Government experiment shows why decentralization makes sense—and why IT innovations may reverse the trend

A farmer in Colonia Independencia, Paraguay drives his tractor over a wooden bridge. Berkeley Haas researchers worked with the Paraguayan government on an experiment that gave cell phones to government workers who advise farmers on marketing their crops. (Credit: Jan-Schneckenhaus for iStock/Getty Images)

A major question in organizations is whether to keep decision-making power concentrated at the top, or to cede control to lower-ranked managers. Yet whether a decentralized approach is more effective than top-down management depends on whether middle managers have additional knowledge that is lacking among those at the top.

“There’s this robust idea in economics that social systems in general and organizations in particular have valuable information dispersed throughout the organization— it’s why we think markets have an edge over central planning, why there is federalism in government structures, or why we see so much delegation in private firms,” says Berkeley Haas Prof. Ernesto Dal Bó. However, he notes that there is little empirical research documenting whether that’s true, and moreover, whether organizations take advantage of that information when they decentralize.

In a study forthcoming in Econometrica, a leading journal in economics, Dal Bó and Prof. Frederico Finan conducted a field experiment with the Paraguay government to assess whether this idea holds true in a government setting. Along with co-authors Nicholas Li and Laura Schechter, they analyzed the tradeoffs of decentralized versus centralized control.

Their study focused on a Paraguayan government program that provides business support to local farmers. Government workers, known as agricultural extension agents (AEAs), help farmers with tasks such as marketing and diversifying their products. Each agent reports to a supervisor who oversees their work. To improve the performance of the AEAs, the central government gave them GPS cell phones to track their locations and monitor their work—with the goal of reducing shirking among those prone to slack off.

In order to compare the efforts of agents who were given the GPS technology versus those who were not, the research team worked in connection with the government to randomly distribute cell phones among the agents. Before the rollout of the study, the authors also asked supervisors to choose which agents they would give cell phones in the absence of the randomized process. This allowed the researchers to analyze whether supervisors have, and make use of, valuable information about their workers.

They found that agents with cell phones improved their performance relative to those who didn’t have them. Those with cell phones visited a significantly higher share of farmers in their assigned area over a given week (33%) than those lacking the new technology (27%). Likewise, farmers reported they were more satisfied with their interactions with the agents who had the phones. Interestingly, the researchers found that this effect was entirely driven by improvements among agents who the supervisors said they would have selected. Those agents increased farmer visits by 15 percentage points, while the agents who the supervisors would not have chosen did not change their behavior in a meaningful way. This suggests that the middle managers knew which agents would benefit from greater oversight, and how to best reduce shirking.

Based on those results, the researchers analyzed the efficacy of decentralized control, which they note depends on how much information the central government already has at its disposal. It’s possible, for example, that the centralized government could have effectively selected the agents if their performance was correlated with measurable traits, such as age or cognitive ability. In the experimental trial, however, they found that these observable characteristics were not correlated with the middle managers’ selection of promising agents, implying that the supervisors relied on information not easily collected by a survey or administrative records.

The research team then modeled alternative scenarios to estimate an optimal policy for rolling out new technology. They estimated that, under decentralized control, the best policy would be to give cell phones to 75% of the agents, due in part to the fact that phones not only did not improve some workers’ performance, but appeared to backfire, actually reducing some people’s performance However, the researchers also found that the government could do even better than the decentralized approach by running a small pilot experiment, and then using the results to learn which individual employee data best predicts response to monitoring. This indicates that the benefits of decentralization decreases as communication technologies improve, since the availability of data reduces the information gap between different levels of the federal government.

The results suggest that, when a centralized authority lacks information, a decentralized system can leverage the expertise of middle managers and offer improvements over the top-down approach. However, when a government has access to data and uses experimental methods, it has the potential to put even more effective policies in place.

“We are in a very particular moment, where the ability to gather and use data is changing rapidly, even in middle- and low-income countries,” Dal Bó says. “Because of this, the optimal decision to decentralize may change too. We have seen a worldwide trend toward decentralization in the last 40 years or so, but the tide may be changing right under us.”