Widows’ Land Rights and Agricultural
Investment
⇤
Brian Dillon
†Alessandra Voena
‡
July 23, 2018
Abstract
This paper examines the connection between widows’ land inheritance rights and landinvestments in Zambia. We study whether the threat of land expropriation uponwidowhood deters households from fallowing, applying fertilizer, and employing labor-intensive tillage techniques. Variation in inheritance by widows is based on customaryvillage practices, which we observe in surveys of village leaders. Controlling for possibleconfounding factors, both OLS and IV estimates show lower levels of land investmentby married couples in villages where widows do not inherit. Concern over prospectiveloss of land by the wives reduces investment in land quality even while the husband isalive.Keywords: land tenure security; widowhood; land investment; gender discrimination;African development; farm productivityJEL codes: O13, J16, D23
⇤We are grateful to Anthony Chapoto, Thom Jayne and the Food Security Research Project in Lusaka forsharing their data. We thank the editor Andrew Foster, two anonymous referees, Kate Baldwin, DingiswayoBanda, Lorenzo Casaburi, William Clark, James Fenske, Rachel Heath, Kelsey Jack, Simone Lenzu, NathanNunn, Salla Rantala, and participants in workshops at Harvard Kennedy School, IFPRI, UChicago, UW,IHME, and WGAPE for helpful comments. Michael Mulford, Holly Hickling, Sarojini Rao, Noemi Noceraand Sarah Co↵ey provided excellent research assistance. This project did not receive any specific grantfrom funding agencies in the public, commercial, or not-for-profit sectors. However, this work was partiallyconducted while the authors were Giorgio Ru↵olo Post-doctoral Fellows in the Sustainability Science Programat Harvard University, for which support from Italy’s Ministry for Environment, Land and Sea is gratefullyacknowledged. We are joint first authors on the paper.
†University of Washington. Email: [email protected]. Corresponding author.‡University of Chicago and NBER. Email: [email protected].
1
1 Introduction
Land productivity in Sub-Saharan Africa (SSA) is far lower than in other regions of the
world. Intensive farming coupled with insu�cient investment in restoring land quality has
led to a gradual decline in soil productivity across the region (Barrett, Place and Aboud,
2002; Morris, 2007). This has severe implications for agricultural output and food security.
Agriculture provides the majority of employment and accounts for a major share of GDP in
most countries in SSA. Any impediment to agricultural productivity has important, negative
consequences for welfare.
Why is land investment so low in SSA? The fragmented system of land tenure sys-
tems in the region is believed to be one critical factor. Most agricultural land is governed
by traditional systems of communal control. These traditional systems may involve allo-
cation or revocation of land use rights by community leaders, restrictions on the sale of
land, restrictions on the rental of land, land inheritance along strictly matrilineal or strictly
patrilineal lines, or various other departures from the Western concept of property rights.
Policymakers, researchers, and outside observers have long been concerned that limited or
opaque rights to land use may be a deterrent to investment.
In this paper, we examine the link between an important feature of land tenure in
Zambia – land inheritance rules – and farmers’ investments in soil quality. The large majority
of agricultural land in Zambia is still governed by local custom. We focus on a widow’s right
to maintain control over the household land after her husband dies, which is determined
at the village level. Concern over land expropriation upon widowhood is serious: recent
work in the same setting shows that on average, the death of a male household head leads
to a one-third reduction in the the land area cultivated by the widow’s household, after
accounting for the reduction in household labor supply (Chapoto, Jayne and Mason, 2011).
This aspect of land governance is of particular importance in modern-day Zambia, where
a high infectious disease burden coupled with the HIV crisis have contributed to relatively
high mortality among prime age males.
Our empirical work is motivated by a simple dynamic model of agricultural investment
that allows for localized heterogeneity in inheritance norms. Investment – for now, think of
2
fallowing (as in Goldstein and Udry (2008))– raises future land productivity. As long as the
wife has positive weight in the household planning problem, preventing her from inheriting
land lowers investment relative to a situation in which the household expects that she will
inherit. The distortion is exacerbated by higher mortality risk.
We test the predictions of the theory using a 2008 agricultural household survey in
Zambia. The data set includes standard agricultural, economic, and demographic informa-
tion at the plot, household, and community level. Data on the land inheritance policy for
widows comes from a survey of village leaders conducted concurrently with the household
survey and designed specifically to measure land allocation and inheritance norms, among
other topics. Over a third of households in the data live in an area where widows do not
inherit their deceased husband’s lands.
We use this large micro data set to test whether couple-headed households – those
for whom inheritance-by-widows policies are potentially relevant – invest less in land quality
when they live in a community where women do not inherit land if their husband dies. The
analysis is repeated for three di↵erent types of land investment: fertilizer application, fallow-
ing, and the use of labor-intensive tillage practices aimed at reducing erosion and run-o↵ (in
Section 3 we explain why even a short- to medium-term investment such as fertilizer might
be a↵ected by the policy). Across a variety of specifications we find strong evidence of lower
land investment in areas where widows do not inherit. Couples in non-widow-inheritance
villages apply 13–18% less fertilizer, fallow 4–5% less land area, and use intensive tillage
techniques on 3–5% fewer hectares, relative to the averages among households with positive
levels of each activity. (Those figures are 37–50% for fertilizer, 12–16% for fallowing, 7–11%
for intensive tillage if we include households with zeroes). These findings are economically
significant, although they do not represent a wholesale abandonment of land quality. Rather,
they suggest that when households observe their leaders expropriating widows’ land, they
make small but important adjustments to invest less in land quality. These findings are
robust to detailed wealth, geography, and household controls. Instrumental variables esti-
mates, using ethnic group shares as instruments, show even larger e↵ects. This IV strategy
a�rms our interpretation of the previous literature and of our own results: variation in eth-
nic group norms regarding widows leads to the observed variation in inheritance-by-widows
3
rights.
We implement a number of extensions to the main results. Falsification tests show
that norms governing widow land inheritance only a↵ect investment in households headed
by a couple. This suggests that the e↵ect is not driven by general variation in the strength
of property rights. When we use additional information about who inherits, other than the
widow, we find the pattern one would expect to find if the results are causal: investment
is highest when the widow inherits, weakly lower when someone in her family inherits, and
lowest when the land reverts to the chief or to a family member not directly related to the
widow. Finally, we show that farm revenues are weakly lower in areas where widows do not
inherit (the point estimates are negative, but not always statistically significant). We do not
observe farm costs, so we cannot determine whether profits are also lower in non-inheritance
areas. But the weakly negative e↵ect on revenues suggests that insecure tenure for widows
forces households to reallocate resources away from the optimum that would be achievable
if widows could inherit land.
Our findings contribute to three strands of literature. The first is the substantial body
of work on the link between land investment and land tenure security at the household level in
developing countries (see Place (2009) and Fenske (2011) for recent reviews). Not surprisingly
given the diversity of tenure regimes across SSA, the existing evidence on this link is mixed.
While a number of studies have found, in line with this paper, a positive relationship between
tenure security and land investment (Besley, 1995; Gavian and Fafchamps, 1996; Place and
Otsuka, 2002b; Benin, 2006; Deininger and Jin, 2006; Goldstein and Udry, 2008), others
have not (Place and Hazell, 1993; Brasselle, Gaspart and Platteau, 2002; Place and Otsuka,
2002a; Pender et al., 2004; Amsalu and De Graa↵, 2007). Thanks largely to the richness of
the Zambia data, our approach combines many of the strengths of the papers in this domain
(Fenske, 2011). First, we have a relatively large and nationally representative data set –
almost 6,000 households for our main specifications. Second, we test impacts on multiple,
continuous measures of investment, rather than on binary outcomes. And third, our measure
of tenure security is not a de jure legal statute or a self-reported plot characteristic, but a
community-level policy variable that reflects the de facto prevailing norms.
We also contribute to this literature by focusing on tenure security as it relates to
4
inheritance rights. Previous work has not focused on the interaction between widows’ rights
and agricultural investment before widowhood. Our findings suggest that the dynamic di-
mension of property rights is important. Households are forward-looking enough to incorpo-
rate the probability of retaining control over land into their investment decisions. While it
has been shown that the onset of widowhood leads to reductions in wealth and asset-holding
in many communities in sub-Saharan Africa (Peterman, 2012), the impact of this form of
insecurity on the resource allocation decisions of intact couples is not well understood.
This paper also contributes to the important literature on gender-related di↵erences
in land tenure or women’s ownership of land (Besley, 1995; Udry, 1996; Yngstrom, 2002;
Allendorf, 2007; Goldstein and Udry, 2008; Ubink and Quan, 2008; Kumar and Quisumbing,
2012; Ali, Deininger and Goldstein, 2014; Doss et al., 2015), and more generally on the e↵ect
of women’s rights on development (Duflo, 2012; Doepke, Tertilt and Voena, 2012; Fernandez,
2014). In this sense, our empirical tests can be interpreted as joint tests of the importance of
women in the household planning problem and of the e↵ect of non-inheritance on investment.
Our results suggest that women do matter in household decisions, and that depriving them
of secure access to land in the future can distort present-day investments.
Lastly, we contribute to the growing body of work in economics that explores the role
of institutions in determining economic outcomes (Acemoglu, Johnson and Robinson, 2001;
Rodrik, Subramanian and Trebbi, 2004; Acemoglu, Johnson and Robinson, 2005; Tabellini,
2010). That local norms and institutions matter for economic development is hardly a new
idea. Our findings demonstrate a subtle, previously unexplored channel by which these
e↵ects operate. Households do not abandon land quality entirely when custom dictates that
widows do not inherit. Yet, they respond in an empirically detectable fashion by reducing
their investments in an asset that may be subject to expropriation. This suggests both a
clear understanding of how local leaders operate, and a forward-looking set of responses.
