent on daughters’education by 0:2%. This pattern is also seen in
the e¤ect of elderly male and female members of the household. Together, these results suggest that
there is strong preference for educating own-sex children in households in Malawi. Regressions (f)
and (g) show that there is little e¤ect of demographic composition on men’s and women’s clothing.
There is evidence of increasing and decreasing economies of scale to household size: while there are
increasing economies of scale to the purchase of food, there are decreasing economies of scale to
education.
The results demonstrate that patrilineal households spend a greater share of their expenditure
on their sons’education and men’s clothing. However, the magnitudes are small and the coe¢ cients
are only signi…cant at the 10% level. This suggests that the correlation between kinship and the
intra-household allocation of private goods is weak. This supports the assumption of the framework
that land rights ( ) and how consumption goods are shared in marriage ( ) are independent.
The results also demonstrate that there is strong gender bias in the intra-household allocation of
children’s education in Malawi.
35
-16.275
(2.351)
1.402
(0.461)
1.669
(0.480)
1.000
(0.551)
1.723
(0.579)
2.904
(1.104)
-1.174
(1.024)
7135
Ln(HH size)
# children (m)
# children (f)
# adults (m)
# adults (f)
# elderly (m)
# elderly (f)
-0.119
(0.179)
7135
-0.002
(0.228)
0.516
(0.122)
0.557
(0.105)
-0.249
(0.086)
-0.274
(0.088)
1.909
(0.369)
0.092
(0.094)
0.199
(0.106)
7135
-0.493
(0.173)
0.720
(0.093)
-0.195
(0.055)
-0.066
(0.059)
-0.239
(0.059)
0.978
(0.241)
-0.007
(0.047)
-0.453
(0.116)
7135
0.330
(0.134)
-0.264
(0.069)
0.591
(0.078)
-0.254
(0.053)
-0.118
(0.051)
1.323
(0.233)
0.110
(0.063)
0.094
(0.239)
7135
-0.158
(0.257)
-0.186
(0.136)
-0.292
(0.116)
-0.178
(0.107)
-0.204
(0.105)
1.537
(0.579)
-0.150
(0.183)
Table 12: The e¤ect of descent on intra-household allocation
(b)
(c)
(d)
(e)
Educ (%) Daughters’educ (%) Sons’educ (%) Clothing (%)
0.332
0.133
0.320
0.891
(0.232)
(0.175)
(0.118)
(0.365)
0.118
(0.099)
7135
0.233
(0.111)
0.166
(0.057)
0.038
(0.053)
0.069
(0.049)
0.041
(0.046)
-0.664
(0.251)
-0.101
(0.075)
(f)
Women’s clo (%)
0.150
(0.128)
0.111
(0.086)
7135
0.023
(0.085)
0.001
(0.051)
0.001
(0.042)
-0.009
(0.039)
-0.011
(0.039)
-0.088
(0.231)
0.091
(0.054)
(g)
Men’s clo (%)
0.261
(0.152)
Ln(various prices) (see footnote 45). Standard errors are reported in parentheses.
denotes signi…cance at 1% level,
Controls included: Basic, Region, Geography, Village economy, Employment, Gender, # uncategorised HH members,
at 5% level and
at 10% level.
Ln(pc real expend) instrumented with number of livestock and construction material of dwelling. F-stat on excluded instrument: 140.6; partial R2 : 0.10.
N
-0.885
(0.818)
Patrilineal
Ln(pc real expend)
(a)
Food (%)
-13.190
(1.155)
7
Conclusion
The aim of this paper has been to estimate the impact of spouses’outside options on productivity.
Matriliny and patriliny have a key distinction in that descent and therefore land rights pass through
the female line in matriliny but through the male line in patriliny. This implies that women have
stronger land rights than their husbands in matrilineal households, while men have stronger land
rights than their wives in patrilineal households. Detailed household data on Malawi, where both
kinship systems co-exist, have been used to provide a causal regression analysis of the impact of
descent on consumption, where descent is interpreted as capturing spouses’relative outside options.
Historically, matriliny has been criticised because the fact that men do not own land could lead
to disincentives for long-term investment and, as a result, poor economic performance. This paper
shows that while matrilineal men are disincentivised from agricultural labour, this leads to better
economic outcomes. Matrilineal households have higher consumption, on average. More generally,
this demonstrates that the size of the household pie is not invariant to spouses’ outside options.
The existing literature on intra-household allocation needs to take into account that productivity
and thus total resources may change with bargaining power.
