The effects of health shocks on labor market outcomes: evidence from UK panel data

The effects of health shocks on labor market outcomes: evidence from UK panel data This study examines the link between health shocks and labor market outcomes in the United Kingdom. For sample periods of up to 9 years, I use longitudinal data from the British Household Panel Survey to test how sudden health shocks affect a number of labor market outcomes, such as labor and household income, employment status, and hours worked. Addition- ally, the study examines potential mechanisms underlying the link between health declines and labor market outcomes. By estimating propensity score matching difference-in-differences models, the study shows that sudden health declines lead to significant and persistent reductions in earnings. The effects are strongest for individuals experiencing severe health shocks, males, individuals with higher education and those working in managerial jobs. When examining potential channels, I provide evidence that increased health care expenditures and health care usage as well as reduced work productivity can explain the observed effects on labor market outcomes. Keywords Health shocks · Labor market · Mechanisms · United Kingdom JEL Classification C23 · I10 · I12 · J60 Introduction employment and earnings, policymakers should discuss ways to provide improved assistance to workers experienc- Despite being previously examined by several research- ing sudden declines in health in dealing with the situation ers, there is still uncertainty about the relationship between and allowing them to return to work. Additionally, there labor market and health outcomes. While the majority of is little evidence about the potential mechanisms through existing studies has focused on examining health effects which health shocks could affect immediate and long-term following changes in income and employment, there has labor market outcomes. Using longitudinal data from the been growing interest in learning more about the effects British Household Panel Survey (BHPS) for up to 9 years, of health on labor market outcomes in recent years. Given this study examines the effects of health shocks on several the close link between employment status and both health labor market outcomes and explores potential channels and health insurance coverage, an improved understand- through which labor market outcomes are affected follow - ing of how labor market outcomes are affected following ing health shocks. health shocks is relevant for policymakers. If health shocks The analysis of this paper adds to the existing literature have significant long-term negative spillover effects on in two ways. First, the study uses longitudinal data from the UK. I am following the individuals over four different sam- ple periods: 3, 5, 7 and 9 years. Besides testing for the imme- Electronic supplementary material The online version of this diate ee ff cts of health shocks on labor market outcomes, this article (https ://doi.org/10.1007/s1019 8-018-0985-z) contains supplementary material, which is available to authorized users. also allows me to provide evidence whether people adapt to health shocks over time or whether the effects are persistent. * Otto Lenhart These findings could indicate whether there are potential ottolenhart@gmail.com labor market institutions in place that make it challenging to Department of Economics, Duncan Wing, Strathclyde reintegrate workers into the workforce following the health Business School, University of Strathclyde, 199 Cathedral issues. Street, Glasgow, UK Vol.:(0123456789) 1 3 O. Lenhart Second, besides testing for the presence of a causal link conditions and health outcomes. These findings are mixed: from health to labor market outcomes, this study examines earlier work provides evidence that economic downturns potential mechanisms through which health shocks can actually improve health outcomes [40, 44, 51], while more affect labor market outcomes. An examination of potential recent work suggest that health declines along with the channels is important from a policy and a social welfare economy [15, 38, 39]. perspective since it provides evidence for how policymak- A large number of studies have examined the relation- ers can potentially help prevent substantial losses in pro- ship between income and health. Following the pioneer- ductivity following health shocks. This study examines the ing study by Case et al. [10], several papers have also role of three mechanisms: (1) changes in the frequency of provided evidence for the presence of a strong positive health care usage. This can impact labor market outcomes association between household income and health [4, 13, by taking time away from work and work-related activi- 14, 34, 47, 48]. This phenomenon has become known ties; (2) changes in the likelihood of paying for health care. as the income gradient in health. In more recent year, While universal health coverage is provided in the UK by researchers have expressed the need to test for the causal the National Health Service (NHS), individuals have the nature between income and health by pointing out that option to purchase additional private care to forego long the presence of a positive association could be the result waiting times and receive potentially lower quality care; (3) of third factors that inf luence both health and income or changes in the worker’s productivity. Observing changes due to reverse causality, which exists if health outcomes is the productivity of workers who previously experienced influence people’s employment status and, therefore, negative health shocks could suggest that employees should their income. find ways on how to allow workers to be better reintegrated Several previous studies have examined the effects of to the workforce, while policymakers should create an envi- health shocks on labor market outcomes. The majority of ronment that mitigates the risks to employees experiencing early work on the topic has focused on elderly groups of sudden health shocks. the population and the effects of health on retirement [ 7, 9, The study finds that health shocks, captured by declines 18, 26, 31, 32, 49, 54, 60]. These studies established that in self-reported health status and the onset of health condi- older adults are significantly less likely to be employed and tions, significantly affect labor market outcomes. Negative more likely