What happened and what’s next for DPPD

--

tl;dr: we were able to demonstrate the viability of the method and will continue to work on the widespread adoption of DPPD. If you want to get involved, reach out at mail[at]datapoweredpd.org

Where we started

Work in the data for development space tends to focus on the aggregate — the collective behavior of individuals and groups — while discarding individual observations, including outliers.

In April 2020, we — the GIZ Data Lab, Pulse Lab Jakarta, the UNDP Accelerator Labs Network, and the University of Manchester Centre for Digital Development — set out to change this.

Together, we built on Albanna and Heeks’ early work on the Data Powered Positive Deviance method and initiated a series of pilots covering a variety of domains, countries, data sources and types to determine whether digital data can be used to identify local innovations, and to help innovators scale these innovations to address challenges in areas like deforestation, food insecurity and gender-based violence. We also sought to answer if this could be achieved at a lower cost and at a larger geographical and temporal scale, and faster than more conventional approaches.

Map of DPPD pilots

Testing a method that existed on paper only and which required access to a range of data sources and a set of skills not easily available — all while engaging with various types of stakeholders and coordinating a network of partners from Ecuador to Indonesia — proved challenging.

To overcome these complications, we were intentional in the steps we took to conduct this work. That began with bringing partners with complementary strengths to the table: experience in running data projects with partners across the globe (GIZ Data Lab and GIZ in-country teams); in-depth expertise on asset-based approaches (UNDP Accelerator Labs); pioneering research on the use of big data in the positive deviance approach (University of Manchester Centre for Digital Development); data analytics experience and data partnerships (Pulse Lab Jakarta).

Next, we ensured that our pilot projects were set up across different geographies and domains, and used different data types, allowing us to test the method under varying conditions. This was only possible thanks to the committed and capable in-country pilot teams. While we did have a central team that advised the pilot teams on applying the method and led in the documentation of learnings and the write-up of the DPPD handbook and other materials, the pilot teams worked largely independently and the initiative was mainly remote-only. We developed and refined the method in a back and forth between the central team and the pilot teams. The central team drafted frameworks, methods and tools; shared them with the pilot teams for application; and reworked them based on their feedback. In addition, pilot teams developed their own specific tools where needed, some of which were ultimately included in the handbook.

Overview of the seven DPPD pilots

What we learned

We learned quite a lot over the past two years and were able to document some of these learnings in a report and an academic paper:

  • DPPD is not universally applicable: Digital data is rarely available at the level of individuals, mainly due to the need to de-identify and aggregate digital observations to preserve privacy. This makes DPPD better suited to development challenges where aggregated units are of relevance, e.g. geographic units such as villages or urban blocks.
  • Niche know-how is needed: A combination of country-specific domain knowledge and domain-specific data knowledge is crucial in the very early stages of a DPPD project. The former is needed to understand the normative behaviors in certain geographies and the contextual and structural factors that enable the successful practices applied by positive deviants. The latter is required to identify relevant performance indicators as well as mapping out suitable data sources that could be used.
  • Adopt a holistic approach when understanding deviance: DPPD provides an opportunity to uncover factors beyond the individual positive deviant that could be modified and transferred, such as access to markets or the provision of government subsidies. This can inform the design of nuanced interventions and thereby increase their effectiveness and contextual fit.
  • Earth observation data is the low hanging fruit for DPPD: Earth observation data turned out to be the most viable non-traditional data source that can be employed in the DPPD method. It can be acquired at low cost, over long periods of time, and thanks to recent advances in remote sensing technologies, it is witnessing a growing availability at a high resolution, including coverage of lowest-income countries where other datasets are lacking.

What we achieved

We’re excited about the progress we made over the past two years. While there remain open questions to investigate to improve the method, we were able to demonstrate its viability to identify locally sourced solutions by combining non-traditional and traditional data sources.

For example, the team in Niger was able to document that successful farmers use so-called Zaï holes or stone cords to harness rainfall and avoid rain water runoff. They sow in times when sowing is likely to result in higher yields and reduced seed loss due to climatic factors (they have a notion of “useful rain” when at least 14mm of water have fallen). What’s more, positive deviants do not systematically clean their fields after harvesting; they keep millet stalks in the field to protect the soil from wind erosion and for fertilization to restore the organic matter. They also use a blend of active planting (e.g. Gao trees that help regenerate and fertilize the soil when they lose their leaves) and passive restoration to increase their yield. Such Assisted Natural Regeneration techniques result in better water infiltration, reduced soil evaporation and recovery of degraded land.

The DPPD team in Mexico City was invited to present their recommendations on how to create safe public spaces for women to 16 ministries and government entities of Mexico City during a meeting of the Cabinet Council for Monitoring Substantive Equality Public Policies. As a result, the team is working with the technical teams of various ministries to apply recommendations on green areas, informal commerce and other characteristics into construction work currently underway on the multimodal transportation station of Indios Verdes and the surrounding neighborhoods.

