Development studies often focuses on the negative: constraints, challenges, negative impacts, etc. But what if we could use new digital datasets to identify positive deviants: outlier individuals, households, districts and others that outperform their peers in achievement of development goals?
In this episode, Basma Albanna and Richard Heeks discuss the “Data-Powered Positive Deviance” (DPPD) programme. The programme built on an original idea by GDI researcher, Basma Albanna, that was fleshed out in a paper co-authored with Richard Heeks, GDI’s Professor of Digital Development. It argued that traditional methods of identifying positive deviants relied on costly and time-consuming primary data-gathering from the field. Instead, it might be possible to identify outliers in the growing number of digital datasets already available.
Basma Albanna studied for her PhD at the Global Development Institute. She is now a lecturer at Ain Shams University and a Consultant for the GIZ Data Lab
More about Basma Albanna:
Richard Heeks is Professor of Digital Development in the Global Development Institute, part of the School of Environment, Education and Development. He is Director of the Centre for Digital Development.
More about Richard Heeks:
More about the“Data-Powered Positive Deviance programme
Intro music Anna Banana by Eaters
Find out more about the Global Development Institute:
Intro music Anna Banana by Eaters
Development studies often focuses on the negative: constraints, challenges, negative impacts, etc. But what if we could use new digital datasets to identify positive deviants: outlier individuals, households, districts and others that outperform their peers in achievement of development goals?
In this episode, Basma Albanna and Richard Heeks discuss the “Data-Powered Positive Deviance” (DPPD) programme. The programme built on an original idea by GDI researcher, Basma Albanna, that was fleshed out in a paper co-authored with Richard Heeks, GDI’s Professor of Digital Development. It argued that traditional methods of identifying positive deviants relied on costly and time-consuming primary data-gathering from the field. Instead, it might be possible to identify outliers in the growing number of digital datasets already available.
Basma Albanna studied for her PhD at the Global Development Institute. She is now a lecturer at Ain Shams University and a Consultant for the GIZ Data Lab
More about Basma Albanna:
Richard Heeks is Professor of Digital Development in the Global Development Institute, part of the School of Environment, Education and Development. He is Director of the Centre for Digital Development.
More about Richard Heeks:
More about the“Data-Powered Positive Deviance programme
Intro music Anna Banana by Eaters
Find out more about the Global Development Institute:
Intro music Anna Banana by Eaters
Intro Welcome to the GloTal Development Institute podcast based at the University of Manchester, where Europe's largest research and teaching institute is addressing poverty and inequality. Each episode will bring you the latest thinking, insights and debates in development studies for.
Richard Heeks So welcome to this great podcast. I'm Richard Hicks. I'm a Chair in Digital Development at the Institute and I'm with Basma Albanna.
Basma Albanna Who used to be a student at GDI at the Centre for Digital Development.
Richard Heeks So Basma, we worked together on your PhD, but can you explain a little bit about your PhD research and what initially attracted you to work on this idea of data power of positive deviance?
Basma Albanna Okay. So my, my PhD research was on developing a methodology that uses digital data, such as remote sensing data and mobile data to identify outliers who succeed against odds. And the point is to try to uncover their underlying practises and strategies and trying to skill them within their communities. And it's an idea that was based on or is based on a development approach, which is called the positive deviance approach that was devised by the Sternin's in the 1990s. But the thing is that they used to do this identification of outliers using more traditional data such as interviews and surveys and. And the innovation in my research was mainly into looking into digital data sources that are being increasingly available and trying to capitalise on them to identify those outliers or outperformers. And the first time I learnt about this concept was one year before applying for my PhD. So I was taking a course on design thinking. And there I one of the examples that that were used was on the positive deviance approach. And the that time I, I thought to myself that this concept is very intriguing and how I haven't heard about it before. And what what's really interesting about it is that it's different than the standard model of development where you instead of looking for failures, you look for success. And instead of relying on external expertise, you're building on local wisdom and know how. And the following year, when I was applying for my PhD and I wanted to do something in in data science and development. I couldn't help but remember the concept of positive deviance and how the analogy between the outliers in the positive deviance approach and the outliers in statistics or in data science. And that's why I thought, why not merge?
Richard Heeks Okay. Yeah, no, that's great. So you've talked quite a bit about about outliers and development, about Sternin's work. And maybe could you give us an example, maybe start with an example of the traditional approach to positive deviance and an example of what an outlier would represent?
