Déterminants Des Revenus Ruraux Au Kenya

Journal of African Economies Advance Access published October 21, 2011
Journal of African Economies, Vol. 0, number 0, pp. 1 –35

doi:10.1093/jae/ejr039
Determinants of Rural Income: The Role

of Geography and Institutions in Kenya
Maren Radenya,b and Erwin Bultea,c, *
a
Development Economics Group, Wageningen University, PO Box 8130, 6700 EW
Wageningen, Netherlands
b
International Livestock Research Institute, PO Box 30709, 00100 Nairobi, Kenya
Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

c
Department of Economics, Tilburg University, Tilburg, Netherlands
* Corresponding author: Erwin Bulte. E-mail: erwin.bulte@wur.nl
Abstract
We revisit the debate about the root causes of income divergence, and ask
whether geographical variables or institutions are the main determinants
of income. Complementing earlier cross-country work, we focus on the
local level and seek to explain within-country income differences. Analysing
Kenyan household data, we find that certain geographical variables appear
to be more important drivers of per capita income levels than local insti-
tutions. Once we control for geography, our measures of community-level
institutions do not seem to explain within-Kenya differences in income.
JEL classification: 012, B52, 040, D72
1. Introduction
A rapidly growing literature explores the factors explaining differences in
per capita income across countries. Two main hypotheses have emerged
from this literature. One identifies geographical variables as the main
determinant of long-term income, and the other points to institutions as
the key driver.1 In the light of ongoing efforts to enhance the effectiveness
and efficiency of development assistance, the question which hypothesis is
correct is clearly an issue of first-order relevance. If geographical factors are
1
Alternative explanations for income differences include policies and, for the African con-
tinent, the history of slave raids (Nunn, 2008).
# The author 2011. Published by Oxford University Press on behalf of the Centre for the
Study of African Economies. All rights reserved. For permissions, please email:
journals.permissions@oup.com
2 | Maren Radeny and Erwin Bulte
the main impediment to economic growth, then ‘big push, top-down’

approaches based on a careful diagnosis of the geographical impediments
combined with a blueprint of technical fixes may potentially be
appropriate to lift societies out of poverty. Matters are arguably more
complex if it were true that ‘institutions rule’. Then, institutional reform
may be a prerequisite for successful follow-up intervention (e.g., Easterly,
2006). But our understanding of which institutions to reform, and how
to go about reforming them, is quite imperfect (Rodrik, 2006).
Most research into the root causes of income inequalities focus on a

global sample of countries. Such studies face several challenges, and
Brock and Durlauf (2001) note ‘This literature does rely on assumptions
that may be argued to be [. . .] dubious and whose implausibility renders
the inferences typically claimed by empirical workers to be [. . .] suspect’
(p. 232). Among the potential problems, consistent measurement of rel-
evant variables, especially institutional proxies, is difficult. Also, the open-
endedness of growth theories (so that validating one theory does not imply
falsifying another one) implies a risk of omitted variables—variables not
included in the regression models, but correlated with both regressors
and the dependent variable (biasing estimates). This point is aggravated
by the simple observation that the set of countries included in most
income regressions is quite diverse. In addition, one may doubt whether
the ‘one-size-fits-all’ approach of most linear income models adequately
captures the diversity of mechanisms linking institutions and geography
to economic outcomes (the concern of parameter heterogeneity). For
example, the link between malaria ecology and incomes varies with the
quality of health interventions in countries.
Recent evidence by Bhattacharyya (2009a) suggests that the nature of the
geography-versus-institutions debate varies with the aggregation level of
the analysis. Specifically, while the empirical evidence for global samples
typically points to institutions as the main determinant of income, geo-
graphical factors (notably malaria prevalence) best explain income differ-
ences in a sample of African countries. Regardless of the sample,
however, cross-country studies gloss over within-country income differ-
ences, and take mean per capita income as the key dependent variable.
The main objective of this paper is twofold. First, using the local data
that we collected in central, western and eastern regions in Kenya, we
revisit the geography-versus-institutions debate at the local level, and
aim to explain within-country income differences. Secondly, recognising
the multidimensionality of the institutional framework, we aim to ‘unbun-
dle’ institutions and distinguish between a number of institutional proxies.
Determinants of Rural Income | 3
To attenuate concerns about the endogeneity of our institutional variables,

we also seek to identify an exogenous variation in local institutions by
using various instrumental variables. Our main results provide support
for the geography-based perspective on underdevelopment—at least
within Kenya. Both geography and institutions may therefore matter for
promoting growth and reducing income inequality, depending
on whether the focus is on international or domestic income differences
(or rather: domestic income differences for certain countries).
The paper is organised as follows. In Section 2, we briefly summarise

the theories and evidence underpinning the perspectives that, respectively,
geography or institutions are the key determinants of income.
In Section 3, we introduce the setting. In Section 4, we outline the identi-
fication strategy and introduce our data. Section 5 contains our empirical
results, including a robustness analysis. Finally, in Section 6, we discuss
our findings and place them in perspective.
2. Geography versus institutions: Theory and evidence

The geography-based view on underdevelopment identifies two broad cat-
egories of geographical factors that determine long-term income growth:
transport costs and intrinsic productivity. Although transport cost may
not be invariant with respect to institutional variables, it is evident that
geographical variables affect the ease with which regions or countries can
trade with the rest of the world. Variables such as coastal access, the pres-
ence or absence of navigable rivers, distance to major consumer or produ-
cer areas, but also mountain ranges, potentially have a large effect on
transport cost. Africa, for example, has few navigable rivers and many land-
locked countries, and is relatively far away from ‘core markets’ in Europe
(Gallup et al., 1999).
The geographical determinants of intrinsic productivity are potentially
varied. For example, prevalence of infectious diseases, such as malaria,
affects human productivity and lowers life expectancy (attenuating incen-
tives to invest in various forms of capital). The prevalence of malaria and
other infectious disease is to an important extent determined by the
ecology of the parasites (different species of malaria Plasmodia) and the
vectors (different species of Anopheles mosquitoes). Geography also
affects productivity via the direct link between agro-ecological conditions
and agricultural output. Tropical agriculture suffers a productivity loss of
30–50% compared with temperate-zone agriculture, after controlling for
other factor inputs (Gallup et al., 1999; Gallup and Sachs, 2001; Sachs and
Malaney, 2002; Sachs, 2003). Lastly, resource endowments may affect
income, albeit not necessarily in a positive sense (see Brunnschweiler
and Bulte, 2009 on the resource curse).
How may institutions matter? Institutions refer to the social, economic,
legal and political organisation of a society. They shape the incentive struc-
ture of economies, and determine economic performance via their impact
on accumulation and investment decisions of economic agents.
Institutions affect information flows, transaction costs, investment risk

and the ability of societies to coordinate on collective action to address
social dilemmas. For example, secure private land rights may spur invest-
ments in land via a so-called ‘assurance effect’ (agents invest if they expect
to collect the future returns); a ‘collateralisation effect’ (security enhances
the scope for using land as collateral—facilitating access to credit and
enabling productive investments) or a ‘realisability effect’ (capturing that
investments in land may be encouraged if land can be rented out or
sold). But the concept of institutions is broader, and also includes a politi-
cal dimension (e.g., checks on the executive, degree of authoritarianism),
governance-type of variables (e.g., corruption, rent seeking) and social
capital (shared norms and values, trust).
Sometimes the dividing line between institutions (rules of the game)
and policies (play of the game) is fuzzy. Glaeser et al. (2004) make the
point that some of the so-called institutional variables commonly used
in empirical studies reflect policy choices as much as they reflect con-
straints on decision-making. For example, Bhattacharyya (2009b) shows
that both strong market-creating institutions, including effective contract
enforcement, as well as market-stabilising institutions, capturing macroe-
conomic stability and distortionary policies, are growth enhancing.
However, such market-creating and stabilising institutions arguably
reflect both constraints for policy-makers as well as the choices made by
them. That is, they are both inputs and outputs of the political decision-
making process.
No one doubts that many African countries face both geographical and
institutional challenges. While the ‘geography school’ points to malaria,
depleted soils and prohibitive transport costs as the main impediments
to development, political scientists like Rowley (2000) emphasise the link
between political culture and economic performance. Many economic pro-
blems of Africa have their roots in public-choice impediments (see also
Bates, 2006, 2008). According to this view, African leaders and the elites
who support them have deliberately crafted governance systems to promote

rent-seeking (see also Congdon Fors and Olsson, 2007).
Which perspective is predominantly correct? As hinted above, the evi-
dence supporting one view versus another appears to vary with the aggre-
gation level. Cross-country evidence from a sample of former European
colonies suggest that institutions predominantly explain cross-country
variation in per capita incomes (e.g., Hall and Jones, 1999; Acemoglu
et al., 2001; Easterly and Levine, 2003; Rodrik et al., 2004). After controlling
for (predicted) institutional quality, these studies find no evidence that

geography or policies matter. However, Bhattacharyya (2009a) pits geogra-
phy versus institutions in the context of a much smaller, and more hom-
ogenous, sample. Focusing on African countries, he examines the relative
contributions of the history of slave trade, the legacy of extractive colonial
institutions and the prevalence of malaria in explaining variation in
incomes. Interestingly, for this sample, malaria risk is a significant determi-
nant of long-term economic development (or the lack of it), and the
remaining factors—including the quality of institutions—do not enter sig-
nificantly. The channel via which malaria affects growth could be reduced
savings. Increased mortality and morbidity raise current consumption at
the expense of savings and investments.
Scaling down the aggregation level further, the logical next level of analy-
sis focuses on intra-country income differentials. Since the local data on
income, geography and institutions are not readily available, combined
with the fact that there are no ‘routine’ instrumental variables for insti-
tutions, it is no surprise that few such studies exist. Yet, it is likely that
studies based on local variation in institutions are especially informative
for policy-makers. Grimm and Klasen (2008) focus on the local income
determinants in Indonesia. This paper identifies tenure security as the criti-
cal institutional link fostering investment and higher incomes. Hence,
moving from the global sample of former colonies via a sample of
African nations and onwards to samples of villages within specific
nations, we observe that the nature of the evidence swings—from
institutions to geography, and then back to institutions, as the main deter-
minant of income levels.
3. Geography and institutions in Kenya