Apart from developing the method and testing it through different pilots, we set out to promote it so that others can use and build upon it. We published a handbook that serves as a practical guide for development practitioners to identify and scale innovative, successful local practices. We also published the method in a peer-reviewed journal, a sign of its scientific rigor. As we developed the method, we documented and shared what we learned with the wider data community through reports, webinars and blog posts.

The challenges we faced

We view the pilot projects as a crucial, successful first step. However, we did stumble a few times along the way. One obstacle was that we were not able to identify key differentiators of positively-deviant behavior in some of the pilot projects. Data quality could be to blame. For example, the control data in one of the pilots was relatively old (dating to 2012), and we found that it falsified our predictions (of positive deviance) when we realized its inaccuracy after going into the field.

In another pilot, the digital data we used underrepresented the actual measure it is supposed to capture, leading to the identification of false positive deviants, which made it harder to identify significant differences between the two groups (the group we identified as positive deviants and the other group of non-positive deviants). For example, in the Mexico safe public spaces pilot, crime investigation reports were used to measure crimes in the various urban blocks of Mexico City. These reports only capture a small proportion of the actual crimes happening, but we still used them because they were the closest available indicator of crime in the city. This led to the identification of positively-deviant areas not because of lower crime rates, but because of fewer crimes reported.

Sample size proved to be another challenge. In comparing the groups of positive deviants and non-positive deviants, we used qualitative methods (e.g. interviews, community participatory approaches), which limited the sample size to an average of 10 units per group. The choice of the small sample was also driven by resource constraints, as pilot projects could not afford large scale surveys. This made it harder to identify statistically significant differences (as the size of the sample increases, so too does the ability to detect significant differences).

What’s more, we wanted each pilot to advance further in the DPPD process within the timeframe of the initiative. We aimed for at least some of the pilots in which differentiators of positively-deviant behaviors were identified to reach Stage 4 (Design and Implement Interventions, see the figure below), where we would work with positive deviants to help others in their community apply their practices or to inform policy interventions. However, this is only now happening in the pilots.

The five stages of the DPPD method

What’s next

We have taken an important first step in demonstrating that digital data can indeed be used to identify positive deviants, and to some extent scale their practices. ​​We have also shown that digital data can be used not only for top-down identification and tackling of development challenges, but as a tool to identify and diffuse local practices and strategies.

That said, we know that a handful of pilots, a paper and a method handbook are not enough — so the work continues.

On the ground, we are pursuing our pilot work in Niger, Mexico and Somalia to advance to Stage 4 (Design and Implement Interventions). We are also supporting a second generation of DPPD pilots: in India on early sown wheat, in North Macedonia on financial governance, and in Panama on deforestation. In India, for example, we are combining remote sensing data and large scale surveys to identify and characterize farmers who have adopted an early sown wheat innovation that aims to reduce the negative impacts of rising temperatures at the end of the season.

To build further adoption of the method, we are also developing training materials that are tailored to different audiences. We have also been in conversation with a number of organizations that want to learn more about the method following the launch of the handbook — and we are aiming to build a community of DPPD practitioners to further test and refine the method.

A new setup

We are outlining a series of questions to improve and adapt the method and are moving into a new setup that will allow us to further this investigation. A network of researchers and practitioners applying the tools and learnings from the DPPD handbook will connect with and learn from one another through a dedicated community of practice.

A team of DPPD experts, led by Basma Albanna and Andi Pawelke, will provide advice, guidance and training to teams that need more than what the handbook can provide. A data scientist and a mixed methods field researcher will join the team and a pool of domain-specific data experts will provide subject matter expertise based on the various domains of application. In addition to the project efforts, the DPPD team will continue its work on frameworks, methods and tools. Representatives of the founding organizations of the DPPD initiative will serve as an advisory group to help scale the DPPD method and grow the DPPD community.

Get involved

We hope you will continue to engage with us through the next stage of this endeavor. Here’s how you can get involved:

  • Apply the method: explore the application of DPPD in your work. Learn about it from our paper, see how others are doing it on our blog, use the handbook to apply it yourself, or message us for support.
  • Provide data access: we are interested in developing DPPD use cases using social media and mobile data. If you have access to such data and are interested in the method, please reach out to us.
  • Contribute with your expertise: are you an expert in data for development or community participatory approaches? Join the DPPD community or write to us to learn more.
  • Support us: Reach out to learn how to partner with us.

Written by Andreas Pawelke and Basma Albanna on behalf of the DPPD Initiative with inputs from Catherine Vogel, Jeremy Boy and Richard Heeks

--

--

Data Powered Positive Deviance DPPD

We are an international collective that is dedicated to utilizing big data to find effective locally developed solutions to complex problems.