Basma Albanna Okay. So one example is the very famous story of Vietnam, which is that when Sternin was at that time heading the Save the Children Foundation in Vietnam, and he was asked to solve the problem of malnutrition, and instead of going for the supplemental feeding programmes, he he came across this concept of positive deviance, which is looking into the ones who are doing better and trying to understand what are the uncommon strategies and practises that they're doing and using it and as the starting point for looking for solutions. And he did that in Vietnam. He looked for extremely poor villages that had well-nourished children. And by doing so, he was able to identify some of those villages. And he went there. He did some ethnography and qualitative research, and he found out that there are few families that had those well nourished children. And what they did is that mothers did very simple things, such as adding sweet potato tops and shrimps, that the rice paddies to their children's meal, or they're having more frequent meals than then, then, then the fewer meals over the day and washing their children's hand more frequently, and by skilling those simple, extremely simple practises, he was able to reduce and rehabilitate an estimated of 80,000 children and was able also to reduce malnutrition rate drastically and by from from that he started it thinking why not making it a tool that can be used in development and it's an asset based approach tool. And he used it to he used it in other domains and other applications, too, to scale the method.
Richard Heeks So can you give a couple of other examples of other development challenges or development problems that the positive deviance approach has been applied to?
Basma Albanna The positive approach was applied in child in and in female genital and FGM in Egypt. It was also applied in retaining school retention, which was applied in agriculture to identify farmers who are having higher agriculture productivity, in reducing infections at hospitals. So those are examples of how it has been applied previously.
Richard Heeks So that last example, they will they be finding out hospitals with much lower infection rates than peer hospitals or similar hospitals and then going in and investigating how it was, what were the factors that underlay the fact that they had much lower infection rates and then try to scale those out to other hospitals in order to reduce infections more generally? Okay, that's fine. I get it. So that sounds like a very valuable development technique. So what were some of the issues that you were identifying with positive deviance that made you think, well, maybe big data or other digital data sources might offer something additional or better than what's being done in the traditional positive deviance approach?
Basma Albanna The main issue was that when you're relying on traditional data and primary data collection, there are limitations to the sample size because the cost and the time is directly proportional to the sample size. So so in the traditional approach, you are looking into a small population and within the small population you're trying to look for outliers. And this reduces the number of outliers that you can identify, because normally there are around from 2 to 10%. And when you have a small sample, it's very hard to generalise their strategies and practises on larger population so that the there was a limitation on the sample size that they can take due to the time and cost constraints. And also there are others, there are other issues like risk of of going to places where it's unsafe to do this kind of qualitative enquiry. And again, the problem of trying to scale those practises to larger populations so that they can inform policy and community interventions. So that's for the main challenges when it comes to the positive deviance approach. And and that was the point where I thought to myself, how about using digital data that is available, increasingly available and using this data, readily available data to identify those outliers at a larger population with very little initial time and cost and those outliers or positive deviance will be the entry points for the qualitative enquiry, which is the expensive part. But you will have larger samples of those kind of outliers.
Richard Heeks Yeah. So I guess, I mean, think about it. If you were doing something with hospitals, maybe with traditional field work where you have to go in and you have to interview a lot of people in each hospital, maybe you might have, I don't know, at best, a few dozen hospitals, whereas I guess the big data sets, maybe they contain details about hundreds or thousands of hospitals or whatever the other units of analysis there might be. And therefore, you're going to have a much greater set of data to be working on. And as you said, you're going to have much greater number of in absolute terms, a much greater number of outliers that you can look at and analyse the factors. And instead of it being maybe just a small handful of outliers, hospitals, that maybe they have some exceptional features about them, you can't generalise. You're going to have maybe again, dozens or even hundreds of these positive deviants. And then when you look at them, you're really going to be able to generalise from those their features. Okay. That's that's clear. So you've had this idea essentially alright positive deviance is a great thing in development, but there are some challenges around time and and risk and the sample sizes and so on and big data might might help. So so kind of what was your next step with having had this nice idea, this potentially great idea? What was your next step in taking that forwards with your PhD research?