This paper aims to contribute new micro-level evidence to the
geography-versus-institutions debate. We examine how geography and
institutions affect income levels among rural households across different

villages in central, western and eastern regions in Kenya. Earlier studies
have explored the link between geography and poverty in Kenya (e.g.,
Kristjanson et al., 2005; Okwi et al., 2007), but the role of institutions
has largely been ignored.
In the 1980s and 1990s, the Kenyan economy grew slowly and erratically.
At the dawn of a new millennium, the new government embarked on an
economic recovery process through a broad nationwide development fra-
mework. While Kenya’s economic growth accelerated between 2003 and

2007, with growth in real GDP estimated at about 7% in 2007, the
growth momentum faltered in 2008.2 Not surprisingly, low mean
incomes are also reflected in a high incidence of poverty. Official statistics
indicate that the incidence of rural poverty was 49% in 2005–06 (KNBS,
2007), with significant regional and sub-regional differences (see also
CBS, 2003; Okwi et al., 2007; Suri et al., 2009). What explains these
differences?
Geographical and agro-ecological conditions display a significant vari-
ation across Kenya. These include rainfall, soil fertility and altitude, but
also infrastructural access to markets and public facilities, and the presence
or absence of natural resources (forests or water bodies). Kristjanson et al.
(2005) use spatial data analysis to examine the geographical correlates of
meso-, or community-level poverty incidence in Kajiado district, Kenya.
They identified pasture potential (normalised differential vegetation
index (NDVI)), livestock density, distance to a major town, road density,
access to education and security, soil fertility and agricultural potential
as relevant factors explaining variation in poverty levels across sub-
locations. Okwi et al. (2007) also relate the incidence of rural poverty to
geographical conditions, and find mixed effects of geographic variables
at regional (provincial) levels. Slope, soil type, distance or travel time to
public resources, elevation, type of land use, demographic and income
inequality variables significantly explained spatial patterns of poverty.
However, controlling for these variables, provincial dummies remained sig-
nificant. This suggests that provinces in Kenya are heterogeneous along
multiple relevant dimensions—perhaps not just geographical ones.
Burke and Jayne (2008) examined the relative importance of
spatial-versus-household-level factors in explaining variation in wealth
2
Restrained by internal and external factors—including the 2008 post-election conflict, the
global financial crisis and high fuel and food prices—economic growth slowed from 7.1%
in 2007 to 1.7% in 2008.
and poverty across regions, communities and households in Kenya. They

provide evidence that spatial factors—remoteness and isolation, weak
economic integration and, in particular, agricultural potential—play a
substantial role herein. Poverty was prevalent in areas with binding land
constraints and areas with relatively low agricultural potential.
Household-specific factors explain an equal proportion of the variation
in household wealth as spatial factors.
The latter finding suggests that although geographical variables are
important, other factors may also matter. Possibly institutional proxies

are among such other factors. Obviously, our ‘micro-level’ analysis fails
to pick up any effects due to ill-functioning (political) institutions at the
macro level. National institutions affect all villages alike, and the counter-
factual (i.e., an alternative set of Kenyan macro institutions) does not
exist.3 However, the relevant institutional framework encompasses national
as well as local institutions, and arguably there is a variation in the quality
of institutions at the local level. This is what we set out to explore in this
paper, and we distinguish between four dimensions of institutional quality
at the community (village) level. A detailed description of our institutional
variables is provided in the next section.
4. Data and empirical strategy

In 2009, we collected data from 430 households in twenty-eight commu-
nities, across four agro-regional zones in Kenya. These data are summarised
in Table 1. Agro-regional zones represent a cluster of areas with similar
broad climatic conditions, agricultural activities and rural livelihood strat-
egies. In addition, the zones reflect diversity in market access and popu-
lation density. The study sites are located in central, western and eastern
parts of Kenya and represent high- and low-potential agro-ecological
zones. Within each community, households were randomly selected from
a list of all households in the community. The survey covers various
3
One referee correctly pointed out that national institutions need not necessarily imply
that all villages are affected alike. Kenya has a ‘winner takes all’ political system at the
national level, which significantly impacts on the distribution of resources at the local
level. In so far as access to resources is significantly influenced by the presence or the
absence of matching ethnic identities of the incumbent government and local commu-
nities, Kenya’s national patronage system is likely to matter for income distribution at
the local level. In that case, national institutions governing the distribution of public
goods along ethnic lines may be an important factor. However, our data do not permit
us to explore this further—we focus exclusively on local institutions.
Table 1: Summary of the Data
Variable type Description Mean SD
Income variables
Household-level annual Income from crops, livestock and off- 120,794 123,327
income farm (salary income, remittances,
business income and income from
casual Labour and dividends) in KShs.
Household-level annual Household annual income per adult 32,564 42,550
per capita income equivalent unit (in KShs.)
Geographical variables

Mean main season Average main season rainfall from 1997 563 178
rainfall to 2007 (mm)
Mean maximum Mean long-term maximum tempera- 26.86 2.33
temperatures (8C) tures (8C)
Altitude Altitude (m)—household level 1492 299
Distance to market Distance to the nearest market (km) 2.52 1.27
Distance to major town Distance to the nearest major town 20.39 9.41
(km)
Distance to a good road Distance to all-year motorable road 8.56 4.33
(km)
Mountains, scarps and Landscape dummy—mountains and 0.15 0.36
hiUs major scarps, hiUs and minor scarps
Volcanic foot ridges Landscape dummy—volcanic foot 0.17
ridges
Uplands Landscape dummy—upper and lower 0.26
middle-level uplands
Lower level upland Landscape dummy—lower level upland 0.14
Lacustrine plains Landscape dummy—lacustrine plains 0.28
Phaeozems Soil type dummy—Phaeozems 0.11
Greyzems Soil type dummy—Greyzems 0.03
Podzols Soil type dummy—Podzols 0.28
Regosols Soil type dummy—Regosols 0.17
Rankers Soil type dummy—Rankers 0.40
NDVI Mean normalised differential 168 14.46
vegetation index (1997–2009)
Malaria risk dummies Based on P. falciparum parasite rate
(PfPR)—(Noor et al., 2009)
Low malaria risk PfPR ranges is ,1% 0.44 0.50
Medium malaria risk PfPR ranges from 1 to ,5% 0.18 0.38
High malaria risk PfPR ranges from 20 to 40% 0.17 0.38
Very high malaria risk PfPR is .40% 0.21 0.41
Household controls
Male household head Household head is male 0.68
(continued on next page)

Table 1: Continued
Education of household Household head has secondary and 0.26

head above education
Age of household head Age of household head 58.55 14.20
Land size (acres) per Total land holding in acres per adult 0.86 1.10
capita equivalent
Land tenure Household has title deed (yes/no) 0.75
Cash crop farmer Household is a cash crop farmer 0.49
(tea, coffee and sugarcane)

Income from salaries Household received income from 0.55
salaries or wages
Community controls
Population density Population density (per km2)—based 427 159
on 1999 census
Regular transport Availability of regular transport 0.64
available (yes/no)
Primary school Primary school located within the 0.65
community
Distance to health centre Distance to the nearest health facility 4.54 4.05
(km)
Clean water available Availability of clean water for domestic 0.58
use
Years of mobile phone Number of years of mobile phone 4.43 2.48
coverage
Proportion of Catholics Proportion of adults who are Catholics 0.26 0.23
(community level)
Proportion of Protestants Proportion of adult who are 0.53 0.23
Protestants (community level)
Proportion of Pentecost Proportion of adult who are 0.17 0.14
Pentecostals (community level)
Colonial settlement Colonial settlement (dummy) 0.29 0.45
Institutions variables
Governance quality Based on the rating by the community 0.68 0.26
on a number of indicators of level of
performance of the sub-chief
Trust index Trust index based on a household’s rat- 0.67 0.09
ing of their level of confidence in (a)
household, (b) members of the
extended family, (c) members of the
community. The indices were scaled
on a 1 (very low)– 10 (very high) point
scale. The sub-indices scores were
added and normalised to a 0– 1 scale
and aggregated at the community
level

Table 1: Continued
Cooperation Dummy variable, taking a value of 1 if a 0.83

household responded positive to the
question ‘Do most people in your
village help you when you are in
need?’, otherwise zero.
Political participation Average proportion of adults in our 0.59 0.13
(2005 referendum) sample who voted during the 2005
referendum