The paper proceeds as follows. In the following section we provide more details about
the setting. We then develop a theoretical framework for the analysis, in Section 3. Section
4 describes the empirical specifications, and Section 5 describes the data. In Section 6 we
report the main OLS and IV results, as well as a number of extensions. We conclude with
a discussion in Section 7.
5
2 Background and Setting
In much of sub-Saharan Africa, control of rural farmland is governed by customary insti-
tutions rather than formal, statutory law. In Zambia, the village headman plays a central
role in land allocation decisions. The extent of the chief’s control over land can vary across
ethnic groups and regions, with family groups exerting greater influence in some areas than
in others. Yet, as a general rule, women do not formally own or exercise substantial control
over land (Shezongo-Macmillan (2005), as cited in Chapoto, Jayne and Mason (2011)).
Widows’ land rights have always been an area of concern, but they rose to prominence
in Zambia in the 1990s and 2000s, when the HIV/AIDS epidemic led to a sharp increase
in deaths of prime age males (Strickland, 2004). In the nationally representative data set
used in this paper, 15% of households are headed by a widow. Among some ethnic groups
in Zambia, support for widows has traditionally included the practice of levirate marriage,
i.e., re-marriage of a widow to the brother of her deceased husband, in part to keep land
within the family. Recent evidence suggests that this practice has become less common since
the onset of the HIV crisis, because of concerns about disease transmission (Gausset, 2001;
Malungo, 2001).
Local rules governing land inheritance by widows are rooted in longstanding village
traditions. This deference to local determination was codified in the Lands Act of 1995,
which allows village leaders to decide who inherits customary land (which covers 94% of the
country). While the Constitution rejects gender discrimination in many domains, it does not
regulate practices pertaining to many personal and family matters, including bequests. The
1989 Interstate Succession Act does not grant equal inheritance rights to women and has lim-
ited enforceability, particularly when it is in conflict with customary practices (Richardson,
2004). The National Gender Policy of 2000 and subsequent initiatives by women’s advocacy
groups have not changed the fact that women in most rural communities do not challenge
their inequitable positions under customary norms (Veit, 2012).
Because Zambia is one of the most ethnically diverse and highly fractionalized coun-
tries in sub-Saharan Africa (Taylor and Hudson, 1970; Posner, 2004), local institutions re-
garding widow land inheritance can vary even within a relatively small geographical area.
6
There is substantial evidence of widows losing control of land in some places. Malungo (2001)
describes the challenge of distinguishing between property grabbing and support for widows
in the form of re-marriage. Kajoba (2002) documents specific instances from Chibombo dis-
trict in which women lose access to the household land after their husband dies. Chapoto,
Jayne and Mason (2011) show that on average across Zambia, the death of a male household
head leads to a one-third reduction in the land area cultivated by the widow’s household.
The distribution is roughly bi-modal: almost half of widows experience a landholding decline
of 50% or more, while a third experience no decrease.
In Zambia, as in much of sub-Saharan Africa, discussions about tenure security and
control over land take place alongside a longstanding narrative of declining soil quality (Bar-
rett, Place and Aboud, 2002; Gilbert, 2012; FAO, 2016). Agricultural lands across the conti-
nent have lower levels of soil organic matter and lower nutrient content than elsewhere in the
world. This is largely due to a lack of investment in soil health, i.e., too little use of restora-
tive crops, fallowing, fertilizer, and investments to protect against erosion and run-o↵. The
problem is severe enough that in 2006, Ministers of Agriculture from the Africa Union states
signed the Abuja Declaration on Fertilizer for the African Green Revolution. The agreement
aimed to catalyze greater investment in restoring soil quality across the continent.
This paper deals with only a subset of the important issues regarding widows’ rights,
tenure security, and land investment. However, what is clear is that these themes represent
the intersection of critical policy issues in modern day Zambia.
3 Theoretical framework
To capture the impact of women’s inheritance rules on agricultural investment, we develop
a simple non-separable discrete time model of a farming household making agricultural pro-
duction choices related to land quality. We abstract from numerous complicating factors in
order to focus on the most relevant features of the problem.
Consider a household composed of two agents, a husband (H) and wife (W ). Each
person derives utility from consumption ct according to the spot utility function ln(ct), where
t is the period and i 2 {H,W}. The couple solves an intertemporal collective household plan-
7
ning problem, making Pareto e�cient decisions about resource allocation using the Pareto
weights µi for each person i, withP
i µi = 1.
The household faces uncertainty about the survival of its members, each of whom
have probability ⇡t of surviving to period t when alive in period t� 1. The problem in any
time period is denoted by {B,H,W} to indicate who is still living: “both”, “husband only”,
or “wife only.” Define pBt+1 = ⇡2t+1. This is the probability that both spouses survive until
period t+ 1, conditional on having survived to period t. Furthermore, ⇡t+1(1� ⇡t+1) is the
probability that only one spouse dies at the end of period t.
The household farms a fixed amount of land A, with average land quality defined
by the state variable Qt. In every period t, the household divides its resources between
consumption by living members, cit, and agricultural investment, qt. For the moment, qt is
a generic investment, which encompasses fallowing, fertilizer application and labor-intensive
land tillage. The price of one unit of investment q over land A is denoted as d, and the price
of output is normalized to 1.
The farm production function is F (A,Qt), with output increasing in both arguments.
Assume first that investment only raises output in the next period (an assumption we relax
below). The equation of motion for land quality is:
Qt+1 = qt + (1� �)Qt,
where � is the rate of land quality depreciation. We further suppose that F (0, Qt) = 0 8Qt,
and that widows who lose their land enjoy a fixed, positive consumption floor c > 0.
The local inheritance policies are given by the parameter �, which take a value of 1 if
the wife can inherit the land in the event that the husband dies, and 0 otherwise. In keeping
with the prevailing norms in Zambia, the man keeps his land if his wife dies.
Consider the forward-looking infinite horizon problem of this household. A couple
8
solves the following problem:
V Bt (Qt(�),�) = maxc,q�0
0
@X
i2{H,W}
µiln(cBit )
1
A (1)
+ �
2
4pBt+1VBt+1(�, Qt+1(�)) + pt+1
X
i2{H,W}
µiV it+1(Q
Bt+1(�),�)
3
5
s.t. cBt F (A,Qt(�))� dqBt , cBt = cBHt + cBW
t
qBt = QBt+1 � (1� �)Qt(�)
A widower solves the following problem, which does not actually depend on � once land
quality is accounted for:
V Ht (Qt(�)) = maxc,q�0 ln(cHt ) + ⇡t+1�V
Ht+1(Qt+1(�)) (2)
s.t. cHt F (A,Qt(�))� dqHt
qHt = QHt+1 � (1� �)Qt(�)
A widow solves the following the problem
V Wt (Qt(�),�) = maxc,q�0 ln(cWt ) + ⇡t+1�V
Wt+1(Q
Wt+1(�),�) (3)
s.t. cWt �⇥F (A,Qt(�))� dqWt
⇤+ (1� �) · c
qWt = QWt+1 � (1� �)Qt(�)
Assuming an interior solution, from the first order conditions of problem (1) we have that
investment and consumption satisfy:
d
cBt= �
pBt+1
d(1� �) + FQ(A,QBt+1)
cBt+1
+ pt+1
✓µH d(1� �) + FQ(A,QB
t+1)
cHt+1
(4)
+ �µW d(1� �) + FQ(A,QBt+1)
cWt+1
◆�.
Hence, at the optimum, the marginal cost of land investment is equal to the expected value
of the marginal benefit in the next period, depending on the realization of the mortality
9
shock.
Equation 4 further implies the following, for a given level of Qt:
(i) if µW = 0, the inheritance-by-widows policy � does not a↵ect land investment
(ii) if µW > 0, a household facing � = 0 will choose lower investment and higher present
consumption than a household facing � = 1.
Point (i) states that the inheritance rule only a↵ects investment if the wife has a
strictly positive weight in household decision making, which may be the result of altruism
on the part of the husband, or of e�cient intrahousehold bargaining. Point (ii) states that as
long as the wife has some weight in the planning problem, the household will undertake less
investment, and hence future land quality will be lower, if she loses access to the household
land upon the death of the husband.
Condition 4 holds for concave production functions, as well as for convex ones, as long
as the value function is concave.1 In fact, when land quality is unobserved, as in our case, the
e↵ect of the inheritance regime will be further amplified whenever FQQ(A,Qt) > 0. Output
is unlikely to be globally convex in soil quality, but it may be convex over a range. This is
a direct implication of a foundational theory in agricultural chemistry. Von Liebig’s “law of
the minimum” states that there are inherent complementarities between soil nutrients and
other soil quality components (especially soil organic matter) (Von Liebig, 1840). Economic
evidence from western Kenya is supportive of von Liebig’s claims. Marenya and Barrett
(2009) find that the marginal returns to fertilizer (land investment) are lower on plots with
lower levels of soil organic matter (a land quality state variable). Convex returns to soil
quality are not only possible in Zambia, they may be the norm, given the low levels of land
quality in the region (Burke, Jayne and Roy Black, 2017). Under these circumstances, land
1Concavity of the value function requires FQQ(A,Qt+1) ct+1 < [FQ(A,Qt+1) + (1� �)d]2. For a generic
utility function, it requires FQQ(A,Qt+1) < �u00(ct+1)u0(ct+1
[FQ(A,Qt+1) + (1� �)d]2, and hence is less stringent
as absolute risk aversion increases. What these conditions state, intuitively, is that even if output is convex
in soil quality over a range, farmers will still choose Qt in the convex part of the domain if it is too costly
to invest up to the point where output is concave in quality. The maintenance of soils across sub-Saharan
Africa in a state of perpetual low quality suggests that this is highly plausible.
10
quality in any period will be systematically lower in areas with an inheritance regime that
penalizes widows.