to retire following the occurrence of a health health events are shown to reduce labor earnings, household shock. Other studies additionally show that health shocks income, and the likelihood of being employed. The negative have negative labor market effects for younger individuals effects are found for all four variations of the sample length, by examining several types of health shocks. These include which suggests that health shocks have lasting impacts on the presence of permanent health conditions [46], reduced labor market outcomes. When examining the effects across psychological health [12], injuries from road accidents different subgroups of the population, I find larger effects [16], the onset of disability [11, 36], reduced physical for males, higher educated individuals, and those working health [22, 23] and sudden illness [24]. managerial and professional jobs. When examining poten- Van Doorslaer and Koolman [58] find that income- tial mechanisms, the study provides evidence that increased related health inequalities in the UK are larger than in health care usage and health care expenditures as well as most other European countries, while García-Gómez et al. reduced work productivity can explain the observed per- [24] suggest that differences in the provision of disability sistent negative effects of health shocks on labor market benefits could explain these differences across nations. outcomes. The authors argue that relating the size of benefits to previous earnings, as done in the Netherlands, reduces the average income loss from health shocks compared to Labor market outcomes and health: when benefits are paid at a flat rate like in the UK. Besides previous evidence examining the effects of negative health events on labor market outcomes, this study tests for additional potential A number of previous studies have examined health effects mechanisms that can explain health-related inequalities in of negative employment shocks on people’s health. It has the UK (health care usage, health care expenditures, and been established that negative employment events such as work productivity). mass layoffs, plant closings and job loss have significant negative effects on health outcomes of affected individuals [20, 21, 37, 52, 53, 56]. Furthermore, other studies have examined the association between worsened economic 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 1 DD model setup Treatment group Control group Health Employment Health Employment Panel A 3-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (shock) Fair/poor/very poor Working Excellent/very good Working  2002 (post) Fair/poor/very poor Excellent/very good Panel B 5-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (shock) Fair/poor/very poor Working Excellent/very good Working  2003 (post) Fair/poor/very poor Excellent/very good  2004 (post) Excellent/very good Panel C 7-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (pre) Excellent/very good Working Excellent/very good Working  2003 (shock) Fair/poor/very poor Working Excellent/very good Working  2004 (post) Fair/poor/very poor Excellent/very good  2005 (post) Excellent/very good  2006 (post) Excellent/very good Panel D 9-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (pre) Excellent/very good Working Excellent/very good Working  2003 (pre) Excellent/very good Working Excellent/very good Working  2004 (shock) Fair/poor/very poor Working Excellent/very good Working  2005 (post) Fair/poor/very poor Excellent/very good  2006 (post) Excellent/very good  2007 (post) Excellent/very good  2008 (post) Excellent/very good the self-reported health measure is reduced since each Data individual’s health is only compared to his or her own prior assessment. This allows controlling for the fact that This study uses data from waves 10–18 (2000–2008) of each respondent may have their own scales in ranking their the British Household Panel Survey (BHPS), a nationally health (reference bias). Furthermore, in comparison to the representative panel survey of private households in Great two other commonly used UK datasets with detailed infor- Britain that started interviewing 10,300 individuals from mation on labor market outcomes (Labor Force Survey and 5,500 families in 1991. The use of the BHPS provides New Earnings Survey), the BHPS also provides informa- several advantages for the purpose of this study. Due to tion on several health outcomes. Finally, the BHPS gives it longitudinal nature, the dataset allows accounting for a complete representation of incomes across the pay dis- time-invariant unobserved heterogeneity and composi- tribution since it questions all individuals above 15 years tional selection. The potential for measurement error in of age who live in the household at the time of the inter- view. Given that individuals in the UK become eligible to receive state pensions at the age of 65, the sample is Taylor (1998) provides a full description of the sampling strategy restricted to all individuals between the ages 18 and 64 in applied in the initial wave in order to design a nationally representa- tive sample of the British population. 1 3 O. Lenhart the surveyed households for whom information on health off from work to help their family member with the doc- and labor market outcomes is available. tor appointments. Given that the analysis uses four different The analysis uses two different types of health shocks. In sample lengths, it is able to provide evidence whether any the main specification, which is presented in Table  1, health potential changes in health care usage only occur immedi- shocks are defined as a decline in self-reported health sta - ately after the health shock or whether these changes persist tus. Self-assessed health is categorized from 1 (= excellent) for several years. to 5 (= very poor) in the BHPS. This measure of health While health shocks can reduce the labor force partici- has been shown to be a good predictor of other health out- pation of people who can no longer work, another channel comes, including mortality [29], future health care usage through which adverse health events can affect earnings is [57] and hospitalizations [45]. To remove concerns about by reducing labor productivity. Workers might not be able potential reporting heterogeneity of health status, Johnston to perform the same tasks or might need longer to complete et al. [33] suggest the additional use of more objective