Published on Development Impact

What Drives Technology Adoption in Agriculture? Disentangling regulation and rising wages in Brazil: Guest Post by C. Austin Davis

This page in:

This is the fifth in our series of posts by Ph.D. students on the job market this year
Something dramatic happened in Brazilian agriculture between 2007 and 2013: the previously-steady labor intensity of a major crop, sugarcane, fell by 70 percent (see Figure). This drop was the result of the rapid, widespread adoption of mechanical harvesting.  My job market paper, “Why Did Sugarcane Growers Suddenly Adopt Existing Technology,” studies how mechanization was achieved.

In stakeholder interviews, I heard two explanations.  The first is environmental regulation.  Sugarcane harvesting was the target of environmental regulation because manual harvesting is preceded by burning the field; at its peak, sugarcane growers were burning an area the size of New Jersey every year.  First passed in 2002, strengthened in 2007, and expanded in 2008, state-level restrictions on burning effectively mandated mechanization.
I evaluate the regulations in two complementary ways.  I take advantage of an area threshold that exempted growers with less than 150 hectares.  With confidential, establishment-level data from the 2006 Census of Agriculture, I use a regression discontinuity (RD) approach to compare the harvesting practices of regulated and unregulated growers on either side of the threshold.  Since the RD estimate is primarily informative about establishments near the threshold, I also test the regulations via a difference-in-differences (DiD) approach that captures the behavior of larger growers.  Here, I use county-level labor intensity as a measure of mechanization, constructed from confidential administrative data on all formal employment in Brazil.  I estimate how changes in the stringency of regulation affect changes in labor intensity.
Both approaches give the same results: regulation has no statistically significant effect on harvesting practices or labor intensity.  Even if the true effect of regulation lies at the extreme of the 95 percent confidence interval, regulation can account for only a quarter of the observed change in outcomes.
If regulation played a limited role, what drove mechanization?  Industry participants cited rising wages as a second potential explanation.  Referring back to the figure, manual harvesting became more expensive as real wages for harvest workers increased by over 50 percent between 1999 and 2013.  Brazil was enjoying a boom during this period, with economy-wide increases in labor demand.  Wages were rising across all sectors, especially for unskilled workers, and employment increased in every sector except agriculture.
But the timing does not obviously support wages as a driver of mechanization; wages increased steadily for many years before mechanization took off. I reconcile these dynamics in a simple, frictionless neoclassical model.  In the model, there is a threshold wage for each parcel of land, above which a profit-maximizing grower will harvest mechanically.  The thresholds may be different for each parcel, depending on land characteristics like steepness.  If wages are well below the threshold for the majority of parcels, wages may increase steadily with no change in harvesting technique.  Eventually, as wages rise, they will cross the thresholds of many parcels, causing widespread mechanization.  Thus, steady increases in wages can lead to the seemingly sudden adoption of mechanical harvesting.
To empirically assess the importance of wages, I use an instrumental variables strategy to estimate the elasticity of labor demand.  Sugarcane is grown in a fertile area alongside other major crops (maize, soy, coffee, and oranges). All five crops employ large numbers of demographically-similar workers. Fluctuations in the international market for these other crops, combined with local variation in the importance of each crop, provide plausibly exogenous shifts to the labor supply available to sugarcane growers.  I use these shifts to identify the wage elasticity of labor demand in sugarcane.
My elasticity estimates imply that wages alone are sufficient to explain mechanization.  The estimates of the elasticity of labor demand are large, in the range of -2 to -3.  I multiply the estimates by the observed rise in wages to predict how labor intensity would have declined due to wage changes alone.  The predicted decline is similar in magnitude to the observed decline in labor intensity, and the difference is not statistically significant.
In richer countries, fewer people work in an increasingly productive agricultural sector. This relationship holds across countries and within a country over time; its ubiquity makes it one of the fundamental facts of development.  But the nature of the relationship is uncertain.  Do improvements to agricultural productivity stimulate growth in other sectors?  Or is it the other way around?
My work disentangles the effects of regulation and rising wages.  At least in this context, the push of regulation seemed to have little impact on changes in agriculture.  The pull of outside opportunities, on the other hand, appears to have played a dominant role.  My analysis does not rule out some third factor as a cause of mechanization, but no third factor is necessary.  The mechanization we see is exactly what we would have expected, given the increase in wages.
Another way to view these results is as a success of sustainable development. It is often assumed that more economic output means more pollution.  But, in this context, major development markers, like wages for the poor and agricultural productivity, improved alongside environmental outcomes.  Mechanization has significantly curtailed the air pollution associated with sugarcane harvesting even as agricultural workers earn substantially more. Development may well have led to a better environment in this case.
Austin is a Ph.D. candidate at the University of Michigan.  You can see more about his work at

Join the Conversation

The content of this field is kept private and will not be shown publicly
Remaining characters: 1000