18 Conclusions and Looking Forward

This volume has sought to contribute to improving the practice of measuring poverty and wellbeing in developing country contexts. The contributions include: two sets of software code designed to provide an advanced yet flexible basis for consumption (PLEASe) and multidimensional (EFOD) poverty analysis; a review of the theoretical foundations underlying the methods expressed in the code; discussion of practical issues encountered in the analysis of poverty and wellbeing both in general terms (Chapter 4) and in the eleven country cases included here; a synthesis of general lessons emerging from the case studies as a group (Chapter 16); and an extension to the analysis of inequality (Chapter 17). The hope is that this combined package will facilitate the conduct of rigorous analysis of poverty and wellbeing and enhance transparency and reproducibility.

The volume was not designed to suggest an exact cookbook approach to conducting analysis or to permit analyses to be produced more quickly. Rather, the analytical packages are meant to permit the analyst to spend more time thinking, cross-checking, and judging and less time on mechanical tasks. As emphasized, the default code sets, particularly for the estimation of consumption poverty (PLEASe), are unlikely to correspond to country circumstances. Analysts are near certain to be obliged to modify the PLEASe code to account for specific country circumstances. For both PLEASe and EFOD, careful checking that peculiarities of the circumstances/data are not generating erroneous results is required.

As noted in Chapter 1, many countries remain strongly dependent on external assistance for the conduct of their own assessments of poverty and wellbeing. This is, in no small measure, a reflection of the complexity of the task. While it is hoped that the materials contained in this book will help developing country analysts to grapple with this complexity, many of the challenges posed in rigorously measuring poverty and wellbeing are essentially irreducible. As a consequence, the attainment of one of the overarching goals of the volume, to facilitate locally produced analysis of poverty and wellbeing, will only be achieved via the development of a community of skilled analysts in developing countries.

Training of individuals in the theory and practice of the measurement of welfare, as in this volume, is clearly necessary. However, it is not likely to be sufficient. The shape of the overall programme for monitoring and evaluating wellbeing is likely to strongly influence the rate of growth of local capabilities. Our experience, combined with the experiences presented in the case chapters, points to a series of design choices for programmes of measurement of wellbeing that can improve the quality of analysis and prospects for capacity-building often without a significant increment to the financial resources allocated to the task. We consider four choices—or maybe better areas of concern—in sections 18.2.1 to 18.2.4.

The frequency of consumption-based surveys and data is often insufficient. Conduct of a consumption survey once every five or six years, as is frequently the case in Africa, allows too much time to elapse between surveys. This is true purely for analytical reasons. As discussed in Chapter 16, nearly all the evidence for poor regions of developing countries indicates that the determinants of welfare are frequently volatile. Hence, measured welfare can be expected to shift fairly dramatically as a consequence of shocks. A smaller number of measurement points increases the difficulty of differentiating between long-term trends and short-term shocks. Is a decline in measured consumption poverty the result of a negative shock to welfare in the initial period and a positive shock to welfare in the subsequent period or the result of real gains registered as part of an ongoing development process? This separation of shocks and trends is more difficult with fewer data points.

Capacity-building considerations provide further impetus for more frequent periodicity of consumption surveys. If consumption surveys occur only once every five or six years, then the development of a functional community of analysts becomes very difficult. As emphasized, the task of consumption poverty measurement is challenging. In developing countries, demand for individuals with the talents to undertake this analysis is invariably high. Once the analysis of a survey is complete, these individuals will be pulled into other areas in the absence of a coherent and relatively continuous programme of welfare monitoring. After four or five years have passed, reassembling the core team members that conducted the previous analysis is practically impossible. As a result, the analysis of each survey is often undertaken with a brand new team and an associated need for external assistance.¹

Hence, both for the information provided and for capacity-building, countries should plan to conduct a household consumption survey approximately once every two to three years and explicitly consider the modes for the development of the necessary community of analysts. Moreover, the degree of complexity of the more frequent surveys proposed here needs careful reflection and must in practice be established in light of the specific financial and human resources available in concrete country contexts.

For the large majority of national statistical agencies in developing countries, obtaining an accurate picture of consumption for a household is challenging. Complicating the task with a series of additional objectives, such as sources of income (including a detailed look at agricultural production), migration dynamics, time use, and (occasionally) all of the above and more in a panel dataset, risks undermining the attainment of all of the targeted objectives due to proliferation of non-sample error.

The utility of data that permits comprehensive and cross-referenced analysis of household behaviour broadly defined is not in doubt. And answering many important questions requires adequate panel data. These advantages of more complicated data collection efforts are well established. At the same time, it is also perfectly clear that statistics agencies in developing countries frequently struggle with the challenge of adequately capturing consumption. Adding tasks may detract from attaining this basic objective.

A greater degree of collaboration between national statistical services and universities and/or other research/training-oriented organizations likely offers the most promising path forward. While national statistical services are in the process of developing and consolidating capacity to generate the fundamental statistics necessary to manage a country, including welfare statistics, advanced efforts to collect broader arrays of data, whether in panel or not, should be located in universities or other non-government organizations (NGOs).² Properly organized, these data collection and analysis efforts, housed outside of the national statistics agency, should also serve as human resource development centres for the statistics agency.

A basic challenge in undertaking meaningful consumption-based poverty analysis is the critical importance of prices, i.e. of coming to grips with how nominal consumption aggregates are converted into measures of real consumption over time and space as well as among different population groups. It is well established that aggregate inflation measures may not capture the rate of change in the prices actually faced by poorer households. Several of the case studies in this volume and the companion volume by Arndt, McKay, and Tarp (2016) illustrate this conundrum. A similar challenge exists in relation to ensuring methodological comparability over time in more general terms.