health the same tasks compared to before the health shock. Without outcomes. In an additional specification, health shocks are empirically testing for the presence of this channel, García- defined as the onset of a new health condition. In each wave, Gómez and López-Nicolás [22] point out that productiv- the BHPS asks respondents whether they suffer from any ity losses could either be absorbed by the employer or by of the following 15 health conditions: body pain, migraine, the inability to work extra time. Using a sample of 2264 skin issues/allergy, asthma/chest pain, anxiety, heart or workers, Myde et al. [43] provide evidence for a strong link blood pressure, hearing problems, stomach/liver/kidney between health risks and self-reported work productivity. pain, seeing problems, epilepsy, diabetes, alcohol or drug To capture whether changes in work productivity are a problems, stroke, cancer or other conditions. As indicated by potential mechanisms underlying the relationship between García-Gómez [23], using the onset of health conditions as a health shocks and labor market outcomes, this analysis health shock could provide evidence regarding any potential examines four proxies for work productivity: (1) hourly anticipation effects [23]. wages of workers, (2) the likelihood of reporting that cur- Besides examining the short- and long-run effects of rent work is limited by one’s health, (3) the likelihood of health shocks on earnings and employment, this study addi- having difficulty to concentrate, and (4) the likelihood of tionally tests for the role of potential mechanisms underlying constantly feeling under strain. While hourly wages is the the link between sudden declines in health and labor market most direct way of measuring productivity, the other three outcomes. According to the Grossman [25], health can be outcomes should provide additional evidence for poten- viewed as both a consumption and investment good since it tial changes in labor productivity following the onset of not only makes people feel better, but it also increases the health shocks. Antikainen and Lönnqvist [5] suggest that, number of healthy days to work and to earn income. Gross- especially in “knowledge-intensive” organizations, where man [25] states that to keep certain levels of health capital, knowledge has more importance than other inputs, work individuals invest into their health through channels such as performance can be negatively affected by health problems market inputs of health care, diet and exercise. or other personal issues because they are highly dependent The first mechanism examined is changes in health care on the ability to concentrate. Using a factor analysis model, usage, which is captured by three outcomes: (1) the likeli- Halkos and Bousinakis [27] provide empirical evidence that hood of having more than five annual doctor visits, (2) the increased stress leads to reduced work productivity. Using likelihood of having spent a night at a hospital in the previ- the four proxies of work productivity listed above, my study ous year, (3) the likelihood of having used a number of other tests whether individuals who suffered health shocks are not health services, and (4) changes in the likelihood of having able to perform the same tasks compared to prior to the paid for health care services in the previous year. While the experiencing the sudden health decline. NHS provides universal coverage to all individuals in the UK, two serious issues that the program has been dealing with are the quality of care and long waiting times [59]. To Econometric methods avoid these problems, individuals have the option to pur- chase additional private care to forego long waiting times DD matching models before seeing a doctor in some cases. Persistent increases in out-of-pocket expenditures on health can be associated with Similar to García-Gómez and López-Nicolás [22], this labor market outcomes in two ways: (1) earned income of study estimates propensity score matching combined with individuals recovering from health shocks could be reduced difference-in-differences (DD) models to estimate the Aver - due to time spent away from work for doctor visits, and (2) age Treatment Effect on the Treated (ATET). This empirical household income could be affected to a larger extent than strategy allows me to compare the distributions of outcomes individual income if other household members take time for treated individuals (who suffer the health shock) with 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data the distributions of outcomes of matched individuals in propensity score, and (2) kernel matching on the propensity the control group, without having to make any functional score. Since there are no reasons to expect one of the meth- form assumptions. As pointed out by García-Gómez and ods to be preferable to the other, the use of both methods López-Nicolás [22], matching frameworks are often criti- allows the analysis to test for the robustness of the observed cized for assuming away potential biases that might exist effects. When examining the ATET of health shocks on due to unobserved heterogeneity. The authors argue that one labor market outcomes, the analysis examines four different solution to remove concerns about such biases is the use of outcomes: (1) total annual labor income, (2) total annual longitudinal data that includes data from before and after the household income, (3) the probability of being employed, health shock. Using longitudinal data from the BHPS, this and (4) hours worked per week. Standard errors are obtained study is able to first difference the outcomes of the treated following recent findings by Abadie and Imbens [3 ], who and the controls to eliminate any unobservable fixed effects established how to take into account that propensity scores that influence the selection into the groups as well as the are estimated in the first stage. They show that ignoring outcomes of interest [22]. Thus, the estimated ATET’s are this fact when estimating average treatment effects on the weighted averages of the differences in differences between treated in the second stage may lead to confidence inter - each of the treated individuals and his/her matched control. vals that are either too large or small. By showing that the The study uses estimated propensity scores, which cal- propensity matching estimator have approximately Normal culate the probability of