Abstract and Keywords
This synthesis chapter seeks to draw general lessons from the case studies presented in the book. It does not include a review or summary of each chapter. Instead, we revert to themes that emerged from earlier in the book. Specifically, we argue that appropriately assessing living standards is challenging, and focus on the different nature of the challenges for consumption poverty line estimation and multidimensional poverty measurement. The six case studies uniformly indicate that the process of drawing appropriate absolute poverty lines is not straightforward and cannot be done mechanically. This is so principally due to five key factors: heterogeneity, volatility, vulnerability, data, and theory. The first three factors are usefully grouped together. In all of the case countries considered, poor people are heterogeneous, frequently live in environments with strikingly high levels of volatility, and are, almost by definition, vulnerable to shocks.
This synthesis chapter seeks to draw general lessons from the case studies presented in Part II. It does not include a review or summary of each chapter. For this, the reader is referred to the chapter abstracts. Instead, we revert to themes that emerged from Part I. Specifically, we argue that appropriately assessing living standards is challenging, and we focus on the different nature of the challenges for consumption poverty line estimation and multidimensional poverty measurement. Conclusions are provided in Chapter 18.
16.2 Absolute Poverty Lines
The six case studies uniformly indicate that the process of drawing appropriate absolute poverty lines is not straightforward and cannot be done mechanically. This is so principally due to five key factors: heterogeneity, volatility, vulnerability, data, and theory. The first three factors are usefully grouped together. In all of the case countries considered, poor people are heterogeneous, frequently live in environments with strikingly high levels of volatility, and are, almost by definition, vulnerable to shocks. The upshot of these combined factors is a high level of variation in living standards, particularly at lower levels of aggregation.
In Madagascar, an ongoing political crisis strongly and negatively impacted living standards in urban zones while the overall rural poverty rate stagnated at a high level (see also Stifel et al. 2016). Climate shocks powerfully affect welfare, negatively and positively, particularly in rural zones. In large countries, such as Ethiopia, Mozambique, and Tanzania, climate shocks can be (p.270) geographically concentrated with some domains experiencing positive shocks while other domains experience negative ones. The Mozambique case illustrates that the combination of abnormally high global food and fuel prices with local negative weather shocks can be particularly powerful. This underlying variability of living standards (across space, through time, by household characteristics, by nature of the shocks experienced, etc.) complicates essentially all aspects of the analytical task from sample design to the process of analysis.
The next challenge relates to data. Household consumption information is far too expensive to collect for every household in a population. Hence, all countries rely on randomly selected samples. Samples inherently limit the scope for specificity (as emphasized in Chapter 4) and add sample variation to the fundamental variation in welfare outcomes discussed in the preceding paragraph. More perniciously, datasets in developing country settings almost invariably suffer from a fairly high level of non-sample error. While problems with units are particularly common (e.g. Malawi), they are by no means the only problem encountered. Unfortunately, there is no substitute for knowing the data well, cross-checking with other sources, and making careful deliberate choices.
Finally, while consumer theory provides an elegant grounding for welfare analysis, it provides little firm guidance across a vast array of practical choices. The consistency versus specificity debate discussed in Chapter 2 is just one salient example. There is no substitute for careful consideration of the circumstances. For this reason, the case country applications almost invariably modify the PLEASe code in order to handle local specificities. And, because country circumstances vary greatly, the modifications imposed vary substantially across the country cases as well.
With these challenges recognized and with the allocation of an appropriate level of effort, the results of the studies presented here can be highly informative. For example, in the case of Ethiopia, discussed in Chapter 5, official results, which depict substantial declines in poverty, are largely confirmed. For Madagascar, Chapter 6 finds that rural poverty rates were likely more stable than official estimates, which showed an aggravation of rural poverty. In Malawi, the authors of Chapter 7 argue that poverty rates likely fell by more than the official estimates indicate between 2004/5 and 2010/11. In Chapter 8, official estimates of poverty trends through time in Mozambique are confirmed and sensitivity of the regional poverty profile to methodological choices is discussed. The analysis of Pakistan in Chapter 9 provides the greatest divergence between official and revised estimates. The chapter shows that consumption poverty has likely been increasing through time rather than declining as official estimates suggest. Chapter 10 argues that, in Uganda, an updated bundle is required, resulting in an altered regional poverty profile and generally higher poverty levels corresponding to revised basic-needs baskets.
(p.271) We argue that these are all exceedingly useful insights derived with due consideration to country circumstances and available data. As emphasized in Chapter 1, these insights refer to private consumption possibilities with a particular focus on households living ‘near’ the reference welfare level targeted by absolute poverty lines. We argue as well that private consumption results are, on their own, not sufficient to make fully general statements about the evolution of wellbeing. In addition, as emphasized in the companion volume to this book (Arndt, McKay, and Tarp 2016), a broader analysis can help to develop a coherent narrative. Ideally, this narrative both helps to explain why living conditions are evolving in the ways observed and enhances confidence in overall conclusions as observations across multiple datasets and multiple facets of welfare become mutually reinforcing. Accordingly, the reader is referred to Arndt, McKay, and Tarp (2016) where narratives for five of the six countries with PLEASe applications (Pakistan is the exception) are developed.
In developing these narratives, multidimensional measures can be usefully employed. Furthermore, it is sometimes the case that the data for poverty measurement of private consumption possibilities are inadequate or would require enormous efforts to get into shape for rigorous analysis. In other instances, a focus on multidimensional analysis appears to be a more promising path for advancing knowledge and the state of debate. These observations lead us to our cases where first-order dominance (FOD), operationalized by EFOD, was in focus.