I show that matrilineal households, where women have strong land rights, consume over 10%
more than patrilineal households. These results are con…rmed for per capita and equivalent expenditure. I present a two-stage framework of the husband’s labour allocation decision in rural
Malawi, which explains under what assumptions patrilineal men allocate a greater share of their
labour to agriculture than matrilineal men. Intuitively, the reason for the di¤erence is the asymmetry between labour specialisation and property division following divorce: only men engage in
wage labour; at the same time, they do not have rights to land following divorce under matriliny
but have full rights under patriliny. This incentivises matrilineal men to spend more time on wage
labour. However, because wages are higher than agricultural productivity, matrilineal households
are better o¤. I provide evidence to support this framework by showing that wages are signi…cantly
higher than the average agricultural product, a key assumption of the framework. I show that
patrilineal men spend signi…cantly more time on agricultural labour than matrilineal men and that
they earn signi…cantly less wage income.
Apart from being a goal in its own right, empowering women by increasing their outside options
can have positive side e¤ects, as this paper has demonstrated. In developing countries, providing
men with incentives to move out of agricultural labour and into wage labour can help households
raise their consumption levels. More generally, I have provided evidence to suggest that household
productivity depends on spouses’outside options. This needs to be taken into account in studies
of bargaining power and consumption.
37
Appendices
A
Land Ownership
It is important to provide evidence for the idea that men and women have di¤erential rights to land
in patrilineal and matrilineal communities. This can be demonstrated with the LSMS data analysed
in this chapter. I examine the plots of land that households use for farming, combining plots used
for rainy and dry season cultivation. I calculate the total area of land owned solely by the husband
or wife, as well as land owned jointly by the spouses and land owned by other members of the
household. The …gures are weighted based on the sampling strategy. Table 13 shows these …gures,
disaggregated by kinship type. The …gures show that women own signi…cantly more land and men
own signi…cantly less land on average in matrilineal than patrilineal communities. Since the total
amount of land owned by households is not signi…cantly di¤erent across these communities, this
implies that women own a greater share of household land on average in matrilineal than patrilineal
communities, while the opposite is true for men. Although the land comes from all sources, including
inheritance, leases and purchases, these …gures do re‡ect land entitlement following divorce. This
is because most plots reported in the sample were inherited. Therefore, there is signi…cant evidence
that women have stronger rights to land following divorce when they are matrilineal, while men
have stronger rights to land following divorce when they are patrilineal.
Table 13: Land ownership, disaggregated by owner and kinship
Husband’s land
Wife’s land
Jointly owned land
Others’land
Total land
N
Matrilineal
Patrilineal
Dual descent
P-value P=M=D
P-value P=M
0:79
1:07
1:16
0:00
0:00
(0:04)
(0:07)
(0:21)
0:56
0:32
0:37
0:00
0:00
(0:03)
(0:04)
(0:09)
0:34
0:29
0:35
0:44
0:27
(0:04)
(0:03)
(0:07)
0:38
0:46
0:46
0:17
0:07
(0:02)
(0:04)
(0:12)
2:06
2:13
2:34
0:45
0:42
(0:07)
(0:06)
(0:24)
4409
2448
279
7136
6857
This table reports mean (standard error). The unit of measurement is acres.
P-values are reported for the rejection of the null hypothesis.
38
B
Proof of Special Case and Propositions
B.1
Special Case
Suppose Assumptions 1 and 2 hold. Then there is overinvestment in agriculture
Special Case
( h1 >
hE
1)
exists a
= 1 and underinvestment in agriculture ( h1 < hE
1 ) when
when
2 (0; 1) such that h1 =
= 0: Moreover, there
hE
1.
Proof. It is easy to show that the planner’s objective function HW is strictly concave in the
…rst-period labour choice h1 : Given Assumptions 1 and 2, the husband and wife’s value functions
in the married state are identical, or V2M = U2M . Thus, the planner’s …rst-order condition is
@HW
@h1
=
First, suppose
1 0 e1
u ( )w1 + f 0 (h1 ) U20S (L2 ) + (1
) U20M (L2 )
2
2
1 0 e1
+ (1
)
u ( )w1 + f 0 (h1 ) V20S (L2 ) + (1
) U20M (L2 )
2
2
:
= 1: In order to show that that the husband’s choice h1 is too high, it is
su¢ cient, due to the concavity of HW , to show that
wife has no land in the divorced state, so that
@HW
@h1
V2S0 (L2 )
< 0 at h1 . Note that when
= 1, the
= 0. Evaluating the derivative at h1 , the
…rst term is zero by the husband’s …rst-order condition. It is then su¢ cient to show that the second
term is strictly negative, or
1 0 e1
u ( )w1 + f 0 (h1 ) (1
2
2
) U20M (L2 ) < 0:
This inequality is sati…ed by the husband’s …rst-order condition. Therefore,
h1 >
@HW
@h1
< 0 at h1 and
hE
1.
