The Agricultural Productivity Gap
in Developing Countries
Douglas Gollin
Williams College
David Lagakos
Arizona State University
Michael Waugh
New York University
World Bank ABCDE Conference,
Paris
June 1, 2011
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
1 / 16
Agricultural Productivity Gaps
In most sub-Saharan countries, agriculture is a large sector in terms of
employment and output.
But typically agriculture's share of employment is much higher than its
share of GDP.
Is this evidence of allocative ineciency?
McMillan and Rodrik (2011): report potential gains from moving
workers out of agriculture and into other sectors.
Potentially important policy implications.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
4 / 16
Some Questions about Productivity Gaps
This paper asks what we actually know about these productivity gaps.
I
I
I
I
Are the gaps measured accurately?
How could large gaps persist?
Can labor market rigidities possibly explain gaps of the magnitude
found in the data?
What other explanations might be relevant?
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
5 / 16
Relevant Literature
Productivity gaps are widely noted in the literature as far back as
Kuznets and, of course, Lewis.
I
I
I
Gollin, Parente, and Rogerson (2004) documented productivity gaps for
a large set of countries.
Caselli (2005) argued that cross-country income dierences must be
closely related to dierences in agricultural productivity.
Related work by Restuccia, Yang, Zhu (2008), Vollrath (2009).
Micro data systematically nds evidence of lower living standards in
rural areas.
I
Ravallion, Chen, and Sangraula (2007)
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
6 / 16
A First Look at the Data
What do the data show?
I
I
Agriculture's share of employment high.
Share of value added lower than share of employment.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
8 / 16
Agriculture Shares of GDP and Workforce
60
MYA
LAO
MWI
TZA ETH
SLE CAF
40
BDI
MLI
UZB
KGZ
NGA
GUY
TON KHM
SYR
BTN
SWZ
PRY
MNGPAK
FJI
ARM
BGD
NIC
STP
DMA
MDA
MAR
SRB
BOLCHN
SEN
PHL
BLR BLZ
EGY
IDN
TUN
GTM HND
CPV ROM
LKA
MKD
SLV
THA
MNE
URY
IRN
DZA
NAM
VCG
MHL TUR
COL
CRI
MYS
UKR
IRQ
BGR
GAB
GRD
KAZ AZE
DOM
ECU
PER
ARG
PAN
JAM
SUR
MDV
MUS
BRA CUB
LCA
RUS
LBN
VEN
POL
LTU
BWA
CHL
MEX
JORZAF
0
20
VA share of Agriculture
80
100
Figure: Agriculture shares in developing countries
0
20
40
RWA
NPL
SDNLBRPNG
BFA
TCD
KEN
IND
MDG
UGA
GMB
CIV
ZMBGIN
CMR WSM
ALB
VNM
TJK
ZWE
YEM
LSO
GEO
GHA
60
80
100
Empoyment Share of Agriculture
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
9 / 16
Implications of the Raw Data
Taken at face value, these data imply that value added per worker
(
VA )
L
is lower in agriculture than in the non-agricultural sector.
The implied productivity gaps are large.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
10 / 16
Agricultural Productivity Gap
We dene the Agricultural Productivity Gap (APG) to be
APG
n /Ln
≡ VA
VAa /La
Simple two-sector model says APG should be 1.
Typical developing country has APG of 4.
Some (particularly in sub-Saharan Africa) have APGs of 8 or more!
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
11 / 16
Simple Two-Sector Model
◮
Technologies
Ya = Aa Lθa Ka1−θ
and
Yn = An Lθn Kn1−θ
◮
Competitive labor markets
◮
Households can supply labor to either sector
◮
Then, in equilbrium:
APG ≡
VAn /Ln
Yn /Ln
=
= 1.
VAa /La
pa Ya /La
”Raw” Agricultural Productivity Gaps
Measure
Weighted
Unweighted
5th Percentile
1.7
1.1
Median
3.7
3.0
Mean
4.2
3.6
95th Percentile
5.4
8.8
Number of Countries
112
112
Sub-Saharan Productivity Gaps
In general, the APGs are largest for the poorest countries.
They are even larger, however, for sub-Saharan countries.
A number of African countries have APGs greater than 10.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
2/5
”Raw” Agricultural Productivity Gaps
6
4
0
2
Number of Countries
6
4
2
0
Number of Countries
8
Asia
8
Africa
4
8
12
16
0
4
8
APG
APG
Americas
Europe
12
16
12
16
6
4
0
2
Number of Countries
8
6
4
2
0
Number of Countries
8
0
0
4
8
APG
12
16
0
4
8
APG
Possible Implications of Large APGs
APGs appear to explain much of cross-country disparities in GDP
per worker.
Appear to suggest massive misallocation.
Policy debate: encourage movement out of agriculture?
