DISABILITY, GENDER AND THE LABOUR MARKET IN WALES
Melanie K. Jones, Paul L. Latreille and Peter J. Sloane
WELMERC, Department of Economics, University of Wales Swansea
March 2004
ABSTRACT
Wales exhibits high rates of disability and inactivity, and a higher incidence of mental health problems than other parts of Britain. Using data from the Welsh Local Labour Force Survey 2001, our results indicate that the low participation rate of the disabled in Wales is partly attributable to their having fewer qualifications; marginal effects suggest education could be a potent remedy for improving their labour market status. In terms of the pay differential between disabled and non-disabled individuals, it would appear that disabled women in Wales suffer disproportionately to disabled men.
JEL Classification: I1, J2, J3
Keywords: Disability, gender, employment, wage discrimination, Wales.
Acknowledgements
Material from the Quarterly Labour Force Surveys is Crown Copyright, has been made available from the Office for National Statistics (ONS) through the UK Data Archive and has been used by permission. Material from the Welsh Local Labour Force Survey is also Crown Copyright and has been made available by the Welsh Assembly Government through the ONS. The ONS, the Data Archive and the Welsh Assembly Government bear no responsibility for the analysis or interpretation of the data reported here.
1
1.Introduction
Increasing political attention has been focused on the disabled with the passing of the Disability Discrimination Act (DDA) in 1995 and the subsequent formation of a Disability Rights Commission, alongside other policy measures such as the the Disabled Person’s Tax Credit and the New Deal for Disabled People. These measures reflect an awareness by policy-makers of the problems faced by disabled people, who constitute a substantial, important and increasing section of society. Using a broad definition of disability that includes individuals with a long-term (12 months or more) health problem covered by the DDA and/or that limits the kind or amount of work that an individual can do, and based on the autumn quarter of the 2001 Labour Force Survey (LFS), Smith and Twomey (2002) report that nearly one in five people of working age in Britain have a current disability. This figure conceals substantial regional differences, with disability rates being highest in the North West[1] and in Wales (24.2% and 23.0% respectively) and lowest in the South East (16.3%) (see Sly, 1996 for an earlier discussion of inter-regional variation).
In the present paper we adopt the narrower ‘work-limiting’ definition of disability (see Section 2 for details), which is generally regarded as more appropriate in research considering labour market issues. The effect of adopting this alternative definition is to reduce the ‘headline’ disability rate by around 3½ percentage points to just under 16% of the working age population[2]. As shown in Table 1, which also details the figures separately for males and females, substantial regional variation is again evident using the work-limiting disability measure. Thus, the incidence figure for the South East is just 13.19%, compared to 19.85% in Wales; a figure exceeded only by the North at 21.11%. The composition of disability/health problems also exhibits some variance by region as shown in Table 2 (again see Section 2 for data details), the most common disability/health problem in all regions being that affecting limbs, followed by skin, breathing and organ problems. Especially noteworthy is that mental health problems are more prominent in Wales and Scotland than elsewhere, while the opposite is true in the South West.
As Smith and Twomey surmise:
“the reasons for regional variations in disabilities… are likely to be associated with regional variation in: the distribution of industries; the availability of, and access to healthcare and adequate housing; lifestyle and dietary behaviour; levels of education; and the age distribution of the population.” (p. 418)
Whatever the cause, the consequences are profound. As can be seen in Table 3, the employment rate for the disabled across Great Britain as a whole is marginally below 40%, and just half the non-disabled rate. However, there is also substantial variation in the percentage of the disabled who are in employment, from 26.65% in Wales to 49.78% in the South-West. These differences are much more marked than differences in ILO unemployment rates. Part of the lower employment rates noted above can be attributed to differences in activity rates: while fewer than half the disabled are inactive in three regions, the figure exceeds 60% in four others, including Wales, where there are particularly high levels of economic inactivity more generally[3]. Previous work has indicated that high levels of disability are a contributory cause to this last phenomenon (Blackaby et al., 2003).