For example, Chapter 14 noted that in Tanzania, price inflation as measured by the household budget survey differs drastically from inflation rates derived from the published consumer price index (CPI) and the GDP deflator. It also made the observation that to complicate matters further the World Bank (2015) assessment of consumption poverty trends over the most recent period (2007–11/12) includes changes in the data collection methods employed in 2011/12 compared with all earlier surveys. The Bank assessment also took the opportunity to apply a series of methodological changes to the computation of the nominal consumption aggregate and the poverty lines. These differentials in effect render the analyses of the 2011/12 non-comparable with published analyses from 2007 and earlier; and in order to account for these differences, a series of steps were taken to revise 2007 data and calculations.

The revisions to the 2007 data are considerable. World Bank (2015: 2) reports as noted in Chapter 14 that ‘consumption per adult rose by almost one-third’. The poverty line was also adjusted upward substantially, leaving the measured poverty rate at the national level essentially at the same value as reported in previously published assessments. Nevertheless, the issue of achieving comparability and transparency in data and methods applied clearly dominates any analysis of consumption poverty trends over the 2007 to 2011/12 period.

Chapter 17 provides another illustration of comparability issues when estimating inequality. First, poorer people allocate much more of their budget to food, particularly basic foods. If the price of basic foods rises relative to other prices, the poor will be disproportionately affected. Second, poor people may tend to purchase goods in smaller quantities at higher unit prices than those who are better off. These specific effects, which have implications for inequality, are held in focus in Chapter 17.

On this background, household-specific deflators are estimated in Chapter 17 using fifteen surveys collected in six countries in the period 1999–2011 and analysed as an integral part of the UNU-WIDER Growth and Poverty Project. In some countries (Mozambique, Tanzania, Malawi, and Pakistan), measured inequality is higher when these two additional factors are considered. In other countries (Ethiopia and Madagascar), no differences are found. The analysis suggests that poverty estimation based on national accounts consumption means and estimates of inequality from consumption surveys should account for these two effects. Had Pinkovskiy and Sala-i-Martin (2014) accounted for these effects, the adjustments would, by and large, increase their estimated levels of poverty and reduce the rate of decline in poverty over time. The magnitude of the adjustment is—as noted in Chapter 17—country- and year-specific, stressing the need for better data and understanding of this area of poverty and inequality inquiry.

The desirability of a multiplicity of poverty measures, reflecting the numerous facets of wellbeing, is by now well established (Ravallion 2016; Alkire et al. 2015). Successful survey programmes, such as the demographic and health surveys that are regularly conducted across the developing world, provide a wealth of non-monetary indicators that furnish critical insights into any broad-based assessment of wellbeing. The first-order dominance methods (FOD) held in focus in Chapters 11–15 provide one useful means for deriving general conclusions across a series of indicators. As discussed in Chapter 3, the FOD approach identifies welfare differentials between populations using multiple binary welfare indicators without imposing weighting schemes or making assumptions about preferences for each indicator. Other techniques, such as those elaborated in Alkire et al. (2015), provide a series of alternatives. As revealed in the Tanzania case (Chapter 14), application of multiple techniques can enhance insight (in this case about the multidimensional welfare of two subgroups of children) even if the same set of indicators are employed.

When a variety of methods are employed, the result is a dashboard of basic welfare indicators that should provide reasonably detailed insight into a series of important dimensions of welfare (Stiglitz et al. 2009). This dashboard has been proposed as a substitute for ‘mashup’ indices that collapse a series of non-monetary indicators into a single number analogous to the consumption poverty headcount (Ravallion 2010). However, we see no reason why ‘mashup’ indices, such as the Alkire–Foster (AF) multidimensional index, should not form a part of a comprehensive dashboard of quantitative indicators, particularly if the limitations of each indicator in the dashboard are clearly elucidated.

It is perhaps useful to note that Ravallion (2010) defines a ‘mashup’ index as one that is not ‘informed by theory or practice’. The FOD, given its basis in theories of dominance, would not qualify as a ‘mashup’ index.

These elements—a regular household consumption survey, coming to grips with price trends and differentials, concerted efforts to monitor non-monetary indicators such as those in focus in demographic and health surveys, and a series of more advanced and pointed surveys including panel elements, likely conducted from a university base—provide ample raw material for the emergence of a healthy and active community of quantitative analysts.

The cost of collecting this information is not likely to be substantially different from the amounts currently allocated. While increasing the frequency of consumption surveys clearly increases costs, the associated call for reduced complexity reduces costs. In addition, the capacity-building gains associated with greater frequency apply equally well to data collection as to data analysis capabilities, opening the door to more cost-efficient as well as higher-quality data collection.³

If the quantitative information both informs and is informed by a similarly active qualitative research programme, many of the desiderata of an idealized welfare monitoring programme will have been fulfilled. Indeed, there exists a reasonably large international community of analysts who work largely within these confines.

There are, however, good reasons to extend this scope of activity. Welfare outcomes, such as the consumption poverty rate, are macroeconomic variables similar to gross domestic product (GDP). Triangulation of welfare outcomes from the surveys with national accounts, price data, trade data, weather outcomes and more provides at least two distinct benefits. First, if indicators derived from within the survey are broadly confirmed by indicators external to the survey, confidence in both indicators is enhanced. Second, recourse to a broader array of data can help to generate a much better understanding of the reasons for movement (or lack thereof) in welfare indicators, thus helping to distinguish, for example, between transitory shocks to welfare and more permanent shifts.

Drawing conclusions based on this triangulation across multiple sources is the focus of the companion volume to this book (Arndt, McKay, and Tarp 2016). As the techniques in focus in the present volume for country-focused poverty analysis become internalized, readers are referred to the companion volume for broader-based welfare assessment and triangulation.