Our view: gaps may reect measurement issues; need to ask more
questions.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
12 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
What Do Agricultural Productivity Gaps Reect?
Sector dierences in hours worked per worker?
Construct measures of hours worked by sector for 56 countries
Sector dierences in human capital per worker?
Construct measures of human capital by sector for 127 countries
Urban-rural dierences in cost of living?
Use cost-of-living data for 87 countries from World Bank
Measurement error in national accounts data?
Use household income/expenditure surveys from 20+ countries
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
13 / 16
Preview of Results
After adjustments, APG in average developing country reduced from 4
to 2.
Household survey data also suggest big sector income gaps.
Puzzling that gaps are still so large.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
14 / 16
”Simple” Measurement Questions
◮
National accounts exclude production of agriculture for own consumption?
◮
Rural people counted as agricultural workers?
”Simple” Measurement Questions
◮
National accounts exclude production of agriculture for own consumption?
No, it is included.
◮
Rural people counted as agricultural workers?
”Simple” Measurement Questions
◮
National accounts exclude production of agriculture for own consumption?
◮
Rural people counted as agricultural workers?
No, only ”economically active” persons should be included.
”Simple” Measurement Questions
◮
National accounts exclude production of agriculture for own consumption?
◮
Rural people counted as agricultural workers?
In practice, official estimates similar to our surveys.
100
80
TZA
UGA
GIN
VNM
60
NPL
KEN
GHA
40
SEN
ARM
ROM
20
ECU
PER
ZAF
BOL
CUB
COL
CHL
JOR VEN
BLR
0
Employment share of Agriculture (Our Estimates)
Agriculture Sector in Developing Countries
0
20
40
60
80
Empoyment Share of Agriculture (Official Data)
100
Sector Differences in Hours Worked
◮
Average hours worked per worker might differ across sectors
◮
We construct average hours worked per worker by sector for 56 countries
- Population census micro data or labor force surveys
- All employed or unemployed persons 15+ years old
- Industry of primary employment (employed); industry of previous
employment or rural/urban status (unemployed)
- Hours worked in reference period (usually one week)
2.0
1.0
TUR
KEN FJI
MYS SWZ
PHL
CRI
RWA
IDN
LBR MEX
SYR PER
KHM ZAF
BOLSLE
PAN
CHL
BGD
BWA
PAK
DOM VEN
BTN
NPLLKA
CHN
TONLCA ECU
ROM NGA
BRA
TZA
50
40
1.5
ARM
JOR
LSO
ZMB
IRQ JAM
GHA
ETH
30
UGA
20
Hours Worked in Non−Agriculture
60
Sector Differences in Hours Worked
20
30
40
Hours Worked in Agriculture
50
60
Sector Differences in Hours Worked: Summary
◮
Explains on average a factor 1.2
◮
Only a few countries above 1.5
◮
Unlikely to be the main cause of APGs in developing countries
Sector Differences in Human Capital
◮
Relevant for U.S. development experience (Caselli & Coleman, 2001);
developing countries today (Vollrath, 2009)
◮
We construct human capital per worker by sector for 127 countries
- Same sources, sample selection as for hours
- Years of schooling measured directly when available
- Impute years of schooling using educational attainment otherwise
- Baseline: assume 10% rate of return on year of schooling
(Psacharoplous & Patrinos 2002; Banerjee & Duflo, 2005)
2.0
1.5
1.0
UKR
GEOARM
KGZ
KAZ
ROMMDA
UZB
CUB BLR
AZETON
CHL SRBGUY
JOR TJK
PAN
MHL
LTU
MKD
ALB
PERARG
JAM
ZAF
COL
PHLZWE
ECU
SWZ BLZ FJI
BOL
CHN
NAM
CRI
MEX
IDN
NGA
ZMB
DOM
VNM
VENSLV
THA
EGY
SUR
MDV
LSO
UGA
PRY
HND
MYS
PNG
SYR
MWI
CMR
BRA
YEM
GHA
IND
MDG
GTM
BWA
TZATUR
IRQ
NIC
GAB
LAO
LBR