16.3 First-Order Dominance
Multidimensional methods, such as the first-order dominance approach in focus here, are also complementary to the drawing of absolute poverty lines in terms of methods. In many ways, the practical implementation of many multidimensional methods is substantially more straightforward than the evaluation of private consumption. This divergence begins with data. Frequently, the indicators employed for multidimensional analysis are much easier to observe than household consumption patterns and, as a consequence, very plausibly less subject to non-sample error. For example, important indicators of asset quality, such as the type of roof on a house, are typically quite easy to observe. In contrast, consumption poverty requires information on consumption values, quantities, and estimated prices at detailed product levels. Furthermore, in many households, consumption expenditures are diffused across multiple members. In some cases, one or more household members do not wish to disclose categories of expenditure to other household members, substantially complicating the task of the enumerator.
(p.272) It is not the case that all possible multidimensional measures are easy to observe. For example, one could employ the household’s consumption poverty status as an indicator in an FOD analysis. Obviously, then all of the difficulties associated with estimating absolute poverty lines also apply to the multidimensional analysis. Nevertheless, the five case countries considered develop a series of reasonably robust and informative indicators even in relatively data-poor environments, such as the Democratic Republic of the Congo and Nigeria.
In addition, once the series of indicators has been selected and appropriate cut points between deprived and non-deprived populations have been determined, the implementation process that follows tends to be much more routine. This feature of FOD analysis helpfully puts the accent on the choice of indicators and choices for cut points within indicators. For example, suppose that, as in Ghana and Tanzania, we would like to use anthropometric data to develop one of a number of indicators for considering the welfare of children aged zero to 60 months. If we are to employ only one anthropometric indicator, then the standard measures associated with stunting, wasting, and underweight must be combined. We might define a child as deprived if it is considered one or more of stunted, wasted, or underweight, using standard definitions.
While certain technical considerations do enter the appropriate choice of indicators and cut points (see section 4.3.1 in Chapter 4), it is imminently possible to engage in a broad debate across stakeholders with respect to the choice of appropriate indicators and cut points. There is no general rule that necessarily prioritizes one particular indicator over another, and cut points that define one subgroup as deprived and not deprived for particular indicators are clearly open for discussion in the right contexts and circumstances. This has the salutary effect of opening the potential for a reasonably inclusive analytical process. In contrast, the potential for an inclusive process is much more circumscribed with respect to consumption poverty. The efforts of this book to lower the barriers to entry to poverty analysis notwithstanding, the technical choices involved in poverty line estimation will remain exactly that, technical choices, with more limited scope to benefit from broad-based inputs.
While data collection for multidimensional analysis, such as FOD, and its subsequent analytical implementation are frequently more straightforward than poverty line estimation, the interpretation of results can be somewhat less direct. The idea that a poverty line can divide households into poor and non-poor groups on the basis of total private consumption is fairly easy to grasp and has been around for a long time.1 The concepts driving FOD results, while (p.273) reasonably intuitive, are not nearly as clear-cut. Work must be done to explain what is meant by first-order dominance and by ‘net probability of domination’.
There is also a need to build up a corpus of experience in order to properly interpret results. For example, is a 50 per cent probability of advance in five key indicators at the national level over a five-year period a good or a bad performance? Based on the results for the five cases presented here, a 50 per cent probability of advance would appear to be a reasonably favourable result. This is so due to the stringency of the FOD criteria, which require advance across all indicators that is broadly shared across the full population.
Finally, particularly with respect to spatial comparisons, mainly indeterminate outcomes (A does not net-dominate B, nor does B net-dominate A) require some further analysis in order to determine the root of the indeterminacy. Welfare comparisons between two regions may yield indeterminate outcomes because the two regions are very similar. It could also be because they are very different with one region lagging in one indicator and the other region lagging in a different indicator. It is normally not difficult to ascertain the nature of an indeterminate outcome—a look at the mean values of the indicators by spatial domain is often (but not always) sufficient; but this analytical work needs to be done and then explained.
Overall, we find that the cases illustrate that FOD analysis, as implemented by EFOD, represents a powerful addition to the analytical toolkit. Similar to many other multidimensional approaches, FOD shares the desirable properties that data challenges are frequently relatively mild and implementation is straightforward. These two features allow for a focus on choice of indicators and cut points. With some technical guidance, a relatively open and inclusive process involving key stakeholders in choices of indicators and cut points is imminently possible. At the same time, effort is required to adequately interpret and explain FOD results. This is especially true at this point in time when experience with multidimensional measures in general and FOD in particular is relatively limited.
Arndt, C., A. McKay, and F. Tarp (eds) (2016). Growth and Poverty in Sub-Saharan Africa. Oxford: Oxford University Press.
Stifel, D., T. Razafimanantena, and F. Rakotomanana (2016). ‘Utility-Consistent Poverty in Madagascar, 2001–10: Snapshots in the Presence of Multiple Economy-Wide Shocks’, in C. Arndt, A. McKay, and F. Tarp (eds), Growth and Poverty in Sub-Saharan Africa. Oxford: Oxford University Press, 370–92.
(1) While the poverty headcount (e.g. the share of the population living below the poverty line) is straightforward to grasp, the concepts behind the poverty gap and squared poverty gap are less intuitive and less well understood across the broad community that uses/consumes poverty analysis.