What happens if the Pareto weights vary across inheritance regimes, i.e., when µi(� =
1) 6= µi(� = 0)? This seems like an intuitive possibility, if the inheritance regime is related
to other aspects of women’s standing in the community. In fact, when � = 1, the widow’s
and the widower’s problem are identical, allowing us to rewrite the Euler equation so that
the Pareto weight only enters decision making when � = 0.2 This means that allowing the
Pareto weight to vary with the inheritance regime does not a↵ect the implications of the
model. This is not surprising: a plausible alternative model would likely have lower Pareto
weights for women in the areas where widows do not inherit, and hence the prediction of
di↵erential investment would be magnified rather than mitigated.
Finally, we also consider the case in which land investment can have a contemporane-
ous e↵ect on output. This is likely for fertilizer and intensive tillage, though not for fallowing.
The model’s implications are more complex in this case. The production function admits
three arguments: F (A,Qt, qt), with all three inputs having a positive impact on agricultural
output. If the application of fertilizer (or other investment represented by qt) has a concave
e↵ect on output in the same period Fqq(A,Qt, qt) < 0, the Euler equation is as follows:3
d� Fq(A,Qt(�), qBt )
cBt= �
⇢pBt+1
✓FQ(A,QB
t+1, qBt+1)
cBt+1
+ (1� �)d� Fq(A,QB
t+1, qBt+1)
cBt+1
◆(5)
+ pt+1
µH
✓FQ(A,QB
t+1, qHt+1)
cHt+1
+ (1� �)d� Fq(A,QB
t+1, qHt+1)
cHt+1
◆
+�µW
✓FQ(A,QB
t+1, qWt+1)
cWt+1
+ (1� �)d� Fq(A,QB
t+1, qWt+1)
cWt+1
◆��
When the depreciated marginal revenue from investments is small relative to to the
marginal return on land quality in the production function (implying FQ(A,QBt+1, q
Wt+1) >
(1��)[Fq(A,QBt+1, q
Wt+1)�d]), we will again obtain that, for fixed Qt, levels of qt will be lower
2In particular, the Euler equation can be rewritten as dcBt
=
�hpBt+1
d(1��)+FQ(A,QBt+1)
cBt+1+ pt+1
⇣d(1��)+FQ(A,QB
t+1)
cHt+1� (1� �)µW d(1��)+FQ(A,QB
t+1)
cWt+1
⌘ias long as
cWt+1(� = 1) = cHt+1.3Note that the below expression reduces to condition (4) when Fq(A,Q, q) = 0.
11
when � = 0. Furthermore, this result holds unconditionally for all Qt(�) if FQQ(A,Qt, qt) >
0, and FQq(A,Qt, qt) > 0, i.e., if land quality and fertilizer are complements. Again, this
would be consistent with the literature on the low profitability of fertilizer when land quality
is low (Marenya and Barrett, 2009). In fact, what the model reveals is that complementarity
between land quality and fertilizer application may be the primary reason to expect impacts
from a long-run concern related to mortality on a relatively short-term investment such as
fertilizer application. The complementarity with the soil quality state variable strengthens
the e↵ect of land inheritance rules on investments with short- to medium-term returns.
4 Empirical framework
In this section, we first describe the empirical specifications, and then discuss identification.
4.1 Empirical specifications
For the purposes of this study, the key predictions of the above model are captured by
regressions of the following form:
qdsh = �NoWidowInheritanceds + �0Xdsh + ⌫d + ✏dsh (6)
where qdsh is the land investment choice of household h in village s in district d, NoWidowInheritanceds
is a dummy variable that takes a value of 1 if widows do not inherit in the village, Xdsh is
a vector of household and village controls, ⌫d is a vector of district fixed e↵ects to control
for prices and other district characteristics, and ✏dsh is an error term clustered at the village
level. The hypothesis that insecure tenure rights for widows reduces land investment is the
alternative to the null hypothesis H0 : � � 0. As we describe in the next section, because of
data limitations the village-level variables are defined at the level of the standard enumer-
ation area (SEA), a slightly larger level of geographical aggregation that can include more
than one village.
In the model we abstracted from the role of children in the household’s decision
problem. Yet, children are surely important in questions of inheritance and investment. If
12
the husband cares only about children but not about the wife, one would still expect to see
a reduction in land investment in those areas where widows do not inherit, unless the land
will go to another family member who uses the land to the children’s benefit to the same
degree that the widow would. This is the rationale underlying regressions of the following
form:
qdsh = �1 WidowInheritanceds + �2 OtherFamilyInheritanceds + �0Xdsh + ⌫d + ✏dsh (7)
where WidowInheritanceds is the share of headmen in the SEA who state that the widow
inherits the land; OtherFamilyInheritanceds is the share of headmen in the SEA who state
that the heir is one of the children or someone on the wife’s side of the family; and the
excluded group represents the share of headmen in the SEA who state that the land goes
to a member of the husband’s family or goes back to the headman for reallocation. It is
highly likely that both the widow’s and the children’s interests are bundled with the widow’s
interests, hence the e↵ect of inheritance by widows is a combined e↵ect for the widow and
children. If the children’s and/or widow’s interests are at least partially protected when a
child or another member of the wife’s family inherits, we expect �2 to be positive. Hence,
the hypothesis being tested in (7) is that the combined interests of the widow and children
are di↵erentially protected when she inherits, when some other family member inherits, and
when the land reverts to the headman.
4.2 Identification
Why is it that some communities allow widows to inherit, while others do not? The iden-
tification strategy in this paper relies on the assumption that conditional on an array of
community-level and household-level observables, inheritance rights are orthogonal to other
factors that govern investment in land.
In Section 2 we described the context in Zambia, where local rules governing inher-
itance by widows are rooted in longstanding village traditions. Local norms are strongly
influenced by ethnicity (Ashraf et al., 2016). Zambia is an ethnically diverse and highly frac-
tionalised country, even for sub-Saharan Africa (Taylor and Hudson, 1970). Posner (2004)
13
estimates that Zambia has the third highest degree of ethnic fractionalization on the conti-
nent, as measured by the degree of diversity among Politically Relevant Ethnic Groups. The
variation in inheritance practices that we observe in the data stems primarily from historical
traditions at the community level that vary, at least in part, along ethnic lines.
A reasonable concern is that inheritance norms may be correlated with other aspects
of community life that impact land investment. However, we show in the following section
that statistically significant di↵erences in the means of control variables between inheritance
by widows and widow non-inheritance areas largely disappear once we control for district
fixed e↵ects (which is what we do in all empirical specifications). Within a district, villages
that di↵er in inheritance customs do not appear to systematically di↵er on other economic
and demographic variables that are relevant to agricultural production. Inheritance norms
may not be randomly assigned, but with respect to key observables they are close to condi-
tionally independent within-district.
Furthermore, we have found no indication that widows’ land rights have co-evolved
with other norms governing participation by women in agriculture. Our findings are robust to
the inclusion of a historical variable from Murdock’s Ethnographic Atlas, which characterizes
ethnic groups based on whether women traditionally participate in agriculture. This suggests
that gender norms governing land inheritance do not perfectly co-move with those governing
the role of women in agriculture. This should come as no surprise for those who study
gender in agriculture in sub-Saharan Africa. Women traditionally do substantial work on
farms, even in societies where they have few rights. Additionally, we looked for evidence
of widespread ground campaigns to change inheritance-by-widows practices in Zambia. We
found no such evidence for the time period of our data.
If there is reason to be concerned about policy endogeneity, the most likely form of
reverse causality works against our findings. That is, if leaders revoke inheritance rights
from widows in areas where land investment and land quality are high, this would create a
positive correlation between non-inheritance and land investment. That is the opposite of
the empirical relationship in the data. That said, one can imagine other forms of endogeneity,
such as a correlation between inheritance and investment driven by “backwards” leaders or
other forms of disadvantage, that do not necessarily contradict the relationships in the data.
14
Based on the arguments in this section, our preferred empirical approach is to treat
inheritance by widows as conditionally independent within districts, and to interpret OLS
estimates of the a↵ect of widow non-inheritance on land investment as our main findings.
However, we also show IV estimates, using SEA ethnic group population shares as instru-
ments. The IV findings are broadly supportive of the OLS results. We discuss this further
in Section 6.2, when we present the IV results.
5 Data
The primary household data for this paper come from the 2008 Rural Income and Liveli-
hoods Survey undertaken by the Food Security Research Project of Michigan State University
(MSU) and the Central Statistical O�ce (CSO) of Zambia. The data set includes a wide vari-
ety of standard agricultural, economic, and demographic information at the plot, household,
and community level. In 2008 the MSU and CSO team conducted a separate survey with
1,043 village chiefs (“headmen”) in the study area. The headmen survey covered a range of
village characteristics, including norms regarding inheritance, distances to trading centers,
business activities, access to government extension and input subsidization programs, trends
in community well-being, and others. We use data from both of these surveys.
The household data contains information for 8,094 households. After dropping house-
holds with no agricultural activities, and a small number of additional households that are
missing key variables, we have a data set of 7,770 households. We limit our analysis to
investment on plots governed by the customary system. This is not very restrictive: of the
27,134 plots in the data set, 94.5% (25,629) are described by households as owned, but with-
out a title deed. These are primarily lands that were allocated to households by the village
headman or by other family members. Also, for most specifications we focus on the 5,803
households that are led by a married couple, because, according to our conceptual frame-
work, policies governing land inheritance by widows will only influence these households. We
consider single-headed households in a falsification exercise.