Furthermore, even when the disabled find employment, they are disproportionately concentrated in less skilled work. This is reflected in substantial differences in relative pay between the disabled and non-disabled. As can be seen in Table 4, disabled hourly pay as a proportion of the non-disabled ranges from 82.4% in London and the South-East to 96.8% in East Anglia for the workforce as a whole. When differentiated by gender, the data confirm that earnings are typically higher for the non-disabled than for the disabled for both men and women, and as might be expected, show also that male average pay typically exceeds that of females in each region. They also indicate that, generally speaking, where earnings for the non-disabled are higher relative to the Great Britain average, so too are those for the disabled, reflecting the relative tightness of regional labour markets. It is unsurpising therefore, that earnings in Wales are among the lowest in Great Britain for each of the sub-groups in Table 4, with the exception of disabled men. However, the figures for the disabled in Wales need to be treated with circumspection (along with the corresponding ratios with the non-disabled), being based on very small sample sizes (n=115 and 108 for males and females respectively)[4]. More reliable estimates can however be obtained using the Welsh ‘boost’ to the LFS, as detailed below. To anticipate the results there, these latter data suggest that the hourly earnings for disabled men (women) reported in Table 5 appear somewhat high (low), and that the more conventional pattern of relative earnings noted above applies for this group also.
In this paper we focus on gender differences in the effects of disability on the labour market focusing specifically on Wales, making use of the Welsh ‘boost’ to the LFS (see Section 2). Since the relative position of women within the labour market in general is inferior to that of men in terms of both occupational attainment and levels of earnings, it is possible that disabled women are at a disadvantage not only relative to the non-disabled but also relative to disabled men. Disaggregating by gender also enables us to distinguish different types of disability and identify both within- and across-group differences. As the preceding discussion indicates, the Welsh labour market is characterised by, inter alia, high rates of disability and inactivity (among both the disabled and non-disabled). These are issues of considerable concern to policy-makers, both within Wales and more widely. To the extent that issues relating to disability and the labour market in Wales are those of other parts of Great Britain ‘writ large’, it is hoped the present paper may provide insights with wide relevance.
The remainder of the paper is structured as follows. In Section 2 we discuss the data to be employed in the estimation work, while section 3 discusses the estimation methodology itself. Results are presented in Section 4, first in terms of the impact of disability on labour force participation (employment), followed by earnings. The impact of different types of health problem is the subject of sub-section 4.3, while decomposition results by both gender and disability are discussed in the following sub-section. Finally, conclusions appear in Section 5.
2.The Data
From March 2001 there has been a Welsh ‘boost’ to the Labour Force Survey, resulting in the Welsh Local Labour Force Survey (WLLFS) dataset (see Hastings, 2003). The main LFS is undertaken quarterly, with a 5-quarter rotation of the sample of private households; in any quarter a fifth of respondents will be having their first interview (‘Wave 1’), while a fifth will be experiencing their last prior to leaving the sample (‘Wave 5’). In contrast, the ‘boost’ is undertaken annually, with households remaining in the sample for four years. The WLLFS dataset contains households from Waves 1 and 5 of the main LFS sample for each quarter, plus the ‘booster’ sample. For the former, the overlap from year to year is 50%, while for the latter it is 75%[5]. The effect of boosting the sample in this manner is that while the main LFS sample covers 4,600 households in Wales per year, the WLLFS contains 21,000 in total, enabling disaggregation down to local, Unitary Authority level.
To elaborate on the definition of disability discussed previously, respondents in the LFS (and the WLLFS) are asked first if they have any health problems or disabilities which would be expected to last more than a year, and second, whether these would affect either the kind or amount of paid work they can do. If positive answers are given to both of these questions we classify individuals as disabled (i.e. the disability is ‘work-limiting’). A further question asks about the type of health problem/disability, split into 17 categories. Where there are multiple disabilities respondents are asked to state which of them is the main health problem/ability. Because of problems of small cell sizes, we group these 17 types of disability into five main categories (as in Table 2) in order to establish if there are significant differences among them in terms of their impact on labour market outcomes.
The basic statistics on employment, unemployment and inactivity from the Welsh ‘boost’ are contained in Table 5, which also shows the figures separately for men and women. As noted previously, the substantial increase in sample size results in data that are more reliable than those for Wales contained in the main Labour Force Survey and reported in Tables 1 – 4, and suggest a slightly better outcome in terms of economic status for the disabled. Despite this improvement however, it remains the case that activity (and accordingly employment) rates for the disabled are very low, at just 36% (31%) and 29% (27%) for men and women respectively. With regard to earnings, those for disabled men appear somewhat high in the main LFS, while the reverse is true for women. For the non-disabled, the figures from the two sources are much closer, reflecting the larger sample size and hence greater reliability of the data.
