Social Research update logo
Issue 16 Spring 1997

Social Research Update is published quarterly by the Department of Sociology, University of Surrey, Guildford GU2 7XH, England. Subscriptions for the hardcopy version are free to researchers with addresses in the UK. Apply by email to

Occupational Gender Segregation

Robert M. Blackburn and Jennifer Jarman

Bob Blackburn is a Reader in sociology and Chairperson of the Sociological Research Group in the Faculty of Social and Political Sciences, and a Fellow of Clare College, in the University of Cambridge. Jennifer Jarman is an Assistant Professor in sociology in the Department of Sociology and Anthropology at Dalhousie University. They share a concern with social inequality in its various forms, particularly gender, ethnicity and social stratification. Both independently and in collaboration, they have written extensively in this area, especially in relation to formal and informal work. Currently they are collaborating in an extension of their work on gender segregation.

Key points

Occupational gender segregation has been at the heart of debates about gender inequality. High levels of segregation have been considered to be a significant factor in the discrepancy between the wages of women and men, to impose constraints on careers, and generally to be at the root of gender inequalities (e.g. Fox and Fox, 1987; Hughes, 1990; Reskin and Roos, 1990). The inequalities of segregation are primarily located in market employment, but they spill over into all aspects of life. Thus, the subject raises significant questions of social justice, of the efficient utilisation of human resources, of the structuring of labour markets, and of wider social aspects of work and family life. While there may be a wide consensus about the importance of the subject, there is less agreement about precisely what the term ‘segregation’ encompasses. This situation can lead to disputes over substantive matters. Furthermore, even when there is agreement about what should be measured, there are disagreements about how best to do this. Accordingly, our purpose here is to clarify some of the conceptual and measurement issues.

This Update first discusses the conceptualisation of segregation. It then presents a short discussion of the major indices that have been used to measure segregation, along with an abbreviated discussion of their shortcomings and our approach to overcoming these. Finally, we reconsider the terms vertical segregation and horizontal segregation, and suggest some new measurement possibilities.


The first step is to clarify the concept of segregation and distinguish it from concentration. Frequently segregation is used in a broad sense, as a term to include both segregation in a narrower sense and concentration. However, following James and Taueber (1985), our practice is to make a firm distinction between the two concepts. Accordingly

Segregation concerns the tendency for men and women to be employed in different occupations from each other across the entire spectrum of occupations under analysis. It is a concept that is inherently symmetrical....Concentration is concerned with the sex composition of the workforce in an occupation or set of occupations. Whereas segregation refers to the separation of the two sexes across occupations, concentration refers to the representation of one sex within occupations.’ (italics in the original, Siltanen et al, 1995: 4-5)

Clearly both the concepts are concerned with the distribution of men and women in occupations. It is easy to see why they have been grouped together; on the other hand it is clear that there are two distinct ideas here. If we were starting from scratch we might well use a different word for one of the senses of segregation. However, the terminology is long established, and here we would just like to point out that segregation (broad sense) includes both segregation (narrow sense) and concentration. We recommend using only the narrow sense, although it has to be acknowledged that not everyone makes these distinctions and this can lead to confusion and misunderstanding.

One important distinction between the narrow sense of segregation and concentration is that the former is symmetrical with regard to men and women while such symmetry is logically impossible for concentration.

Measures of segregation

Segregation measures have generally been expressed as summations over all the occupations under consideration. However, most measures, including all the most popular ones, can be usefully defined in relation to a 2 X 2 segregation table (see Figure 1). Here male occupations are defined as those where the proportion of men is greater than the proportion in the labour force as a whole, with a corresponding definition for female occupations. As well as being more compact, the 2 X 2 table has the important advantage of making clear just what is being measured. The limitations of non-symmetrical measures are then obvious.

Figure 1 Basic segregation table

Occupations Women Men
'Female' Ff Mf Nf
'Male' Fm Mm Nm


N is the total labour force
Nf is the total number of workers in ‘female’ occupations
F is the number of women in the labour force
Ff is the numer of women in ‘female’ occupations,
and so on

In relation to the table we have the following main formulae:

Sex Ratio Equation for Sex Ratio
WE Index Equation for WE Index
IP Index Equation for IP Index
Index of Dissimilarity Equation for Index of Dissimilarity

The Index of Dissimilarity can be seen to be a difference of the column proportions in the table. IP and WE are weighted versions of ID. SR is a weighted version of the difference of row proportions. (It can also be expressed as a more complicated weighting of ID). These weightings vary with the gender composition of the labour force but none is in any way relevant to the degree of segregation. Furthermore, SR and WE are not symmetrical; they are the female versions of indices for which there are corresponding male versions (obtained by reversing M and F in the weightings). These female versions appear to show remarkable declines in British segregation levels since 1950 (over 20 per cent), while the male versions show corresponding increases. We can standardise SR and make the male and female versions coincide by dividing by the irrelevant weighting N/F. We then obtain SR*, the difference of row proportions in the table. In contrast to SR, SR* actually shows a modest increase after 1951. Similar standardisation of IP and WE gives ID, the difference of column proportions. ID also shows a modest change, but in this case a decrease (Blackburn et al 1993, 1995).