15
5.1 Land inheritance by widows
For this paper, the key questions in the headmen survey are: “If a male head of household
dies, who would get the rights to the homestead and crop land assuming he had a wife and
adult children?” and “If a male head of household dies and there are only young children
and a wife, who gets the right to the homestead and crop land?” The options are “Spouse
(wife)”, various other family members such as “Wife’s brother,” “Brothers of deceased” or
“Any adult child,” and “Goes back to the village for headman to re-allocate.” For the main
specifications we code these responses into a “No inheritance” variable that takes a value
of 1 if the widow never inherits the land, and 0 otherwise (in Section 6.3.1 we consider a
finer disaggregation of the inheritance regime, based on who inherits if not the widow). This
widow non-inheritance variable is the key source of policy variation in the paper.
Each headman represents a village, so widow tenure security varies at the village
level. Unfortunately, because of how the data were recorded, it is not possible to match the
household and headmen survey data at the village level.4 Moreover, the headman survey
was conducted in only a subset of all villages. Therefore, we match households to headmen
at the standard enumeration area (SEA) level, one geographical level above the village (but
still highly localized). On average, 3.3 headmen were surveyed in each SEA. Fortunately,
within-SEA responses by headmen are highly correlated. In a majority of SEAs (60%), all
headmen give the same answer regarding the inheritance regime. When there is within-SEA
variation, it usually takes the form of a single dissent from the mode (in only 17% of SEAs
does more than one headman dissent from the mode). Hence, we assign each household the
modal response from all headmen interviewed in the SEA.5
Table 1 shows summary statistics for inheritance norms at the provincial level. Figure
1 maps the same data aggregated at the provincial level. The lack of correlation with agro-
4In both surveys “village” is recorded as a text variable with no corresponding numeric code. Our
attempts to match villages by text name were not successful, with less than 40% matched even after correcting
for spelling variations.5In a robustness check we define this variable using the share of headmen in the SEA who report that
widows do not inherit. This has only a minor e↵ect on findings, with the fertilizer OLS results becoming
less precise. Results available upon request.
16
Table 1: Variation in survey coverage and inheritance norms, by province
ProvinceNumber ofSEAs
Number ofhouseholds
Number ofcouples
Widownon-inheritance (%)
Central 39 787 589 35.9Copperbelt 21 427 330 4.8Eastern 72 1521 1153 37.3Luapula 46 986 743 14.3Lusaka 9 188 139 49.8Northern 76 1525 1153 23.7Northwestern 27 546 398 21.9Southern 49 1017 789 41.3Western 41 815 537 23.0Overall 380 7812 5831 28.0
Notes: Authors’ calculations from MSU/CSO data. Percent widow non-inheritance calculated byfirst averaging responses across headmen within an SEA, then averaging across SEAs within aprovince. Response variable takes a value of 1 if widows inherit, and 0 otherwise.
climatic zone is clear: inheritance norms vary within every province of Zambia.
5.2 Summary statistics and outcome variables
The key outcome variables in this paper are the farmer choices that measure investment in
land. The three that we consider are the amount of inorganic fertilizer applied to the farm,
the share of land fallowed, and the share of cultivated land that is prepared using intensive
tillage techniques. These are the only measures of land investment in the data. There is
insu�cient variation in tree crops, and no information on fencing, security measures, or other
forms of investment.
Fallowing has zero or negative returns in the present period, making it an unambigu-
ously medium- or long-term investment. “Share fallow” is defined as the proportion of land
under the household’s control that is described as “fallow” or “improved fallow” during the
current cultivation period (of which “improved fallow” represents only a very small fraction).
Fertilizer application and intensive tillage can have immediate returns by increasing yields
in the present period, as well as medium- or long-run returns by maintaining soil nutrition
and structure for future years.6 Fertilizer application is calculated as the sum of all fertilizer
6The exact timing of the returns to fertilizer depends on various other inputs and land characteristics.
17
Western'
North'Western'
Southern'
Central'
Lusaka'
Eastern'
Northern'
Luapula'Copperbelt'
23%
41%
50% 36%
22%
24%
14%
5%
37%
Figure 1: Percent of villages where widows do not inherit landNotes: Authors’ calculations from MSU/CSO data.
applied as either basal or top-dressing to cultivated plots.7 The share of intensive tillage
is the percentage of cultivated land that is tilled using planting basins, ridges, or bunds.
These techniques are designed to prevent erosion, absorb water, and improve yields. The
most common alternative tillage techniques, accounting for over 85% of tillage that we term
“non-intensive”, are conventional hand hoeing and plowing.8
Marenya and Barrett (2009) show in Kenya that on plots with insu�cient soil organic matter, nitrogen
fertilizer may not be taken up by the plants and hence have no immediate return. This is an example of
a broader point widely known to agronomists: nutrients can and do remain in soils for many years if not
taken up by plants because of some other growth-limiting factor. Hence, while targeted doses of fertilizer
that deliver the otherwise limiting nutrients to a plot may only a↵ect output for one year, the average farmer
in Zambia, operating without information about current soil quality, will apply nutrient combinations that
cannot all be taken up in the short term and therefore have potentially multi-year e↵ects. It is for this reason
that Gavian and Fafchamps (1996) and others have used fertilizer application or manuring as measures of
investment in other studies on land tenure in sub-Saharan Africa.7In Appendix A we provide results using fertilizer application per cultivated hectare as the dependent
variable, and also discuss why we prefer the measure of total fertilizer for the main results. This decision
has no impact on the findings.8See Ngoma, Mulenga and Jayne (2014) for detailed discussion of conservation tillage methods in the
same setting.
18
Table 2: Household-level summary statistics
N Mean s.d.
Size of household 7812 5.93 2.98Adult equivalents 7812 4.94 2.51Couples (=1) 7812 0.75 0.44Head is widowed (=1) 7812 0.15 0.36Asset index value 7812 -0.04 0.82Matrilineal household (=1) 7807 0.43 0.50HIV prevalence (%) 7712 11.41 5.10No inheritance (=1) 7812 0.34 0.47
Notes: Authors’ calculations from MSU/CSO data.
Table 2 shows summary statistics at the household level. Households are large, with
nearly 6 members on average. 75% of households are headed by a married couple. 15%
of households are headed by a widow, underscoring the importance of policies related to
widowhood. The asset index is a standardised wealth measure formed from the first principal
component of a vector of assets and dwelling characteristics (Sahn and Stifel, 2003). Slightly
less than half of households are self-described as matrilineal, i.e., tracing lineage through
female ancestors and descendants. The mean HIV prevalence rate is 11.4%. Approximately
a third of households live in a village where widows do not customarily inherit land.
In Table 3 we report household descriptive and agricultural statistics by inheritance
regime, for couple-headed households. Column 3 shows the raw di↵erences between groups.
Households in non-inheritance areas are slightly larger, have a little more land, live in commu-
nities with greater access to government fertilizer subsidies, are less likely to be matrilineal,
and have more cattle. Although many of these di↵erences are statistically significant, most
are small enough to be of little economic significance. More importantly, the statistically
significant di↵erences between groups largely disappear when we look within district and
cluster standard errors at the SEA level (column 4). The only economically meaningful
di↵erence in a control variable is for the fertilizer subsidy policy variable, and in that case
the sign works against the hypothesis of the paper: households in non-inheritance areas have
slightly greater access to a fertilizer subsidy. Finally, the unconditional and within-district
mean levels of fallowing, fertilizer and intensive tillage – the key outcome variables in the
data set – are all significantly higher in areas where widows inherit. This is consistent with
19
Table 3: Household summary statistics by inheritance regime, couples only
Widow caninherit
Widowcannotinherit Di↵erence
Di↵erencew/ districtf.e.
(1) (2) (3) (4)Control variablesSize of household 6.44 6.64 -0.20* -0.05Adult equivalents 5.36 5.52 -0.16* -0.04Asset index value 0.03 0.03 0.00 0.08*Matrilineal household (=1) 0.43 0.40 0.03** -0.01HIV prevalence (%) 11.02 12.19 -1.17*** n/aNumber of plots 3.76 3.73 0.03 -0.03Total land area (Ha) 3.85 4.37 -0.52 -0.74Share maize 0.39 0.45 -0.07*** 0.002Fertilizer subsidy available (=1) 0.63 0.71 -0.08*** -0.09*Female agriculture (=1) 0.79 0.90 -0.10*** -0.004Number of cattle 2.74 3.54 -0.80** 0.53
Outcome variablesTotal fertilizer applied (kgs) 173.83 151.07 22.76 74.4***Share fallow 0.12 0.09 0.02*** 0.02**Intensive tillage (share) 0.33 0.28 0.05*** 0.03*N 3839 1992
Notes: Authors’ calculations from MSU/CSO data. Entries in columns 1 and 2 are mean values.
Column 3 is the di↵erence (column 1 – column 2). Entries in column 4 are regression coe�cients
on a dummy for “Widow can inherit” with district fixed e↵ects and standard errors clustered at
the SEA level. HIV prevalence is measured at the district level, so is not identified in column 4.
*** p <0.01; ** p <0.05; * p <0.10.
the central hypothesis of the paper, and previews the results to follow.
The “Female agriculture” binary variable is from a di↵erent data set, the Ethnographic
Atlas of Murdock (1967). We assign this variable a value of 1 if the tribe of the household
head is traditionally associated with female participation in agriculture, and 0 otherwise.
We include this dimension of between-ethnic-group variation in the light of the evidence
on strong a link between modern gender roles and the traditional involvement of women in
agriculture (Alesina, Giuliano and Nunn, 2013).9 As we see in Table 3, households in non-
inheritance areas actually have higher rates of traditional female participation in agriculture.
9Because some ethnic groups in the data set are not represented in Murdock (1967), the female agriculture
variable is only available for two thirds of households.
20
This lends further support to the identification discussion of the previous subsection: the fact
that traditional norms governing participation by women in farming do not strongly co-vary
with the inheritance by widows norm lends support to the idea that the rules governing land
transfer can evolve independently of those governing agricultural practices.
6 Results
In this section, we first present OLS regression results based on equation (6) for the three
land investment decisions: fertilizer application, fallowing, and intensive tillage. We then
present the IV results, and a series of empirical extensions.
