Starting small with big data: ‘Boots on the Ground’ and Big Data are complements not substitutes
This post was co-authored by CEGA affiliate Travis J. Lybbert of UC Davis and Oscar Barriga Cabanillas, PhD student in Agricultural and Resource Economics at UC Davis.
Innovative and powerful tools, such as those provided by big data and machine learning, offer researchers new ways to study individual, collective, and market-level responses to poverty alleviation interventions and related economic development outcomes. However, these tools only reach their full potential when they are informed by and reflect sufficient appreciation for local context.
In an ongoing study funded by CEGA’s Digital Credit Observatory (DCO), we are deploying these new empirical tools to evaluate the welfare impacts of small, short-term digital loans in Haiti. In this post, we describe the role that initial Focus Group Discussions (FGDs) have played in the conceptualization and refinement of our research design, which mixes machine learning with traditional household surveys.
Our DCO study contributes to the design of a mobile money product slatted to launch this year in Haiti that provides small, 30-day loans ranging from $2 to $20 dollars. These loans, disbursed and repaid via mobile money, use Call Detail Records (CDR) to determine client credit limits. The study further aims to rigorously evaluate the welfare impacts of this new product. To better understand the potential customer base, we conducted an exploratory study to better understand the likely client for this ‘nano-loan’ product.
Through a series of FGDs, we discovered that such a nano-loan product appeals to a well-defined set of prospective clients. In the context of urban, peri-urban, and rural settings in Haiti, our FGDs signaled the following:
- Potential customers are likely to be vulnerable and food insecure and will leverage these nano-loans to smooth consumption in the face of minor shocks
- Potential customers are less likely to be micro-entrepreneurs with short-term working capital constraints
- This group is mostly self-employed and lacks access to formal credit sources
- In times of need, they rely on family and friends (especially popular with women) and informal pawnbrokers (especially popular with men) to access stop-gap credit for food, transportation, or school fee expenses
Our experiences during these FGDs have informed our research design and data collection strategy in important ways. First, the insights they generated made clear that a traditional approach of a baseline and final survey might not be ideal for capturing small-lived changes in consumption patterns. We opted instead for a five-round phone survey over 12 weeks to better capture the welfare impacts of the product.
Moreover, structured surveys conducted before each FGD session gave us an understanding of the size and distribution of the consumption levels of potential adopters. We have decided to complement the phone surveys with higher frequency SMS-based mini-surveys to better track how debt portfolios, income, investment, and food security evolve for our respondents over time. We will randomize which day these SMS surveys are conducted for each individual, and will aim to conduct these surveys more intensively during critical periods of the loan cycle.
Insights from the FGDs also raised concerns about offering consumption loans to poor and vulnerable households since it introduces the potential for overborrowing. If nano-loans are used to smooth short-term consumption shocks but simultaneously contribute to deepening debt cycles, then evaluating the overall welfare effects of such a digital credit product is complicated. Our modified research design and data collection strategy seek to address these complications. With more rigorous insights on risks of triggering debt cycles, we hope future research will be able to incorporate behavioral and other design features to address this potential issue.
Finally, the results from the FGDs showcased to our collaborating Mobile Network Operator the value of having outside partners that can provide a different perspective on issues that their internal teams might overlook. Such complementary insights open not only strengthen our partnership but open the door for future collaborations. This exploratory work highlighted the importance of complementing the use of big data with direct conversations with potential beneficiaries. The admonition to “know thy context” is surely as true for big data applications as it is for any empirical analysis that aims to sharpen our understanding of a complex world!