treatment given a vector of observ- distributions, Abadie and Imbens [3] provide evidence that able variables, to match individuals who receive a health the matching on estimated propensity score is more efficient shock to individuals that are similar but do not experience than matching on the true propensity score in large samples. the health shock. The propensity scores are based on pre- treatment variables and are estimated using probit models. Assignment of treatment and control groups Observable characteristics that are included to obtain the propensity scores are age, gender, household size, educa- This assignment of individuals into treatment and control tional attainment, and area. In additional specifications, I group used in this study is similar to previous work by furthermore include information on lagged health status as García-Gómez and López-Nicolás [22] and García-Gómez a covariate when estimating the propensity scores. Follow- [23] as well as Lechner and Vázquez Álvarez [35]. A crucial ing Rosenbaum and Rubin [50], the use of a function of X, challenge when estimating the effects of health shocks on called the propensity scores P(X), rather than a potentially labor market outcomes is the fact that many health and labor high-dimensional vector of covariates implies that: market outcomes are potentially jointly determined by many people. The use of propensity score matching DD model can E Y  D = 1, P(X) = E Y  D = 0, P(X) , (1) 0 0 overcome this concern by identifying arguably exogenous where Y denotes the untreated state, D = 1 indicates treat- 0 health shocks that are independent of employment status. ment and D = 0 indicates non-treatment. The analysis of this Table 1 shows the setup for the two groups used in the DD study follows Heckman et al. [28] difference-in-difference models estimated in this study, which analyzes four varia- matching methods, which uses both comparisons between tions of sample length to test for both immediate and the treated and non-treated, and differencing over time. Thus, long-term effects of adverse health events on labor market the conditions needed to identify the ATET using the differ - outcomes. Despite different sample periods, all three models ence-in-difference matching estimator is: share the following characteristics: E Y − Y  D = 1, P(X) = E Y − Y  D = 0, P(X) , � � 0,t 0,t 0,t 0,t 1. Individuals from both treatment and control group are (2) in excellent or very good health and are working in the where t and t′ represent the post- and pre-treatment peri- pre-treatment period (Pre). ods, respectively. Thus, the ATET provides a weighted aver- 2. Individuals forming the treatment group experience a age of the differences in differences between individuals in health shock in the treatment period (Shock), meaning the treatment and the control group and it is obtained by their health status drops to fair, poor or very poor. Indi- estimating the following equation: ATET = E Y D = 1, P(X) − E Y D = 0, P(X) The analysis is conducted using the “teffects psmatch” command in DID 1t 0t Stata, which incorporates the findings by Abadie and Imbens [3] and − E Y D = 1, P(X) − E Y D = 0, P(X) � � 1t 0t takes into account potential estimation errors in the propensity score (3) due to the fact that the propensity scores were estimated in the first stage. In an earlier paper, Abadie and Imbens [2] show that bootstrap- In the empirical analysis, the study uses two alternative ping, a technique that had previously often been used to obtain stand- methods when matching treated individuals with those in ard errors for propensity score matching estimators, is generally not the control group [8]: (1) nearest neighbor matching on the valid for matching estimators. 1 3 O. Lenhart Table 2 Sample sizes for Health shock: drop in health status Health shock: onset of health condi- treatment and control group tion Treated Control Total Treated Control Total 3-year sample 591 9720 10,311 1620 5034 6654 5-year sample 585 11,155 11,740 1525 5190 6715 7-year sample 504 11,760 12,264 1001 5061 6062 9-year sample 315 11,268 11,583 432 4635 5067 viduals in the control group remain in excellent or very that are present in all waves of the corresponding sample good health. All members of the treatment and the con- periods. Given that the onset of new health conditions occur trol group are working during the period in which the less frequently than declines in health status, the sample treatment occurs. sizes are smaller when using health condition as the health 3. Self-reported health status of individuals in the treat- shock. ment group remains in fair, poor or very poor health in Given that the analysis in this study only includes indi- the first year after the health shock, while individuals in viduals who are present in all survey waves for each sample the control group remain in excellent or very good health length period (either 3, 5, 7 or 9 years), attrition could pose a throughout the post-treatment period (Post). potential issue. The obtained treatment effect estimates could potentially be biased if people drop out of the survey due to Additionally, using the same setup as shown in Table 1, health-related problems. Given the longitudinal data set of I use the onset of a health condition as an alternative health the BHPS, I am able to identify individuals who drop out of shock. Individuals forming the treatment group report the the survey and compare them to those who remain in it and onset of a health condition in the treatment period, while are used in the analysis. Online Appendix Table A1 shows those in the control group report no health conditions comparisons of descriptive statistics for the two groups for throughout the study period. Again, all individuals work in the sample periods of 5, 7, and 9 years. It is noticeable that both the pre-treatment and the treatment period. Given that there are only very small differences in health status between information regarding the presence of health conditions are people remaining in the survey and those who drop out at potentially less subjective than self-assessed health status, some point. Individuals who stay in the BHPS for all the the findings from this additional health shock can remove years in each of the three period analyzed are shown to be concerns about potential reporting heterogeneity of health slightly more likely to be employed and have higher labor status [33]. and