Second, suppose
so that
@HW
@h1
U2S0 (L2 )
= 0. Note that in this case, the husband has no land in the divorced state,
= 0. In order to show the choice of h1 is too low, it is su¢ cient to show that
> 0 when the derivative is evaluated at h1 . Evaluating the derivative at h1 , the …rst term
is zero by the husband’s …rst-order condition. It is then su¢ cient to show that the second term is
strictly positive, or
1 0 e1
u ( )w1 + f 0 (h1 )
2
2
V20S (L2 ) + (1
) U20M (L2 ) > 0:
This inequality is satis…ed by the husband’s …rst-order condition.
Finally, since the planner’s …rst-order condition is a continuous function of
E
at
2 (0; 1) such that
E
@HW
@h1
, there exists a
= 0 when the derivative is evaluated at h1 : This implies that h1 = hE
1
.
39
B.2
Proposition 1
If Assumption 3 holds, an increase in the share of land accruing on divorce leads
Proposition 1
dh1
d
to an increase in …rst-period agricultural labour at the expense of …rst-period wage labour:
Similarly, second-period agricultural labour is increasing in
in both states:
dhS
2
d
;
dhM
2
d
> 0.
> 0.
Proof. The …rst-order condition for the choice of h1 is
(
u0 (cS2 )
@g( L2 ; hS2 ) 0
f (h1 ) + (1
@L2
)u0 (cM
2 )
@g(L2 ; hM
2 ) 0
f (h1 ))
@L2
Totally di¤erentiating both sides with respect to
w1 u0 (c1 ) = 0:
yields
@g( L2 ; hS2 )
@L2
2
@ g( L2 ; hS2 )
@g( L2 ; hS2 ) 2 00 S
+ L2 (u0 (cS2 )
+
(
) u (c2 )))
@L2
@L22
dh1 2 2 00
+
[ w1 u (c1 )f 0 (h1 )
d
@g( L2 ; hS2 ) 2 0
+ ( ( u00 (cS2 )(
) (f (h1 ))2
@L2
@ 2 g( L2 ; hS2 ) dhS2
@ 2 g( L2 ; hS2 )
+ u0 (cS2 )(f 0 (h1 ))2 (
+
)
@L22
@L2 @hS2 dL2
f 0 (h1 )(u0 (cS2 )
@g( L2 ; hS2 ) 00
f (h1 ))
@L2
@g(L2 ; hM
2 ) 2 0
) (f (h1 ))2
+(1
)((u00 (cM
2 )(
@L2
2
M
M
@ 2 g(L2 ; hM
0
2 @ g(L2 ; h2 ) dh2
2 )
u0 (cM
)(f
(h
))
(
+
)
1
2
2
M
dL2
@L2
@L2 @h2
+u0 (cS2 )
+u0 (cM
2 )
@g(L2 ; hM
2 ) 00
f (h1 ))]
@L2
= 0:
This total di¤erential consists of a constant term, which is the …rst two lines of the expression,
and the derivative
dh1
d
multiplied by several terms. The expression can be rewritten more generally
as
C +A
where C =
f 0 (h1 )(u0 (cS2 )
@g( L2 ;hS
2)
@L2
dh1
= 0;
d
+ L2 (u0 (cS2 )
@ 2 g( L2 ;hS
2)
@L22
the term in the square bracket. In order to derive the sign of
+(
@g( L2 ;hS
2 ) 2 00
) u (cS2 )))
@L2
dh1
d ,
and A is
the signs of C and A need
to be derived. I begin with A. Since the functions u( ); f ( ) and g( ) are concave, their …rst
derivatives are positive but their second derivatives are negative. Therefore, all terms in A have
40
an unambiguously negative sign, apart from u0 (cj2 )
dhj2
dL2
@ 2 g(Aj ;hj2 ) dhj2
(f 0 (h1 ))2
@L2 @hj2 dL2
= Kj , because the sign of
is unknown. The sign can be derived by totally di¤erentiating the second-period …rst-order
condition with respect to L2 . I will focus on the condition in the divorce state; the following
calculations also hold for the married state. The …rst-order condition is
@g( L2 ; hS2 )
= w2 :
@hS2
The total derivative is
@ 2 g( L2 ; hS2 ) dhS2
@ 2 g( L2 ; hS2 )
= 0:
+
dL2
@(hS2 )2
@hS2 @L2
Therefore,
@ 2 g( L2 ; hS2 )=@hS2 @L2
> 0:
@ 2 g( L2 ; hS2 )=@(hS2 )2
dh2
=
dL2
This implies that Kj has a positive sign. However, it can be shown that the sum of Kj and another term in A is de…nitely negative. In particular, I take the sum of KS and u0 (cS2 )
u0 (cS2 )
@ 2 g( L2 ;hS
2)
(f 0 (h1 ))2 :
@L22
2
S
@ 2 g( L2 ; hS2 ) dh2 0
2
0 S @ g( L2 ; h2 ) 0
(f (h1 ))2
(f
(h
))
+
u
(c
)
1
2
@L22
@L2 @hS2 dL2
