Professional Documents
Culture Documents
http://dx.doi.org/10.1257/aer.104.5.165
Economists have long recognized that dwarf the (already large) productivity differ-
cross-country differences in aggregate labor ences in the aggregate.
productivity are enormous. Recently, Caselli In this paper we re-examine the cross-country
(2005) and Restuccia, Yang, and Zhu (2008), agricultural productivity data using new evi-
among others, have shown that these differences dence from disaggregate sources. We focus on
have a strong sectoral dimension. In particu- physical measures of labor productivity for the
lar, differences in agricultural labor productiv- worlds three staple grainsmaize, rice, and
ity are far larger than those of the aggregate. wheatwhich together account for roughly half
Caselli (2005), for example, reports that the of the calories consumed by the average individ-
ratio of labor productivity in the ninetieth and ual. Because productivity in these crops is easy
tenth percentiles of countries is a factor of 22 in to measure, we avoid the black-box nature of
the aggregate, and a factor of 45 in agriculture. output data from aggregate sources. Moreover,
Because developing countries have most of their because crop yields are observed at many lev-
workers in agriculture, understanding why pro- els, from individual production units to national
ductivity differences in agriculture are so large aggregates, we are able to cross-check macro-
is key to understanding world income inequality. level productivity statistics with micro-level
There are a number of reasons one may be estimates of productivity for these crops.
skeptical, however, about the agricultural pro- We find that cross-country differences in the
ductivity data underlying these conclusions. quantity of grain produced per worker are enor-
One may worry that agricultural output data may mous, and at least as large as those of the agricul-
be badly measured, particularly in the worlds tural sector as a whole. Moreover, we show that
poorest countries, where statistical agencies independent micro-level estimates of grain yields
often have limited resources (Jerven 2013). The correspond very closely to their counterparts in
international prices used to aggregate agricul- aggregate data. We conclude that the large dis-
tural goods to the country level may be better parities in agricultural labor productivity are real,
suited to richer countries than poorer countries, at least for staple grains, and are not merely an
since international prices are quantity-weighted. artifact of mismeasurement or poor data quality.
More generally, skepticism may arise simply
from the sheer magnitudes in agriculture, which I. Cross-Country Differences in Agricultural
Output per Worker
*Gollin: University of Oxford, Department of We begin by presenting the most recent avail-
International Development, Queen Elizabeth House, 3
Mansfield Road, Oxford OX1 3TB, United Kingdom (e-mail: able data on cross-country agricultural produc-
Douglas.Gollin@qeh.ox.ac.uk); Lagakos: University of tivity differences, derived from aggregate data
California-San Diego, University of Economics 0508, 9500 from the Food and Agriculture Organization
Gilman Drive, La Jolla, CA 92093, and NBER (e-mail: (FAO) for 2007. These data provide a measure
lagakos@ucsd.edu); Waugh: New York University, Kaufman of gross output per worker in agriculture, where
Management Center, 44 West Fourth St. 7-160, New York,
NY 10012-1126 (e-mail: mwaugh@stern.nyu.edu). We output is valued at international prices. These
thank Kevin Donovan, Diego Restuccia, Nora Trapani, and numbers do not correspond to the national
Chris Udry for helpful comments, and we thank Jonathan accounts concept of value added, as they do not
Greenland and Glenn Farley for excellent research assis- adjust for intermediate inputs used in agricul-
tance. All potential errors are our own.
Go to http://dx.doi.org/10.1257/aer.104.5.165 to visit tural production. The international prices used
the article page for additional materials and author disclo- here are only for outputs. We draw on data for the
sure statement(s). 80 countries for which data on gross agricultural
165
166 AEA PAPERS AND PROCEEDINGS MAY 2014
output per worker and GDP per capita are avail- Table 1Tons Produced per Hectare and Hectares per
able, and in which at least one hundred thousand Agricultural Worker
people are employed in agriculture.
Tons produced Hectares
The results are striking. Countries in the top per Hectare per worker
10 percent of the world income distribution
produce 50.1 times as much agricultural output Maize Rice Wheat
per agricultural worker as countries in the bot- Top 10 percent 9.2 8.1 4.9 44.6
tom 10 percent. Countries in the top quarter of Bottom 10 percent 2.0 2.9 2.0 1.4
the income distribution produce 29.9 times as Ratio of top to bottom 4.7 2.8 2.5 31.2
much agricultural output per worker as coun- 10 percent
tries in the bottom quarter. The United States
in particular produces more than 100 times Notes: Land is measured as hectares of arable land. Workers
are agricultural workers and are measured as the total num-
as much agricultural output per worker as the
sub-Saharan countries in the data (Ethiopia,
ber of economically active persons involved in agriculture.
