Mental Health and Employment: A Bounding Approach Using Panel Data*
| Published date | 01 October 2022 |
| Author | Mark L Bryan,Nigel Rice,Jennifer Roberts,Cristina Sechel |
| Date | 01 October 2022 |
| DOI | http://doi.org/10.1111/obes.12489 |
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 84, 5 (2022) 0305-9049
doi: 10.1111/obes.12489
Mental Health and Employment: A Bounding
Approach Using Panel Data*
MARK LBRYAN,† NIGEL RICE,‡ JENNIFER ROBERTS† and CRISTINA
SECHEL†
†Department of Economics, University of Sheffield, Sheffield, UK (e-mail:
m.l.bryan@sheffield.ac.uk)
‡Centre for Health Economics and Department of Economics and Related Studies, University of
York, York, UK (e-mail: nigel.rice@york.ac.uk)
Abstract
The effect of mental health on employment is a key policy question, but reliable causal
estimates are elusive. Exploiting panel data and extending recent techniques using selection
on observables to provide information on selection along unobservables, we estimate that
transitioning into poor mental health leads to a 1.6% point reduction in the probability
of employment; approximately 10% of the raw employment gap. Selection into mental
health is almost entirely based on time-invariant characteristics, rendering fixed effects
estimates unbiased in this context, meaning researchers no longer have to rely on the
narrow local average treatment effects of most health/work IV studies.
I. Introduction
An individual’s relationship to the labour market is a key determinant of their financial
security and a source of broader well-being (Black, 2009). In most countries, people with
health problems have a much lower employment rate than the rest of the population.1
In the United Kingdom, every year 300,000 people stop work and become reliant on
health-related benefits, costing the government £13 bn and employers another £9 bn
(Black and Frost, 2011). Recent work by Jones, Rice, and Zantomio (2020) shows that
acute health shocks substantially increase the probability of exiting the labour market
and reduce hours and earnings. Adverse mental health (MH) seems to be particularly
pernicious in its labour market effects. The employment rate for people with an MH
problem is only 35% (Oakley, 2016), and the disability employment gap between those
with and without an MH problem is around 40 percentage points (Munford, Rice, and
JEL Classification numbers: I12, J14, J24.
*We are grateful to the Health Foundation for funding this research under their Social and Economic Value of
Health programme, Award No. 751630. The UKHLS data (University of Essex, 2018) were accessed via the UK
Data Service. Some of the data used are provided to the authors under Special Licence arrangements. We are
also grateful to Andrew Jones, Maarten Lindeboom, Arne Risa Hole, and two anonymous reviewers for helpful
comments on earlier versions of this article.
1https://ilostat.ilo.org/how-do-people- with-disabilities-fare- in-the-labour- market/
1018
©2022 The Authors. Oxford Bulletin of Economics and Statistics published by Oxford University and John Wiley & Sons Ltd.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and
reproduction in any medium, provided the original work is properly cited.
Mental health and employment: a bounding approach using panel data 1019
Roberts, 2016). Common MH problems, like anxiety and depression, account for over
40% of UK disability insurance claims (McInnes, 2012). MH is neglected in terms of
health spending, and often hidden in the workplace due to stigma and discrimination
(WHO, 2013). Internationally, the World Health Organization (WHO, 2008) estimate
that MH disorders comprise around 13% of the global burden of disease; and the OECD
estimates that MH problems affect more than one in six people across the European Union
in any one year (OECD/EU, 2018).
There is a complex relationship between MH and work. Work is generally good
for MH (Waddell and Burton, 2006), but there can also be adverse effects from long
hours, stress, and job insecurity (WHO, 2000). MH is also an important determinant of
an individual’s labour market situation, affecting the chances of obtaining employment,
‘good work’, and adequate reward. This complex relationship poses a number of problems
for the estimation of causal effects. Frijters, Johnston, and Shields (2014) summarize these
as: reverse causality (since health affects work and vice versa); measurement error (as we
do not observe the true health stock of an individual); and endogenous selection (since
unobserved characteristics and circumstances that affect health outcomes are also likely
to be related to labour market outcomes). Our study focuses on the latter problem, but we
also employ methods that aim to reduce the biases arising from the first two issues.
Causal estimation of the effect of an individual’s MH status on their chance of
being in employment requires independent variation in MH. However, many of the tools
that are often used to create a pseudo-experimental framework for estimation of causal
effects (such as exogenous policy changes or other ‘shocks’) are not valid, or have
only weak validity, in the context of MH and work. Most of the recent econometric
evidence relies on instrumental variable (IV) estimation and/or longitudinal data with
fixed effects (FEs) in an attempt to deal with endogenous selection. Few of the IV studies
are satisfactory; the instruments used have little theoretical support and virtually none of
the studies provide convincing empirical evidence on instrument validity. Further, where
the identification strategy is arguably stronger, the results often provide very specific
estimates of a local average treatment effect (LATE), which in most cases is derived from
an arbitrary exogenous shock (e.g. the death of a close friend used by Frijters et al.,2014).