6.1 Main results
Table 4 shows the main coe�cient of interest from OLS estimates of equation (6). Regressions
are limited to couple-headed households, and include district fixed e↵ects as well as controls
for landholdings, household size, and presence of a government fertilizer subsidy program in
the SEA. Columns 2 and 3 add additional controls to the baseline specification in column
1. Column 4 adds a control for the historical role of women in agriculture from Murdock
(1967). This variable is not available for all ethnic groups, which reduces the size of the
estimation sample in column 4.
Results in panel 1 are from regressions with fertilizer application as the dependent
variable. The coe�cient of interest is highly statistically significant, and of significant eco-
nomic magnitude. Farming households apply 61-88 fewer kilograms of fertilizer in SEAs
where widows do not inherit. This is equivalent to 37-50% of the mean rate of fertil-
izer application for couple-headed households, or 13-18% of the mean rate (478 kg) among
couple-headed households with non-zero levels of fertilizer application (approximately 35%
of households apply some fertilizer). The coe�cient estimates are remarkably stable across
specifications, even as we include additional control variables that increase R2 (from 25% in
column 1 to 35% in column 4).
Panel 2 shows estimates of the coe�cient of interest from regressions with the share of
land fallowed as the dependent variable. Regressions are identical to those from panel 1, apart
21
Table 4: Land investment and inheritance by widows, OLS
(1) (2) (3) (4)
Panel 1
Dependent variable: Fertilizer application (kg)No inheritance (=1) -81.01*** -73.36*** -61.11*** -88.06***
(22.60) (22.11) (18.50) (25.56)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5843 5842 5830 3865R-squared 0.25 0.29 0.34 0.35Mean of dep. variable 165.76 165.58 165.87 177.24
Panel 2
Dependent variable: Share of land fallowedNo inheritance (=1) -0.013* -0.014* -0.016** -0.016*
(0.007) (0.007) (0.007) (0.010)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5529 5528 5517 3621R-squared 0.17 0.17 0.18 0.18Mean of dep. variable 0.107 0.107 0.107 0.102
Panel 3
Dependent variable: Share of land under intensive tillageNo inheritance (=1) -0.034 -0.035* -0.035* -0.020
(0.021) (0.021) (0.021) (0.019)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5476 5475 5464 3599R-squared 0.37 0.37 0.37 0.42Mean of dep. variable 0.32 0.32 0.32 0.30
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households headed by a couple.
All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding size,
household size (adult equivalent), and presence of fertilizer subsidy program in the sampling unit. Standard
errors in parentheses, clustered at the SEA level. *** p <0.01; ** p <0.05; * p <0.10
22
from the di↵erent dependent variable. The e↵ect of non-inheritance is statistically significant
at 5 or 10% in all regressions. The e↵ect is again of substantial economic magnitude and
stable across specifications. A policy of widow non-inheritance is associated with fallowing
rates that are lower by 0.013-0.016 percentage points, representing 12-16% of the mean
fallowing rate, or 4-5% of the mean among couple-headed households with non-zero levels of
fallowing (30% of the sample).
Panel 3 shows the results using intensive tillage as the dependent variable, with the
same patterns of controls. The signs of the estimated coe�cients are consistent with the
fertilizer and fallowing results: investment is lower in areas where widows do not inherit.
Conditional intensive tillage rates are lower by 0.020-0.035 percentage points, in non-widow-
inheritance communities. This represents a reduction of 7-11%, or 3-5% for the average
household with non-zero levels of intensive tillage (43% of households in our sample). The
key coe�cient is statistically significant at 10% in columns 2 and 3, and very close to statis-
tically significant in column 1 (p-value = 0.101). Across these three specifications, the point
estimates are stable. The estimate is attenuated in column 4, for which the sample includes
almost 2,000 fewer households. Overall, the e↵ects in panel 3 of Table 4 are smaller in mag-
nitude and less precisely estimated than those in panels 1 and 2, but remain important and
in the hypothesized direction.
6.2 IV results
We turn now to the instrumental variables results. As discussed in Section 4.2, we believe
policy endogeneity is unlikely to drive our findings, so we view the IV estimates as a ro-
bustness check. For instruments, we use ethnic group population shares at the SEA level.10
Zambia is well suited to this IV strategy, as there is substantial mixing of ethnic groups at
the local level (Ashraf et al., 2016). In fact, there are so many ethnic groups in the data,
10We examined two other instruments that we hypothesized would be predictive of inheritance by widows
rights, but not otherwise directly related to agricultural inputs. These were (i) SEA-level rates of matrilin-
eality as reported in the household surveys, and (ii) the percentage of households in inheritance by widows
areas who are of the same ethnicity as a given household, but live in other villages (to isolate the ethnic
variation). Neither performed as well as the results shown, in that F-stats were substantially lower.
23
most with only scant representation, that including all of them would lead to a large and
unwieldy set of instruments. Instead, we estimate IV regressions using the shares of the five
largest ethnic groups, which account for roughly half of sample households.11 These instru-
ments reflect the underlying source of di↵erent traditions regarding inheritance by widows:
variation in cultural norms and widows’ rights by ethnic group.
Table 5: Land investment and inheritance by widows, IV
(1) (2) (3)
Dependent variable: Fertilizer (kg) Share fallowShare intensivetillage
No inheritance (=1) -168.325** -0.061* -0.498***(83.202) (0.036) (0.121)
Observations 3865 3621 3599R-squared 0.35 0.17 0.23First stage F-stat 14.65 13.94 13.33Over-identification test (p-val) 0.40 0.68 0.41
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households headed by a couple.
All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding size,
household size (adult equivalent), presence of fertilizer subsidy program in the sampling unit, asset index
value, number of cattle owned, and female agriculture variable from Murdock (1967). Robust standard errors
in parentheses (we use robust standard errors so that we can explicitly test for over-identification). First-
stage F statistic refers to the first stage IV regression. P-value is from a Sargan test for overidentification.
*** p <0.01; ** p <0.05; * p <0.10
IV estimates are based on the column 4 specifications from Table 4, which include
the largest set of controls. For this IV strategy to satisfy the exclusion restriction, it is not
necessary that ethnicity a↵ects agricultural outcomes only through widow land inheritance.
Rather, the instruments must be conditionally uncorrelated with the outcomes, after con-
trolling for the large set of covariates included in our regressions. Indeed, we find that with
the substantial set of controls in the IV regressions, the instruments are conditionally valid:
over-identification tests are easily satisfied for all three outcome variables.
Table 5 shows the coe�cient of interest from the IV regressions. In all cases the
results are greater in magnitude than the OLS results, statistically significant, and in the
11Results are broadly similar if we use the 10 or 15 largest ethnic groups, although the ethnic group shares
become small quickly. Over-identification becomes a concern as the number of instruments increases.
24
hypothesized direction. The first stage F-statistics are all well above 10. These regressions
provide clear support for the OLS results in section 6.1. Land investment by couple-headed
households is substantially lower in area where widows do not inherit.
We should note that IV results are not as strong for fallowing and intensive tillage
if we exclude the “Female agriculture” variable from Murdock (1967). This underscores the
importance of the control variables in identifying the causal e↵ects on land investment.
6.3 Empirical extensions
So far, the empirical results present a consistent story: couple-headed household invest less
in their land in areas where widows do not inherit. This relationship is economically and
statistically significant, and robust to the use of the population shares of major ethnic group
as instruments for widow non-inheritance.
We now consider a number of extensions. We divide these into three groups: testing
for heterogeneity based on who inherits land (if not the widow), falsification tests using
non-couple-headed households, and a study of the implications for farm revenues.12
6.3.1 Variation in who inherits
As described in Section 4.1, the headmen survey data contains additional information on
who, other than the widow, inherits land in the event that the male head of household dies.
In this subsection we test for meaningful variation based on additional details about the
identity of the heir. If households are aware of this variation and incorporate it into their
subjective beliefs about the likelihood of the wife or her family retaining the land, then we
expect to see the highest level of investment in areas where the widow inherits, followed by
areas where a child or widow’s family member inherits, followed by the excluded group. We
again use the full set of controls, from column 4 of Table 4, as the basis for these regressions.
Estimates of equation (7) are shown in Table 6. All point estimates are positive,
12We also examined whether the deterrent e↵ect of widow non-inheritance is greater in areas with higher
rates of HIV mortality, which proxies for the probability of widowhood. Data on district-level HIV rates
for 2007 are from the CSO (Central Statistical O�ce, 2005). Those results are imprecise and di�cult to
interpret, so we have provided them in Appendix section B.
25
Table 6: Land investment and additional details about the heir, OLS
(1) (2) (3)
Dependent variable: Fertilizer (kg) Share fallowShare intensivetillage
Widow inherits 192.565*** 0.050** 0.059*(56.231) (0.024) (0.033)
Other family member inherits 111.321* 0.064** 0.018(61.815) (0.026) (0.044)
Cattle owned Yes Yes YesAsset index Yes Yes YesFemale agriculture Yes Yes YesObservations 3865 3621 3599R-squared 0.35 0.18 0.42Mean of dep. variable 177.24 0.10 0.30
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households headed by a couple.
All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding size,
household size (adult equivalent), presence of fertilizer subsidy program in the sampling unit, asset index
value, number of cattle owned, and female agriculture variable from Murdock (1967). Standard errors
clustered at SEA level. *** p <0.01; ** p <0.05; * p <0.10
as expected when we define variables according to inheritance rather than non-inheritance.
Five of the six inheritance variables are statistically significant, indicating that investment
is higher when the widow or her family inherits. For fertilizer and intensive tillage, the
ordering of the magnitudes of the inheritance variables is as hypothesized, with the greatest
investment in areas where the widow herself inherits the land. The di↵erence between the
coe�cients on “Widow inherits” and “Other family member inherits” is statistically signif-
icant for fertilizer (p-value = 0.076). In column 2 we cannot reject the hypothesis that the
inheritance coe�cients are equal.
