Jump to comment:

• Methodological pitfalls in the measurement and decomposition of socioeconomic inequality of smoke exposure
  Guido Erreygers
  Published on: 28 February 2017

• Published on: 28 February 2017

  Methodological pitfalls in the measurement and decomposition of socioeconomic inequality of smoke exposure

  • Guido Erreygers, Professor

  NOT PEER REVIEWED I would like to point out a few disturbing inaccuracies in the methodology and interpretation. Since the health variable is binary, the authors apply "Wagstaff's correction" to the Concentration Index. This is a perfectly legitimate decision, but the authors mistakenly suggest that this correction can be applied to both the relative and the absolute version of the index, yielding two normalized indices. In fact,...

  Show More

  NOT PEER REVIEWED I would like to point out a few disturbing inaccuracies in the methodology and interpretation. Since the health variable is binary, the authors apply "Wagstaff's correction" to the Concentration Index. This is a perfectly legitimate decision, but the authors mistakenly suggest that this correction can be applied to both the relative and the absolute version of the index, yielding two normalized indices. In fact, there is only one Wagstaff index, which can be expressed as W = RC/(1-m) = AC/[m(1-m)], with m = prevalence. All the results which are presented under the heading "Absolute concentration index" in Table 2 and as "AC" in Table 3 are therefore irrelevant. These results refer to the index mW which nobody has ever used in the literature. It follows that the decomposition formula (6) of the paper is redundant. Similar remarks hold for the application of the alternative "Erreygers correction". The correction leads to one index, not two as suggested by the authors. The index can be written as E = 4mRC = 4AC. As a consequence, the results presented under the heading "Absolute concentration index" in Table A.1 of the supplementary online material are irrelevant. They refer to the index mE, which is not the one defined by Erreygers. After a brief comparison of the Wagstaff and Erreygers indices the authors conclude that "the calculated RC and AC informed qualitatively similar inference" (p. 9). Since E = 4m(1-m)W, the values of E and W clearly will be positively correlated, and the more strongly so if the variation in prevalence is limited. For the whole population, the (unweighted) average of m is 0.28, and the standard deviation 0.19. As expected, the correlation of E and W is high, but not perfect: the coefficient of correlation is 0.8815, and the rank correlation coefficient 0.8726. Nevertheless, countries such as Liberia, Bangladesh and Benin, make large jumps in the rankings if inequality is measured by one index rather than the other. At least for these countries, it may be doubted whether the results are "qualitatively similar". When it comes to the calculation of the between-group and within-group shares, it makes no difference whether the Wagstaff or Erreygers index is used. The main problem here is that any rank-dependent index is not subgroup decomposable. There is always a residual term, which may be quite large. Because of this, the decomposition results presented in the paper are unreliable.

  Conflict of Interest:

  None declared

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top