= u0 (cS2 )(f 0 (h1 ))2 (
@ 2 g( L2 ; hS2 ) dh2
@ 2 g( L2 ; hS2 )
+
)
@L22
@L2 @hS2 dL2
= u0 (cS2 )(f 0 (h1 ))2 (
@ 2 g( L2 ; hS2 ) @ 2 g( L2 ; hS2 )=@hS2 @L2 @ 2 g( L2 ; hS2 )
+
)
@L22
@L2 @hS2
@ 2 g( L2 ; hS2 )=@(hS2 )2
= u0 (cS2 )(f 0 (h1 ))2 (
@ 2 g( L2 ;hS
2) 2
)
@L2 @hS
2
@ 2 g( L2 ; hS2 )=@(hS2 )2
(
+
@ 2 g( L2 ; hS2 )
):
@L22
In order for this term to be negative, the expression inside the brackets needs to be negative:
@ 2 g( L2 ;hS
2 ) )2
(
@L2 @hS
2
S 2
@ 2 g( L2 ;hS
2 )=@(h2 )
+
@ 2 g( L2 ;hS
2)
@L22
< 0. It can be shown that this is the case as the long as the Hessian
matrix of g( ) has a positive determinant, which is always true if g( ) is concave:
@ 2 g( L2 ;hS
2) 2
)
@L2 @hS
2
@ 2 g( L2 ; hS2 )=@(hS2 )2
(
+
@ 2 g( L2 ; hS2 )
@L22
<
,
0
gL2
S
2 h2
+ gL2 L2 < 0
ghS hS
2
2
, gL2 L2 ghS hS
2
2
gL2 2 hS > 0;
2
where the sign changes in the last line because ghS hS is negative. The last line is the determinant
2
41
2
of the Hessian matrix of g( ), which is always positive if g ( ) is concave, which I have assumed to
be the case. Therefore, the expression is always negative and A is unambiguously negative.
The constant term C is unambiguously positive if we assume that the rate of diminishing
marginal product of g( ) and rate of diminishing marginal returns of u( ) are low. The term C is
f 0 (h1 )(u0 (cS2 )
@g( L2 ; hS2 )
@ 2 g( L2 ; hS2 )
@g( L2 ; hS2 ) 2 00 S
+ L2 (u0 (cS2 )
+
(
) u (c2 )));
@L2
@L2
@L22
whose sign is given by the expression inside the brackets, since
f 0 (h1 ) is always positive. The
…rst term inside the brackets is positive while the second and third are negative. In order for the
sum of these to be positive, it is required that
L2 (
where g 0 denotes
@g( L2 ;hS
2)
,
@L2
g 00 denotes
00
g 00
0u
+
g
) < 1;
g0
u0
@ 2 g( L2 ;hS
2)
,
@L22
(8)
u0 denotes u0 (cS2 ) and u00 denotes u00 (cS2 ).
In order to understand the implications of the condition more fully, I assume a Cobb-Douglas
production function and iso-elastic utility: g( L2 ; hS2 ) = ( L2 ) h2 ; 0 < ;
< 1 and u(c2 ) =
c12
.
Condition (8) simpli…es to
c2
>
g( L2 ; hS2 )
, w2 (1
where
hS2 ) > (
1)g( L2 ; hS2 );
is the coe¢ cient of relative risk aversion. The condition is likely to be satis…ed when
the intertemporal elasticity of substitution is high ( is low) or wage income is signi…cantly larger
than agricultural income. Relating this to Condition (8), at low levels of agricultural production
the marginal product of agricultural labour is likely to be high and the condition on the shape of
g( ) will be satis…ed. A low
will satisfy the condition on the shape of u( ).
1
Recalling that A is negative and that C + A dh
d = 0, it must be unambiguously true that if the
…rst derivatives of g( ) and u( ) do not diminish at too fast a rate, C is positive and
dh1
> 0:
d
The e¤ect of
on hS2 and hM
2 is found by totally di¤erentiating the …rst-order conditions that
de…ne these optimal choices. For example, in the case of hS2 :
@ 2 g( L2 ; hS2 ) 0
@ 2 g( L2 ; hS2 ) dhS2
dh1 @ 2 g( L2 ; hS2 )
+
f
(h
)
+
L2 = 0:
1
d
d
@(hS2 )2
@hS2 @L2
@hS2 @L2
Thus,
42
dhS2
=
d
1
@ 2 g( L2 ;hS
2)
2
@(hS
2)
which is positive as long as
de…nes the optimal choice of
hM
2
dhM
2
=
d
(
@ 2 g( L2 ; hS2 ) 0
dh1 @ 2 g( L2 ; hS2 )
f
(h
)
L2 );
+
1
d
@hS2 @L2
@hS2 @L2
dh1
d
> 0. Similarly, di¤erentiating the …rst-order condition that
yields
1
@ 2 g( L2 ;hM
2 )
2
@(hM
2 )
which again is positive as long as
B.3
> 0.
