https://orcid.org/0000-0001-8185-1808
![Sample our Global Development journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5929/TOC-Subject-Sample-2021-Global-Development.jpg"})
ABSTRACT
Poverty reduction is a key objective of development interventions. Evaluating the effectiveness of policies and programmes thus requires practical, reliable and context-relevant measures of poverty. This article is the first to compare the new Extreme Deprivation Index (EDI) framework with the increasingly used global Multidimensional Poverty Index (MPI) framework. Locally adapted versions of both non-monetary poverty measures were calculated for each household using an original survey in Rwanda’s main coffee-producing region (a high deprivation context) and another in Laos’s main coffee-producing region (a relatively low deprivation context). We examine the resulting poverty profiles and discuss implications for policy design and evaluation. We find that, despite limited overlap, in both contexts each index identifies households that are consistently worse off on multiple key markers of poverty and can therefore be considered valid measures. In addition, known key markers of poverty can predict adjusted global MPI status better than EDI status in Laos, whereas the EDI framework performs best in Rwanda. We conclude that the EDI framework provides a quick and reliable way to identify households with very low standards of living in high deprivation contexts. It is particularly useful for programmes with limited resources operating in comparatively poor rural settings.
Acknowledgments
We are grateful to the Alliance of Bioversity International and International Center for Tropical Agriculture (CIAT) in Rwanda, the National University of Laos (NUoL) as well as the local authorities in both countries for hosting and facilitating our project. We sincerely thank our research teams in Laos and Rwanda, especially the enumerator teams and above all our gifted research assistants Phothong Chanthavilay, Rénovat Muhire and Mukamana Theonille for all their tireless efforts. Thanks also to Eva Ming and Outhoumphone Sanesathid for invaluable assistance with the data cleaning in Laos, as well as to the CDE office in Laos for all their administrative support. We would like to also acknowledge Prof. John Sender and Prof. Christopher Cramer for helpful inputs to the EDI construction and Prof. Ben Jann and Prof. Daniel Gatica-Perez for useful advice on statistical procedures. We are grateful to Tina Hirschbuehl for editing the manuscript with great attention to detail. We also thank the Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS) programmes and the National Institute of Statistics in Rwanda for allowing us to use some of their data sets. We also gratefully acknowledge two anonymous reviewers for a close reading of the manuscript and their thoughtful suggestions for improvement. Finally, we sincerely thank all our research participants for offering their time.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. This research is part of the FATE (Feminisation, Agricultural Transition and Rural Employment) project. We refer to our surveys as FATE surveys to distinguish them from other data sources.
2. We selected two sectors close to Erlebach’s (Citation2006) study site. This allows for some comparisons across time.
3. One selected village in Laos is close to, but not part of, the Bolaven Plateau. Some inhabitants used to grow coffee in the past but production is now dominated by rubber plantations.
4. While statistical techniques such as principal component analysis (PCA) can be helpful to identify a subset of a large number of eligible goods, it is more important to rigorously justify the selection process in the terms outlined here. The same goes for the assignment of weights to the index. Cramer, Sender, and Oqubay (Citation2020, 205) note that ‘unweighted indices of socio-economic status have often been found to perform just about as well in identifying low socio-economic status rural households as the indices constructed using PCA to estimate weights’.
5. Since the focus is on private consumption, goods received as gifts or donations should ideally not be counted. This is a limitation of the FATE surveys which did not ascertain how goods were obtained.
6. In the two instances where households reported having goods requiring reliable electricity but did not have access to electricity, we did not count these goods (i.e. these households count as non-owning). One of the households simply remained EDI-poor whereas the other changed status from non-EDI-poor to EDI-poor.
7. For this reason, we combined the categories basic mobile phones and smartphones into whether the household has any type of mobile phone or not (landlines being irrelevant for private households in our sample). On the other hand, we included radios and torches separately from mobile phones as households in our sample frequently own them together with mobile phones.