Quality of government Institutional quality index based on a 0.42 0.08
institutions household’s rating of their level of
confidence in (a) local justice, (b) local
police, (c) local political authorities.
The indices were scaled on a 1 (very
low) –10 (very high) point scale. The
sub-indices scores were added and
normalised to a 0 –1 scale and
aggregated at the community level.
important aspects of household livelihoods. First, we collected a compre-

hensive measure of household income, Yi, for household i. Household
income sums crop income (value of total output including own consump-
tion, less input costs), livestock income (value of livestock sales and total
value livestock products less production costs) and off-farm income (salar-
ied income, remittances, business income and income from casual labour
and dividends). To control for differences in family size across households,
we computed per capita income levels. The average per capita income in
2009 was KShs 32,564 (USD 434). We use the log of per capita income
as our dependent variable.
Next, we collected information on four dimensions of institution
quality; vector Ij for village j. To attenuate endogeneity concerns in our
income model and to reflect that institutional quality is a characteristic
of the community (and not of individual villagers), we consistently
measure institutions at the village level—sometimes by aggregating house-
hold responses and creating a community average. Since institutions are
only observed at the level of twenty-eight communities, we cluster standard
errors at the community level in what follows. Vector Ij includes measures
of (i) the quality of local governance, (ii) social capital, (iii) political par-
ticipation and (iv) trust in local government institutions. This is perhaps
an ad hoc list—it is certainly not an exhaustive list of proxies for local insti-
tutions. But these variables have the advantage that the required data were
relatively easy to collect. We hope vector Ij captures enough of the relevant
institutional spectrum to be informative.4
Cross-country studies indicate that good governance is important for
economic growth and provision of public goods. The local governance
quality (i) influences the enforcement of the rule of law, including
by-laws that protect private property, and (ii) increases the efficiency of
the use of public resources and provision of public goods and services.

This variable should also pick up satisfaction with respect to conflict
management and corruption. Information on governance quality was
collected through focus group discussions (FGDs). We asked respondents
to rate, on a five-point scale, the performance of the sub-chief5 regarding
(a) working in the best interest of the community; (b) working in the
best interest of the people in general and (c) the level of confidence in
the sub-chief.
Social capital is a fuzzy term, and typically captures trust, norms and
networks that enable collective action (Knack and Keefer, 1997; Bowles
and Gintis, 2002). We quantified two dimensions of social capital;
community-level trust and cooperation. Trust lowers transaction cost
and facilitates exchange of information. In high-trust communities,
households are more likely to spend fewer resources in protecting them-
selves from exploitation and unlawful or criminal violation of property
rights. High-trust communities likely depend less on formal institutions
to enforce agreements. Consequently, informal credit markets that
depend on interpersonal trust can flourish in high-trust communities,
facilitating investments especially in areas with limited access to financial
markets. Social norms influence people’s preferences and constraints.
Cooperation norms, for example, act as constraints on narrow self-
interest and can lead to collective action in the provision of various
public goods and act as informal insurance. In addition, trust and
social norms may improve economic outcomes indirectly through politi-
cal channels (Knack and Keefer, 1997). They can affect the level and
4
Earlier research has emphasised the importance of tenure security and property rights as a
driver of development. In our study, we control for property rights (title deeds) at the
household level—as a control variable rather than a community-level institutional
proxy. However, the formal property rights variable does not enter significantly in any
of our models.
5
A sub-chief is a government representative at the lowest administrative unit—the
sub-location. A sub-location is made up of a cluster of villages or communities.
character of political participation resulting in improved government

performance and improved quality of economic policies. To measure
trust, we followed the World Value Survey approach, and asked respon-
dents to rate their level of trust in household members, extended family
and community members. Trust was measured on a ten-point scale, and
scores for the sub-indices were added, normalised between zero and one,
and aggregated at the community level. On average, the trust level was
rather high with a score of 0.7. To measure the level of cooperation,
respondents were asked whether most people in their village would

help them in case of need. On average, 83% of the households provided
a positive response.
Political participation may matter for development as well (for a dis-
cussion, refer to Blattman, 2009). Arguably, political participation fosters
accountability of village leaders, and facilitates the flow of information
from the people to the leaders. Political participation is an important
potential check on local politicians, making them more accountable
and responsive to the needs of the electorate. Accountability is impor-
tant for the efficient use of local development funds and resources,
and thus increases the efficiency in provision of public goods and ser-
vices. In addition, political participation is likely to affect government
policies which in turn affect income distribution. We measure partici-
pation as the average proportion of adults that voted in the 2005 refer-
endum and 2007 general elections. Because these two measures are
highly correlated (r ¼ 0.91), we use participation in the 2005 referen-
dum in our models.
Our final institutional dimension—trust in local government
institutions—is based on the stated level of confidence in the justice
system, police and local political authorities. These were measured at
the household level on a ten-point scale, and complement the FGD
information on local governance discussed above. The scores for the
sub-indices of justice system and police were added and normalised
between zero and one, and aggregated at the community level.
Similarly, the score for the local political authorities was aggregated at
the community level. The average score across communities equals
0.42 (in both cases), indicating modest trust. The quality of justice
system and police affect the enforcement of the rule of law. The
quality of the local political authorities is likely to affect provision of
public goods and services as discussed earlier. While decision-making
with regard to such decisions is centralised, there is considerable vari-
ation at the local level because implementation depends on individual
politicians, police officers and judicial officers, whose capabilities and

incentives are not identical. Moreover, resources available at the local
level vary from location to location, affecting the efficiency and effec-
tiveness of (and, hence, the appreciation for) local institutions and
authorities.
To what extent do these different institutional variables indeed capture
different dimensions of the institutional framework? To probe into this
issue, we computed the correlation between these variables, and found
that they were not highly correlated (Supplementary data, Table SA1,

Panel A). The correlation coefficient (r) ranges from 20.004 to 0.698,
suggesting that most variables pick up something that is ‘distinct’. As men-
tioned earlier, this does not necessarily imply our list of proxies is ‘com-
plete’. The greatest correlation exists between local justice and police and
local political authorities (r ¼ 0.698, significantly correlated at the 1%
level). Hence, problems due to multicollinearity may emerge when both
proxies are included in one regression model. We therefore enter them
sequentially.6
Next, we turn to the set of geography variables. As discussed pre-
viously, geography can affect income directly through its effects on agri-
cultural productivity, morbidity and natural resource endowments. Our
vector with geography variables (Gi) contains data from primary and
secondary sources. We collected GPS coordinates for each household
in 2009, and used these coordinates to generate a number of geography
variables from existing secondary data. Following the literature,
we include altitude, rainfall, temperature, indicators of soil quality
(different landscape attributes and soil types), NDVI, malaria risk, dis-
tance to markets, distance to major towns, distance to good road and
population density (from the 1999 census). Rainfall and soil quality,
for example, determine agricultural productivity through their
effects on land productivity. The disease environment affects human
productivity and life expectancy. While some geographical variables
such as distance to markets, distance to major towns, distance to
good road and altitude vary somewhat within villages, others such as
rainfall, malaria risk, indicators of soil quality and temperature vary
6
We have also explored correlations between household-level appreciation of institutions
and household characteristics (see Supplementary data, Table SA2). The results show
that, depending on the institutional variable considered, the level of education of the
household head, cash crop farming and wage income are correlated with institutional
quality. The emerging patterns in the data do not cause us to reject the hypothesis that
our institutional variables are able to capture something meaningful.
only across villages. The rainfall data were collected by the National
Weather Service Climate Prediction Centre as part of a Famine Early
Warning System.7 We calculated the average main season rainfall for
each household as well as the variance of rainfall. However, when con-
trolling for rainfall levels, the variance never enters significantly (so this
variable is dropped in what follows). Soil data are based on the
Exploratory Soil Map and Agro-climatic Map of Kenya.8 We used the
new malaria risk data based on the work by Noor et al. (2009),
which is exogenous to per capita incomes as it is based on climatic

conditions. Correlations between geographical variables are provided
in Panel B of Supplementary data, Table SA1. Overall, these
correlations suggest that main season rainfall, altitude and maximum
temperature are highly correlated. To attenuate multicollinearity con-
cerns, we therefore include only maximum temperature in subsequent
analyses.
We also collected information on a number of household and commu-
nity controls. Household controls, vector Xi, are the age, gender and edu-
cation of the household head, per capita land holdings (in acres), source of
income and types of crops cultivated (cash versus food crops). The vector
of community controls (Cj) includes population density, ethnicity, avail-
ability of clean water, distance to health facilities, availability of regular
transport, having a primary school within the community, number of
years of mobile phone coverage, religious composition of the community
and history of colonial settlement.
The religious composition variables warrant extra discussion. Below
we explore whether the religious composition of a village affects the
quality of local institutions. There are various mechanisms via which
religion may affect local institutions. For example, as religion may pre-
scribe certain behaviours and impose norms of conduct, it helps villa-
gers to form consistent expectations about the responses of their
peers—facilitating cooperation and trust. Religious values or traditions
can also influence the views of individuals about community
members, and indeed people in general. Moreover, participation in reli-
gious groups (direct interaction) builds social capital. Religion can,
therefore, via several channels, enhance intra-group trust and facilitate
7
These data have been compiled by the Tegemeo Institute of Agricultural Policy and
Development.
8
For details on the actual soil types and the associated information, see the documentation
‘Exploratory Soil Map and Agro-climatic Zone Map of Kenya, 1980’.
collective action (see also Welch et al., 2004).9 Religion can also poten-
tially influence civic and political participation via certain social norms.
In addition, churches are avenues for providing political information. In
Kenya, some churches are more vocal in political matters than others. In
particular, the non-Catholics (Protestants and Pentecosts) under the
National Council of Churches of Kenya (NCCK) have been actively
involved in political issues. Religious institutions, therefore, hold the
potential to reconnect people to politics, and provide political infor-
mation and participatory opportunities to their followers (Greenberg,