Proposition 2
If Assumption 3 holds, then
Proposition 2
then
dh1
d
@ 2 g(L2 ; hM
dh1
2 ) 0
f (h1 )
= 0;
M
d
@h2 @L2
de2
d
de1
d
< 0. If, in addition, Assumption 4 holds,
< 0.
Proof. The derivative of …rst-period consumption with respect to
de1
=
d
which is always negative as long as
with respect to
dh1
d
w1
is
dh1
;
d
is positive. The derivative of second-period consumption
is
M
dh1 @g(L2 ; hM
@g(L2 ; hM
de2
2 ) dh2
2 )
= f 0 (h1 )
(
+
M
d
d
dL2
@L2
@h2
w2 ):
1
Since f 0 (h1 ) dh
d is positive, a su¢ cient condition for this to be negative is
M
@g(L2 ; hM
@g(L2 ; hM
2 ) dh2
2 )
+
M
dL2
@L2
@h2
w2 < 0;
which is satis…ed by Assumption 4.
C
Variables
Tables 14 and 15 below provide details of the variables used in this chapter. Geographical variables
are explained separately in Table 16.
43
44
Indicator
Continuous
Continuous
Continuous
Continuous
Continuous
Continuous
Continuous
Indicator
Indicator
Indicator
Immigration
Dist to nearest road
Dist to nearest pop centre
# Own Children/Children
# Adults
# Elderly
# Uncategorised
Pc/eq real exp
Women’s group exists
Semi-permanent/Traditional
Language (e.g. Chewa)
Indicator
Indicator
Indicator
Continuous
Indicator
Patrilocal
Matrilocal
Other residence
Divorce rate
Any business/wage empl.
Continuous
Continuous
Indicator
Continuous
Indicator
Indicator
Indicator
Indicator
Age
Any schooling
HH Size
South/Centre/North
Patrilineal
Matrilineal
Dual Descent
Crop prop
Variable type
Continuous
Variable name
Land
Table 14: Variables
Description
Amount of land owned by household (HH), separated by type of
cultivation (dry/rainy), in acres
Individual’s age
= 1 if individual ever attended school, = 0 otherwise
Number of members of HH
= 1 if HH is in the South/Centre/North, = 0 otherwise
= 1 if HH resides in a patrilineal community, = 0 otherwise
= 1 if HH resides in a matrilineal community, = 0 otherwise
= 1 if HH resides in a community with both patrilineal and
matrilineal descent, = 0 otherwise
= 1 if HH lives in the husband’s natal village, = 0 otherwise
= 1 if HH lives in the wife’s natal village, = 0 otherwise
= 1 if HH is neither patrilocal nor matrilocal, = 0 otherwise
% of household heads who report being divorced or separated in district
= 1 if any type of business/wage employment is listed as one of three
main sources of village employment, = 0 otherwise
Proportion of HHs interviewed in village that farm this crop,
excluding respondent HH
= 1 if individuals come to village at certain times of the year to work,
= 0 otherwise
Distance (km) from HH to nearest road
Distance (km) from HH to nearest town with population > 20 000
Number of own children/children, of any age/aged between
0-14 years, that are members of the HH
Number of adults, aged between 15-59 years, that are members
of the HH
Number of elderly, aged 60 years or over, that are members of
the HH
Number of individuals whose age was unreported, that are
members of the HH
Per capita/equivalent real expenditure
= 1 if a women’s group in the village exists, = 0 otherwise
= 1 if house is made of semi-permanent/traditional material,
= 0 otherwise. Excluded group: permanent.
= 1 if Language is most spoken in community, = 0 otherwise
45
Food/Education/Clothing (%)
Wage earnings
Wage labour
Agric labour
Ganyu labour
Income
Variable name
Total labour
Table 15: Variables cont.
Variable type Description
Continuous
Total number of hours spent last week on agricultural,
wage, ganyu, business and unpaid work
Continuous
Total number of hours spent on wage work last week
Continuous
Total number of hours spent on agricultural work last week
Continuous
Total number of hours spent on ganyu work last week
Continuous
Total earnings of HH in past 12 months, consisting of
salaries, income from crop sales, pro…t from business
and remittances from children and others
Continuous
Total earnings of husband from all wage work in
past 12 months
Continuous
Share of total HH expenditure spent on
food/education/clothing
46
Continuous
Continuous
Indicator
Indicator
Indicator
Indicator
Greenness
Soil quality
Soil quality
Soil quality
Agro-ecology
Continuous
Continuous
Continuous
Continuous
Continuous
Continuous
Temperature
Temperature
Rainfall
Rainfall
Rainfall
Greenness
Greenness
Variable type
Continuous
Continuous
Category
Temperature
Temperature
Table 16: Geographical variables
Reference period
Description
1960-1990
Average daily range: mean of max. temp.- min. temp.
