An Empirical Validation Method for Narrowing the Range of Poverty Thresholds

Determining the threshold separating poor from non-poor populations is one of the most influential choices when measuring poverty. Commonly used selection criteria leave considerable room for discretion or are not appropriate as academic standards. This vacuum in academic guidance leads to arbitrary and/or ideologically driven choices. This can greatly influence the measurement of a societal phenomenon, and thus research and public policy decisions over a period that extends well beyond the mandate of those making that threshold decision. This paper sets out an empirical validation method that contributes to reducing the range of thresholds and thereby aids decision-makers in making that normative decision. Our method uses an absolute concept of empirical validity and requires that the microdata for measuring poverty hold additional information closely associated with poverty. The method builds on insights from theory on measurement error that, for any given threshold, some persons are wrongly identified as poor (false positives) and others are wrongly identified as not poor (false negatives), and that the reduction of one error can only be attained by increasing the other. Our method uses the additional microdata to disaggregate the population into (likely) false positives and (likely) false negatives and analyzes marginal changes in this composition as the poverty threshold becomes stricter. Using Canadian data, we show that this approach substantially narrows the range of thresholds for two unidimensional poverty indicators, namely a material deprivation and an income poverty indicator. The underlying principle of the method extends to other (unidimensional) social indicators. The analysis itself can also serve other purposes, such as deepening our understanding of poverty and the cost-effectiveness of policies to reduce it.