PAKLCA
ETH
BDI RWA
KEN
IRN
BGD
KHMBTN
SLE
STP
CAF
MAR
GMB
NPL
CIV
SDN
BFA
TCD
SEN
GIN
5
10
MNG
MLI
0
Years of Schooling in Non−Agriculture
15
Sector Differences in Years of Schooling
0
5
10
Years of Schooling in Agriculture
15
10
2.0
5
SWZ
ZWE
NAM
NGA
ZMB
EGY
UGA LSO
MWI
CMR
GHA
MDG
BWA
TZA
GAB
LBR
ETH
BDI RWA KEN
SLE
STP
CAF
MAR
GMB
CIV
SDN
BFA
TCD
SEN
GIN
1.5
1.0
ZAF
MLI
0
Years of Schooling in Non−Agriculture
15
Sector Differences in Years of Schooling: Just Africa
0
5
10
Years of Schooling in Agriculture
15
2.0
1.5
ROM MDA
1.0
UKRGEO ARM
KGZ
KAZ
MNG
UZB
BLR
CUB
SRBGUYAZE TON
TJK
JOR
PAN
LTU
MHL
MKD
ALB
PER
JAM
ARG
ZAF
COL
PHLZWE
ECU
SWZ BLZ FJI
BOL
CHN
NAM
CRI
MEX
IDN
NGA
DOM
ZMB
VNM
VEN
SLV
THA
EGY
MDV
LSO
SUR
UGA
HND
PRY
MYS
PNG
MWI
SYR
CMR
BRA
YEM
GHA
IND
BWA
G
TM
MDG
TZA
IRQ
TUR
GAB
NIC
LAO
LCA
ETH LBR
PAK
BTN
BDIRWA
KEN
BGDSTP
IRN
KHM
CAF
SLE
MAR
GMB
NPL
SDN
BFA CIV
TCD
SEN
GIN
MLI
2
3
CHL
1
Human Capital in Non−Agriculture
4
Sector Differences in Human Capital
1
2
3
Human Capital in Agriculture
4
Quality Differences in Schooling
◮
Rural schools often of lower quality than urban schools (Williams, 2005;
Zhang, 2006)
◮
Potentially overestimate human capital among agriculture workers
◮
We use literacy data to adjust for schooling quality
Uganda: Literacy by Years of Schooling Completed
("
Literacy Rate
!#'"
!#&"
!#%"
!#$"
!"
!"
$"
%"
&"
'"
Years of Schooling
)*)+,-".*/01/2"
,-".*/01/2"
(!"
Measuring Quality Differences in Schooling
◮
Given literacy rates by years of schooling: ℓni (s) and ℓai (s) for s = 1, 2, ...
◮
Assume that one year in rural school is worth γ years in urban school
◮
For each country i, solve for γi that solves
min
γ
¯
s “
X
ℓ˜ni (γs) − ℓ˜ai (s)
”2
s=1
where ℓ˜ni (·), ℓ˜ai (·) are polynomial interpolations of ℓni (·), ℓai (·) for s ∈ [0, ¯s ].
Measuring Quality Differences in Schooling
Table 3: Rural-Urban Education Quality Differences
Country
γˆ
Argentina
0.87
Bolivia
0.95
Brazil
0.89
Chile
0.92
Ghana
0.90
Guinea
0.62
Malaysia
0.93
Mali
0.89
Mexico
0.77
Panama
0.87
Philippines
0.80
Rwanda
0.88
Tanzania
1.25
Thailand
0.90
Uganda
0.82
Venezuela
0.78
Vietnam
0.74
Average
0.87
1.80
PAN
1.60
MEX
VEN
PHL
BOL
UGA
GHA
BRA
ARGCHL
MYS
VNM
1.40
RWA
THA GIN
1.20
MLI
TZA
1.00
Human Capital Ratio, Quality Adjusted
2.00
Quality-Adjusted and Unadjusted Human Capital
1.00
1.20
1.40
1.60
Human Capital Ratio, Unadjusted
1.80
2.00
Sector Differences in Human Capital: Summary
◮
Explains on average a factor 1.4
◮
Range (10th-90th percentile) from 1.2 to 1.6
◮
Quality adjustments using literacy data don’t change results much
◮
Even assuming rural years of schooling are worth 1/2 as much as urban,
still get only a factor 1.6 on average, maximum of 2.1
15
10
5
0
Number of Countries
20
25
Cost-of-Living Differences
0.9
1.0
1.1
1.2
1.3
1.4
1.5
Urban/Rural Cost of Living
1.6
1.7
1.8
”Adjusted” Agricultural Productivity Gaps
Measure
Countries w/ Complete Data
All Countries
5th Percentile
1.1
0.8
Median
2.3
1.8
Mean
2.0
1.9
95th Percentile
2.6
2.6
Number of Countries
34
112
10
8
6
4
RWA
ZMB
JAM
2
LSO
IDN
TUR CHN UGA
PAK
ROM
BRA
BGD
ARM
MEX
ZAF
TZA
CHL
DOMNPL
BWA
ETH
KEN
BOL
NGA
VEN
PAN PHL
GHA
PER
KHM
CRI
ECU
SWZ
JOR
0
Adjusted APG
12
14
16
Raw vs Adjusted Gaps
0
2
4
6
8
Raw APG
10
12
14
16
4
3
JAM
LSO
1
2
IDN
CHN
TUR
PAK
ROMBRA
BGD
ARM
MEX
ZAF
TZA
CHL
DOM
BWA
ETH
KEN
BOL
NPL PHL
NGA
VEN
GHA PAN
PER
KHM
CRI
ECU
SWZ
JOR
UGA
0
Adjusted APG
5
6
7
Raw vs Adjusted Gaps
0
1
2
3
4
Raw APG
5
6
7
Raw vs Adjusted Gaps
4
8
12
30
20
16
0
4
8
12
Raw APG
Complete Data, Adjusted
All Countries, Adjusted
20
0
10
Number of Countries
12
8
4
0
16
30
Raw APG
16
0
Number of Countries
10
0
4
8
12
Number of Countries
16
All Countries, Raw
0