household incomes than those who drop out of the sur- By examining a sample of individuals who are employed vey. Table A1 shows that the attrition rates were between 13 during both the pre-treatment period and the year of the and 15% for the sample periods shown in Online Table A1. health shock, the potentially simultaneous determination One of the main assumptions of estimating propensity of health and labor market outcomes is accounted for and score matching DD estimates is that the overlap assump- allows testing for the effects of experiencing health declines tion, which is satisfied when there is a chance of seeing on labor market outcomes in the post-treatment period. This observations in both control and treatment groups at each framework ensures that the observed effects are not the result combination of covariate values. As highlighted in the refer- of reverse causality, which would exist if changes in labor ence manual for Stata Treatment Ee ff cts, the analysis cannot market outcomes lead to the health shock. One assumption predict or otherwise account for the unobserved outcomes of this framework is that there are no anticipation effects, of some individuals if the assumption is violated [55]. Fig- meaning that people report declines in health because they ure  1a–d provide plots of the estimated densities of the expect negative employment shocks to occur in the future. propensity scores for both treatment and control groups for all four sample periods. All the graphs present clear evi- Descriptive statistics dence that the overlap assumption is satisfied since there are chances of seeing observations in both groups at each Table 2 presents sample sizes for the four different sample combination of covariate values. length for each of the two health shocks that examined in Table 3 provides results from covariate balance tests for the study. For the health status shock in the shortest period the matching conducted in the analysis. In well specified of study (2000–2002), the control group consists of 9720 matching models, the covariates should be balanced, which observations and the analysis includes 591 observations for allows for the outcomes to be conditionally independent of the treatment group. The analysis includes only individuals the treatment when conditioning on covariates [55]. The left 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Fig. 1 Density of propensity scores, a 3-year sample, b 5-year sample, c 7-year sample, d 9-year sample side of Table 3 shows the balance test results for the main individuals. The nearest neighbor matching estimates find analysis, while the right side shows differences between reductions in annual labor income in the range of £1181.40 raw and matched data when lagged health information is for the year after the shock in the 3-year sample (p < 0.01) included as a covariate. Overall, the balancing results indi- to £4432.32 for the 9-year sample (p < 0.01), which exam- cate that the matching succeeds in balancing the covariates ines the effects for up to 4 years after the health shock. and reducing the standardized differences between the two The immediate effects on earnings in the year following the groups. drop in health status is slightly smaller in magnitude than estimated by García-Gómez and López-Nicolás [22], who find a decline in income of €1118 (measured in 2001 Euros, Results which corresponds to £1763.31 using the 2001 €/£ conver- sion rate) using Spanish data and a 3-year sample. The effects of health status shocks While the kernel matching estimates also provide evi- dence for declines in labor income in all four periods, two of Table 4 presents propensity score matching DD effects of the effects are statistically insignificant. Overall, given that sudden declines in health status on four labor market out- losses in labor income are larger in magnitude for the longer comes and four different sample lengths. The estimates in the sample periods, the results do not suggest that individuals first two columns provide evidence that health shocks have adapt to the health shock. On the contrary, it appears that substantial negative effects on labor earnings of affected individuals struggle to be reintegrated into the labor force following the declines in health. The next two columns show Table  3 shows balancing test results for the Nearest Neighbor matching analysis. The results are consistent for the kernel matching Income results are adjusted for inflation using the U.K. Consumer analysis. These results are not shown, but are available upon request. Price Index and 2000 as the base year. 1 3 O. Lenhart Table 3 Covariate balance tests Main analysis Analysis with controls for lagged health Standardized differ - Variance ratio Standardized differ - Variance ratio ences ences Raw Matched Raw Matched Raw Matched Raw Matched 3-year  HH size − 0.0175 0.0024 0.9420 0.9163  Education 0.1734 − 0.0018 0.9906 0.9687  Age 0.1071 − 0.0186 1.0413 1.0315  Gender − 0.0022 0.0275 1.0014 1.0016  Area − 0.0335 − 0.0127 0.9285 0.9490 5-year  HH size − 0.0672 0.0417 1.0414 1.1571 − 0.0672 − 0.0310 1.0414 1.1079  Education 0.3427 0.0149 0.9780 0.9559 0.3427 − 0.0757 0.9780 0.9161  Age 0.1533 0.0432 0.9215 0.8796 0.1533 0.0243 0.9215 0.8692  Gender 0.1038 − 0.0715 0.9996 0.9896 0.1038 − 0.0017 0.9996 0.9999  Area − 0.0769 − 0.0320 1.0018 0.9967 − 0.0769 − 0.1255 1.0018 0.9490  Lagged health status – – – – 0.6766 0.0187 0.6010 0.9708 7-year  HH size − 0.1542 0.0182 1.1199 1.2669 − 0.1542 − 0.2428 1.1199 0.9570  Education 0.1278 − 0.0738 0.9103 0.8231 0.1278 0.4209 0.9103 1.3985  Age − 0.0831 − 0.0123 1.0280 1.0092 − 0.0831 0.1321 1.0280 1.1516  Gender − 0.0428 − 0.0892 1.3490 1.9217 − 0.0428 − 0.2305 1.3490 1.1363  Area 0.2269 0.0432 0.7737 0.8212 0.2269 − 0.0232 0.7737 0.7864  Lagged health status – – – – 0.7242 − 0.0782 0.5197 1.0818 9-year  HH size 0.0455 0.0744 0.8414 0.7221 0.0455 0.0440 0.8414 0.9859  Education 0.2734 − 0.0498 1.0289 1.1293 0.2734 0.0232 1.0289 0.8383  Age 0.1377 0.0050 0.8731 0.8611 0.1377 − 0.1599 0.8731 1.0662  Gender − 0.0170 − 0.0145 1.4092 2.0770 − 0.0170 0.0592 1.4092 0.9511  Area 0.2399 0.0667 0.8979 0.9703 0.2399 0.2130 0.8979 0.8791  Lagged health status – – – – 0.7109 − 0.0410 0.6144 0.8386 Table 4 Effects of health shocks on labor market outcomes (health status) Total labor income (£ per year) Total HH income (£ per year) Employed Weekly work hours NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel ing ing ing matching 3 year sample − 1181.40*** − 769.08 − 2834.63*** − 3355.70*** − 0.0068 − 0.0186* − 0.06 (0.57) − 1.14 (0.72) (430.54) (621.24) (756.41) (1065.85) (0.0073) (0.0109) 5-year sample − 3041.75*** − 3948.23*** − 4362.41*** − 4255.36*** − 0.0356*** − 0.0370*** − 1.17** − 0.57 (0.65) (462.60) (752.80) (777.40) (1063.76) (0.0103) (0.0149) (0.58) 7-year sample − 2097.46*** − 671.85 − 3025.02*** − 4677.16*** − 0.0378*** − 0.0268* 0.93 (0.57) 0.39 (0.78) (437.16) (720.86) (715.66) (1345.02) (0.0128) (0.0149) 9-year sample − 4432.32*** − 3345.17** − 5005.84*** − 4871.36*** − 0.0052 − 0.0159*** 0.35 (1.07) 0.51 (0.88) (810.96) (1583.87) (842.24) (1683.60) (0.0071) (0.0052) Robust standard errors, based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 5 Annual treatment Total labor income (£ per year) Employed effects 5-year sample (health shock in 2002)  Treat*2000 456.45 (571.33) 0.0089 (0.0182)  Treat*2001 29.13 (548.56) 0.0056 (0.0152)  Treat*2003 − 496.32 (526.55) 0.0017 (0.0192)  Treat*2004 − 1913.57** (763.25) − 0.0734*** (0.0310) 7-year sample (health shock in 2003)  Treat*2000 695.75 (873.84) − 0.0090 (0.0236)  Treat*2001 124.97 (762.76) − 0.0083 (0.0194)  Treat*2002 1336.91 (846.87) 0.0063 (0.0213)  Treat*2004 − 1358.27* (781.94) 0.0568 (0.0367)  Treat*2005 − 2078.62** (923.18) − 0.0572 (0.0361)  Treat*2006 − 2498.40** (1022.74) − 0.1212*** (0.0431) 9-year sample (health shock in 2004)  Treat*2000 592.78 (2527.47) − 0.0005 (0.0077)  Treat*2001 882.40 (2567.48) 0.0004 (0.0048)  Treat*2002 2161.92 (2325.20) 0.0021 (0.0054)  Treat*2003 128.95 (2301.63) − 0.0009 (0.0057)  Treat*2005 385.63 (2226.96) 0.0145 (0.0259)  Treat*2006 − 948.35 (2033.88) − 0.0198*** (0.0063)  Treat*2007 − 1551.61 (2298.16) − 0.0319*** (0.0066)  Treat*2008 − 1625.28 (2229.07) − 0.0421*** (0.0071) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in paren- theses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 the effects on health declines on annual household income. evidence that there are any changes on the intensive margin For all sample periods and both matching techniques, I find of employment. While four of the eight estimates for the even larger reductions in household income than for labor likelihood of being employed are statistically significant at income. For the 9-year sample, the results suggest that the 1% level, the immediate effects are substantially smaller household incomes are reduced by £5005.84 and £4871.36 than those observed by García-Gómez and López-Nicolás following the health shock (both p < 0.01). A potential expla- [22] for Spain. When re-estimating the analysis with only nation for the difference in magnitudes for total household individuals who remained at work throughout the sample income and individual labor earnings is that other household periods, I find very similar declines in labor earning and members take time away from work to assist the individuals household income. This suggests that changes in employ- recovering from health shocks. ment are not the only driver of the observed income losses. Appendix Table A2 furthermore provides matching DD The later part of the study examines some other potential results for the effects of sudden declines in health status on mechanisms through which health shocks can affect labor the natural log of both total labor and household income. market outcomes. Consistent with the results in Table 4, all estimates show that health shocks negatively affect labor earnings and total Annual treatment effects household incomes of affected individuals. While it should be noted that two of the eight labor income estimates are Table 5 shows annual estimates for the effects of health imprecisely estimated, Online Table A2 confirms that the shocks on total annual labor income and the likelihood of observed treatment effects are robust to the measure of income used in the analysis. Table 4 additionally shows the effects of health shocks on Appendix Table  A3 furthermore shows DD matching estimates the likelihood of being employed and weekly hours worked. when lagged health status is included as a covariate to obtain the pro- My analysis finds that individuals reduce their labor mar - pensity score values. The results are consistent with the main results ket activity on the extensive margin, while there is little from Table  3, providing further evidence that negative health events affect labor market outcomes. 1 3 O. Lenhart Fig. 2 Annual treatment effects on labor income, a 5-year sample. b Annual treatment effects on employment, 5-year sample. c Annual treat- ment effects on labor income, 7-year sample. d Annual treatment effects on employment, 7-year sample being employed, which are obtained by interacting each Mild vs. severe health shocks year with the treatment indicator. Since this test includes effects during pre-shock periods, it provides a test for the The longitudinal nature of the BHPS furthermore allows parallel trends assumption made in the main DD model. me to identify individuals who experienced large changes Given that this analysis is not feasible in the 3-year sam- in self-reported health as well as others whose health status ple, Table 5 only shows treatment effects for sample peri- only slightly declined. Information on self-reported health ods of 5, 7 and 9 years. status in the BHPS is provided on a scale from 1 (= excel- For all the sample periods, no statistically significant lent) to 5 (= very poor). For this analysis, I define the two differences are estimated during the years before the onset types of treatments the following way for all sample lengths: of the health shocks. Furthermore, none of the pre-shock (1) mild health shocks are average declines in self-reported treatment effects that are shown in Table  5 are jointly health by at most one point on the scale between the pre- significant. This provides suggestive evidence that the and post-shock period; (2) severe health shocks are average parallel trends assumption is satisfied. The estimates for declines in self-reported health by more than one point on both labor income employment status become larger in the scale between the pre- and post-shock period. Individu- magnitude several years after the shocks, indicating that als in the control group are those whose average health status the effects of health shocks on labor market outcomes remained the same across both periods. Compared to the are