1960-1990
Temperature seasonality: standard deviation of monthly
climatology
1960-1990
Minimum temperature of coldest month
1960-1990
Average temperature of wettest quarter
2008-2009, 2009-2010 Average 12-month total rainfall, July-June
2008-2009, 2009-2010 Average total rainfall in wettest quarter, July-June
2008-2009, 2009-2010 Average start of wettest quarter in dekads, from July onwards
2008-2009, 2009-2010 Total change in greenness within primary growing season,
averaged by district
2008-2009, 2009-2010 Onset of greenness increase in day of year, starting July 1st,
averaged by district
2008-2009, 2009-2010 Onset of greenness decrease in day of year, starting July 1st,
averaged by district
N/A
Nutrient availability: 7 categories de…ning how serious
this is as a constraint
N/A
Rooting conditions: 7 categories de…ning how serious
this is as a constraint
N/A
Excess salts: 7 categories de…ning how serious this
is as a constraint
N/A
Agro-ecological zones created from WorldClim climate data
D
Language
Table 17: Language
(11) + language
Ln(real expenditure)
0:086
(0.034)
Patrilineal
Tonga (M = 14:0%; N = 192)
(11) cont.
Ln(real expenditure)
-0.036
(0.101)
Chewa (M = 76:6%; N = 3859)
0:139
(0.081)
Other (M = 3:2%; N = 165)
-0.118
(0.101)
Yao (M = 85:0%; N = 644)
0:210
(0.091)
Lambya (M = 12:8%; N = 70)
-0.153
(0.096)
Nyanja (M = 80:2%; N = 377)
0.102
(0.084)
Nkhonde (M = 4:5%; N = 67)
-0.034
(0.109)
Lomwe (M = 81:8%; N = 308)
0.148
(0.094)
Sukwa (M = 0%; N = 35)
-0.055
(0.104)
Sena (M = 7:7%; N = 307)
-0.026
(0.121)
Nyakyusa (M = 0%; N = 20)
-0.094
(0.140)
Ngoni (M = 72:4%; N = 272
0.062
(0.104)
Senga (M = 3:2%; N = 7)
0.098
(0.098)
7136
0.368
N
R2
M indicates % of households speaking this language who are matrilineal. The excluded language
is the predominantly patrilineal Tumbuka, N=797, M=4.6%. Controls included: Basic, Region,
Geography, Village economy, Employment, HH Composition and Gender. Standard errors are
in parentheses.
denotes signi…cance at 1% level,
at 5% level and
47
at 10% level.
E
A Model of Divorce Probabilities
Let us consider a world with a continuum of measure 1 of individuals. The individuals are in…ntely
lived and are either married or divorced in any given period. At any time t = 1; 2; :::, the proportion
of married individuals is Mt and the proportion of divorced individuals is Dt = 1
individuals divorce with probability
Mt . Married
per period and divorced individuals remarry with probability
per period. Divorces and remarriages are identical and independent across individuals and time
periods.
By the law of large numbers, the evolution of the proportions Mt and Dt can be described by
a Markov chain with the following transition matrix T :
Mt
Mt+1
Dt
1
Dt+1
1
In the long run, the proportions of married and divorced individuals converge to the stationary
distribution of this Markov chain. This is a vector p that satis…es the following condition,
T p = p;
or,
(T
I)p = 0:
In fact, p is an eigenvector of T with eigenvalue equal to 1, normalised so that the elements add
up to 1. Using these facts gives the following solution for p:
p = [
;
]
+
+
= [pm ; pd ];
where pm and pd are the equilibrium proportions of married and divorced individuals in the
population respectively. This can be used to solve for
pm
:
pd
=
48
as a function of , pm and pd :
(9)
Therefore, as long as ; pm and pd are known, the divorce probability
can be calculated. I
assume that each time period t represents one year; this is consistent with the theoretical framework,
where husbands engage in agricultural labour that yields produce at the end of the season. I use
the weighted values of pm and pd observed in the LSMS sample. That is, I use the proportion
of household heads who are either married or divorced/separated, calculated separately for the
patrilineal and matrilineal groups. I exclude individuals who report being widowed or having never
married; however, there is no need to normalise pm and pd . The values of pm and pd for patrilineal
and matrilineal households are given in Table 7.
To calculate
, I use the Malawi Longitudinal Study of Families and Health (MLSFH). This
survey asks individuals about their marriages and divorces. As years are reported for each marriage
and divorce, I am able to calculate the number of years each remarriage takes. I use the 2010 wave
of the data as this matches the LSMS sample most closely; I restrict the sample to men only, as the
theoretical framework focuses on their decisions. The annual remarriage probability is calculated
as the proportion of men who remarry within one year of their divorce. This calculation is carried
out separately for matrilineal and patrilineal men, where I identify lineage through the men’s
tribal a¢ liation. Together, these calculations yield the remarriage probabilities for matrilineal and
patrilineal men in Table 7. Using this information in formula (9) along with pm and pd , I am able
to calculate the value of
for patrilineal and matrilineal husbands.
