![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
Abstract
The paper reviews different approaches to, and current knowledge of the measurement of occupational segregation, using the case of gender segregation. It shows that most popular segregation ‘indices’ are actually statistics of association in a 2x2 table, often with distorting weightings. The dimensions of segregation comprise a vertical dimension measuring inequality and an orthogonal horizontal one measuring difference without inequality. Together, the dimensions make up segregation as generally understood; so segregation and its dimensions require consistent measurements. Conditions for suitable measures are considered, and the limitations of the various measures noted. The alternative conception of segregation, where all occupations are treated as though they were the same size, is shown to be seriously flawed. The most useful measures are selected and shown to be related as Lorenz curves. Since all segregation measures vary with the number of occupations considered, standardisation on 200 occupations is introduced for the chosen measures.
Acknowledgement
I am indebted to Jennifer Jarman for her assistance and advice. I am grateful for ESRC support in Research Grant RES-000-22-2779.
Notes
1. Segregation is often used to indicate total separation (the sheep from the goats, etc). However, the technical term, as used here, refers to a variable ranging over values from 0 to total separation (from 0 to 1, or sometimes 0–100%). A population comprised solely of monks and nuns, for example, would have total segregation and so a value of 1.
2. See Charles and Grusky (Citation2004) and Anker (Citation1998) bibliographies.
3. Here we are concerned with segregation of two groups. To explore the separation of several groups at once requires a different methodology, see for instance Elliot (Citation2005), Silber (Citation1992).
4. As a measure of inequality, the vertical dimension cannot be gender symmetrical, unlike the resultant segregation.
5. A measure exists composed of the type of tenure of the house, whether centrally heated, the density of people per room and the number of cars available to the household (Blackburn, Dale and Jarman, Citation1997, p. 255). The measure is based on the data available in the 1991 British Census.
6. It was actually introduced in the working paper Blackburn, Siltanen, and Jarman (Citation1990), but the journal article was not published till 1993.
7. IP may be expressed as IP = 2α/N, where α is the difference between observed and expected values in the table (expected being the values if there were no relationship). Unlike the other measures this appears to be independent of the marginal totals. However, this is an illusion as the possible size and the significance of α depend on the expected values, and so on the marginal totals.
8. For precise matching the cutting point may require splitting an occupation (Siltanen et al., Citation1995) though with many occupations this is hardly necessary. This also applies to determining p and Ni above.
9. These values of D c and D r should not be confused with ID and SR∗ as they are based on a different table.
10. Lieberson’s (Citation1976) index of net difference is the same as Somers’ D, and so the Gini coefficient.
11. Recall that ID has traditionally been expressed as summation of terms for every occupation and MM can be similarly expressed.
12. Exceptions include Blackburn and Jarman (Citation1997, Citation2004, Citation2006), Blackburn et al. (Citation2001), Brooks, Jarman, and Blackburn (Citation2003).
13. For extensive discussion of such criteria see James and Taeuber (Citation1985), Siltanen et al. (Citation1995). The present brief discussion updates these accounts.
14. For technical limitations see also Siltanen et al. (Citation1995), Watts (Citation1998a, Citation1998b).
15. Strictly the ordering of occupations to create the axes is not possible with no segregation, but any ordering would suffice. The diagonal is the limiting position of the curve as segregation tends to zero.
16. The number of occupations must be less than or equal to the number of workers. If occupations equaled workers there would be total segregation (triangle OAB) while more occupations is impossible. In practice 200 or more occupations gives a good approximation to the curve.
17. A detailed discussion of the estimation process for standardizing MM may be found in Jarman, Blackburn, Brooks, and Dermott (Citation1999, Appendix).
18. We should note that this meets three basic criteria: MM E = 0 when n = 1; MM E increases as n increases; and MM E → 1 as n → ∞. The third criterion here is not precisely what is required, but the difference is negligible.
19. Earlier standardization for Canada and UK were estimated on limited data. The formula given here is more soundly based and so is to be preferred. The result is not greatly different but is more precise.