How Does Population Density
Affect Agricultural Intensification
and Household Well-being in
Africa?
Insights from 5 countries
Rui Benfica, Jordan Chamberlin, Derek Headey,
Thom Jayne, Anna Josephson, David Mather,
Milu Muyanga, Jacob Ricker-Gilbert
Why does Population Density Matter?
• Population in SSA expected to grow by 500
million in next 20 years.
• Many rural households live in areas of relatively
high population density.
• large tracts of un-used land in many countries.
• What does this mean for agriculture and food
Zambia
Kenya
security?
2
Descriptive evidence reveals that larger farms have higher
incomes.
Log(Per Capita Income)
6.2
ETHIOPIA
9.8
KENYA
4.4
RWANDA
4.2
9.4
5.8
4.0
5.4
0
4.0
.25
.5
Ha
.75
1
MOZAMBIQUE
9.0
0
3.8
.25
.5
Ha
.75
1
3.8
0
.25
.5
Ha
.75
ZAMBIA
3.6
3.5
3.4
3.0
0
.75 1
.25 .5
Ha
3.2
0
.25 .5
.75 1
Ha
Per Capita Land Access (Ha)
Relationships between farm size and
household income
3
1
Boserupian theory predicts that population
density drives intensification
Research questions:
1) Through what channels does population density
drive intensification?
2) What are the constraints to smallholder
intensification in areas of high population
density?
3) Is there a population density threshold beyond
which farmers are no longer able to intensify
production?
4
Pathways Through which Population Density
Affects Household Outcomes
observed
unobserved
observed
observed
INDIRECT EFFECTS
Land prices
Non-market
institutions
Landholding
Output
prices
Population
density
Wage
rates
Input
demand
Output
supply
Income
Information flow,
Institution development,
Transaction costs
DIRECT EFFECTS
5
Household Intensification Under
Population Density
Yit = α1Dt + α2D2t + Ptβ + ρLit + Xit𝜕 + ci + vit
Y: Outcome measure of interest
D: population density and its square.
P: factor and output prices
L: landholding
X: other household, and community factors
c: unobserved time-constant effects
v: unobserved time-varying effects
�𝟏 , α
�𝟐 = 0, tests the direct effect of pop den on Y
H0 : α
6
Indirect Partial Effect + Total Partial Effect
Landholding (farm size)
Lit = η1Dt + η2D2t + Mit𝜕 + μit
Prices
Pit = ζ1Dt + ζ2D2t + Zit𝜕 + εit
Calculating Total Effect (combine D and D2)
Yit = αDt + (Dt)Pitβ + ρLit(Dt) + Xit𝜕 + ci + vit
Total Partial Effect
𝜕Yit
𝜕Yit
𝜕Yit dPit
𝜕Yit dLit
=
+
∗
+
∗
𝜕Dt
𝜕Dt
𝜕Pit dDt
𝜕Lit dDt
=
INDIRECT EFFECT
or DIRECT EFFECT +
𝜕Yit
= α� + β� (ζ1+ 2ζ2Dt) + ρ� (η1+ 2η2Dt)
𝜕Dt
TOTAL EFFECT
7
Is population density endogenous?
• Maybe
– Due to omitted variables
– Or due to reverse causality
• correlation between covariates and unobserved
heterogeneity ci controlled for using the correlated
random effects (CRE) estimator.
ci = Ψ + 𝑋�𝑖 δ + ai; where ai = (o, σ2)
• Some countries used IV methods with control
function approach.
8
Estimation Procedure
• All models estimated linearly.
• Pooled CRE
• Estimate via Seemingly Unrelated Regression
(SUR).
– Models are linked through their error terms, so allows
us to directly compute total partial effects.
– Efficiency gain over equation-by-equation.
9
Case Studies
Ethiopia
Kenya
Zambia
Malawi
Mozambique
“Land abundance”
(high land/labor ratios)
Zambia Northern Mozambique
“Land constrained”
(low land/labor ratios)
Southern Malawi
1010
Kenya
Gridded Population Data
• High-resolution gridded estimates of rural
population distributions
– GRUMP (Global Rural-Urban Mapping Project, Balk
and Yetman 2004)
– AfriPop project (Linard et al. 2012)
• Significant improvements over earlier databases
– input statistical data are at fairly high levels of
disaggregation
– reporting units further disaggregated spatially
11
12
13
14
INDIVIDUAL COUNTRY CASE
STUDIES
15