2000). They can make demands upon the state, and influence political
outcomes so that its potential effect on income may be indirect. We
will empirically explore these issues below.
We calculate the proportion of the adult population in our sample
belonging to different religious groups: Catholics and non-Catholics
(especially Protestants, Pentecost and Muslims). These data were collected
at the household level and aggregated at the community level.
In addition to controlling for the religious composition of villages, we
also tried to control for the ethnic composition. Ethnolinguistic diversity
(fractionalisation or polarisation) may directly hinder economic develop-
ment and indirectly shape the underlying institutions and policies that
influence economic development. Evidence suggests that ethnically frag-
mented societies tend to suffer from reduced social cohesion and a
smaller supply of public goods.10 In rural Kenya, however, villages tend
to be ethnically homogenous, even if the ethnic identity of villagers
varies from one region to the next. Hence we do not include an ethnic frac-
tionalisation variable. However, we did set out to control for the ethnic
identity of the villages. Our study sites spread across four major ethnic
communities: Kikuyu (Central region), Luhya and Luo (Western region),
and Kamba (Eastern region). The Kikuyu, Luyha and Luo communities
comprised some 28% each of our sample, while the rest (16%) were
Kamba. When including ethnicity dummies to represent these groups,
we found that they were correlated with some geographical variables.
9
In Italy, the prevailing lack of trust towards others in the South has been attributed to the
strong Catholic tradition, which emphasises the vertical bond with the Church and tends
to undermine the horizontal bond with fellow citizens.
10
The incumbent leader may implement policies aimed at expropriating resources from
ethnic losers, restricting the rights of other groups and discouraging the growth of indus-
tries or sectors that might threaten the position of the ruling group (e.g., Alesina et al.,
1999).
Hence we do not include them explicitly in most analysis (but including

them does not affect our main results; details available on request).
Finally, we turn to our identification strategy, which is inspired by stan-
dard macro-level income regressions. Indeed, we follow Pande and Udry
(2005) who recommend to exploit synergies between research on specific
institutions (based on micro data) while addressing the ‘big macro ques-
tions’ of growth. We first explore correlations between geography, insti-
tutions and income using simple ordinary least squares (OLS) models,
specified as:

Y ij = b1 + b2 Gij + b3 I j + b4 X ij + b5 C j + 1ij (1)
where Yi refers to per capita income of household i and i ¼ 1, . . . , 430. Ij
refers to a vector of community-level institutional variables and Gij is a
vector of geography variables. Finally, Xij and Cj are vectors of household
and community controls, respectively. We enter Ij and Gij separately and
in combination. We cluster standard errors at the community level to
control for correlation between households in a community as almost all
institutional variables (and most of the geography variables) only vary
across communities.
We believe virtually all our geography variables are exogenous to house-
hold income.11 However, this is perhaps not true for our institutional vari-
ables. While income is measured at the household level and institutions are
measured at the community level—attenuating concerns about reverse
causality: the potential concern that household income affects community-
level institutions—we cannot rule out endogeneity concerns due to
measurement error or omitted variables. For that reason, we augment
our OLS analysis with a series of 2SLS models. For the instrumental vari-
able (IV) model, our first-stage regression is specified as:
I j = g1 + g2 Gij + g3 X ij + g4 C j + g5 V j + mij , (2)
where Vj refers to a vector of excluded instruments. Our identifying
assumption is that the religious composition of the village is exogenous
to household income, and does not influence income other than
through its impact on institutions (i.e., our religion variables are correlated
with local institutional quality, but not with the error term of the income
11
We acknowledge that geography may not be entirely exogenous to household income if
migration responds to geographic conditions. However, our study sites are located in
rural villages where migration is very limited. In Kenya, migration is mainly from the
rural to urban areas. Apart from the settlement schemes, migration across rural villages
is rare.
model). We explore the empirical basis for these assumptions below. The
second stage of the IV model is akin to model (1), but replaces our
vector of institutional variables Ij by their predicted values, Ij* (as predicted
in accordance with the model in (2)).
As mentioned above, when testing for multicollinearity we found high
correlations between some community control variables, notably popu-
lation density, ethnicity, water source, distance to health facilities, avail-
ability of transport and geography variables (e.g., malaria endemicity).
We included only community controls that are not highly correlated

with the geography variables. Since some of our soil variables were
highly correlated, we estimated models with soil-type and landscape attri-
butes separately. Based on the Akaike information criterion and the
Bayesian information criterion, we decided to retain models with land-
scape attributes. However, our results are robust to including all these
additional controls.
Finally, and as a robustness analysis, we have also done a factor analysis
to reduce the dimensionality of our institutional vector. We selected three
factors for inclusion in a series of estimates. As an extra robustness analysis
we have also estimated models with region dummies to capture geophysical
effects. These results are generally consistent with the ones presented below,
and are included in Supplementary data, Tables SA1 and SA2.12
5. Regression results
We first explore simple correlations between income, geography and insti-
tutions, and start by examining the correlation between income and
geography or institutions separately. Later, we proceed by combining
geography and institutions together in one model. Table 2 explores the
link between geography and per capita income. Column 1 is a parsimo-
nious model with only geographical variables (and a constant). In
Column 2, we add household controls, and in Column 3 we add commu-
nity controls. Column 4, finally, is the complete geographical model.
In Column 2, several geographical variables are significantly correlated
with per capita income. Specifically, maximum temperature, distance to
12
As an extra robustness analysis, we have also estimated a series of models at the village
level, explaining average income by our vectors of institutions and geographical variables,
and controls. Qualitative results are similar to the ones presented below, explaining
household income, and are made available in Supplementary data, Table SA9.
Additional results of village-level models (OLS specifications, etc.) are available on
request.
Table 2: Geographical Variables and Household Income in Kenya—OLS Results
Dependent variable: log OLS (1) OLS (2) OLS (3) OLS (4)
of per capita income
Geography variables
Mean maximum temp- 20.133*** 20.126*** 20.039 20.041
eratures (8C) (0.044) (0.028) (0.047) (0.038)
Distance to the nearest 20.114** 20.045 20.070 20.010
market (km) (0.052) (0.044) (0.055) (0.042)
Distance to the nearest 20.040*** 20.040*** 20.053*** 20.047***
major town (km) (0.008) (0.007) (0.006) (0.008)

Distance to a good 0.004 20.014 0.024 0.003
road (km) (0.026) (0.019) (0.019) (0.015)
Mountains, scarps and 0.256 20.120 0.247 20.107
hiUs (landscape) (0.384) (0.476) (0.243) (0.423)
Volcanic foot ridges 0.104 20.376 0.547*** 20.135
(landscape) (0.399) (0.464) (0.177) (0.322)
Upper and lower 0.269 20.021 20.514* 20.586
middle-level uplands (0.339) (0.440) (0.292) (0.479)
(landscape)
Lower level upland 20.137 20.509*** 20.469** 20.876***
(landscape) (0.168) (0.169) (0.218) (0.207)
Mean NDVI (1997– 0.512 0.796 5.949*** 4.839***
2009)—log (0.901) (0.787) (1.068) (0.885)
Medium malaria risk 20.402 20.656** 22.241*** 22.072***
(0.298) (0.265) (0.402) (0.454)
High Malaria risk 20.402 20.683 22.402*** 22.132***
(0.354) (0.516) (0.471) (0.748)
Very high Malaria risk 20.606 20.450 22.990*** 22.177**
(0.429) (0.549) (0.564) (0.814)
Household control variables
Male household head 0.187* 0.194*
(0.103) (0.102)
Education level of 0.354*** 0.342***
household head (0.112) (0.111)
Age of household head 20.010*** 20.010***
(0.003) (0.003)
Land size (acres) per 0.313*** 0.312***
capita (0.045) (0.044)
Household has title 20.028 20.021
deed (0.121) (0.131)
Cash crop farmer 0.256 0.360**
(0.156) (0.166)
Income from salaries or 0.603*** 0.559***
wages (0.114) (0.127)

Table 2: Continued
Dependent variable: log OLS (1) OLS (2) OLS (3) OLS (4)
Community control variables

No. of years of mobile 0.033 0.011
phone coverage (0.020) (0.024)
Proportion of Catholics 20.230 20.039
(0.224) (0.220)
Colonial settlement 22.270*** 21.651***
(dummy) (0.302) (0.413)