The inclusion of FE eliminates endogenous selection bias arising from time-invariant
unobserved variables (such as childhood circumstances) that influence both health and
work outcomes. Also, FE may give a more relevant policy parameter, because these
models estimate the average effect on work outcomes for those whose MH changes, rather
than a more narrowly defined LATE. However, these models cannot deal with unobserved
effects that vary over time (such as changes in work relationships); if these are present,
they will bias the estimated effect of health on work, providing a misleading basis for
policy formulation. Practitioners face a dilemma given the difficulty of finding suitable
instruments for MH and the need for reliable quantitative evidence. In this context,
the use of FE models without instrumentation warrants deeper scrutiny.2This is now
possible by exploiting the methods developed by (Altonji, Elder, and Taber, 2005; Altonji
2Technically FE is also an IV estimator, with deviations from the means used as the instruments (see Verbeek, 2012,
p. 387-8). Thus, the effect identified (the average affect for the subgroup whose MH changes) is also a LATE.
However, for clarity, when comparing our work with the existing literature, we reserve the term LATE for explicit
IV methods.
©2022 The Authors. Oxford Bulletin of Economics and Statistics published by Oxford University and John Wiley & Sons Ltd.
1020 Bulletin
et al.,2011), who use selection on observable characteristics to provide information on
selection along unobservable factors; and in particular Oster (2013b,2019) who extends
and generalizes this method to enable the estimation of an unbiased treatment effect in the
presence of unobserved confounders.
We make two key contributions to the literature. Firstly, we fill a number of important
gaps in the evidence base by providing quantitative estimates of the effect of MH on the
employment of prime age adults. This is important evidence for social and economic policy
across all countries. The vast majority of existing evidence on the relationship between
health and work considers either physical health or general measures of overall self-
assessed health (see Ghatak, 2010 for a review). Where MH is considered, the estimates
often pertain only to severe health problems. In contrast, we use measures of general MH
derived from two psychometric instruments; the General Health Questionnaire and the
Short Form-12 health survey. These measures are good proxies for the true MH stock:
they are designed to provide information on all aspects of MH, and are less likely to
suffer from the reporting biases that are present in simple overall evaluative measures
(Bound, 1991; Bound et al.,1999; Lindeboom and Kerkhofs, 2009). To date, almost all
of the existing evidence comes from the United States: a country that has very different
health and welfare systems to many other countries, and in particular to the universal
health care coverage of the UK National Health Service. Our estimates for England and
Wales contribute to a very small pool of UK evidence, and provide important context
to the current policy priority to increase the number of disabled people in work by one
million over 10 years (DWP, 2017). In addition, much of the evidence on the impact of
health on labour market outcomes is for older workers, since this is where the burden
of most physical ill-health is felt. In contrast, MH disorders are particularly prevalent
in prime age workers (Kessler et al.,2005), so evidence is needed for this key group.
As well as estimating average effects for our sample of prime age individuals, we also
explore how both the health– employment relationship and any bias in the estimates vary
across a number of sub-groups differentiated by sex, age, education, physical health, and
household income. These results will also make a valuable contribution to the economic
analysis of the cost-effectiveness of health care interventions that are expected to have
important labour market effects3; for example, the Improving Access to Psychological
Therapies (IAPT) initiative that has been rolled out in England and Wales from 2008 to
help people who suffer from anxiety and depression.
Secondly, the vast majority of evidence comes from cross-sectional studies, whereas
our longitudinal analysis can control for individual unobserved factors that confound the
relationship between employment and health. Further, we explore any remaining biases
that are not removed by the inclusion of FE, by employing Oster’s method (Oster, 2013b;
Oster, 2019) to deal with unobservable selection. Ours is the first study to use this method
with individual longitudinal data incorporating FE.4We estimate the bias that arises from
3This is an important area for health policy; for example, Public Health England have recently commissioned a
model to estimate the cost-effectiveness of health interventions that are expected to have significant labour market
effects https://www.gov.uk/government/publications/health-matters-health- and-work
4In the only panel data applications of the Oster method of which we are aware, Hener, Rainer, and Siedler (2016)
and Cattan et al. (2017) use individual-level data with sibling FE, and Black et al. (2014) use firm-level data. By
assuming that the FEs capture all time-invariant variables and have no unobserved counterparts, we apply Oster’s
©2022 The Authors. Oxford Bulletin of Economics and Statistics published by Oxford University and John Wiley & Sons Ltd.
Get this document and AI-powered insights with a free trial of vLex and Vincent AI
Get Started for FreeStart Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
Start Your Free Trial of vLex and Vincent AI, Your Precision-Engineered Legal Assistant
-
Access comprehensive legal content with no limitations across vLex's unparalleled global legal database
-
Build stronger arguments with verified citations and CERT citator that tracks case history and precedential strength
-
Transform your legal research from hours to minutes with Vincent AI's intelligent search and analysis capabilities
-
Elevate your practice by focusing your expertise where it matters most while Vincent handles the heavy lifting