Number of Countries
Complete Data, Raw
0
4
8
Adjusted APG
12
16
0
4
8
Adjusted APG
12
16
Comparing National Accounts & Household Surveys
◮
National accounts might underestimate agriculture value added/income
◮
Can compare with household income/expenditure survey evidence
◮
Use World Bank’s Living Standards Measurement Surveys (LSMS)
◮
Explicit goal of LSMS: household income and expenditure measures
Comparing Macro to Micro Sector Income Measures
◮
Agricultural value added, household i
VAa,i = ya,i − INTa,i
◮
”Agricultural revenue”
ya,i =
J
X
“
”
pj xihome
+ ximarket
+ xiinvest
,j
,j
,j
j=1
◮
Non-agriculture value added
VAn,i = yn,i − INTn,i
Comparison of Macro and Micro APG
Agriculture Share (%) of
Employment
Country
APG
Value Added
Micro
Macro
Micro
Macro
Micro
Cote d’Ivoire (1988)
71.0
32.0
37.7
4.3
4.0
Guatemala (2000)
40.2
15.1
16.9
3.4
3.3
Pakistan (2001)
57.8
25.8
20.5
4.2
4.3
South Africa (1993)
11.8
4.3
8.2
1.6
1.5
Computing Income, Expenditure
◮
Deaton (1997), others: consumption expenditures more reliable than
income
◮
Robustness: compare VA/worker to Income/Worker and
Consumption/Worker
Sector Differences in VA, Income, and Expenditure from Micro Data
Country
VA/Worker
Income/Worker
Consumption/Worker
Cote d’Ivoire (1988)
4.0
3.7
3.5
Guatemala (2000)
3.3
3.3
2.6
Pakistan (2000)
4.3
5.3
1.9
South Africa (1993)
1.5
1.6
1.7
Different Labor Shares Across Sectors?
◮
Production functions with different labor shares
Ya = Aa Lθa a Ka1−θa
◮
and
Yn = An Lθnn Kn1−θn
In equilibrium
APG =
Yn /Ln
θa
=
pa Ya /La
θn
Different Labor Shares Across Sectors?
◮
Production functions with different labor shares
Ya = Aa Lθa a Ka1−θa
◮
Yn = An Lθnn Kn1−θn
In equilibrium
APG =
◮
and
Yn /Ln
θa
=
pa Ya /La
θn
Macro evidence on θa , θn
- Employment share of agriculture varies a lot across countries;
- Aggregate labor share of GDP doesn’t (Gollin, 2002)
- Suggests θa isn’t much higher than θn
Different Labor Shares Across Sectors?
◮
Production functions with different labor shares
Ya = Aa Lθa a Ka1−θa
◮
Yn = An Lθnn Kn1−θn
In equilibrium
APG =
◮
and
Yn /Ln
θa
=
pa Ya /La
θn
Macro evidence on θa , θn
- Employment share of agriculture varies a lot across countries;
- Aggregate labor share of GDP doesn’t (Gollin, 2002)
- Suggests θa isn’t much higher than θn
◮
Micro evidence on θa , θn
- Sharecropping arrangements suggest θa ∼ 0.5
- Econometric estimates: θa ∼ 0.5 − 0.7
Conclusions
Measurement problems seem to play a large role in explaining the
APGs.
Adjusting for simple and obvious observables reduces the APGs
substantially.
But large productivity gaps remain even after all our adjustments.
If these numbers are accurate, they suggest a major puzzle:
I
Why are people not moving out of rural areas as fast as they possibly
can?
I
What barriers or labor market rigidities could possibly keep urban
populations so small?
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
3/5
Possible Explanations?
Harris-Todaro models suggest that riskier income in urban areas might
keep people from moving; but only at entirely implausible levels of risk
aversion.
I
(And do we really think that rural livelihoods are less risky than urban
livelihoods?)
Are there some unmeasured amenities to rural life?
I
Should we be trying to measure the value of leisure in rural areas?
I
Is there a value to food security and self-reliance?
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
4/5
Directions for Further Research
Continue pursuing measurement issues.
Look for evidence of
measurable
rigidities in labor markets and
migration costs.
Consider plausibility of alternative models and explanations.
Gollin, Lagakos & Waugh ()
Agricultural Productivity Gaps
ABCDE 2011
5/5