Disability, employment and earnings: an examination of heterogeneity

Abstract

This article uses information from an ad hoc module on disability in the 2002 UK Labour Force Survey to identify the heterogeneity that exists within the disabled group and examine its impact on labour market outcomes. After controlling for a range of personal characteristics, the type, severity, duration and cause of the disability are found to be important determinants of employment, but there is less evidence to support the influence of within group heterogeneity on earnings.

Acknowledgements

The author would like to thank two anonymous referees and Peter Sloane and Phil Murphy for very helpful comments on earlier drafts, and Jack Kneeshaw (Data Archive) and Lawrence Brooker (ONS) for information regarding data access. 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.

Notes

¹ There remains significant debate about the most appropriate definition of disability and average employment rates vary considerably by definition (Berthoud, 2003).

² The module was introduced in the European LFS to provide comparable information on the labour market situation of people with disabilities in the EU in preparation for the 2003 European Year of People with Disabilities (Dupre and Karjalainen, 2003).

³ Gunderson and Hyatt (1996) consider the impact on earnings using a specialized dataset from Ontario in the early 1980s. They find that the proportion of the cost of the accommodation passed on to workers is higher if they are injured at another firm.

⁴ The additional questions relating to the duration, cause and severity of the disability were only asked to those with a long-term health problem. Those with a long-term health problem represent 28% of the male working age sample and 27% of the respective female sample. Whilst this definition does not coincide with more standard work limiting or DDA definitions, it maximizes the number of observations for the analysis and considers the entire heterogeneity within the broadest definition of disability. It is reassuring to note that the main conclusions are robust to restricting the sample to alternative (work-limiting or DDA) definitions of disability.

⁵ Employment is defined using International Labour Organization (ILO) definitions in the LFS and therefore includes, employees, the self-employed, those on government training schemes and unpaid family workers. The nonemployed include both the unemployed and the inactive.

⁶ These variables all relate to the current period and therefore no information is available on how the severity or type of disability has changed over time. The impact of the duration of disability may depend on the severity at onset and how severity changes over time, however, no information is available to control for this.

⁷ All the models are also estimated without the controls for the dynamic influences (duration and cause) but the results are not sensitive to this.

⁸ Changes in the benefit regime, retirement and social norms may all give rise to cohort effects in this context. However, cohort effects cannot be identified without repeated cross-sectional data.

⁹ As expected, having another disabled individual in the household is a significant determinant of individual disability status, since it captures common lifestyle and environmental influences. It potentially could affect employment through the role of caring; however, the presence of another disabled member in the household is insignificant in a simple employment probit model. The sensitivity of the results is tested by using the presence of a previous long-term health problem for identification. However, this information is only available for a subset of respondents. The results were not sensitive to the change in identification strategy.

¹⁰ The evidence, however, is mixed. Several studies find that the nonemployed tend to over-report ill-health or disability (e.g. Kerkhofs and Lindeboom, 1995; Kreider and Pepper, 2007) whilst others find that self-reported information is an unbiased measure of true health or disability (e.g. Dwyer and Mitchell, 1999; Benitez-Silva et al., 2004).

¹¹ The variables included in Y_i and Z_i are typical in the literature and are assumed to be exogenous. However, when examining the labour market impact of epilepsy, Famulari (1992) considers the endogeneity of education in an earnings equation and finds that failing to account for this endogeneity underestimates the impact of severity on earnings.

¹² Information about earnings is only asked to employees in waves 1 and 5 of the LFS. The results from the selection equations are qualitatively similar to the employment equations above, but are not presented here. None of the identifying variables is significant in the simple earnings equation.

¹³ For identification, these variables need to have a significant impact in their respective equation but no influence on earnings. All these variables are insignificant when included in a simple earnings equation. Unemployment 12 months ago is not significant at the 5% level in a simple probit model of disability status. Moreover, the key results are not sensitive to modifications of the exclusion restrictions, when the selection terms are created from two independent probit models or to estimation without the selection correction. Variables included in the wage but not the employment or disability equations (for example, industry) are defined only for the employed.

¹⁴ This is consistent with Madden (2004) who finds the controls for selection into disability are insignificant in participation equations. The positive sign on ρ indicates that unobservables that are positively correlated with disability are positively correlated with employment amongst the disabled, which appears counterintuitive. The correlation in a standard bivariate probit model with disability and employment (without selection) is, however, negative and strongly significant.

¹⁵ It may be that the nature of birth onset disability differs from age onset disability; however, a range of controls for the type, severity and duration of the disability are already included in the model to capture these influences.

¹⁶ Onset at birth also has a limited effect on observable characteristics such as education. Indeed, there is no descriptive evidence to suggest those who are disabled at birth are significantly less likely to hold educational qualifications. It should, however, be noted that hereditary conditions which are included in this group may not affect an individual prior to labour market entry.

¹⁷ A more restricted specification was also estimated, which excluded the controls for type of employment, since the within group differences may affect earnings through occupational choice. The results are fairly robust to the inclusion of the additional controls and so they are not presented here, but are available from the author on request.

¹⁸ The negative influence of short durations would be consistent with the effect of adaptation through time outweighing the negative influence of time out of the labour market or reductions in human capital investment.