Basma Albanna So the next step was to look for like viable case studies that where I can test out the method and test the viability of those data sources for application. And initially I wanted to try the three main datasets that are very known in their use in big data for development. So I wanted to try remote sensing data, online data and mobile data, but somewhere in the middle of the year I realised that the mobile data is extremely challenging to actually access the data. So I limited my initial research at the very beginning to online and mobile data online and remote sensing data. Sorry. And to give you an example of the kind of cases that I that I worked on. So the first case that I worked on was on researchers in information systems, Egyptian researchers in the information systems field that are having publication out performance when compared to their peers. And for that I used online readily available data such as Google Scholar data and and other data and online digital datasets to identify those using their citation matrices such as each H index and HI each index and stuff like that. And after identifying them using those digital data, I had to do a fieldwork in Egypt to an interviewed the outlier researchers to uncover some of their practises and strategies. So that was this the first case that I worked on.
Richard Heeks And let me just I mean, picking up on one of those things because because you already mentioned it, I mean, a couple of things to bring out of that. First is data for development has a great deal of promise, but when you're actually going out there, it can be quite a challenge to find the datasets, accessibility, data quality, all of these things are actually, you know, can be quite serious challenges. And the other thing you identified is you've gone in hoping that this might be a sort of cheaper, quicker way of of doing positive deviance. But actually, from if I understand correctly, from what you're saying, you can do that for identifying the positive deviants. But then actually it turns out you have to kind of generally at the moment at least getting on the ground and still go and talk to them in order to find the factors that are underlying their positive performance. Is that right?
Basma Albanna Yes. Yes. So the digital data can only tell you what is happening like the what, but would it be able to explain to how and why? And that's the part where you would need what we call thick data, which is data collected through qualitative approaches and ethnography. But but the beauty of this method is that it provides a systematic way to actually integrate those two types of data. And this was something that missing that was missing in the data for development practise, finding those kind of ways to actually complement those two different types of data. Because I strongly believe that big data, the value from big data cannot be actually obtained unless it's complemented with traditional data. So it's not supposed to substitute it, supposed to just compliment it. But as you mentioned, Richard, there were a lot of challenges when it comes to accessing such data. But once you are able to access it, you can start to gain the value of using it, which is identifying them at large scale and over long periods of time. So it's not just the. So you're not only able to identify deviants at large scale, but you can also identifiy deviance over time. And this was something that's very hard to actually achieve with traditional data because you would need like perspective studies. And so so it was also one of the now.
Richard Heeks And I mean, I don't know whether that was the case with the with the research. So just thinking about sort of development challenges, looking at the research is one of the key challenges in research has been that there are biases, all the various other contextual factors, which means that research in the Global South doesn't really get the profile that it that it might otherwise get. And what you're looking at is helping to understand, well, how is it that some researchers in the Global South are able to get that profile and citations and impact for that for their research? In terms of that longitudinal picture of positive deviants, they put some particular examples that you came across in your work of how that might be useful in addressing development challenges that couldn't otherwise be addressed. Or that was that's something for the future, if you like, rather than for the particular cases that you looked at.
Basma Albanna It was important in identifying or in this kind of longitudinal coverage was extremely important. And other cases that I've worked on as part of a of a global collaborative that I was part of. And just to give you an example, so in one of the case studies that we were looking on, what we're looking into are looking into and rangelands and communities in Somalia that are able to preserve their rangelands despite frequent droughts and the the hardest drought and the most climatic stress that they were they were they experienced was in 2016 and 2017. And in the analysis of the remote sensing data, we had to track the health of the rangelands from 2016 and 2017 until 2020. And based on it, deviance was identified. So our baseline was the rangelands before the drought. And, and based on it, we were able to identify the rangelands that are having an enhanced vegetation health despite what they witnessed in 2016 and 17, something that just wouldn't have been possible if it weren't for the remote sensing data that was able to cover this period of time. A similar case was in Ecuador when we looked into deforestation rates and we wanted to see the deforestation rates in the past five years because if I'm not looking over a long period of time, they might not be deforested this year, but they already finished all the forest in the past five years. So here we were also looking into rates of of deforestation over a period of time.
Richard Heeks Yeah. So those are those are two good examples of where this new approach is enabling you to get insights into understanding outperformers. In terms of these are cattle pastoralists.
Basma Albanna Yes, yes. Cattle ranchers.
Richard Heeks Preserving their lands. And in Ecuador, equally, that's cattle farmers as well.
Basma Albanna Yes. So those were cattle farmers in Ecuador and pastoralists in Somalia.