Although both ID and SR* are symmetrical and free from distorting weightings, neither is satisfactory. This has been pointed out often with respect to ID (e.g. Duncan and Duncan, 1955; Tzannatos, 1990; Watts, 1992; Blackburn et al, 1995). SR* has comparable weaknesses, but as no one has advocated its use, these have not been discussed. Essentially the problem is that there are no statistics of association which are independent of the marginal totals of a table. It is true that each difference of proportions has what is known as ‘marginal independence’ with respect to one pair of marginal totals, but this only applies to the socially unlikely process of the relevant row or column being multiplied throughout by a constant. In the real world, social processes are quite different and in particular people move among occupations according to demand for labour.

The marginal matching statistic, MM, is designed to meet these problems. The segregation table is modified to be symmetrical at all times. The number of workers in female occupations is ‘matched’ (made equal) to the number of women in employment, and similarly for male occupations (Nf = F and Nm = M). Several statistics of association, including the two differences of proportions and tauB, now coincide. This resulting measure is known as MM. Because the relativities of the marginals (F/M : Nf/Nm) remain constant, MM is not affected by changes in the gender composition or occupational structure of the labour force, but only by the distribution of men and women across occupations, i.e. by the level of segregation. (For a fuller explanation see Siltanen et al, 1995; Blackburn et al, 1993). Despite a tendency to overweight the extremes, the gini coefficient is a useful alternative, as explained below.

Segregation and inequality

Frequently segregation, in the stricter, more narrow sense, is regarded as evidence of inequality, or even as directly measuring inequality. This is clearly not correct, as concentration of the sexes in different occupations does not necessarily mean that either is disadvantaged. We can address the issue of inequality through a distinction between the vertical and horizontal dimensions of segregation.

Here it is necessary to clarify some further confusion in current approaches. What we have discussed so far was originally simply called ‘segregation’, but is now frequently referred to as ‘horizontal segregation’ despite the fact that it has an integral vertical component. For example, in this usage ‘horizontal’ segregation refers to the extent to which men and women are located in different occupations such as company directors and childminders; yet on any conventional vertical measure, the company director has a higher position. Less obvious, but still important are the differences among occupations which are closer in stratification level such as building labourer and nursery attendant. ‘Horizontal’ segregation, as conventionally defined, is not strictly horizontal.

Turning to the ‘vertical’, we find a somewhat different sort of problem. The classic definition states:

Vertical occupational segregation exists when men and women both work in the same job categories, but men commonly do the more skilled, responsible or better paid work. For example the majority of school heads may be men while the majority of teachers are women, the majority of hospital consultants may be men while the majority of nurses may be women. (our emphasis) (Hakim, 1981: 521).

Here there is a clear vertical dimension, but the problem is that we are limited in our consideration of vertical inequalities by the restriction to comparisons within very limited areas of the labour force (workers within education to other workers within education, workers within the health service to other workers within the health service, and so on). So an exploration of a vertical dimension ends up by illustrating small amounts of inequalities across the occupational spectrum, but with no consistent vertical dimension emerging. We also want to know whether, for example, the female-dominated occupations in health are higher than the male-dominated occupations in education. The vertical dimension of inequality covers all occupations, but the usual definition of vertical segregation does not. In this conception, vertical segregation is vertical, but is severely restricted in that it is only a dimension in a very limited sense. Hakim (1981) was aware of this limitation but at the time saw no solution.

We recommend redefining vertical and horizontal segregation in such a way as to capture more adequately vertical and horizontal dimensions. The key point of our definitions is that they are in line with every-day and mathematical conceptions. Having recognised that conventionally defined ‘horizontal’ segregation includes both horizontal and vertical components (in our sense), we now conceptualise it as ‘overall’ segregation - the resultant of horizontal and vertical components. This can be visualised as forming the hypotenuse of a right-angle triangle. The other two sides now represent the vertical and horizontal dimensions of segregation (See Figure 2). Clearly the three sides are related, so that the length of one side can be deduced from knowledge of the lengths of the other two. Overall segregation (O) may be expressed as the vector sum O = V + H (V and H being the vertical and horizontal components respectively).

Figure 2 Dimensions of segregation

Figure 2

By conceptualising overall segregation in this way we have incorporated our new definitions of vertical and horizontal segregation. Now vertical refers to a single dimension of inequality, with advantage running from low to high. Aspects of vertical inequality include social stratification, power, skill and earnings. All are related and may be taken as indicators of social advantage. Information from all occupations can be included in order to assess the amount of vertical inequality in an occupational structure, as opposed to a situation where information could only be obtained from small sections of the occupational structure. This integrates the study of gender inequality in occupations into the more general area of the study of social inequalities.

