The Agricultural Productivity Gap in Developing Countries

This study seeks to understand why, in most developing countries, measured value added per worker is so much lower in agriculture than in other sectors of the economy. Simple two-sector models predict that value added per worker should be equal in agriculture and “non-agriculture,” and yet in the average developing country, national accounts data show that value added per worker is on average 4 times higher in non-agriculture than agriculture. In numerous countries, particularly in Sub-Saharan Africa, this “agricultural productivity gap” (APG) is much larger, as high as a factor of 10. Taken at face value, this deviation between theory and data suggests that living standards are much lower for agricultural households, and that some economic, social, or political “barrier” is preventing people in developing countries from moving out of agriculture. The implication for policy is that identifying and removing these barriers could lead to enormous increases in well-being for residents of developing nations. On the other hand, differences in value added per worker at the sector level could reflect differences in worker characteristics, cost-of-living differences between agricultural and non-agricultural areas, or some other type of mis-measurement. Our study asks whether the large APGs implied by national accounts data are still present once we apply better measures of inputs and outputs by sector. In particular, we assess the role of sector differences in hours worked, human capital, capital-intensity of production, and differences in the cost of living in explaining the large measured APGs. We do so using a new database that we are building from micro censuses and surveys from 127 developing countries. Our data is based partly on census data assembled by the International Integrated Public Use Microdata Series (I-IPUMS), partly on a database of schooling attainment from the Education Policy and Data Center (EPDC), partly from the World Bank’s Living Standards Measurement Studies (LSMS), and partly from individual household surveys conducted independently by national statistical agencies. Preliminary results suggest that three factors — differences in human capital per worker, average hours worked per worker, and regional costs of living — account for roughly half the benchmark APGs for our sample. The average APG falls from about four to about two, and medians and other percentiles of the distribution fall as well. Thus, the APGs are reduced in large part when better measurement of labour input and the real value of output are taken into consideration. Still, a puzzlingly large gap is left unexplained, which, taken literally, implies that real living standards are roughly one half as high among agricultural workers than among non-agricultural workers. This motivates us to ask whether these gaps are present in other sources of data besides national-accounts data. In particular, we ask whether large APGs are found in household surveys in developing countries. In particular, we plan to use LSMS household income surveys to compute average income per worker by sector, and to compare these averages to those in the national accounts. The LSMS data provide an ideal setting to perform these measurements, as the studies explicitly attempt to provide detailed measures of income and consumption. For farm households, for example, the surveys measure the quantity of agriculture product produced, product by product, whether or not the products are produced for market sale or home consumption. They also collect data on prices for each agricultural product, allowing us to measure an overall value of output produced by each household. We plan to compute income per agriculture worker and income per non-agriculture worker using every LSMS survey for which such calculations are possible. It seems that this will be possible in at least 20 countries. Preliminary results from one such country, Cote d’Ivoire, suggest that the large APG implied by the national accounts data is largely confirmed by the micro data (the gaps in both data sources are just over 3.) Further work will continue to collect and analyse data to try to understand why real value added produced per agricultural worker is so much lower than by workers in non-agricultural sectors of the economy.

Outputs