8. We thank an anonymous reviewer for this suggestion. After analysing wall, floor and roof materials in each sample, we included the indicator with the largest variation in each case: walls in Laos and floors in Rwanda.
9. We thank Prof. John Sender for this suggestion.
10. Despite the central importance of employment to poverty, we have not added any employment indicators into our MPIs and EDIs for four main reasons. First, the concept and design of the EDI framework is based upon private consumption only. Specifically, it derives from Engel-type expectations about the division of consumption between necessities and more luxurious goods and aims to identify people with extremely low living standards (see Sender, Cramer, and Oya Citation2018). Therefore, employment indicators have no part here conceptually. Second, the aim of the EDI framework is to provide a practical way of identifying the most deprived using easily verifiable answers. Most items in the EDIs are tangible and visible goods. Employment data is much more difficult to assess not least due to the informal and dynamic nature of most rural labour relations as well as occupational multiplicity and questions about household membership. This does not mean that employment data cannot or should not be collected. Quite the contrary, our article underlines its relevance. However, it is more difficult and resource-intensive to capture employment relations accurately (e.g. requiring more enumerator training and probing as well as a more complex survey design) and, therefore, it is not conducive to the aims of the EDI framework. Third, most global and national MPIs do not use employment data. In fact, the MPI framework neither requires nor precludes the inclusion of employment indicators. This not only highlights the flexibility of the MPI framework but also underlines a certain conceptual arbitrariness that is not present in the EDI framework. Fourth, as employment is neither inherently required in the MPI nor the EDI frameworks, it is revealing to leave it out in both indices and to assess the extent to which these measures help us understand the employment characteristics of poor households.
11. We thank an anonymous reviewer for this observation.
12. Variables that are part of the MPI framework have been omitted as there are significant differences by design: the reason that EDI-poor-only households are not also MPI-poor is mostly because many are not deprived in schooling and/or nutrition which are heavily weighted in the MPI framework.
13. To create a dichotomous indicator, we exclude child-headed or male-headed-only households as they are negligible.
14. Classification analysis is typically used in machine learning and its application to poverty research is relatively new (Gao et al. Citation2020). Whereas these machine learning classification models are built on algorithms that are trained and then tested on separate data sets with the same predicted variables, our goal here is not to train a machine learning algorithm but simply to evaluate the confusion matrices obtained from the regression models with known markers of poverty.
15. Differences in the understanding of what constitutes ownership may however limit test-retest reliability.
16. Since 2018, the Oxford Poverty and Human Development Initiative (OPHI) has replaced the flooring with a housing indicator in the global MPI (Alkire, Kanagaratnam, and Suppa Citation2018). It counts a household as deprived if either the floor is made of natural materials or the roof or walls are made of natural or rudimentary materials including reused wood, wood planks or plywood. While it does make sense to consider walls and roofs as well, it does not seem adequate to include some of these materials in the deprived category and this level of detail was not differentiated in the FATE surveys.
17. When calculated on the DHS and MICS data respectively for Nyamasheke (using the adjusted destitution measure) and for rural areas in Laos (using the adjusted global MPI), the multidimensional poverty headcount ratio and the MPI changed by less than 5 percentage points in either country when excluding the water indicator.
18. For the adjusted global MPI in Laos, households were counted as deprived if either household head indicated that they sometimes (during two to four months in the last 12 months) worried about not having enough food for the household and that they sometimes (during two to four months in the last 12 months) did not manage to buy the type of food they wanted to eat. Households were also considered as deprived if either household head said that either of these occurred often (during more than four months in the last 12 months, i.e. significantly more than during the entire lean season). For the adjusted destitution measure in Rwanda, households were counted as deprived only if either household head indicated that both of these occurred often (during more than four months in the last 12 months, i.e. significantly more than during the entire lean season). In cases where only one household head was available (e.g. widowed households), we based the indicator only on her or his responses for both the adjusted global MPI and the adjusted destitution measure.