The relative magnitudes of the estimated coe�cients for fertilizer and tillage are
consistent with households considering the wife’s potential for long-term control of the land
when making investment decisions. The evidence is far from definitive. Yet, the intuition
underlying these results aligns with those of the previous section, and gives further reason
to interpret the relationship between the inheritance regime and land investment as causal.
26
6.3.2 Falsification tests
How confident can we be that the e↵ect we are measuring relates specifically to widow
non-inheritance, and not to some generalized di↵erences in the security of property rights
across communities? If the e↵ect we document were due to the latter rather than to the
mechanism proposed, it would have a negative impact on the land investment choices of all
types of households, not just those headed by couples. To test this, we conduct falsification
tests of the widow non-inheritance variable on the land investment choices of households
headed by a single man or woman. The majority of households in these regressions are
headed by a woman (83%), reflecting the relative ease with which widowed men re-marry.
Table 7: Falsification tests, non-couple headed households, OLS
Fertilizer(kg)
Sharefallow
Shareintensivetillage
No inheritance (=1) -25.439 -0.017 -0.027(18.113) (0.014) (0.024)
Cattle owned Yes Yes YesAsset index Yes Yes YesFemale agriculture Yes Yes YesObservations 1293 1205 1188R-squared 0.55 0.25 0.43Mean of dep. variable 71.74 0.12 0.26
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households not headed by a
couple. All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding
size, household size (adult equivalent), and presence of fertilizer subsidy program in the sampling unit.
Standard errors in parentheses, clustered at the SEA level. *** p <0.01; ** p <0.05; * p <0.10
Table 7 shows the results. The specifications match those from column 4 of Table
4. None of the results are statistically di↵erent from zero. The point estimate for fertil-
izer is significantly smaller than the OLS estimates for couple-headed households, and the
coe�cients for all three outcomes are much smaller than the IV results for couple-headed
households. If we use the specification from column 3 of Table 4, which excludes the Female
Agriculture variable and thereby increases the number of observations by roughly 40%, the
results are even weaker (in that point estimates are smaller and p-values larger). Households
27
not headed by a couple do not appear to be directly a↵ected by the inheritance-by-widows
regime. Taken collectively, we interpret Table 7 as evidence that the main e↵ects we report
in Section 6 are indeed likely to operate through a perceived lack of future tenure security
for widows on the part of couple-headed households.
These tests have some limitations. The falsification groups include a substantial
number of widows who might be impacted by other policies related to widowhood. But if
anything, we would expect such policies to co-move with the widow non-inheritance policy, so
that the expected direction of bias would be to exaggerate rather than attenuate the e↵ects
in Table 7. A related concern is that widows in non-inheritance areas who cultivate their own
plots are clearly not selected at random. This concern is mitigated by the fact that, in some
instances, a widow who does not inherit from her husband may move and obtain new land in
her community of origin (Kajoba, 2002). Finally, when we attempt to pool the sample and
estimate the impact on couple-headed households using a di↵erence-in-di↵erence framework,
the result is only significant for fertilizer. This underscores the conditional nature of our
main results, and the role of the control variables in absorbing other important sources of
variation in land investment.
6.3.3 Implications for revenues
If widow non-inheritance reduces investment in land quality, is there a corresponding e↵ect on
household earnings? A strength of our main analysis is that it focuses on input or investment
demand equations, not on a stochastic outcome such as revenues that is a function of many
other variables. Nevertheless, soil degradation threatens food security largely by reducing
land productivity, so it would reasonable to hypothesize that average revenue may be lower
in areas with lower land investment. To test this proposition we estimate regressions with
log of agricultural revenues as the dependent variable, using the same set of specifications
that we used for the land investment variables in Table 4.
Table 8 shows the results. The coe�cient of interest is negative in all columns, in line
with the proposed hypothesis. Yet, with the inclusion of any additional control variables, the
widow non-inheritance variable is not statistically significant. In our preferred specification,
column 4, the variable is far from significant.
28
Table 8: Agricultural revenue and widow land inheritance, OLS
Dependent variable: Log of total agricultural revenue(1) (2) (3) (4)
No inheritance (=1) -0.094* -0.081 -0.057 -0.045(0.056) (0.056) (0.053) (0.064)
Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5231 5230 5219 3495R-squared 0.42 0.44 0.46 0.47Mean of dep. variable 13.8 13.8 13.8 13.9
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households headed by a couple.
All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding size,
household size (adult equivalent), and presence of fertilizer subsidy program in the sampling unit. Standard
errors in parentheses, clustered at the SEA level. *** p <0.01; ** p <0.05; * p <0.10
It is possible that reduced output results in reduced consumption from own produc-
tion, which is not reflected in revenues. Likewise, with a negative but imprecise e↵ect on
revenues, it is possible that profits are lower in non-inheritance areas. We cannot test the
e↵ect on profits because we lack comprehensive data on costs. Nonetheless, if costs are higher
for farmers who alter their investments in response to non-inheritance of land by widows,
then in combination with the imprecisely estimated but negative e↵ects on revenues it is
plausible that farm profits are lower when widows do not inherit.
7 Discussion
In this paper we have demonstrated a strong, arguably causal connection between the security
of widows’ land inheritance rights and investments in land productivity in Zambia. The idea
that local institutions matter for economic outcomes is not new in the social sciences, and
is the subject of a rapidly growing literature in economics. Yet, the forces underlying our
findings are surprisingly subtle. For us to find the e↵ects that we do, households must
perceive their leaders’ attitudes toward widows, and worry enough about mortality of the
household head that it impacts their land investment decisions. The former is highly likely.
Widowhood is commonplace, and there are widows in every village. Households would have
29
had numerous opportunities to observe how widows are treated by their leaders.
What about the latter e↵ect? Is the death of a household head likely enough to
support our interpretation of the estimated results? We believe so. Although the mortality
rate (due to illness) for male household heads between 2001 and 2004 was only 1.7% in the
study communities, the mortality rate for all prime-age adults over the same period was
greater than 10.1% (Chapoto, Jayne and Mason, 2011).13 The HIV crisis was nearing its
peak during the years preceding collection of the data used in this paper, and mortality
from malaria and other infectious diseases remained a constant concern. Households faced
a real possibility that prime age household members might pass away. Additionally, even
if one considers these annual mortality rates to be modest, it would be consistent with a
wide literature in psychology and economics for households to over-estimate and over-react
to the empirically “small” probability of head mortality (Kahneman and Tversky, 1979).
Furthermore, the agronomic and economic evidence discussed in Section 3 suggests that
complementarities between land quality and land investments will magnify the e↵ects of
future insecurity on current-day investments. Finally, while the estimated e↵ects are robust
and statistically significant, they are not overwhelmingly large. We reported in subsection 6.1
that in comparison to the mean levels of investment by households with non-zero investment
of each type, the negative e↵ects of widow non-inheritance are 13-18% for fertilizer, 4-5%
for fallowing, and 3-5% for intensive tillage. Widow non-inheritance leads households to
re-balance their investments away from land, not to abandon land quality entirely. Our
findings of negative but statistically non-significant impacts on revenues are supportive of
this interpretation.
In our view, the modest size of the estimated e↵ects also serves to underscore the main
contribution of this paper. While we believe the reductions in investment to be important
for long-run productivity, we are not claiming to have identified a silver bullet for increasing
farm output in sub-Saharan Africa. Rather, our findings are best interpreted as a view into
one of the many subtle but important ways that insecure rights for women can undercut
13Chapoto, Jayne and Mason (2011) report that 10.1% (542/5342) of households experienced at least one
death of a prime-age member of various demographic subcategories. Multiple deaths within each category
are not counted, so the actual mortality rate from illness is bounded below by 10.1%.
30
growth and productivity. Indeed, widow non-inheritance is not the first place one looks
when trying to find variation in land rights. None of the papers in recent literature reviews
utilizes variation of this nature (Place, 2009; Fenske, 2011), and we would not have thought
to write this paper were it not for the widespread land expropriation from widows that led
to the work of Chapoto, Jayne and Mason (2011). Yet, while widows’ access to land may not
be the only determinant of agricultural investment decisions, households are clearly sensitive
to the how their local leaders view widows, and they take these prospects into account when
allocating scarce resources.
The debate over customary versus statutory approaches to land management often
treats local communities as unitary objects. We have shown that customary systems in
Zambia can in fact exhibit important heterogeneity in the degree to which they safeguard
the interests of widows. When women are systematically disadvantaged by local leaders,
their households may be impacted even while their husbands are alive. This suggests a
forward-looking and dynamic dimension to agricultural investment that is usually di�cult
to observe. More generally, this finding serves as a reminder that any comparison of land
tenure regimes should consider the broad range of dimensions along which tenure security
may or may not be protected (Whitehead and Tsikata, 2003).
As researchers and policymakers continue to search for ways to increase land in-
vestment and land productivity in SSA, findings such as these serve as a reminder that
households are solving a complex and often poorly understood optimization problem. Low
rates of program uptake and unexpectedly small responses to seemingly well-designed inter-
ventions may stem from unobserved variation in property rights and security of assets that
leads households to make unexpected, but nonetheless optimal, resource allocation decisions.
31
Appendix - intended for online publication only
A Alternative fertilizer measure
The main results for fertilizer are based on total fertilizer applied to the farm, with exten-
sive controls for household landholding size. We prefer this fertilizer measure for the main
specifications because it matches the literature (e.g., Gavian and Fafchamps (1996)), and
because construction of an intensive measure of fertilizer application involves dividing to-
tal fertilizer by planted acres, which could be endogenous to realized or expected fertilizer
availability (especially for basal fertilizer applied at planting). Despite these concerns, as
an additional robustness check we re-estimate the fertilizer specifications using kilograms of
fertilizer per cultivated hectare as the dependent variable. Results are provided in Table
A1. The broad findings are unchanged from the main results: across all specifications, less
fertilizer is applied in areas where widows do not inherit.