49
References
[1] Ali, D. A., K. Deininger and M. Goldstein. "Environmental and Gender Impacts of Land
Tenure Regularisation in Africa: Pilot Evidence from Rwanda." World Bank Policy Research
Working Paper 5765 (2011).
[2] Angrist, J. D. and Jörn-Ste¤en Pischke. Mostly Harmless Econometrics. Woodstock: Princeton
University Press, 2009.
[3] Attanasio, O. and V. Lechene. "Conditional Cash Transfers, Women and the Demand for
Food." IFS Working Paper 10/17 (2010).
[4] Besley, T. "Property Rights and Investment Incentives: Theory and Evidence from Ghana."
Journal of Political Economy 103.5 (1995): pp. 903-937.
[5] Brasselle, A.-S., F. Gaspart and J.-P. Platteau. "Land Tenure Security and Investment Incentives: Puzzling Evidence from Burkina Faso." Journal of Development Economics 67.2 (2002):
pp. 373-418.
[6] Chiappori, P.-A. "Rational Household Labour Supply." Econometrica 56.1 (1988): pp. 63-89.
[7] Chirwa, E. W., P. M. Mvula and B. M. Dulani. "The Evaluation of the Improving Livelihoods
through Public Works Programme." (2004). Mimeo.
[8] Davison, J. Gender, Lineage and Ethnicity in Southern Africa. Oxford: Westview Press, 1997.
[9] Deaton, A. The Analysis of Household Surveys: A Microeconometric Approach to Development
Policy. Baltimore, Maryland: Johns Hopkins University Press, 1997.
[10] Deaton, A. and S. Zaidi. "Guidelines for Constructing Consumption Aggregates for Welfare
Analysis." Living Standard Measurement Study Working Paper 135 (2002).
[11] Deininger, K. and S. Jin. "Tenure Security and Land-Related Investment: Evidence from
Ethiopia." European Economic Review 50.5 (2006): pp. 1245-1277.
[12] Dercon, S. and P. Krishnan. "In Sickness and in Health: Risk Sharing within Households in
Rural Ethiopia." Journal of Political Economy 108.4 (2000): pp. 688-727.
[13] Doss, C. R. "Women’s Bargaining Power in Household Economic Decisions: Evidence from
Ghana." University of Minnesota, Department of Applied Economics Sta¤ Papers 13517
(1996).
[14] Du‡o, E. and C. Udry. "Intrahousehold Resource Allocation in Cote D’Ivoire: Social Norms,
Separate Accounts and Consumption Choices." National Bureau of Economic Research Working Paper 10498 (2004).
50
[15] Dyson, T. and M. Moore. "On Kinship Structure, Female Autonomy and Demographic Behaviour in India." Population and Development Review 9.1 (1983): pp. 35-60.
[16] Dzimadzi, C. and B. Chinsinga. "The Listenership, Readership, Viewership Survey of the
Malawi Social Action Fund." (2004). Mimeo.
[17] Gollin, D., D. Lagakos and M. E. Waugh. "The Agricultural Productivity Gap in Developing
Countries." CGEB Working Paper Series 03 (2012).
[18] Hirschmann, D. and M. Vaughan. "Food Production and Income Generation in a Matrilineal
Society: Rural Women in Zomba, Malawi." Journal of Southern African Studies 10.1 (1983):
pp. 86-99.
[19] Hoddinott, J. and L. Haddad. "Does Female Income Share In‡uence Household Expenditures?
Evidence from Côte D’Ivoire." Oxford Bulletin of Economics and Statistics 57.1 (1995): pp.
77-96.
[20] Johnson, G. D. and L. Hendrix. "A Cross-Cultural Test of Collins’s Theory of Sexual Strati…cation." Journal of Marriage and the Family 44.3 (1982): pp. 675-684.
[21] Johnson, M. M. Strong Mothers, Weak Wives: The Search for Gender Equality. Berkeley:
University of California Press, 1988.
[22] Kerr, R. B.. "Food Security in Northern Malawi: Gender, Kinship Relations and Entitlements
in Historical Context." Journal of Southern African Studies 31.1 (2005): pp. 53-74.
[23] Kerr, R. B. "Informal Labor and Social Relations in Northern Malawi: The Theoretical Challenges and Implications of Ganyu Labour for Food Security." Rural Sociology 70.2 (2005): pp.
167-187.
[24] Kishindo, P. "The Marital Immigrant. Land and Agriculture: A Malawian Case Study." African
Sociological Review 14.2 (2010): pp. 89-97.
