Vulnerability to poverty: an empirical comparison of alternative measures

Abstract

This article compares empirically the several measures of individual vulnerability to poverty proposed in the literature, in order to understand which is the best signal of poverty that can be used for policy purposes. To this aim, the Receiver Operating Characteristic (ROC) curve, the Pearson and Spearman correlation coefficients are used as precision criteria. The results show that two groups of indexes can be identified, high and low performers, and, among the former, that proposed by Dutta et al. (2011) is the most precise.

Keywords:

• poverty
• risk
• vulnerability
• receiver operating characteristic curve

JEL Classification::

• D63
• I32

Acknowledgements

This article was partly based on work carried out during a visiting to the European Centre for Analysis in the Social Sciences (ECASS) at the Institute for Social and Economic Research, University of Essex and supported by the Access to Research Infrastructures action under the EU Improving Human Potential Programme. Data from the British Household Panel Survey (BHPS) were supplied by the UK Data Archive. Neither the original collectors of the data nor the archive bear any responsibility for the analysis or interpretations presented here. I thank an anonymous referee, Giorgio Brunello, Conchita D'Ambrosio, Isabella Procidano, Luca Salmasi, the colleagues who took part in the presentation at the University of Padua and the participants of the Sixth Winter School on Inequality and Social Welfare Theory (University of Verona) for their comments and suggestions.

Notes

¹ For a more detailed description of the data, see http://www.iser.essex.ac.uk/bhps.

² The equivalence scales used are the square root of the household size, as well as the Oxford scale and the Organization for Economic Cooperation and Development (OECD)-modified scale, and all values have been expressed in real terms (deflated to January 1998 prices).

³ I selected those households that were present in the panel for at least three periods, with observations in the years 2004 and 2005, since I compare the different vulnerability measures computed for the year 2004 with the poverty status in 2005.

⁴ For a more detailed description of the data, see http://www.diw.de/en/soep.

⁵ Also in this case three different equivalence scales are used, to take into account different degree of equivalence elasticity, i.e. different economies of scale within the household. Real income is deflated to 2005 prices.

⁶ Real equivalized net income is deflated to 1991 prices.

⁷ The data are collected every 2 years from 1987, with an exception for the year 1998 when information was gathered 3 years after 1995.

⁸ Even if the Bank of Italy provides data from 1977, the longitudinal component starts only from 1987, but I restrict the time period analysed to 1989–2004 because, as already pointed out in the literature (Biagi et al., 2009), two few households remain in the panel from 1987 to 1989.

⁹ See http://www.bancaditalia.it/statistiche/indcamp/bilfait for a more detailed description of the data.

¹⁰ I will use the following notation: PC = Prittchett, Suryahadi and Sumarto (2000) and Chaudhuri (2003); C = Chaudhuri (2003); FGT = Foster, Greer and Thorbecke; CD = Calvo and Dercon (2005); DFM = Dutta, Foster and Mishra (2010).

¹¹ In his consumption generating function, Chaudhuri (2003) assumes that the elements of X _h,t are contemporaneously uncorrelated with e _h,t but allows for potential correlation between X _h,t and lagged consumption shocks. If this is the case, the standard within-estimator cannot be used, that is the reason why Chaudhuri (2003) uses first differences of consumption and instruments the changes in the predetermined variables using lagged changes and levels of the same variables. In this case, if income is used rather than consumption, the correlation between X _h,t and lagged shocks should not be an issue.

¹² As in Brandolini et al. (2010), if the 50% of the median equivalized disposable income poverty threshold is used, for Italy I find that about the 12% of households is poor.

¹³ The corresponding FNs are, respectively, 15%, 20% and 25%.