Availability of regular 0.185* 0.226***
transport (0.093) (0.067)
Constant 11.949** 10.248** 216.656** 211.530**
(4.771) (4.261) (6.074) (4.742)
Adjusted R 2 0.252 0.428 0.263 0.436
Observations 429 429 429 429
Robust standard errors are given in parentheses, clustered at the community level.
***p , 0.01, **p , 0.05 and *p , 0.10.
a major town, distance to a good road, landscape attributes (our soil

proxy) and malaria risk enter significantly. The signs are as expected so
that, for example, proximity to a major town are associated with high-
income levels, and malaria risk and high temperature are associated with
lower income. These results are robust to the inclusion of community con-
trols (Column 3: only the temperature variable no longer enters signifi-
cantly now). While the geography variables are ‘properly exogenous’, this
is not evident for all our controls—for example, livelihood choices (cash
crop farming and off-farm activities) as well as farm size reflect choices
of the household, and such choices could reflect the geophysical
environment.
Of the household controls, age, gender and education level of the house-
hold head, as well as per capita land holdings, are significantly and posi-
tively correlated with income. The same is true for cash crop farmers,
and for farming households with access to off-farm (wage) income. At
the community level, the availability of regular transport is significantly
and robustly correlated with income levels. In what follows, we will not
report the details of these control variables anymore, even if they are
included in the estimations (but of course these estimation results are
available on request).
In Table 3, we explore the correlations between incomes and our

institutional proxies. As before, the first model includes only our
variables of interest (in this case, dimensions of the institutional frame-
work). In Columns 2 and 3, we introduce, respectively, household and
community controls. Column 4 is the complete model. Based on cross-
country evidence, we expect a positive correlation between institutional
quality and income levels. Surprisingly, the reverse seems to be the case
in the countryside of Kenya. Our trust, governance and political partici-
pation variables tend to be negatively correlated with income levels. We

argue that some of these unexpected results may be due to endogeneity
bias.
For example, households with relatively high incomes may feel insecure
as they perceive themselves as attractive targets for robbers (ngangs) or
begging family members. This may explain low levels of trust and levels
of confidence in extended family and community members. Political par-
ticipation and voter turnout may be influenced by responses to changes in
the economy (Radcliff, 1992; Pacek and Radcliff, 1995; Aguiler and Pacek,
2000). In an emerging democracy as Kenya, poor economic performance
may drive voters to the polls (Radcliff, 1992). Economic hardships encou-
rage political participation as a means of seeking redress for grievances. We
return to these concerns below.
Table 4 shows the results of a series of OLS models that simultaneously
includes geography as well as institution quality variables (and household
and community controls). We include indicators of institutional quality at
a time. Across all models, proximity to the nearest major town has a posi-
tive and significant correlation with income. Other geography variables
(mean temperature, landscape attributes, malaria risk and NDVI) are sig-
nificantly correlated with income when controlling for various institutional
proxies (but not others). The evidence for the institutional variables is
mixed. Cooperation and political participation enter with a negative
sign, governance now enters positively and there does not appear to be a
correlation between income and trust.
For various reasons—omitted variables and possible measurement
error—the above relationships cannot necessarily be interpreted as
causal. We next follow the literature and use an IV estimation strategy to
attenuate endogeneity concerns. Specifically, we adopt a 2SLS framework,
and use as excluded instruments two types of variables: (i) a religious com-
position variable (proportion of Catholics in the community) and (ii) a
sub-set of geography variables. For both types of variables, it is not a
priori evident that the exclusion restriction is satisfied. For example, if
Table 3: Institutions and Household Income in Kenya—OLS Results
Dependent variable: OLS (1) OLS (2) OLS (3) OLS (4)
log of per capita
income
Institution variables (community level)

Trust index 22.744*** 22.752*** 20.986 21.653**
(0.862) (0.638) (0.989) (0.743)
Cooperation 1.371*** 0.411 1.303** 0.660
(0.477) (0.404) (0.580) (0.576)
Local justice and 22.341** 21.020 20.336 0.151

police (1.101) (0.860) (0.968) (0.890)
Subjective assess- 20.435* 20.132 20.934*** 20.663**
ment of local gov- (0.226) (0.225) (0.318) (0.293)
ernance quality index
Political participation 21.494* 20.431 22.488*** 21.198**
(2005 referendum) (0.755) (0.547) (0.442) (0.486)
Household control variables
Male household 0.258** 0.197*
head (0.115) (0.107)
Education level of 0.411*** 0.385***
household head (0.102) (0.098)
Age of household 20.005 20.007**
head (0.004) (0.003)
Land size (acres) per 0.291*** 0.283***
capita (0.047) (0.043)
Household has title 20.090 20.084
deed (0.114) (0.111)
Cash crop farmer 0.420*** 0.193
(0.122) (0.129)
Income from salaries 0.699*** 0.663***
or wages (0.123) (0.116)
Community control variables
No. of years of 0.022 0.007
mobile phone (0.026) (0.024)
coverage
Proportion of 0.440 0.254
Catholics (0.394) (0.350)
Colonial settlement 1.099*** 0.796***
(dummy) (0.211) (0.182)
Availability of regu- 0.733*** 0.577**
lar transport (0.262) (0.237)
Constant 12.641*** 11.313*** 10.601*** 10.264***
(1.100) (0.739) (0.870) (0.668)
Adjusted R-square 0.130 0.359 0.227 0.404
***p , 0.01, **p , 0.05, *p , 0.10.
Table 4: Geography, Institutions and Income at Household Level—OLS Evidence
Dependent vari- OLS (1) OLS (2) OLS (3) OLS (4) OLS (5)
able: log of per
capita income
Geography variables
Mean maximum 20.011 20.036 20.085* 20.068* 20.201**
temperature (8C) (0.040) (0.038) (0.047) (0.039) (0.096)
Distance to the 0.005 20.028 20.020 20.012 20.003
nearest market (0.039) (0.050) (0.044) (0.040) (0.051)
(km)

Distance to the 20.046*** 20.048*** 20.054*** 20.047*** 20.060***
nearest major (0.008) (0.007) (0.008) (0.008) (0.010)
town (km)
Distance to a 0.002 0.008 20.001 20.001 20.032
good road (km) (0.012) (0.017) (0.016) (0.013) (0.027)
Mountains, 20.168 20.113 20.059 20.208 20.248
scarps and hiUs (0.402) (0.419) (0.427) (0.440) (0.406)
(landscape)
Volcanic foot 20.057 20.131 20.317 20.410 20.956*
ridges (0.311) (0.318) (0.326) (0.427) (0.549)
(landscape)
Upper and lower 20.590 20.629 20.602 20.612 20.567
middle-level (0.449) (0.482) (0.474) (0.474) (0.442)
uplands
(landscape)
Lower level 20.729*** 20.861*** 20.908*** 20.862*** 20.886***
upland (0.196) (0.208) (0.210) (0.190) (0.208)
(landscape)
Mean NDVI 5.349*** 5.013*** 4.386*** 3.711*** 2.027
(1997– 2009)— (0.698) (0.875) (0.911) (0.966) (1.684)
log
Medium malaria 22.484*** 22.151*** 21.811*** 21.756*** 21.119
risk (0.416) (0.467) (0.450) (0.481) (0.786)
High Malaria risk 22.597*** 22.216*** 21.907** 21.843** 21.308
(0.739) (0.751) (0.724) (0.733) (0.875)
Very high 22.683*** 22.321** 21.986** 21.834** 21.207
Malaria risk (0.774) (0.838) (0.777) (0.787) (1.042)
Trust index 20.638 0.476
(0.655) (0.963)
Cooperation 20.830* 20.940**
(0.411) (0.417)
Local justice and 0.451 20.912
police (0.738) (0.978)

Table 4: Continued
Dependent vari- OLS (1) OLS (2) OLS (3) OLS (4) OLS (5)
able: log of per
capita income
Subjective assess- 0.360* 0.827**

ment of local (0.203) (0.374)
governance
Quality
Political partici- 20.645 21.033**
pation (2005 (0.407) (0.477)

referendum)
Household and Yes Yes Yes Yes Yes
community con-
trols included?
Constant 213.518*** 212.647** 28.118 24.935 8.013
(3.805) (4.806) (5.314) (5.364) (10.911)
Adjusted R 2 0.436 0.435 0.436 0.435 0.435
Observations 429 429 429 429 429
***p , 0.01, **p , 0.05, *p , 0.10.
religion promotes certain behaviours (e.g., savings, hard work) or enhances

somebody’s trustworthiness and makes him a more attractive trading
partner, then it is easy to conceive how religious identity may affect
income via other channels than the quality of local institutions (e.g., Tu
and Bulte, 2011 for an analysis of religion and income among herders in
rural Tibet). However, it is an open question whether Catholicism
encourages different behaviours than other religions in our sample. It is
also not evident that specific religions promote differential access to valu-
able trading networks. We view these as empirical questions and let the
data speak. Above we demonstrated that the fraction of Catholics in the
community is not correlated with household income (Table 2). Below
we demonstrate that the Sargan’s test confirms that the exclusion restric-
tion holds (Table 5).
We adopt a similar approach to argue that a subset of specific geographi-
cal variables may be used as instruments. We select variables that are never
significantly correlated with income, and that satisfy the Sargan test.
However, it is perhaps not evident why geography might affect institutional
quality. Various authors have established such a link in macro studies (see
Acemoglu et al., 2001; Engerman and Sokoloff, 2002; Rodrik, 2003; Rodrik
Table 5: Geography, Institutions and Household Income—IV Evidence
Dependent variable: log IV (1) 2SLS IV (2) 2SLS IV (3) 2SLS IV (4) 2SLS
Geography variables
Mean maximum temp- 20.048 20.044 20.060 20.052
erature (8C) (0.070) (0.036) (0.073) (0.040)
Distance to the nearest 20.047*** 20.047*** 20.049*** 20.047***
major town (km) (0.007) (0.007) (0.013) (0.007)
Mountains, scarps and 20.051 20.103 20.079 20.139
hiUs (landscape) (0.352) (0.309) (0.309) (0.413)