persistent rather than temporary. Figure 2a–d confirm main analysis in Sect. 5.1, this specification allows using this by providing graphical representations of estimates changes in the entire distribution of health status. Consistent presented in Table 5. with the main DD setup shown in Table 1, all individuals 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 6 The effects of health shocks on labor market outcomes (average differences in health status) Total labor income (£ per year) Total HH income (£ per year) Employed Weekly work hours NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- ing ing ing ing 3-year sample  Mild shock − 565.58** − 568.12* − 368.59 − 584.43 − 0.0097** − 0.0117** − 0.03 (0.27) 0.22 (0.38) (239.51) (322.62) (442.20) (514.00) (0.0041) (0.0053)  Severe − 1743.84*** − 1108.57* − 1411.01 − 1495.33 − 0.0261** − 0.0294*** 1.11** (0.49) 1.58** (0.75) shock (462.67) (598.55) (915.97) (1138.41) (0.0107) (0.0111) 5-year sample  Mild shock − 298.32 − 406.03 372.45 610.45 − 0.0079** − 0.0064 0.17 (0.21) 0.12 (0.27) (234.22) (295.50) (341.97) (452.44) (0.0033) (0.0042)  Severe − 1353.91*** − 1178.19** − 79.18 562.42 − 0.0373*** − 0.0326*** 0.76 (0.51) − 0.10 (0.66) shock (395.69) (515.06) (768.42) (918.08) (0.0097) (0.0109) 7-year sample  Mild shock − 777.23*** − 684.35* − 432.31 − 607.45 − 0.0071 − 0.0012 0.79*** 1.16*** (0.36) (254.05) (379.52) (399.68) (578.83) (0.0044) (0.0059) (0.26)  Severe − 3697.61*** − 2483.74*** − 4366.20*** − 2546.32*** − 0.0652*** − 0.0594*** − 1.52** − 1.90*** shock (362.84) (547.43) (587.85) (1006.68) (0.0132) (0.0119) (0.66) (0.71) 9-year sample  Mild shock − 1739.05*** − 1840.00*** − 3908.70*** − 3758.49*** − 0.0205 − 0.0113** 0.25 (0.23) 0.46 (0.31) (269.20) (341.71) (371.18) (490.90) (0.0142) (0.0053)  Severe − 3335.97*** − 3873.31*** − 5716.03*** − 7504.61*** − 0.0723*** − 0.0619*** − 2.66*** − 1.74** shock (578.20) (832.38) (775.68) (1361.91) (0.0130) (0.0140) (0.68) (0.78) Robust standard errors, cluster by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 Table 7 Effects of health shocks on labor market outcomes (health condition) Total labor income (£ per Total HH income (£ per year) Employed Weekly work hours year) NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- ing ing ing ing 3-year sample − 1049.24*** − 1068.43** − 2107.76*** − 2500.87*** − 0.0066 − 0.0029 − 0.39 (0.37) − 0.83* (0.50) (364.98) (515.82) (580.60) (860.66) (0.0070) (0.0071) 5-year sample − 1653.48*** − 1414.43*** − 2105.80*** − 3490.51*** 0.0020 0.0046 − 0.83** − 1.15** (0.48) (340.84) (521.67) (716.49) (940.21) (0.0022) (0.0029) (0.36) 7-year sample − 3129.55*** − 3292.39*** − 3342.17*** − 4202.84*** − 0.0025 − 0.0035 − 0.30 (0.39) − 0.51 (0.55) (444.22) (698.85) (777.31) (1078.69) (0.0013) (0.0025) 9-year sample − 3482.73*** − 5122.64** − 4097.99*** − 8083.43*** 0.0097 0.0116 0.11 (0.58) 0.45 (0.94) (554.06) (2017.59) (1059.44) (1996.55) (0.0074) (0.0152) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 are still required to be employed throughout the pre-shock for the effects of labor market outcomes following a 5-point period and in the year that the shock occurred. This analysis drop in health satisfaction in the German Socio-Economic is similar to two previous studies that test for employment Panel (GSOEP), which collects self-reported health informa- effects for individuals near retirement with panel data sets. tion on a scale from 0 to 10. Smith [54] data from the Health and Retirement Survey Table 6 presents the results for the two levels of health (HRS) to separately test the effects of experiencing either shocks. As expected, the negative effects on labor income a major or a minor health shock, while Riphahn [49] tests and the likelihood of being employed are substantially larger 1 3 O. Lenhart Table 8 Effects of health shocks on health care usage More than 5 annual doctor visits Spent a night at hospital Used any other health services Paid for any health services NN Kernel NN Kernel NN Kernel NN Kernel 3 years 0.2329*** 0.2555*** 0.0980*** 0.0931*** 0.2403*** 0.2477*** 0.0346*** 0.0261*** (0.0190) (0.0204) (0.0160) (0.0174) (0.0234) (0.0282) (0.0159) (0.0181) 5 years 0.1742*** 0.1906*** 0.0631*** 0.0720*** 0.1397*** 0.1809*** 0.0349*** 0.0570*** (0.0207) (0.0204) (0.0135) (0.0162) (0.0226) (0.0283) (0.0145) (0.0181) 7 years 0.1794*** 0.2104*** 0.0773*** 0.0813*** 0.2058*** 0.2288*** 0.0530*** 0.0489** (0.0202) (0.0218) (0.0167) (0.0161) (0.0261) (0.0308) (0.0183) (0.0197) 9 years 0.1611*** 0.1774*** 0.0345 (0.0219) 0.0556*** 0.1498*** 0.1127*** − 0.0194 − 0.0175 (0.0440) (0.0250) (0.0174) (0.0397) (0.0393) (0.0161) (0.0227) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Examples of health services asked for in the BHPS are usage of a physiotherapist, psychotherapist, health visitor at home and a hospital consultant. Pregnancies are excluded when examining changes in the likelihood of being a hospital in-patient *p < 0.10, **p < 0.05, ***p < 0.01 for individuals who experienced severe health shocks com- the health shock. The observed effects on hours worked are pared to individuals whose average health status declined mixed, with three estimates finding statistically reductions by at most one point. Similar to the previous findings, the in the weekly time spent working following the onset of the results are larger for the two longer sample periods (5 and 7 health condition. Overall, the results in Tables 4 and 7 pro- years), suggesting that the labor market effects are persistent vide consistent evidence that sudden health declines lead to rather than temporary. In the 7-year period, it is observ- substantial and persistent negative effects on labor earning able that individuals with mild health shocks significantly and household income. increase their weekly work hours, while those who experi- enced severe health shocks work significantly fewer hours after the health shock. Overall, the results in Table 6 point Mechanisms out that labor market outcomes are significantly worsened after severe health declines. However, given that labor The effects on health care usage income for those with mild shocks is reduced by £1840 in the 9-year sample (p < 0.01, kernel matching), the results Table 8 presents estimates for the effects of health shocks on also indicate that even relatively small health declines three indicators of health care usage and on the likelihood can negatively affect labor market outcomes