Richard Heeks Yeah. Okay. And we should also mention as well, you talked about this global collaboration that this was involving GIZ, particularly the data lab and the field offices and UNDP Accelerator Labs. Also, I think Global Pulse Lab Jakarta, but particularly there it was GIZ, I think who read an article, a paper that you published that kind of every PhD students dream, read your literature review were really taken by it and funded was a one and a half million euro project as a result which scaled out your ideas to I think six or seven different projects, a couple of which the Ecuador one and the Somalia one you've you've mentioned. So you've kind of got to this wonderful stage that you're doing your own PhD research, but alongside this is this big global project that's that's putting your ideas into, into practise. And where are we kind of up to at the moment with with all of those things, those projects or what what came out of of all of those projects that were funded by GIZ and UNDP.
Basma Albanna So there were four main projects that were somehow co-funded co-funded between the UNDP Accelerator Lab and GIZ Data Lab, one in Somalia, Ecuador, Niger and Mexico. And in those four projects we're currently in the post fieldwork stage and. Some of them went to the stage where they identified those uncommon practises and strategies, and now they're engaging with different stakeholders to inform some maybe policy or community interventions. And maybe I can give you an example of some of the the findings from some of the interesting projects. So in the Niger project, we were looking into crop productivity. Farmers were able to achieve better crop productivity and mainly in a millet and sorghum, so cereal crops. And we were able to identify some of the communities that are achieving better using remote sensing data. And after going there, we were able to identify some findings such as leaving their millet stalks and stems on the ground, which helped reduce so the effects of wind erosion and also to help restore the soil organic matter. They also left the [00:19:13]gough [0.0s] trees which increased soil fertility. There was this notion of use for rain that was common across some of those farmers, which is you don't sow unless the height of the water is 14 millimetres and they also used a technique which is called the [00:19:32]ZI techniques, [0.5s] which also enabled soil regeneration and reducing surface water runoff. So so those were some of the findings that we found in Niger. And now we we're thinking of how to engage with GIZ project there to somehow make use of those kind of findings.
Richard Heeks Okay. So you so you've kind of you've identified the positive deviants. You've been able through fieldwork to identify the factors or the behaviours that underlie that positive deviance across a range of these, these projects. And now moving to the next stage of scaling those out through local actors. What about more generally for the overall project of of DPPD, the data powered positive deviance? What what kind of what do you see as the next stages or the next steps for that beyond those those pilot projects that you mentioned?
Basma Albanna So the initiative and its current form have changed a little bit. So after working with the UNDP accelerator labs and GIZ projects on those different pilots, we're now at a point where it's more of an individual effort. So I'm working now on consulting in DPPD, along with another colleague, Andy, who has been working with me on this. And we're trying to pick up on some consultancy work and training that is related to DPPD I'm working on a project in India to identify communities that are adopting a crop drought resilient, sorry, heat resilient crops. So in India, there's the problem of the terminal heat at the end of the season, organisations such as [00:21:16]Simard [0.0s] introduced early on wheat and we are trying to see who are adopting this kind of innovation and why, why there is better adoption. And another thing that is coming an end of this year is that we're trying to develop a course on DPPD, specifically in climate change mitigation and adaptation. So this is the work that we're currently working on in terms of the DPPD.
Richard Heeks Yeah. And I was going to I was going to say drawing that out, you can see that quite a number of the projects there. Deforestation in Ecuador, the rangelands, Niger, of course, is a very climate challenged location and now the Indian project as well. There's quite a concentration of activity that DPPD can be applied to environmental and climate change related development challenges. But I think you've also shown how it can go broader than that as well, have you? Because Mexico was looking for safer areas for women who are more protected from gender based violence when moving around, moving around in the city. And I think was there also a project around financial governance in North Macedonia that was that was mooted certainly as well. So at least that's showing that it's not just around farming and agriculture and climate change and so on. There's quite a range of different areas and different development challenges that the DPD approach can be applied to. I'm hoping that there's going to be a link somewhere at the end of the of the podcast. But more generally, if people are interested in finding out more about DPPD, what should they do?
Basma Albanna They can check the website that we have. So datapoweredPD.org. And there we have a number of resources so there are the articles, the blog post, the the reports, a handbook which they can start with, which is a great resource which are showing which is showing the different steps and stages of applying DPPD along with very valuable tools. And of course they can get in touch. So there's my email and, and I'm looking forward to see more adopters.
Richard Heeks We all are. DPPD was a great idea, wonderful project. You know that from just a small seed of your own idea some years ago it blossomed into not only a thing for you, but a big global research project and hopefully a really exciting new approach in development. So Basma, thanks very much for being in conversation with me.
Basma Albanna Thank you, Richard. And thanks for all the support in the past four years.
Richard Heeks No problem.