Measurement, Farm in land quality attributes, or from systematic errors in land measurement. The land quality Size and Productivity argument was debunked in a 2010 article by Chris Barrett using laboratory soil testing to By Calogero Carletto, Sara Savastano control for land quality attributes. This paper and Alberto Zezza revisits the land measurement aspect of the controversy by working with data from the 2005 Agricultural economists have long debated the Uganda National Household Survey to compare efficiency and viability of smallholder agriculture. farmers’ own estimates to land measurements While much research has shown that small taken using GPS devices. For the IR to be partially farmers in developing regions are often more or fully explained by errors in land measurement, efficient than larger farmers, some have smaller farmers would have to systematically challenged the validity of that evidence, citing over-report land area with respect to larger potential problems that come with farmers' self- farmers. reporting of land size. If smaller farmers systematically under-report the size of their 5.00 plots, one would observe higher levels of GPS-Self reported farm size production per unit of land that are not linked 4.00 GPS-Self-reported with efficiency in the production process, but to a farm size (acres) 3.00 failure to properly account for the quantity of (acres) land they actually use. 2.00 1.00 In this paper, measurements of land size collected via Global Positioning System (GPS) 0.00 devices are used alongside farmers’ estimates to -1.00 test the validity of that critique. The main result of the analysis is that more accurate measurement of farmers’ plots if anything reinforces, rather than weakening, the existing Deciles of farm size (GPS) evidence of an inverse farm size-productivity relationship. However, contrary to the expectations implicit in the ‘measurement error’ criticism of the inverse Fact or Artifact? farm-size productivity relationship, we find that small farmers tend to systematically under-report Starting with the seminal work of Sen in the the size of their plots, and it is only among the 1960’s, who observed an inverse relationship (IR) top three landholding deciles that farmers tend between farm size and output per hectare in to over-report farm size. This is clearly shown Indian agriculture, a large number of empirical above, where the difference between the GPS studies have presented evidence that appears to measure and the farmers’ self-reported farm size corroborate that hypothesis. A smaller set of is plotted against 10 deciles of farm size, from studies has challenged the validity of that the smallest to the largest. evidence, however, claiming that the observed IR is in fact a mere statistical artifact stemming from One additional issue in our data is the the failure to control for unobserved differences considerable tendency of respondents (or Living Standards Measurement Study Brief Series www.worldbank.org/lsms-isa enumerators) to round their reported plot size to areas. Farms are categorized as small, medium or the nearest acre or half acre. This ‘heaping’ in the large. Small farms, those cultivating landholdings response pattern is not uncommon but it may be smaller than 1.45 acres, exhibit systematically particularly important in the case of land higher yields when area cultivated is measured measurement, since it is bound to matter via GPS as compared to self-reporting. The proportionally more to the left of the difference is reduced for medium farms, whereas distribution, as the same amount of rounding large farms have lower yields measured with GPS represents a larger percentage of the actual size. than those obtained through farmers' estimates. To go beyond these simple, yet telling, descriptives we estimated two versions of a standard model used to estimate the farm-size productivity relationship, one using GPS and the other one with the self-reported land measure. Both estimates supported the IR hypothesis. When more accurate land measures are used thanks to the introduction of GPS devices, the estimated slope of the function becomes steeper, indicating an even stronger IR than what one would conclude based on similar estimates performed using farmers' self-reporting. Key Messages The figure above also shows how the means are The hypothesis according to which the IR would not very different, but at specific points the be a statistical artifact due to small farmers distributions deviate considerably, in a way that under-reporting their farm size is strongly appears to be driven by heaping in the self- rejected by the data. In our sample, small farmers reporting distribution as opposed to a smooth in fact over-report land size, and it is the large curve for the GPS measure. Finally, the famers who are actually more likely to comparison of the two distributions lends underestimate their holdings, which results in support to the case for treating the GPS measure artificially higher yields. as the more accurate of the two. This has clear practical implications for How Does Using GPS Affect Yield policymakers, as it suggests that: (1) policies that Estimates? enable the small farm sector to realize its full potential may be justified by efficiency, as well as The systematic patterns in the difference equity considerations; and (2) it is unlikely that between land measurements we have observed the small-farm sector will rapidly disappear above have the potential to introduce a bias in because of the inefficiency claims alleged by the estimation of agricultural/land productivity. If some analysts. The study also shows that the GPS small farmers report to be cultivating more land technology clearly holds promise for improving than they actually are, their ‘true’ yields are the accuracy in the collection of land size actually even larger than what one would measures in the context of large household compute using self-reported land quantities. surveys. Output Per Acre and Farm Size This brief is based on: Carletto, Calogero, Sara Yields Bias Savastano and Alberto Zezza (2013). Fact or (GPS minus artifact: The impact of measurement errors on the GPS Self-reported self-report) farm size–productivity relationship. Journal of US$/acre US$/acre % Development Economics, Vol. 103, pp. 254–261. Small 236 170 28% Medium 208 193 7% For more information, please visit: Large 77 100 -30% www.worldbank.org/lsms-isa The above table summarizes level of output per Or contact us at lsms@worldbank.org acre computed using GPS and self-reported land Living Standards Measurement Study Brief Series www.worldbank.org/lsms-isa