19. We also compared FATE data to the Lao Population and Housing Census of 2015 to verify the reliability of our data for other indicators: census data report that in the six villages sampled for this study, 94.05% of households have electricity, whereas the FATE survey finds 99.12%. The census puts average household size at 4.98 persons, percentage of households with operational farmland at 94.42, and percentage of the literate population aged 15 years or older at 82.91. The numbers in the FATE survey are 5.24, 91.47 and 77.93 respectively, indicating that our data are reasonably reliable considering different survey designs, sampling procedures and a three-year time gap. Some of these indicators have been used in the poverty analysis above.
20. In Laos, the excluded group is more deprived in asset ownership. All other MPI components or variables of interest are not significantly different between households with and without MPI/EDI data. In Rwanda, on the other hand, excluded households seem to be worse off on a number of indicators such as sanitation and size of operational holding.
Additional information
Funding
Notes on contributors
Patrick Illien
Patrick Illien is a PhD Candidate in Geography and Sustainable Development at the University of Bern. He has undertaken mixed-methods fieldwork in Laos and Rwanda investigating the political economy of agrarian change. Patrick is particularly interested in the relationship between economic growth, labour market dynamics and poverty. He holds an MSc degree in Violence, Conflict and Development from the School of Oriental and African Studies (SOAS) at the University of London.
Eliud Birachi
Eliud Birachi is a research scientist at the Pan-Africa Bean Research Alliance, the Alliance of Bioversity International and CIAT, working in Africa as a market economist. He has a PhD in Agribusiness from the University of Kiel, Germany. His research interests are in linking farmers to markets and especially through farmer collective efforts, value chain analyses and market research, as well as gender mainstreaming of market access work. Recent work focuses on agricultural commercialisation, gender empowerment and rural employment in agricultural dependent systems. He is the FATE project coordinator and lead researcher in Rwanda.
Maliphone Douangphachanh
Maliphone Douangphachanh is a lecturer at the National University of Laos. She was awarded the 2014 Federation PhD Scholarship Prize by the National University of Laos, Laos, and the University of Bern, Switzerland. She defended her PhD thesis in 2020 with the title ‘Gender Division of Labour and Women’s Decision-Making in Coffee Farming Households in Southern Laos’. Maliphone uses mixed-methods approaches. She received her PhD in Gender and Development from the University of Malaya, Malaysia, and is a research assistant in the FATE project, focusing on agricultural commercialisation, gender empowerment and rural employment.
Saithong Phommavong
Saithong Phommavong is a visiting lecturer in the Faculty of Social Sciences, National University of Laos, Vientiane, Lao PDR. He graduated Bachelor of Political Sciences from National University of Laos, Master of Economics from Kobe University in Japan, and PhD of Social and Economic Geography from Umea University in Sweden. His research interests are economic development, pro-poor tourism, political economy, sustainable development, gender relations and land use planning. Recent publications can be found in the International Journal of Culture and Tourism Research, Current Issues in Tourism, Springer and Global Social Welfare, Routledge and IntechOpen. He is a coordinator and senior researcher of the FATE project in Laos.
Christoph Bader
Christoph Bader is a senior research scientist at the Centre for Development and Environment at the University of Bern. An economist by training, he works in the thematic fields of sustainable economies, post-growth societies, poverty and inequality, explaining and fostering sustainable consumption patterns and lifestyles. He also teaches in the CDE study programmes. He received his PhD in Geography and Sustainable Development from the University of Bern.
Sabin Bieri
Sabin Bieri is Director of the Centre for Development and Environment at the University of Bern. As head of the socio-economic transitions cluster, she oversees research and teaching on the social and economic dimensions of sustainability. A social geographer by training, she specialises in questions of work, globalisation and inequality. She is leader of the FATE project, a cross-case study on agricultural commoditisation and rural labour markets in four countries on three continents. She received her PhD from the University of Bern, where she was an awarded member of the Graduate School in Gender Studies.
![Supercharge Your Next Research Paper: Research and write your next paper with Jenni AI.]({"type":"image" , "src" : "/sda/112446/MPU-L1EB.png"})