Volcanic foot ridges 20.105 20.137 20.182 20.214
(landscape) (0.347) (0.289) (0.490) (0.472)
Upper and lower 20.534 20.560* 20.587 20.594
middle-level uplands (0.400) (0.300) (0.408) (0.377)
(landscape)
Lower level upland 20.879*** 20.884*** 20.896*** 20.878***
(landscape) (0.315) (0.208) (0.200) (0.199)
Mean NDVI (1997– 4.661*** 4.723*** 4.664*** 4.509***
2009)—log (0.840) (0.854) (0.817) (0.981)
Medium malaria risk 21.965*** 22.022*** 21.972*** 21.977***
(0.412) (0.318) (0.318) (0.288)
High Malaria risk 22.024*** 22.079*** 22.056*** 22.057***
(0.492) (0.475) (0.451) (0.455)
Very high Malaria risk 22.022*** 22.089*** 22.075*** 22.056***
(0.561) (0.512) (0.492) (0.469)
Trust index 20.079
(1.932)
Cooperation 0.098
(0.886)
Local justice and police 20.268
(0.556)
Subjective assessment 0.130
of local governance (0.559)
quality
Political participation 20.210
(2005 referendum) (0.622)
Household and commu- Yes Yes Yes Yes
nity controls included?
Constant 210.602** 210.808** 210.223* 29.548
(4.399) (4.887) (5.623) (6.095)
Adjusted R2 0.437 0.438 0.439 0.439
Specification tests

Table 5: Continued
Dependent variable: log IV (1) 2SLS IV (2) 2SLS IV (3) 2SLS IV (4) 2SLS
Tests for weak instruments

Stock and Yogo 13.944 165.896 26.302 76.111
(minimum eigenvalue)
Endogeneity tests
Robust DWH 0.583 0.432 0.737 0.488
(p-value)
Over-identifying restrictions test

Sargan’s test 0.776 0.999 0.966 0.978
(p-value)
Score’s test (p-value) 0.773 0.999 0.959 0.970
Excluded instruments are distance to a good road, distance to the nearest market and the
proportion of adults Catholics in the community.
***p , 0.01, **p , 0.05, *p , 0.10.
et al., 2004)13 and in studies on the ‘natural resource curse’ (see Rodrik,
2003; Sala-i-Martin and Subramanian, 2003; Brunnschweiler and Bulte,
2009). We speculate that there may be two reasons why geography might
also shape institutions at the community level. First, geographical endow-
ments and factors (such as vegetation and altitude) may affect the stochas-
ticity of production, which in the context of imperfect markets affects the
incentives and scope for mutual insurance via informal sharing networks
(e.g., Binswanger and McIntire, 1987). Secondly, factor endowments may
affect the potential for accumulation and the degree of inequality in
wealth, human capital and political power, resulting in persistent effects
on local institutions. Hence we assume that, at the local level as well as
at the macro level, geography may shape the evolution of local commu-
nities and their livelihoods, and impact on the dynamics of local govern-
ance and institutions. Of course, it is an open question whether this
assumption is borne out by the data.
13
The most common instrument for institutional quality used in macro studies is settler
mortality in colonial time. Acemoglu et al. (2001) argued that mortality among colonial
settlers was a major factor determining settling patterns, which in turn affected the insti-
tutions that were introduced. Examples of other instruments used in macro studies are
state antiguity (Bockstette et al., 2002), or the fraction of people speaking a European
language (Hall and Jones, 1999).
One important problem of IV models is that of weak instruments. If

instruments are only weakly correlated with the endogenous regressors,
the properties of IV estimator can be poor, and estimations can be
biased (Verbeek, 2004). Several diagnostics and tests for weak instruments
exist, based on analysis of the first-stage reduced-form equations. We use
the minimum eigenvalue test as defined by Stock and Yogo (2005).
Since our number of excluded instruments is smaller than the number of
potentially endogenous variables, we include our institutional proxies
sequentially. Table 5 summarises these regression results, including specifi-

cation tests for endogeneity, over-identifying restrictions for validity of the
instruments and tests for weak instruments (the Stock and Yogo test). Our
minimum eigenvalues for models with the quality of justice system and
police (Model 2), quality of local governance (Model 3) and political par-
ticipation (Model 4) are greater than the critical values proposed by Stock
and Yogo (2005);14 however, this is not true for Model 1. The p-values of
Sargan’s test tend to support the hypothesis that the exclusion restriction is
satisfied. Results of the first-stage regressions for Table 5 are presented in
Supplementary data, Table SA5, and provide a mixed support for the
assumption that our religion and geographical instruments are correlated
with local institutions. All three excluded instruments enter significantly
in the model explaining the quality of local justice and police.
The Durbin –Wu –Hausman test statistic for endogeneity for many of
our institution quality variables is not significant (a ¼ 0.05). Hence we
cannot reject the hypothesis of exogeneity of our main regressors, and
the OLS results may be interpreted as causal. Of course, this conclusion
is valid only for those institutional variables for which our instrumental
variables are appropriate.
The estimation results are robust across columns. Four geographical
variables consistently enter significantly in three models (distance to the
nearest town, NDVI, landscape attributes and malaria risk). We re-run
our IV models using the limited information maximum likelihood
(LIML) estimator that is more robust to weak instruments compared
with the 2SLS estimator (Cameron and Trivedi, 2009). The results are
very similar to those obtained using the 2SLS estimator and are presented
in Supplementary data, Table SA6. Importantly, the IV models again
provide no support for the idea that local institutions matter for
income. None of the predicted institutional variable entered significantly
in the income model.
14
2SLS relative bias critical values.
Finally, and as a robustness analysis, we reduce the number of insti-

tutional variables by performing a factor analysis. Factor analysis aims at
identifying a number of underlying ‘factors’ that explain the variation in
various dimensions of the institutional framework. We find that the
original five institutional variables can be collapsed into three such
factors. These factors capture trust and confidence in government insti-
tutions, political participation and cooperation, and local governance,
respectively. Table 6 presents representative OLS and IV (2SLS) evidence.
The results of the first-stage regressions for Table 6 are presented in

Supplementary data, Table SA7, which are slightly better than the
results for the individual institutional proxies. Specifically, the results
for factors 1 (‘trust in local justice, police, political authorities’) and 3
(‘local governance’) appear to be decent, but factor 2 (‘cooperation
and political participation’) suffers from weak instruments. As before,
the test statistics indicate that our IV approach is appropriate, but
perhaps not necessary in the light of the absence of evidence that our
institutional variables are endogenous.
Also as before, some geographical variables—distance to the nearest
major town, landscape attributes, NDVI and malaria risk—enter signifi-
cantly in all models. Again, none of the institutional factors significantly
affects income levels. Results of this IV (LIML) estimator are presented
in Supplementary data, Table SA8, and are similar to the 2SLS results.
Taken together, we interpret the evidence on these pages as mixed
support for the geography-based perspective on underdevelopment.
Moreover, there is very little evidence that local institutions are an impor-
tant determinant of income differentials within Kenya.
6. Discussion and conclusions

The geography-versus-institutions debate on the root causes of underde-
velopment has raged prominently on the pages of many journals and
books. In this paper, we aim to contribute to this debate by adopting
a micro focus and by unbundling the institutional framework into a
number of important components. Overall, our results provide very
little evidence for the view that institutions are an important determi-
nant of within-Kenya income differences. Instead, we find a number
of geographical factors (related to infrastructure, temperature, malaria
prevalence and landscape attributes) that seem to contribute to such
income differences.
28
Table 6: Geography, Institutions and Income—Robustness Analysis
|
Dependent variable: log OLS (1) OLS (2) OLS (3) OLS (4) IV (1) 2SLS IV (2) 2SLS IV (3) 2SLS IV (4) 2SLS
Maren Radeny and Erwin Bulte

Geography variables
Mean maximum 20.048 20.048 20.075 20.078 20.051 20.054 20.042 20.045
temperature (8C) (0.039) (0.039) (0.048) (0.051) (0.037) (0.045) (0.060) (0.075)
Distance to the near- 20.047*** 20.047*** 20.050*** 20.050*** 20.047*** 20.047*** 20.046*** 20.046***
est major town (km) (0.008) (0.007) (0.007) (0.007) (0.007) (0.007) (0.008) (0.008)
Mountains, scarps and 20.059 20.059 20.066 20.093 20.098 20.150 20.061 20.109
hiUs (landscape) (0.299) (0.395) (0.282) (0.397) (0.305) (0.481) (0.262) (0.507)
Volcanic foot ridges 20.096 20.097 20.196 20.227 20.142 20.193 20.080 20.133
(landscape) (0.286) (0.390) (0.287) (0.402) (0.294) (0.466) (0.289) (0.547)
Upper and lower 20.540* 20.539 20.577* 20.599 20.556* 20.609 20.533* 20.559
middle-level uplands (0.299) (0.368) (0.303) (0.378) (0.298) (0.433) (0.286) (0.470)
(landscape)
Lower level upland 20.879*** 20.877*** 20.817*** 20.826*** 20.853*** 20.906*** 20.890*** 20.874***
(landscape) (0.223) (0.222) (0.235) (0.249) (0.210) (0.213) (0.238) (0.234)
Mean NDVI (1997 – 4.664*** 4.662*** 4.342*** 4.311*** 4.608*** 4.584*** 4.725*** 4.672***
2009)—log (0.877) (0.903) (0.986) (1.014) (0.844) (0.916) (1.066) (1.248)
Medium malaria risk 21.996*** 21.994*** 21.862*** 21.864*** 21.976*** 22.008*** 22.021*** 22.010***
(0.303) (0.314) (0.349) (0.350) (0.286) (0.298) (0.398) (0.364)
High Malaria risk 22.053*** 22.052*** 21.976*** 21.982*** 22.044*** 22.077*** 22.068*** 22.066***
(0.471) (0.493) (0.483) (0.490) (0.450) (0.485) (0.473) (0.480)
Very high Malaria risk 22.048*** 22.048*** 22.007*** 22.012*** 22.062*** 22.067*** 22.057*** 22.073***
(0.500) (0.505) (0.489) (0.500) (0.490) (0.486) (0.481) (0.500)
Trust, local justice, 0.002 0.001 20.025 20.024
police, political (0.055) (0.057) (0.055) (0.068)
authorities