KEN08
IND10
CIV08
in land per worker. Countries at the top of the KEN04
GHA08 MWI00
GHN06
AUS00
KEN02
TAN08
KEN00
income distribution have on average 44.6 hect- BEN08 NIG02
TOG08
1 MWI90
ares per worker, while the countries at the bottom MOZ08
between micro and macro data sources.6 We worker is indeed in the ballpark of one hectare
cannot claim that the selection of data points per worker in developing countries.8
here is statistically representative or that the
FAO yield data are therefore accurate for all
purposes or questions. What we can say, how- III.Conclusion
ever, is that we find essentially no disagreement A recent literature has claimed that
between the FAO yield data and these micro cross-country differences in agricultural labor
estimates of grain yields that we compiled. productivity are even larger than cross-country
What about land per worker? In richer coun- differences in aggregate productivity. In this
tries, there is less concern about the quality of paper we revisit the data underlying this claim.
data on land and total agricultural employment, We focus on the worlds three staple grains
due to the large and regular censuses of agricul- maize, rice and wheatfor which direct
ture and population. For the United States, for measures of physical productivity are readily
example, we found effectively the same land available. We conclude that productivity differ-
per worker when comparing census data from ences in grain are enormous according to both
the US Department of Agriculture (USDA) in macro and micro data. This serves as evidence
the year 2007 to the FAO data reported for the that the large productivity differences in the
United States in 2007. The USDA, which uses agricultural sector are real, and not an artifact
satellite imagery (at least in part) to measure of poor measurement.
crop land (Boryan etal. 2011) show 48 hect- So why is agricultural output per worker so
ares per worker versus 53 hectares per worker much lower in the developing world than in
in the FAO data. rich countries?9 One hypothesis is that policies
As a check of the FAO land-per-worker that distort farm size lead to a misallocation of
data for poor countries, we draw on the Living farmland to farm operators (Adamopoulos and
Standards Measurement Surveys (LSMS) Restuccia forthcoming). Another theory is that
which are regarded as one of the highest quality farm operators in poor countries avoid using
sources of micro data available for developing productivity-enhancing intermediates, such
countries. Furthermore, the LSMS are inde- as fertilizers, because doing so increases their
pendent surveys of households that are gener- consumption risk (Donovan 2013). A third
ally not used in the construction of national theory is that the agriculture sector in develop-
aggregate statistics.7 The LSMS studies show ing countries tends to employ the lowest-ability
very low levels of land per worker in develop- workers (Lagakos and Waugh 2013). Surely
ing countries, as in the FAO data. In Malawi in these are just some of the economic forces at
2010, for example, the LSMS data show that work. There is much room for future research
the average rural household had 0.9 hectares on this important issue.
and 2.3 workers in the labor force. This implies
that in Malawi there are on average 0.4 hectares
per worker in rural areas. Similar calculations
in Ethiopia (2011), Guatemala (2003), Nigeria
(2010), and Tanzania (2010) show 0.3, 1.9, 0.5,
and 0.3 hectares per worker in rural areas. In
short, these LSMS surveys show that land per 8
There are several caveats in order here. First, not all
rural workers are agricultural workers in these countries.
So the above calculations likely underestimate the amount
of land per agricultural worker. Second, LSMS data are the
6
The FAO data are intended to be nationally representa- product of household surveys, which means that they may
tive. Few of our micro studies aim at nationally representative miss large commercial farms. Still, the total acreage devoted
samples though, which may account for some of the discrep- to large commercial farms tends to be a small fraction of
ancies between the two sets of numbers. total farm land in most developing countries.
7 9
Some countries use LSMS data as one of many sources A distinct but related question is why, within the typical
that can provide a cross-check on national aggregate sta- developing country, the value of output produced per worker
tistics, but we are not aware of any countries in which the in agriculture is substantially lower than the value of output
LSMS data are the actual source of national agricultural sta- produced by non-agricultural workers, on average. This is
tistics. Thus, we are confident that the LSMS data provide an the focus of our related work (Gollin, Lagakos, and Waugh
independent measure. forthcoming), and of Herrendorf and Schoellman (2013).
VOL. 104 NO. 5 AGRICULTURAL PRODUCTIVITY DIFFERENCES ACROSS COUNTRIES 169
Suri, Tavneet. 2011. Selection and Comparative E. Giller. 2008. Yield Gaps, Nutrient Use Effi-
Advantage in Technology Adoption. Econo- ciencies and Response to Fertilisers by Maize
metrica 79 (1): 159209. across Heterogeneous Smallholder Farms in
Tittonell, Pablo, and Ken E. Giller. 2013. When Western Kenya. Plant and Soil 313 (12):
Yield Gaps are Poverty Traps: The Paradigm 1937.
of Ecological Intensification in African Small- Williams, John. 1998. Sample Costs to Produce
holder Agriculture. Field Crops Research 143: Rice: Sacramento Valley, Multiple Crop Rota-
7690. tion. Davis, CA: University of California
Tittonell, Pablo, B. Vanlauwe, M. Corbeels, and Ken Cooperative Extension.
Copyright of American Economic Review is the property of American Economic Association
and its content may not be copied or emailed to multiple sites or posted to a listserv without
the copyright holder's express written permission. However, users may print, download, or
email articles for individual use.
Copyright of American Economic Review is the property of American Economic Association
and its content may not be copied or emailed to multiple sites or posted to a listserv without
the copyright holder's express written permission. However, users may print, download, or
email articles for individual use.