Finally, it should be noted that in the remainder of the paper participation is defined as the receipt by an employee of a positive wage. While this understates the degree of participation by treating the unemployed as non-participants, as well as excluding those with missing wage data, the self-employed[6] and persons on government training schemes, such an approach is standard in the literature, and necessitated by the nature of the data and the methodology deployed, to which we now turn.
3.Methodology
We adopt the standard labour force utility maximisation model in which individuals are assumed to maximise their utility subject to budget and time constraints. Health enters the model through the budget constraint, implying a lower wage offer on account of lower productivity of disabled workers and the time constraint as illness leads to more absences and less time for work.
An individual deciding whether or not to enter the labour market will compare the wage offers of potential employers with his or her reservation wage. Low participation rates may result from a combination of high reservation wages associated with certain types of disability resulting from the extra time and energy required to participate in the labour market and/or the presence of disability income transfers.
Let us assume there are two types of individual, the disabled (D) and the non-disabled (N). For each of these types the wage offer equation is given by
[1]
where represents the logarithm of the offer wage, is a vector of the standard productivity related characteristics in the human capital model for individual i of type j, is the associated rate of return, and the error term.
In turn, the reservation wage is given by:
[2]
where represents the reservation wage, the vector Z incorporates the conventional human capital variables, with the addition of variables influencing the value of time (such as the number of dependent children) and is the error term.
The reservation wage is a latent variable, since it cannot be directly observed given the absence of a relevant question in the LFS. Rather it is represented by an indicator variable I, where I equals one if and zero otherwise.
Thus, the probability that an individual works is:
[3]
Assuming that the error terms and are normally distributed, the employment equation may be estimated by a probit specification.
In estimating the wage equation given by [1] it is necessary to correct for sample selectivity, since the employed are unlikely to be a random subset of the total population in terms of their productive characteristics. Accordingly, we utilise a Heckman two-stage procedure in which the probit estimates are utilised to derive the inverse Mills ratio, which is used as an additional independent variable in the wage equation.
Next, we decompose the overall difference between the earnings of the non-disabled and the disabled into explained and unexplained components, utilising a technique developed by Reimers (1983) and applied to disability using US data by Lambrinos (1981) and Baldwin and Johnson (1994), amongst others[7]. The difference in wage differs between non-disabled (N) and disabled (D) employees can be decomposed as:
[4]
The left-hand side of the above equation can be interpreted as the difference in mean wage offers of employers made to non-disabled and disabled employees respectively. The first term on the right hand side of the equation represents that part of the difference in wage offers which is attributable to differences in productivity (i.e. which is non-discriminatory), while the second term represents that part of the wage difference which is unexplained (i.e. which represents the difference in coefficients between the two groups). This will however, only be discriminatory to the extent that there are no unobserved productivity differences between the two groups as a result of types and degrees of severity of disability (for which the number of health problems is included as a proxy). The term is a vector representing the relationship between the observed wage structure and the non-discriminatory norm. Given the typical index number problem (see Oaxaca and Ransom, 1994) can take values varying from zero to one depending on which group is the frame of reference. In the tables below we provide several frames of reference – using the non-disabled as a base (0), the disabled (1), taking the mean of these two results (0.5), taking ratios given by the shares of the non-disabled in the working population (column 4) and finally the figure obtained from a pooled regression (*). It should be noted that were any discrimination to be eliminated, the outcome is likely to be closer to the non-disabled norm given the relative importance of this group in the total population.
Identification is obtained by including a variable for the number of children in the household in the participation (employment) equation if the respondent is a head of household or his or her spouse (otherwise zero) and by including a dummy variable indicating the presence of income earner in the household apart from the respondent. Linear and quadratic terms for age are included in the employment equation, as opposed to experience in the wage equations. Six qualifications and twenty-two unitary authority dummies, together with ethnic origin, type of household tenure and number of health problems appear in both employment and wage equations. The latter also include occupational, industry, small establishment, public sector and part-time dummies, together with a dummy variable for sickness absence in the reference week and the aforementioned tenure variables. The hourly pay variable is based on usual weekly pay divided by usual hours with a variable also included to adjust for the amount of usual overtime, measured in hours. In addition to separate estimation by reported disability status, all equations are estimated separately for men and women, thereby allowing for the probability that some independent variables may have gender-specific effects. Further, we estimate employment and wage equations (for the disabled only) augmented by five health type dummies derived from the 17 main health problems, as outlined earlier, in order to assess the impact of disability types. This has been found to be of considerable import in previous work in the UK, with mental health problems having especially adverse effects (see Jones et al., 2003).