Political participation 0.001 20.016 20.048 20.008
and cooperation (0.111) (0.106) (0.159) (0.199)
Local governance 0.144 0.148 20.030 20.033
(0.102) (0.101) (0.199) (0.211)
Household and commu- Yes Yes Yes Yes Yes Yes Yes Yes
nity controls
included?
Constant 210.578** 210.563* 28.171 27.937 210.196** 29.983* 211.035 210.670
(5.125) (5.315) (5.918) (6.239) (4.940) (5.674) (6.725) (8.215)
Adjusted R2 0.438 0.438 0.439 0.436 0.438 0.438 0.438 0.435
Observations 429 429 429 429 429 429 429 429
Specification tests
Tests for weak instruments
Stock and Yogo 167.653 57.955 48.354 31.247
(minimum
eigenvalue)
Endogeneity tests
Robust DWH 0.429 0.668 0.324 0.709
Determinants of Rural Income

(p-value)
Over-identifying restrictions test
Sargan’s test 0.995 0.976 0.958
(p-value)
Score’s test 0.994 0.974 0.963
(p-value)
Robust standard errors are given in parentheses, clustered at the community level. Excluded instruments for the IV regression are distance to a good
road, distance to the nearest market and the proportion of adults Catholics in the community.
***p , 0.01, **p , 0.05, *p , 0.10.
|
29
Obviously, we need to interpret these regression results with proper

caution. First, the vegetation index, landscape variables, malaria risk and
distance to the nearest town appear as rather robust explanatory variables
for income: the results for the other infrastructural and geophysical vari-
ables are less robust and vary across specifications. Secondly, we aware of
the remaining econometric issues, and admit that we unable to satisfac-
torily instrument for all institutional proxies. Some of our results are
plagued by weak instruments, and even if the instruments ‘pass’ Sargan’s
test, we appreciate that doubts about the exclusion restriction may

remain. Moreover, the lack of panel data is a serious shortcoming, and
despite the chosen IV approach there is of course a potential risk of spur-
ious correlations. The regressions cannot account for village-specific
shocks which may well be correlated with geography, and thus lead to an
unobserved heterogeneity bias in the results. Finally, spatial autocorrelation
may bias some of our estimates (even if the villages in our sample are not
located very close to each other).
Notwithstanding these important caveats, these results appear to conflict
with evidence provided in cross-country studies, typically identifying insti-
tutions as the main driver of long-term economic growth. They also appear
to be inconsistent with existing micro-level evidence by Grimm and Klasen
(2008; for Indonesia), which also implicates institutions (tenure security)
as a key determinant of income. In contrast, our findings match recent evi-
dence provided by Bhattacharyya (2009a) for the sample of African
countries.
But one should be careful with such cross-study comparisons. We focus
on within-Kenya differences in income. The determinants of such
differences may be different from factors explaining why Kenya is poor
relative to the USA or Singapore. For domestic policy-makers and inter-
national development agencies, it is an open question which matter is
most important—closing the gap between disadvantaged communities
in Kenya relative to more advanced ones (by addressing geographical impe-
diments, say, or building infrastructure), or bridging the gap between
Kenya as a whole and the industrialised countries (perhaps by addressing
institutional factors holding back overall development).
Our main result that local institutional variables do not explain income
differences within Kenya could be due to two reasons. First, perhaps in
early stages of development, geography simply matters more than insti-
tutions. In other words, the results may perhaps be interpreted as a
‘stage of development’ effect. At an early stage of development, geography
could be the most powerful force and, at a later stage of development,
institutional variables may become more important. This may be due to

the simple reason that agricultural production constitutes an important
part of total output in early stages of development (and agricultural pro-
duction is likely to be rather sensitive to geophysical conditions).
However, while it is evident that geographical conditions matters for pro-
ductivity in farming, it is not evident that geography matters more for agri-
cultural development than institutional factors (or sectoral or marketing
policies—refer to Bates (1981) or Kirsten et al. (2009) for examples of
work highlighting the importance of policies and institutions in agricul-
tural development).15

Secondly, there is a relatively little spatial variation for our institutional
variables. For example, land tenure (whether or not a household holds a
title deed to the land it cultivates—a household control variable in our
regressions but a key institutional proxy in other studies) never enters sig-
nificantly in our models. Arguably, this simply reflects that the great
majority of our respondents holds title deeds (75% of the farmers), and
the rest has rented land on a rather well-functioning rental market. Of
course, it does not imply tenure security is unimportant.
Similarly, our data suggest a limited variation in cooperation, political
participation or the quality of local government institutions. In this
light, it is not surprising that these variables fail to explain much of the
intra-Kenya income differences. In contrast, geographical conditions vary
widely. Our sample spreads across four agro-ecological zones, displaying
a significant variation in climatic and soil conditions. This is reflected in
the summary of the institutional and geographical variables (Table 1).
This leaves us to speculate that the geography-versus-institutions debate
could be a red herring. For countries like Kenya, domestic inequality and
region-specific poverty can be reduced by tackling the most significant
geographical challenges. But for Kenya as a whole to catch up with the
industrialised world, perhaps institutional reform is necessary. Both
elements should arguably be part of a balanced policy package.
15
We have also estimated a series of models in which we seek to explain agricultural income
by geographical and institutional variables (agricultural income is some 50% of total
income for our sample respondents). Our qualitative results go through: none of the
institutional variables enters significantly and several geographical variables do (albeit
somewhat less robust than before—landscape variables, the vegetation index and
malaria risk are significant in multiple specifications). Distance to the nearest town
does not matter when explaining agricultural income, suggesting that this must be a
key variable to explain non-farm income.
Supplementary data
Supplementary data are available at JAFECO online.
Funding
We thank the Netherlands Organization for Scientific Research (NWO),
the Netherlands Fellowship Programme (NFP), the African Economic
Research Consortium (AERC) and the International Livestock Research
Institute (ILRI) for financial support.

Acknowledgements
We thank three anonymous referees for their helpful comments and sug-
gestions. Remaining errors are our own.
References
Acemoglu, D., S. Johnson and J. A. Robinson (2001) ‘The Colonial Origins of
Comparative Development: An Empirical Investigation’, The American
Economic Review, 91(5): 1369 – 401.
Aguiler, E. L. and A. C. Pacek (2000) ‘Macroeconomic Conditions, Voter Turnout,
and the Working-class/Economically Disadvantaged Party Vote in Developing
Countries’, Comparative Political Studies, 33(8): 995 – 1017.
Alesina, A., R. Baqir and W. Easterly (1999) ‘Public Goods and Ethnic Divisions’,
Quarterly Journal of Economics, 119: 1243 – 84.
Bates, R. H. (1981) States and Markets in Tropical Africa: The Political Basis of
Agricultural Policy. Berkeley: University of California Press.
Bates, R. H. (2006) ‘Institutions and Development’, Journal of African Economies,
15(Suppl. 1): 10 –61.
Bates, R. H. (2008) When Things Fell Apart: State Failure in Late-Century Africa.
Cambridge: Cambridge University Press.
Bhattacharyya, S. (2009a) ‘Root Causes of African Underdevelopment’, Journal of
African Economies, 18(5): 745 – 80.
Bhattacharyya, S. (2009b) ‘Unbundled Institutions, Human Capital and Growth’,
Journal of Comparative Economics, 37(1): 106– 20.
Binswanger, H. and J. McIntire (1987) ‘Behavioral and Material Determinants of
Production Relations in Land-Abundant Tropical Agriculture’, Economic
Development and Cultural Change, 36: 73– 99.
Blattman, C. (2009) ‘From Violence to Voting: War and Political Participation in
Uganda’, American Political Science Review, 103: 231– 47.
Bockstette, V., A. Chandra and L. Putterman (2002) ‘States and Markets: The
Advantage of an Early Start’, Journal of Economic Growth, 7: 347– 69.
Bowles, S. and H. Gintis (2002) ‘Social Capital and Community Governance’, The
Economic Journal, 112(483): F419 – 36.
Brock, W. A. and S. N. Durlauf (2001) ‘What Have We Learned from a Decade of
Empirical Research on Growth? Growth Empirics and Reality’, The World Bank
Economic Review, 15(2): 229 – 72.
Brunnschweiler, C. N. and E. H. Bulte (2009) ‘Natural Resources and Violent
Conflict: Resource Abundance, Dependence, and the Onset of Civil Wars’,
Oxford Economic Papers, 61(4): 651 – 74.
Burke, W. J. and T. S. Jayne (2008) Spatial disadvantages or spatial poverty traps:
household evidence from rural Kenya, MSU International Development