[25] Lamphere, L. "Strategies, Cooperation and Con‡ict Among Women in Domestic Groups"
in Woman, Culture and Society, eds. M. Z. Rosaldo and L. Lamphere. Stanford, California:
Stanford University Press (1974): pp. 97-112.
[26] Lundberg, S. J. and R. A. Pollak. "Separate Spheres Bargaining and the Marriage Market."
Journal of Political Economy 101.6 (1993): pp. 988-1010.
[27] Lundberg, S. J., R. A. Pollak and T. J. Wales. "Do Husbands and Wives Pool their Resources?
Evidence from the United Kingdom Child Bene…t." The Journal of Human Resources 32.3
(1997): pp. 463-480.
[28] Malawi National Statistical O¢ ce. "Note on Construction of Expenditure Aggregate and
Poverty Lines for IHS2." Malawi: National Statistical O¢ ce (2004).
51
[29] Manser, M. and M. Brown. "Marriage and Household Decision-Making: A Bargaining Analysis." International Economic Review 21.1 (1980): pp. 31-44.
[30] Peters, P. E. "Against the Odds. Matriliny, Land and Gender in the Shire Highlands of
Malawi." Critique of Anthropology 17 (1997): pp. 189-210.
[31] Peters, P. E. "Bewitching Land: The Role of Land Disputes in Converting Kin to Strangers
and in Class Formation in Malawi." Journal of Southern African Studies 28.1 (2002): pp.
155-178.
[32] Peters, P. E. "’Our Daughters Inherit our Land, but our Sons use their Wives’ Fields’:
Matrilineal-Matrilocal Land Tenure and the New Land Policy in Malawi." Journal of Eastern
African Studies 4.1 (2010): pp. 179-199.
[33] Phiri, K. M. "Some Changes in the Matrilineal Family System Among the Chewa of Malawi
since the Nineteenth Century." Journal of African History 24 (1983): pp. 257-274.
[34] Phiri, K. M. "Production and Exchange in pre-Colonial Malawi" in Malawi: An Alternative Pattern of Development. Centre for African Studies, University of Edinburgh, Seminar
Proceedings no. 25 (1985): pp. 3-32.
[35] Phiri, K. M. "Pre-Colonial States of Central Malawi: Towards a Reconstruction of their History." The Society of Malawi Journal 41.1 (1988): pp. 1-29.
[36] Place, F. and K. Otsuka. "Tenure, Agricultural Investment and Productivity in the Customary
Tenure Sector of Malawi." Economic Development and Cultural Change 50.1 (2001): pp. 77-99.
[37] Reniers, G. "Divorce and Remarriage in Rural Malawi." Demographic Research Special Collection 1.6 (2003).
[38] Roberts, S. "Matrilineal Family Law and Custom in Malawi: a Comparison of Two Systems."
Journal of African Law 8 (1964): pp. 77-90.
[39] Rostow, W. W. The Stages of Economic Growth: A Non-Communist Manifesto. Cambridge:
Cambridge University Press, 1960.
[40] Schatz, E. "Numbers and Narratives: Making Sense of Gender and Context in Rural Malawi."
University of Pennsylvania PhD Dissertation (2002).
[41] Simpson, J. Some Socio-Economic Factors A¤ ecting the Conservation and Utilisation of Farm
Animal Genetic Resources in Malawi. Malawi: University of Malawi, 2000.
[42] Spring, A. Agricultural Development and Gender Issues in Malawi. Maryland: University Press
of America, Inc., 1995.
[43] Takane, T. "Labour Use in Smallholder Agriculture in Malawi: Six Village Case Studies."
African Study Monographs 29.4 (2008): pp. 183-200.
52
[44] Telalagić, S. "Domestic Production as a Source of Marital Power: Theory and Evidence from
Malawi." Cambridge Working Papers in Economics 1243 (2012).
[45] Udry, C. "Gender, Agricultural Production and the Theory of the Household." Journal of
Political Economy 104.5 (1996): pp. 1010-46.
[46] Udry, C. and M. Goldstein. "The Pro…ts of Power: Land Rights and Agricultural Investment
in Ghana." Journal of Political Economy 116.6 (2008): pp. 981-1022.
[47] Udry, C., J. Hoddinnott, H. Alderman and L. Haddad. "Gender Di¤erentials in Farm Productivity: Implications for Household E¢ ciency and Agricultural Policy." Food Policy 20.5 (1995):
pp. 407-423.
[48] Ulph, D. "A General Non-Cooperative Nash Model of Household Consumption Behaviour."
University of Bristol Working Paper 88.205 (1988).
[49] Vollrath, D. "How Important are Dual Economy E¤ects for Aggregate Productivity?" Journal
of Development Economics 88.2 (2009): pp. 325-334.
[50] World Bank Development Economics Research Group and Malawi National Statistical O¢ ce.
Malawi Living Standards Measurement Study: 2010-2011 IHS3 Survey (2012).
53