![Sample our Global Development journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5929/TOC-Subject-Sample-2021-Global-Development.jpg"})
Abstract
This paper critically reviews conceptual and empirical problems issues with the United Nations Development Programme's two gender‐related indicators: the Gender‐related Development Index and the Gender Empowerment Measure. While supporting the need for gender‐related development measures, the paper argues that there are serious conceptual and empirical problems with both measures that limit the usefulness of these composite indicators. Where appropriate and feasible, the paper suggests modifications to the measures that address some of the identified problems.
Acknowledgements
The author wishes to thank Haishan Fu, Tim Scott and Susana Franco for constructive support and discussions. The author also wants to thank Geske Dijkstra and participants at a UNDP workshop that took place in New York in January 2005 for helpful comments and discussions, as well as Dana Schüler for support in making the alternative calculations and Marc Vothknecht assistance in preparing the manuscript. Funding from the Government of The Netherlands through the UNDP's Gender Thematic Trust Fund in support of this work is gratefully acknowledged.
Notes
1. There had been previous (and eventually discontinued) attempts UNDP to incorporate distributional issues into the HDI (see Anand and Sen, Citation2000).
2. There has been a proliferation of such measures in the academic and policy literature recently, including a Gender Equity Index by Social Watch (Citation2005), a Gender Gap Index by the World Economic Forum (Citation2005), the African Gender Status Index by the Economic Commission for Africa (Citation2004), the Relative Status of Women Index by Dijkstra and Hanmer (Citation2000), and the Standardized Index of Gender Equality by Dijkstra (Citation2002). See Klasen (Citation2006) for a discussion of these measures.
3. See, for example, Bardhan and Klasen (Citation1999), and Dijkstra and Hanmer (Citation2000) for individual contributions, and Schüler's contribution to this JHD special issue for a review.
4. I also want to emphasize that much of the added value provided by the UNDP through these measures is not necessarily the composite index itself but the gathering and presentation of the data on the various dimensions of human development, which allow for a much more careful assessment of human development in a comparative perspective than the HDI by itself (or GDP per capita or income poverty rates, for that matter) could provide.
5. See Grün and Klasen (Citation2003) for a discussion of that literature.
6. For more details, see Bardhan and Klasen (Citation1999), and UNDP (Citation1995, Citation2005).
7. See Klasen (Citation2006) for distinguishing between different types of gender‐related well‐being measures, including gender‐sensitive measures (such as the GDI and the GEM), gender‐disaggregated measures (e.g. male and female HDIs suggested later), as well as gender gap measures (see later).
8. In the Atkinson approach, the welfare “penalty” of inequality (based on existing aversion to inequality) is the measure of inequality itself. Here we describe it as the welfare penalty of inequality as this is really what it is. For details refer to Atkinson (Citation1970).
9. There is also some debate about the exact magnitude of this survival disadvantage, which is likely to be smaller in countries with high overall mortality. See Bardhan and Klasen (Citation1999) for a discussion.
10. For example, the World Bank's World Development Report 2006 Equity and Development takes the position that inequity exists whenever opportunities are curtailed based on ascriptive characteristics of people (such as sex, race, etc.) Using this approach, the biological survival disadvantage of males should be seen as an inequality of opportunities (World Bank, Citation2005). See also the Dijkstra contribution to this JHD special issue for an opposing view.
11. For example, the countries approaching parity in life expectancy are in parts of Sub‐Saharan Africa where females die at higher rates and at younger ages from AIDS, hardly a desirable state of affairs from a policy perspective.
12. If we killed the poor and uneducated, the GDP per capita would rise, as would the education and income components of the HDI.
13. Technically, what counts for the GDI is not the absolute gender gap in earnings, but the gender gap in the earned income indices for males and females, which are based on the log of earnings (under the assumption that only some consumption is human development related, and this portion falls with rising incomes; see Anand and Sen, Citation2000). In the case of Saudi Arabia, the female earned income index amounts to about 50% of the male index. The gap is so much lower as male and female incomes implying that a much higher share of the much lower female earnings is used for human development related consumption, while a lower share of the higher male earnings is used. This generates the fiction that each individual in the household spends their income on their own consumption priorities only, which does not square with the way household decisions are made. Also, even a gap of 50% in nutrition, housing, or clothing spending seems very large.