Working Paper 93. East Lansing: Department of Agricultural, Food, and
Resource Economics and the Department of Economics, Michigan State
University.
Cameron, A. C. and P. K. Trivedi (2009) Microeconometrics Using Stata. Texas:
Stata Press.
CBS (2003) Geographic Dimensions of Well-being in Kenya. Where Are the Poor?
From Districts to Locations. Nairobi: Government of Kenya.
Congdon Fors, H. and O. Olsson (2007) ‘Endogenous Institutional Change after
Independence’, European Economic Review, 51(8): 1896 – 921.
Easterly, W. (2006) ‘Big Push or Déjà Vu: A Review of Jeffrey Sachs’s the End of
Poverty: Economic Possibilities for Our Time’, Journal of Economic Literature,
44: 96 – 105.
Easterly, W. and R. Levine (2003) ‘Tropics, Germs, and Crops: How Endowments
Influence Economic Development’, Journal of Monetary Economics, 50(1): 3– 39.
Engerman, S. L. and K. L. Sokoloff (2002) Factor endowments, inequality,
and paths of development among new world economies, NBER Working
Paper 9259. Cambridge: National Bureau of Economic Research.
Gallup, J. and J. Sachs (2001) ‘The Economic Burden of Malaria’, American Journal
of Tropical Medicine and Hygiene, 64(1, 2 S): 85 – 96.
Gallup, J. L., J. D. Sachs and A. Mellinger (1999) Geography and economic devel-
opment, CID Working Paper 1. Cambridge: Center for International
Development, Harvard University.
Glaeser, E. L., R. La Porta, F. Lopez-De-Silanes and A. Schleifer (2004) ‘Do
Institutions Cause Growth?’ Journal of Economic Growth, 9: 271 – 303.
Greenberg, A. (2000) ‘The Church and the Revitalization of Politics and
Community’, Political Science Quarterly, 115(3): 377– 94.
Grimm, M. and S. Klasen (2008) Geography vs institutions at the village level, IZA
Discussion Paper 3391. Bonn: Institute for the Study of Labor.
Hall, R. E. and C. I. Jones (1999) ‘Why Do Some Countries Produce so much
more Output per Worker than Others?’ Quarterly Journal of Economics,
114(1): 83– 116.
Kirsten, J., A. Doward, C. Poulton and N. Vink (eds) (2009) Institutional
Perspectives on African Agricultural Development. Washington, D.C.: IFPRI.
Knack, S. and P. Keefer (1997) ‘Does Social Capital Have an Economic Payoff? A
Cross-Country Investigation’, The Quarterly Journal of Economics, 112(4):
1251 – 88.
KNBS (2007) Basic Report on well-being in Kenya. Based on the Kenya Integrated
Household Budget Survey 2005/06. Nairobi: Government of Kenya.
Kristjanson, P., M. Radeny, I. Baltenweck, J. Ogutu and A. Notenbaert (2005)
‘Livelihood mapping and poverty correlates at a meso-level in Kenya’, Food
Policy, 30(5 – 6): 568– 83.
Noor, A. M., P. W. Gething, V. A. Alegana, A. P. Patil, S. I. Hay, E. Muchiri, E. Juma
and R. W. Snow (2009) ‘The risks of malaria infection in Kenya in 2009’, BMC

Infectious Diseases, 9 (180): 1471 – 2334.
Nunn, N. (2008) ‘The long term effects of Africa’s slave trades’, Quarterly Journal of
Economics, 123: 139 – 76.
Okwi, P. O., G. Ndeng’e, P. Kristjanson, M. Arunga, A. Notenbaert, A. Molo, N.
Henninger, T. Benson, P. Kariuki and J. Owuor (2007) ‘Spatial determinants
of poverty in rural Kenya’, Proceedings of the National Academy of Sciences of
the United States of America, 104(43): 16769 – 74.
Pacek, A. and B. Radcliff (1995) ‘The Political Economy of Competitive Elections
in the Developing World’, American Journal of Political Science, 39(3): 745– 59.
Pande, R. and C. Udry (2005) Institutions and Development: AView from Below,
Center Discussion Paper 928. New Haven: Economic Growth Center, Yale
University.
Radcliff, B. (1992) ‘The Welfare State, Turnout, and the Economy: A Comparative
Analysis’, The American Political Science Review, 86(2): 444– 54.
Rodrik, D. (2003) ‘Institutions, integration and geography: in search of the deep
determinants of economic growth’, in D. Rodrik (ed.) In Search of Prosperity:
Analytical Country Studies on Growth. Princeton, NJ: Princeton University Press.
Rodrik, D. (2006) ‘Goodbye Washington Consensus, Hello Washington
Confusion’, Journal of Economic Literature, 44: 973– 87.
Rodrik, D., A. Subramanian and F. Trebbi, (2004) ‘Institutions Rule: The Primacy
of Institutions Over Geography and Integration in Economic Development’,
Journal of Economic Growth, 9(2): 131– 65.
Rowley, C. K. (2000) ‘Political Culture and Economic Performance in Sub-Saharan
Africa’, European Journal of Political Economy, 16(1): 133– 58.
Sachs, J. D. (2003) Institutions don’t rule: direct effects of geography on per capita
income. NBER Working Paper 9490. Cambridge: National Bureau of Economic
Research.
Sachs, J. and P. Malaney (2002) ‘The Economic and Social Burden of Malaria’,
Nature, 415(6872): 680– 5.
Sala-i-Martin, X. and A. Subramanian (2003) Addressing the natural resource
curse: an illustration from Nigeria. IMF Working Paper WP/03. Washington,
D.C.: International Monetary Fund.
Stock, J. H. and M. Yogo (2005) ‘Testing for weak instruments in linear IV

regressions’, in D. W. K. Andrews and J. H. Stock (eds) Identification and
Inference for Econometric Models: Essays in Honor of Thomas Rothenberg.
Cambridge: Cambridge University Press, pp. 80 – 108.
Suri, T., D. Tschirley, C. Irungu, R. Gitau and D. Kariuki (2009) Rural incomes,
inequality and poverty dynamics in Kenya. Tegemeo Working Paper 30.
Nairobi: Tegemeo Institute of Agricultural Policy and Development, Egerton
University.
Tu, Q. and E. H. Bulte (2011) ‘Religiosity and Economic Performance:
Micoeconometric Evidence from Rural Tibet’, China Economic Review, 22:

55 – 63.
Verbeek, M. (2004) A Guide to Modern Econometrics, 2nd edn. Chichester, UK:
John Wiley & Sons.
Welch, M. R., D. Sikkink, E. Sartain and C. Bond (2004) ‘Trust in God and Trust in
Man: The Ambivalent Role of Religion in Shaping Dimensions of Social Trust’,
Journal for the Scientific Study of Religion, 43(3): 317– 43.

Déterminants Des Revenus Ruraux Au Kenya

Uploaded by

Document Information

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Déterminants Des Revenus Ruraux Au Kenya

Uploaded by

Copyright:

Journal of African Economies Advance Access published October 21, 2011

Journal of African Economies, Vol. 0, number 0, pp. 1 –35

Determinants of Rural Income: The Role

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

* Corresponding author: Erwin Bulte. E-mail: erwin.bulte@wur.nl

the main impediment to economic growth, then ‘big push, top-down’

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

To attenuate concerns about the endogeneity of our institutional variables,

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

2. Geography versus institutions: Theory and evidence

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

who support them have deliberately crafted governance systems to promote

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

3. Geography and institutions in Kenya

institutions affect income levels among rural households across different

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

and poverty across regions, communities and households in Kenya. They

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

4. Data and empirical strategy

Table 1: Summary of the Data

Variable type Description Mean SD

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

(continued on next page)

Variable type Description Mean SD

Education of household Household head has secondary and 0.26

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

(continued on next page)

Variable type Description Mean SD

Cooperation Dummy variable, taking a value of 1 if a 0.83

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

important aspects of household livelihoods. First, we collected a compre-

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

character of political participation resulting in improved government

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

politicians, police officers and judicial officers, whose capabilities and

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Hence we do not include them explicitly in most analysis (but including

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Table 2: Geographical Variables and Household Income in Kenya—OLS Results

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

(continued on next page)

Community control variables

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

a major town, distance to a good road, landscape attributes (our soil

In Table 3, we explore the correlations between incomes and our

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Table 3: Institutions and Household Income in Kenya—OLS Results

Institution variables (community level)

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Table 4: Geography, Institutions and Income at Household Level—OLS Evidence

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

(continued on next page)

Subjective assess- 0.360* 0.827**

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

religion promotes certain behaviours (e.g., savings, hard work) or enhances

Table 5: Geography, Institutions and Household Income—IV Evidence

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

(continued on next page)

Tests for weak instruments

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

One important problem of IV models is that of weak instruments. If

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011

Finally, and as a robustness analysis, we reduce the number of insti-

Downloaded from http://jae.oxfordjournals.org/ at AGORA Senegal on November 4, 2011