Table A1: Alternative fertilizer measure and widow land inheritance, OLS
Dependent variable: Kilograms of fertilizer per cultivated hectare(1) (2) (3) (4)
No inheritance (=1) -15.91** -15.18** -11.79** -13.38*(6.63) (6.65) (5.77) (7.84)
Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5505 5504 5493 3617R-squared 0.19 0.20 0.27 0.28Mean of dep. variable 60.04 60.03 60.10 63.10
Notes: Authors’ calculations from MSU/CSO data. Regressions restricted to households headed by a couple.
All regression contain district fixed e↵ects, control for the a 4th-degree polynomial in landholding size,
household size (adult equivalent), and presence of fertilizer subsidy program in the sampling unit. Standard
errors in parentheses, clustered at the SEA level. *** p <0.01; ** p <0.05; * p <0.10
32
B Interactions with HIV rates
Recent work has uncovered a number of channels by which HIV/AIDS or HIV/AIDS treat-
ment can impact resource allocation or forward-looking economic behavior by individuals
in low-income settings (Ainsworth, Beegle and Koda, 2005; Delavande and Kohler, 2009;
Gra↵ Zivin, Thirumurthy and Goldstein, 2009; Thornton, 2012; Baranov and Kohler, 2018;
Baranov, Bennett and Kohler, 2015). In our setting, average HIV prevalence is well over
10%, with significant between-district variation. In this section we examine whether the
relationship between inheritance norms and land investment is stronger in areas with higher
HIV rates. In these tests, district HIV rates proxy for uncertainty about the probability of
survival by the male head of household. (Note that identification is not based on the HIV
status of household members).
We estimate regressions of the following form:
qdsh = �NoInheritanceds + �1HIVd + �2HIV 2d (8)
+�3 NoInheritancedsHIVd + �4 NoInheritancedsHIV 2d + �0Xdsh + ↵p + ✏dsh
where HIVd is the district-level HIV rate, ↵p are province dummy variables, and other
variables are as defined in equation (6) in the main body of the paper. Xdsh includes
the full sets of controls from column 4 of Table 4. By plotting linear combinations of
� + �3HIV + �4HIV 2 for counterfactual HIV rates HIVd, we can examine whether the
disincentive e↵ect of widow non-inheritance is greater in magnitude when the probability
of widowhood is higher. The data on annual rates of HIV prevalence are from the Zambia
Central Statistical O�ce (Central Statistical O�ce, 2005). The district-level rates range
from roughly 4-25%. Data from levels below the district were not available.
Figure 2 plots the results and Table A2 reports the corresponding coe�cients.
Figure 2 panel a shows the estimated e↵ects on fertilizer application. The solid line
in the lower panel shows the e↵ect of the rising HIV rate on fertilizer usage in areas with no
inheritance by widows, relative to areas with inheritance by widows. The dashed lines are
standard errors, calculated with the delta method. To aid interpretation, a kernel density
approximation of the HIV distribution is shown in the upper panel. Fertilizer usage declines
33
(a) Fertilizer use (b) Fallowing shares
(c) Intensive tillage
Figure 2: Relationship between HIV rates and household investmentsNotes: Authors’ calculations from MSU/CSO data.
steadily through the first 70% of the HIV distribution, then e↵ectively levels o↵ over the
relevant support. This is weak but intriguing evidence of possible underlying heterogeneity
based on HIV prevalence and the associated sense of heightened survival risk.
In Figure 2 panel b, we show a similar plot based on regressions with the share of land
fallowed as the dependent variable. Unlike with fertilizer application, we find no statistically
significant heterogeneity in the e↵ect of widow non-inheritance based on the HIV rate. We
know from Table 4 that the average e↵ect of widow non-inheritance on fallowing, across HIV
rates, is negative. There is no evidence that the fallowing rate varies with the HIV rate.
Finally, in Figure 2 panel c, we plot the e↵ect of increasing HIV rates on intensive
tillage in areas with a policy of widow non-inheritance. Although the rate of intensive tillage
34
is decreasing over the first 50-60% of the support, the overall relationship is not as clear as
it was in the case of fertilizer. The disincentive to invest in land is increasing in the HIV
rate at lower levels and increasing at higher levels.
Taken together, these three sets of results show at most a weak relationship between
investment disincentives from widow non-inheritance and HIV rates. The strongest indica-
tion of possible heterogeneity by HIV rate is for fertilizer application. Yet we would not claim
that as a finding. If there is underlying heterogeneity based on perceptions of mortality risk,
district-level HIV rates may be a poor measure with which to detect it.
Table A2: Fertilizer application, female inheritance, and HIV, OLS
(1) (2) (3) (4)No inheritance (=1) 98.28 71.43 43.36 -10.98
(91.0) (89.5) (83.9) (107.7)HIV rate 10.60 12.59 13.95 12.47
(8.6) (8.3) (8.6) (10.5)No inher * HIV rate -27.36* -22.44 -17.45 -9.59
(16.0) (15.7) (14.9) (20.2)HIV rate sq. -0.24 -0.30 -0.44 -0.34
(0.3) (0.3) (0.3) (0.4)No inher * HIV rate sq. 0.93 0.80 0.66 0.36
(0.6) (0.6) (0.6) (0.8)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5777 5776 5764 3807R-squared 0.22 0.27 0.32 0.33Mean of dep. variable 167.06 166.88 167.17 179.17
Notes: Regressions restricted to households headed by a couple. All regression contain district fixed e↵ects,
control for the a 4th-degree polynomial in landholding size, household size (adult equivalent), and presence
of fertilizer subsidy program in the sampling unit. Standard errors in parentheses, clustered at the SEA
level. *** p <0.01; ** p <0.05; * p <0.10
35
Table A3: Share of land fallowed, female inheritance, and HIV, OLS
(1) (2) (3) (4)No inheritance (=1) -0.035 -0.031 -0.028 0.025
(0.046) (0.046) (0.046) (0.054)HIV rate -0.002 -0.002 -0.002 -0.001
(0.005) (0.005) (0.005) (0.005)No inher * HIV rate 0.004 0.003 0.003 -0.005
(0.007) (0.007) (0.007) (0.009)HIV rate sq. 0.000 0.000 0.000 -0.000
(0.000) (0.000) (0.000) (0.000)No inher * HIV rate sq. -0.000 -0.000 -0.000 0.000
(0.000) (0.000) (0.000) (0.000)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5470 5469 5458 3570R-squared 0.11 0.12 0.13 0.12Mean of dep. variable 0.11 0.11 0.11 0.10
Notes: Regressions restricted to households headed by a couple. All regression contain district fixed e↵ects,
control for the a 4th-degree polynomial in landholding size, household size (adult equivalent), and presence
of fertilizer subsidy program in the sampling unit. Standard errors in parentheses, clustered at the SEA
level. *** p <0.01; ** p <0.05; * p <0.10
36
Table A4: Share of land under intensive tillage, female inheritance, and HIV
(1) (2) (3) (4)No inheritance (=1) 0.155 0.158 0.158 0.232
(0.140) (0.140) (0.140) (0.155)HIV rate -0.008 -0.008 -0.008 0.001
(0.013) (0.013) (0.013) (0.013)No inher * HIV rate -0.033 -0.034 -0.034 -0.040
(0.022) (0.022) (0.022) (0.024)HIV rate sq. 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)No inher * HIV rate sq. 0.001* 0.001* 0.001* 0.001*
(0.001) (0.001) (0.001) (0.001)Cattle owned No Yes Yes YesAsset index No No Yes YesFemale agriculture No No No YesObservations 5420 5419 5408 3550R-squared 0.29 0.29 0.29 0.35Mean of dep. variable 0.32 0.32 0.32 0.30
Notes: Regressions restricted to households headed by a couple. All regression contain district fixed e↵ects,
control for the a 4th-degree polynomial in landholding size, household size (adult equivalent), and presence
of fertilizer subsidy program in the sampling unit. Standard errors in parentheses, clustered at the SEA
level. *** p <0.01; ** p <0.05; * p <0.10
37
References
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “The Colonial Origins of Com-parative Development: An Empirical Investigation.” The American Economic Review, 91(5): 1369–1401.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2005. “Chapter 6: Institutions asa Fundamental Cause of Long-Run Growth.” In Handbook of economic growth. Vol. 1 of Handbook ofEconomic Growth, , ed. Philippe Aghion and Steven N. Durlauf, 385 – 472. Elsevier.
Ainsworth, Martha, Kathleen Beegle, and Godlike Koda. 2005. “The Impact of Adult Mortalityand Parental Deaths on Primary Schooling in North-Western Tanzania.” The Journal of DevelopmentStudies, 41(3).
Alesina, Alberto, Paola Giuliano, and Nathan Nunn. 2013. “On the Origins of Gender Roles: Womenand the Plough.” The Quarterly Journal of Economics, 128(2): 469–530.
Ali, Daniel Ayalew, Klaus Deininger, and Markus Goldstein. 2014. “Environmental and GenderImpacts of Land Tenure Regularization in Africa: Pilot evidence from Rwanda.” Journal of DevelopmentEconomics, 110: 262–275.
Allendorf, Keera. 2007. “Do Women’s Land Rights Promote Empowerment and Child Health in Nepal?”World Development, 35(11): 1975–1988.
Amsalu, Aklilu, and Jan De Graa↵. 2007. “Determinants of Adoption and Continued Use of StoneTerraces for Soil and water conservation in an Ethiopian highland watershed.” Ecological Economics,61(2): 294–302.
Ashraf, Nava, Natalie Bau, Nathan Nunn, and Alessandra Voena. 2016. “Bride Price and FemaleEducation.” National Bureau of Economic Research Working Paper 22417, Unpublished manuscript.