Somers’ D provides an appropriate measure of vertical segregation. Occupations are ordered on the vertical dimension, and the ‘independent’ variable is gender, with just the two categories of male and female. The formula is:

Equation for Somer's 


where P is the number of consistently ordered pairs and Q the number of inconsistently ordered pairs. In this case, P is all pairs of a man and a woman where the occupation of the woman contains a higher proportion of workers who are women than does the man’s occupation, i.e. the ordering is consistent with segregation. Thus P - Q is the same as the numerator in other measures such as gamma and tauB. In the 2 X 2 case of Figure 1, D = (FfMm - FmMf)/FM = Ff/F - Fm/M, a difference of proportions (= ID), but here we are more interested in including the vertical position of every occupation.

The comparable measure of overall segregation is the maximum potential value of D given the occupational gender distribution, i.e. when occupations are so ordered that P - Q is a maximum. This is when the occupations are ordered from most female to least female, and D is then equal to the gini coefficient (Blackburn et al 1995).

Horizontal now refers to segregation at the same level: it is the extent to which men and women are in different occupations without this giving an occupational advantage to either sex. There is a single horizontal dimension along which all occupations are located. Occupations may, of course, be at different vertical levels, just as the vertical dimension measures occupations independently of their different horizontal positions. This definition of horizontal, together with the meaning we have given to vertical segregation, not only brings the study of segregation within the orbit of more familiar language, but opens the way for systematic analysis of their relationship and combination in overall segregation.

The same degree of overall segregation can be made up of different proportions of vertical and horizontal segregation, as illustrated in Figure 3. Here country A has a much lower level of vertical segregation than country B, indicating a lower level of gender inequality. The Scandinavian countries are among the countries with the highest levels of overall segregation. This has usually been seen as surprising, as these countries are generally regarded as amongst the most socially egalitarian, and this egalitarian image includes gender. In fact the situation is not surprising if the high level of segregation is composed mainly of a large horizontal component. Where there is a high level of horizontal segregation there is more opportunity for women to rise to the top in the ‘female’ careers, thus reducing occupational inequality. Accordingly we hypothesise that Scandinavian countries are rather like country A in Figure 3. Kuwait and Bahrain also have very high levels of overall segregation, but do not have the Scandinavian reputation for social equality. Our hypothesis is that they are more like country B.

Figure 3 Different segregation patterns

Figure 3

The analysis of segregation in terms of its vertical and horizontal components has not previously been attempted in this manner. Yet such an analysis is necessary if we are to understand the relative significance of inequality and difference.


Blackburn, R.M., J. Jarman and J. Siltanen. (1993) ‘The Analysis of Occupational Gender Segregation Over Time and Place: Considerations of Measurement and Some New Evidence’, Work, Employment and Society, 7(3): 335-362.

Blackburn, R.M., J. Siltanen and J. Jarman. (1995) ‘Measuring Occupational Gender Segregation: Current Problems and a New Approach’, Journal of the Royal Statistical Society, Series A, 158: 319-331.

Duncan, O.D. and B. Duncan (1955) ‘A methodological analysis of Segregation Indices’, American Sociological Review 20: 210-217.

Fox, B.J. and J. Fox (1987) ‘Occupational gender segregation of the Canadian Labour Force, 1931-1981’ Canadian Review of Sociology and Anthropology 24: 374-397.

Hakim, C. (1981) ‘Job segregation: trends in the 1970s’, Employment Gazette (December): 521-529.

Hughes, K. (1990) ‘Developments in the Non-traditional Employment of Women and Men in Canada, 1971-1981’, Working Paper No. 1, Cambridge: Sociological Research Group.

James, D.R. and K.E. Taueber. (1985) Measures of Segregation in N.B. Tuma (ed.) Sociological Methodology. San Francisco: Jossey-Bass, 1-31.

OECD (1980) Women in Employment. Paris: OECD.

Reskin, B.F. and P.A. Roos. (1990) Job Queues, Gender Queues. Philadelphia: Temple University Press.

Siltanen, J., J. Jarman and R.M. Blackburn. (1995) Gender Inequality in the Labour Market, Occupational Concentration and Segregation. Geneva: International Labour Office.

Tzannatos, Z. (1990) ‘Employment segregation: can we measure it and what does the measure mean?’ British Journal of Industrial Relations 28 (1): 105-111.

Watts, M. (1992) ‘How should occupational segregation be measured?’ Work, Employment and Society 6: 475-487.

Social Research Update is published by:

Department of Sociology
University of Surrey
Guildford GU2 7XH
United Kingdom.

Telephone: +44 (0) 1 483 300800
Fax: +44 (0) 1 483 689551

Edited by Nigel Gilbert.

Winter 2004 © University of Surrey

Permission is granted to reproduce this issue of Social Research Update provided that no charge is made other than for the cost of reproduction and this panel acknowledging copyright is included with all copies.