This blog is part of the series "Small changes, big impacts: applying #behavioralscience into development".
Access to an extremely large amount of data has enabled us to pursue research endeavors that just a couple of years ago seemed unimaginable. Examples of amazing big data applications in the field of economics are all over the place: using job-portals data to inform labor market policies; analyzing citizens’ reactions to public policies using Twitter; creating daily inflation data using billions of records from online retailers around the world; and even measuring economic growth from outer space!
The data revolution is open to anyone with the right tools, and big data can be useful to answer policy questions. Pairing big data with some of the traditional methods of data gathering such as household surveys can yield timely information and can help shape appropriate policy responses. For instance, traditional household surveys, from which unemployment estimates are calculated could carry outdated employment data by the time they become available. But big data can complement this effort in places where unemployment rates correlate with the frequency with which people use Google to search for jobs, as in the case of Brazil, that could be used to estimate real-time unemployment rates.
Monthly unemployment rate and google searches for “looking for a job” in Brazil, 2006-17
One area where the use of big data is emerging is in the field of behavioral insights. Behavioral sciences have a lot of testable hypotheses, but little data to play around with. Conversely, big data has a lot of information, but needs better questions. Some work has already been done at the intersection of both fields but there is a huge opportunity in this space. Recent research has showed that machine learning techniques can improve human decision-making. For instance, a recent paper analyzes judges' bail decisions in the US, finding that criminality could be reduced by 24 percent if decisions were based on a computation algorithm, instead of relying on judges’ biases. Another strand of behavioral science literature has used big data to predict risky behavior, using network analysis to predict violent crime in Chicago. Finally, some research has used social network data to predict personality traits. One paper finds that, given the information of just 10 Facebook likes, an algorithm can predict your personality more accurately than your work colleagues; with 70 likes, it can predict better than your friends; and with 300 likes, better than your own spouse.
Still, the literature that links big data to behavioral science is scarce and still incipient. Further collaborations between both fields can play out in different ways.
First, big data can be used to create measures of behavioral traits. As research mentioned above suggests, it can be possible to extract patterns from sets of data in order to study the determinants of a given behavior. This approach would have the advantage of using perhaps millions of data points from the real world, instead of relying on relatively fewer observations from a lab experiment. Second, data and behavioral science can be united for prediction purposes. Some emerging research in the field of genoeconomics combines mountains of genetic data to predict outcomes such as risk aversion, financial decision-making, educational attainment, political preferences, and subjective well-being. Finally, measures of behavioral traits could be used to complement other types of more traditional analysis, for example, using behavioral variables to target certain interventions, or to measure the causal impact of a policy.
The potential to harvest big data is particularly high in Latin America and the Caribbean where, in terms of the conventional sources of data, many countries are data-deprived. The irony is that the countries that stand to gain the most from the unconventional sources of data (which in many cases are publicly available), are the ones that have the fewest applications. (The exception is the strand of literature that focuses on predicting poverty with satellite imagery or cellphones’ call records, where the applications have been mostly in developing nations.) There are many reasons why big data hasn’t been as popular in Latin America as in other high-income regions, being perhaps the most important reason that access to internet in the region is far from universal—less than half of the population of Latin America has access to it—and it is also very unequally distributed.
But behavioral insight might be the missing piece that leverages the untapped potential of big data in Latin America. The recent surge in open data initiatives in the region like Google trends, the Ngram viewer, the observatory of economic complexity, DataViva, among many others, can help researchers discern hundreds of stories. By leveraging behavioral insights, these stories can tell us something about the universal aspects of human behavior that ties them together.
As the World Bank, governments, and partners continue experimenting and applying behavioral science in government programs and policies, we will share with you through this series ‘Small changes, big impacts: applying #behavioralscience into development’, the latest development and thinking in the region. Join us and share your thoughts, your work and thinking.
Previous post in the series: Public policy with a true human face | Can behavioral change support water conservation? Examples from the US, Colombia and Costa Rica
Interested on Behavioral Science? Visit our website or sign up for our newsletter to stay connected.