Baranov, Victoria, and Hans-Peter Kohler. 2018. “The impact of AIDS treatment on savings andhuman capital investment in Malawi.” American Economic Journal: Applied Economics, 10(1): 266–306.
Baranov, Victoria, Daniel Bennett, and Hans-Peter Kohler. 2015. “The indirect impact of an-tiretroviral therapy: Mortality risk, mental health, and HIV-negative labor supply.” Journal of HealthEconomics, 44(C): 195–211.
Barrett, C.B., F. Place, and A.A. Aboud. 2002. Natural Resources Management in African Agriculture:Understanding and Improving Current Practices. CABI.
Benin, Samuel. 2006. “Policies and Programs A↵ecting Land Management Practices, Input Use and Pro-ductivity in the Highlands of Amhara region, Ethiopia.” Strategies for Sustainable Land Management inthe East African Highlands, 217–256.
Besley, Timothy. 1995. “Property Rights and Investment Incentives: Theory and Evidence from Ghana.”Journal of Political Economy, 103(5): 903.
Brasselle, Anne Sophie, Frederic Gaspart, and Jean Philippe Platteau. 2002. “Land tenure Securityand Investment Incentives: Puzzling Evidence from Burkina Faso.” Journal of Development Economics,67(2): 373–418.
Burke, William J., Thom. S. Jayne, and J. Roy Black. 2017. “Factors explaining the Low and VariableProfitability of Fertilizer Application to Maize in Zambia.” Agricultural Economics, 48(1): 115–126.
Central Statistical O�ce. 2005. “Zambia - HIV/AIDS Epidemiological Projections 1985-2010.” Republicof Zambia.
38
Chapoto, Antony, T. S. Jayne, and Nicole M. Mason. 2011. “Widows’ Land Security in the Eraof HIV/AIDS: Panel Survey Evidence from Zambia.” Economic Development and Cultural Change,59(3): 511–547.
Deininger, Klaus, and Songqing Jin. 2006. “Tenure security and land-related investment: Evidencefrom Ethiopia.” European Economic Review, 50(5): 1245–1277.
Delavande, Adeline, and Hans-Peter Kohler. 2009. “Subjective expectations in the context ofHIV/AIDS in Malawi.” Demographic Research, 20: 817–875.
Doepke, Matthias, MichAsle Tertilt, and Alessandra Voena. 2012. “The Economics and Politics ofWomen’s Rights.” Annual Review of Economics, 4(1): 339–372.
Doss, Cheryl, Chiara Kovarik, Amber Peterman, Agnes Quisumbing, and Mara van den Bold.2015. “Gender inequalities in ownership and control of land in Africa: myth and reality.” AgriculturalEconomics, 46(3): 403–434.
Duflo, Esther. 2012. “Women, Empowerment, and Economic Development.” Journal of Economic Litera-ture, 50(4): 1051–1079.
FAO. 2016. “Boosting Africa’s Soils.” Food and Agricultural Organization of the United Nations.
Fenske, James. 2011. “Land tenure and investment incentives: Evidence from West Africa.” Journal ofDevelopment Economics, 95(2): 137–156.
Fernandez, Raquel. 2014. “Women’s rights and development.” Journal of Economic Growth, 19(1): 37–80.
Gausset, Quentin. 2001. “AIDS and cultural practices in Africa: the case of the Tonga (Zambia).” SocialScience & Medicine, 52(4): 509–518.
Gavian, Sarah, and Marcel Fafchamps. 1996. “Land Tenure and Allocative E�ciency in Niger.” Amer-ican Journal of Agricultural Economics, 78(2): 460–471.
Gilbert, Natasha. 2012. “African agriculture: Dirt poor.” Nature, 483: 525–527.
Goldstein, Markus, and Christopher Udry. 2008. “The Profits of Power: Land Rights and AgriculturalInvestment in Ghana.” Journal of Political Economy, 116(6): 981–1022.
Gra↵ Zivin, Joshua, Harsha Thirumurthy, and Markus Goldstein. 2009. “AIDS treatment andintrahousehold resource allocation: Children’s nutrition and schooling in Kenya.” Journal of Public Eco-nomics, 93(7-8): 1008–1015.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An Analysis of Decision under Risk.”Econometrica, 47(2): 263–292.
Kajoba, Gear. 2002. “Women and Land in Zambia: A Case Study of Small-Scale Farmers in ChenenaVillage, Chibombo District, Central Zambia.” Eastern Africa Social Science Research Review, 18(1): 35–61.
Kumar, Neha, and Agnes R. Quisumbing. 2012. “Gendered Impacts of the 2007-08 Food Price Crisis:Evidence Using Panel Data from Rural Ethiopia.” Food Policy, 38: 11–22.
Malungo, J.R.S. 2001. “Sexual cleansing (Kusalazya) and levirate marriage (Kunjilila mungaanda) in theera of AIDS: changes in perceptions and practices in Zambia.” Social Science & Medicine, 53(3): 371 –382.
Marenya, Paswel P., and Christopher B. Barrett. 2009. “State-Conditional Fertilizer Yield Responseon Western Kenyan Farms.” American Journal of Agricultural Economics, 91(4): 991–1006.
39
Morris, Michael L. 2007. Fertilizer use in African agriculture: Lessons learned and good practice guidelines.World Bank Publications.
Murdock, George Peter. 1967. Ethnographic Atlas. Pittsburgh:University of Pittsburgh Press.
Ngoma, Hambulo, Brian P. Mulenga, and T.S. Jayne. 2014. “What Explains Minimal Usage ofMinimum Tillage Practices in Zambia? Evidence from District-Representative Data.” Indaba AgriculturalPolicy Research Institute (IAPRI) Working Papers, 82.
Pender, John, Ephraim Nkonya, Pamela Jagger, Dick Sserunkuuma, and Henry Ssali. 2004.“Strategies to increase agricultural productivity and reduce land degradation: evidence from Uganda.”Agricultural Economics, 31(2): 181 – 195. Current Issues in the Economics of Agriculture, Food, andResources: Reshaping Agriculture’s Contributions to Society.
Peterman, Amber. 2012. “Widowhood and Asset Inheritance in Sub-Saharan Africa: Empirical Evidencefrom 15 Countries.” Development Policy Review, 30(5): 543–571.
Place, F., and K. Otsuka. 2002a. “Land Tenure Systems and Their Impacts on Agricultural Investmentsand Productivity in Uganda.” The Journal of Development Studies, 38(6): 105–128.
Place, Frank. 2009. “Land Tenure and Agricultural Productivity in Africa: A Comparative Analysis of theEconomics Literature and Recent Policy Strategies and Reforms.” World Development, 37(8): 1326–1336.
Place, Frank, and Keijiro Otsuka. 2002b. “Land Tenure and Natural Resource Management: A Compar-ative Study of Agrarian Communities in Asia and Africa.” Journal of Development Studies, 38(6): 105–128.
Place, Frank, and Peter Hazell. 1993. “Productivity E↵ects of Indigenous Land Tenure Systems inSub-Saharan Africa.” American Journal of Agricultural Economics, 75(1): 10–19.
Posner, Daniel N. 2004. “Measuring Ethnic Fractionalization in Africa.” American Journal of PoliticalScience, 48(4): 849–863.
Richardson, Abby Morrow. 2004. “Women’s Inheritance Rights in Africa: The Need to Intergrate Cul-tural Understanding and Legal Reform.” Human rights brief, 11(2): 6.
Rodrik, Dani, Arvind Subramanian, and Francesco Trebbi. 2004. “Institutions Rule: The Primacyof Institutions over Geography and Integration in Economic Development.” Journal of Economic Growth,9(2): 131–165.
Sahn, David E., and David Stifel. 2003. “Exploring Alternative Measures of Welfare in the Absence ofExpenditure Data.” Review of Income and Wealth, 49(4): 463–489.
Shezongo-Macmillan, Joyce. 2005. “Women’s Property Rights in Zambia.” Presented at Strategic Liti-gation Workshop, Johannesburg.
Strickland, Richard S. 2004. “To Have and To Hold: Women’s Property and Inheritance Rights in theContext of HIV/AIDS in Sub-Saharan Africa.” ICRW.
Tabellini, Guido. 2010. “Culture and Institutions: Economic Development in the Regions of Europe.”Journal of the European Economic Association, 8(4): 677–716.
Taylor, Charles Lewis, and Michael C Hudson. 1970. “World Handbook of Political and Social Indi-cators II. Section 1. Cross-National Aggregate Data.” DTIC Document.
Thornton, Rebecca L. 2012. “HIV testing, subjective beliefs and economic behavior.” Journal of Devel-opment Economics, 99(2): 300 – 313.
40
Ubink, Janine M, and Julian F Quan. 2008. “How to combine tradition and modernity? Regulatingcustomary land management in Ghana.” Land Use Policy, 25(2): 198–213.
Udry, Christopher. 1996. “Gender, Agricultural Production, and the Theory of the Household.” Journalof Political Economy, 104(5): 1010–1046.
Veit, Peter. 2012. “Custom, Law and Women’s Land Rights in Zambia.” Focus on Land in Africa Brief.
Von Liebig, Justus. 1840. Organic Chemistry in its Application to Agriculture and Physiology. Lon-don:Playfair.
Whitehead, Ann, and Dzodzi Tsikata. 2003. “Policy Discourses on Women’s Land Rights in Sub-Saharan Africa: The Implications of the Re-turn to the Customary.” Journal of Agrarian Change, 3(1-2): 67–112.
Yngstrom, Ingrid. 2002. “Women, Wives and Land Rights in Africa: Situating Gender Beyond theHousehold in the Debate Over Land Policy and Changing Tenure Systems.” Oxford Development Studies,30(1): 21–40.
41