14. This problem was recognized by Anand and Sen (Citation1995) in their technical background note and they said that the gender gaps in earned incomes reflect gender gaps in agency. While this is likely to be the case, it still seems problematic to adjust a proxy of consumption of human development‐related goods by gender gaps in agency to arrive at gender gaps in consumption.
15. These indices of achievement are calculated by taking the difference between the actual achievement (e.g. male life expectancy) and the defined minimum achievement (e.g. 22.5 years in the case of male life expectancy) and dividing it by the difference between the maximum and the minimum achievement. For details, see Bardhan and Klasen (Citation1999) and UNDP (Citation2005).
16. It is not clear how one could allow for compensation of gender gaps in different directions within the current method of calculation in the GDI. Also, one should point that it not obvious that one should allow for full compensation of gender gaps in different directions. Full compensation would particularly lead to the undesirable conclusion that a country with dramatic but equally large gender gaps in opposite directions is as well off as a country with gender equality in all dimensions (see later).
17. The essence of the change was that the inequality aversion calculation is now applied after the log transformation of incomes (rather than applied to unadjusted incomes) to indicate that the declining human development benefit of incomes is not only true for average incomes (as in the HDI), but also for incomes earned by males and females. Dijkstra (Citation2002) and in her contribution to the special issue criticizes the change as it effectively dampens the gender gap in earned incomes and thus reduces the overall penalty for gender inequality. However, I argue that, even if one may see the resulting numbers as misleading (as discussed in Bardhan and Klasen, Citation1999, Citation2000), this change was consistent: if the GDI is designed to be an inequality‐adjusted HDI, one has to treat incomes in both measures the same, and only the procedure used since 1999 ensures this and also avoids some other inconsistencies of the previous method (see Bardhan and Klasen, Citation1999).
18. According to the HDRO, the reason for not reporting trends in the GDI is that there often is no new information on the underlying data, so that such trends would not be meaningful. While this is a serious consideration, I would favour that the series is regularly updated and made available, and those data for which no new information is available should be highlighted. This way, all changes to historical data would be incorporated and researchers and policy analysts could deal with the issue of missing data updates themselves.
19. We do not use the reported HDI but calculate an HDI based on the weighted average of the gender‐disaggregated data. It is very close to, but not identical to, the reported HDI due to rounding errors.
20. There is indeed evidence that gender gaps are smaller in rich countries (for example, World Bank, Citation2001), but substantial gaps remain.
21. For a discussion of how to change the relative importance of gender gaps, see Bardhan and Klasen (Citation1999) and Diskstra (Citation2002). The latter's proposal is to standardize the index in each component by the standard deviation. While this would ensure that in each year each component would have an equal influence on the final measure, such a standardization leads to problems of comparisons over time (and dependence on the number of countries included) as an improvement in education would have a different impact on the GDI depending on the standard deviation of the education index in a particular year, which is a rather undesirable property.
22. Overall labour force participation is, however, indirectly captured in the earned income component.
23. See also Dijkstra (Citation2002), who makes some similar points.
24. This is a point brought to my attention by Richard Leete in the online discussion, who also suggested to simply use income shares rather than income levels for the GEM. Dijkstra (Citation2002) made the same suggestion.
25. An example might illustrate the point. Take the example of Brazil shown in the annex to the Introduction of this JHD special issue. Based on earned incomes of males of $10,963 and females of $4704, the male index is shown to be 0.784 and the female index 0.643. The proposal would be simply to halve the difference (i.e. to make the male index 0.749 and the female index 0.678), based on the notion that this would more adequately reflect gender gaps in consumption.
26. One way to increase them would be to reduce the range of life expectancy values when constructing the indices of achievement (e.g. from current values of 22.5 to 82.5 for males and 27.5 and 87.5 for females to, say, 32.5 to 82.5 for males and 37.5 to 87.5 for females).
28. An example might illustrate the differences between the first and the third methods. Suppose a country has 600 females and 400 males, and 1% of all females (i.e. six females) sit in parliament, compared with 3% of all males (i.e. 12 males). Using the first method, the gender gap would be 0.33 (1% divided by 3%). Note that the total number of parliamentarians is 18, so the female share is 1/3 and the male share 2/3. Thus using the third method, the gender gap would be 0.5 (i.e. 1/3 divided by 2/3). As can be seen, this last method leads to a smaller gender gap as there are simply more females in the country. It seems more appropriate for an indicator of gender equality to consider whether males and females have equal chances to become parliamentarians, for which the first method would be best.