Weighting & Robustness
In constructing multidimensional measures, it is possible to apply weights in aggregating variables (1) within one dimension; (2) across dimensions; and (3) across people.
At each point of aggregation, the parameters that define the marginal contribution of each indicator, dimension, or individual to the overall well-being (or deprivation) need to be determined, taking into consideration also the possible interconnections among them.
OPHI research considers a number of techniques to set weights for multidimensional measures. These include:
- Participatory and expert-based approaches.
- Survey-based methods to elicit directly people’s preferences (standard gamble, visual analogues, and willingness to pay) or making use of subjective-well-being surveys.
- Statistical methods (factor analysis, principal component, latent variable models, and data enveloping analysis).
This work seeks to identify the technical and conceptual strengths and weaknesses of each weighting technique, to clarify which techniques are best suited to set weights in which contexts of multidimensional poverty measurement, and how such weights are to be verified.
In 2008 OPHI held a workshop on weighting in multidimensional measures. For the most up to date products of this workshop, read OPHI’s working papers on this topic. The original materials from the workshop on weighting in multidimensional measures are also available on our workshops page.
Robustness tests assess the sensibility of results to different weights within a given technique and across methods.
OPHI is developing techniques to analyse the conditions under which multidimensional welfare or poverty indicators rank social groups consistently, and the degree of sensitivity of these indicators to changes in their key components. These may include the weights attached to dimensions, poverty lines, k-cutoffs, choices of indicators and any sub-indicator, normalisation or aggregation technique.
In 2009 OPHI held a workshop on robustness tests in multidimensional measures. For the most up to date products of this workshop, read OPHI’s working papers on this topic. The original materials from the workshop on robustness tests in multidimensional measures are also available on our workshops page.| Trackback