of affected with which individuals paid for any health services out of individuals. their own pockets. The first six columns show that, as one could expect, individuals are more likely to have more than The effects of health conditions five annual doctor visits, to spend a night at the hospital and to have used any other services (e.g., physiotherapist, psy- For the results shown in Table 7, I use the onset of a new chotherapist, health visitor at home) over the last 12 months. health condition as an alternative health shock. Given that While the effects are largest in the 3-year sample, where the the presence of health conditions is likely to be more objec- results capture the results in the years immediately after the tive than self-reported health status, these estimates can health shock, the results remain relatively large and statisti- potentially provide additional robustness to the findings cally significant for the longer sample periods. Given that shown in Table 4 by removing concerns about the use of spending a night in the hospital or frequent doctor visits self-assessed health. As shown in Table 2, the number of means lost time at work, the observed changes in health care treated individuals captured with this alternative definition usage can potentially explain the earnings losses to some of health shock is larger than for the drop in health status. extent. Table 7 shows that the negative effects on both labor and The final two columns of Table  8 additionally provide household income are consistent with the results from the evidence that treated individuals are more likely to pay for health status shock, with all effects being statistically signifi- any health care services following the health care shock. The cant at the 1% level. The results for employment indicate the nearest neighbor matching results suggest that the effect is onset of a new health condition did not affect employment on largest for the 7-year sample, again indicating that the effects the extensive margin, which again suggests that other factors on health are persistent. Given that only a small share of explain the losses of earnings and household incomes after individuals in my samples report that they have any health 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 9 Effects of health shocks on work productivity Hourly wage (£ per hour) Work limited by health Having difficulty to concentrate Feeling constantly under strain Nearest neigh- Kernel match- Nearest neigh- Kernel match- Nearest neigh- Kernel match- Nearest neigh- Kernel matching bor matching ing bor matching ing bor matching ing bor matching 3 years − 0.2933 − 0.6774 0.1953*** 0.1915*** 0.1806*** 0.1721*** 0.1493*** 0.1570*** (0.3998) (0.4684) (0.0180) (0.0182) (0.0213) (0.0178) (0.0217) (0.0277) 5 years − 1.4467*** − 1.3654*** 0.0837*** 0.0922*** 0.0809*** 0.0789*** 0.1470*** 0.1535*** (0.1870) (0.3897) (0.0144) (0.0167) (0.0182) (0.0239) (0.0271) (0.0279) 7 years − 0.8068** − 0.0947 0.1060*** 0.1007*** 0.1833*** 0.1643*** 0.1956*** 0.1677*** (0.3794) (0.3334) (0.0158) (0.0200) (0.0240) (0.0241) (0.0239) (0.0291) 9 years − 2.0683*** − 2.0709*** 0.0860*** 0.0857*** 0.0230* 0.0362** 0.1032*** 0.0712*** (0.2474) (0.7561) (0.0277) (0.0234) (0.0120) (0.0180) (0.0338) (0.0338) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses *p < 0.10, **p < 0.05, ***p < 0.01 care expenditures, the increase of paying for health care ser- can provide more evidence on how work performance can vices of 5.30% points (p < 0.01) corresponds to an increase be affected by health shocks. of 52.01% compared to prior to the health shock. The next two columns show that treated workers are These observed changes in health care expenditures significantly more likely to report that their health is lim- could furthermore explain the fact that household income iting their work. Similar to changes in health care usage, reductions following health shocks are even larger than the the effects are largest in the year after the health shock. losses in labor earnings, as previously shown in Tables 4 Using the 3-year sample, I observe a 19.53% point increase and 7. Other household members might reduce their work in the likelihood of reporting health-related work limita- time to support the family members with health issues with tions (p < 0.01). While the effects are smaller for the three their doctor visits, which goes along with increased health longer sample periods, they still show statistically signifi- expenditures. While increases in health care expenditures are cant increases (p < 0.01). The other two proxies of work observable for the first three sample periods, no statistically productivity I examine are reporting having difficulties to significant effects are found for the 9-year sample. concentrate [5] and being constantly under strain [27]. The DD matching estimates obtained for these two outcomes The effects on worker’s productivity provide additional evidence that reductions in work pro- ductivity might explain the losses of labor income to some Another potential channel through which health shocks extent. Again, the effects are quite large and remain persis- can affect labor market outcomes are changes in the level tent across the different sample periods. Overall, the results of work productivity. In Table 9, I show the effects on four in Table  9 suggests that individuals who suffered from a proxies for work productivity for the sample of people who sudden health shock are less likely to perform the same tasks work throughout the sample period. compared to prior to the health shock. First, I examine whether health shocks affect the hourly wages of individuals who remain in the workforce. The DD results provide evidence that wage rates declined substan- Heterogeneous effects tially for workers who experienced adverse health events compared to those who did not. While the estimates for the In a number of additional specifications, I examine whether 3-year period are relatively small and imprecisely estimated, the effects of sudden health declines on labor earnings differ I find that hourly wages are reduced by £2.07 (p < 0.01) across subgroups of the population. Table 10 presents nearest when analyzing the 9-year sample. These estimates suggest neighbor DD matching results across gender, education level, that individuals who remain in the workforce experience job classifications, and age. Using data from the Netherlands, less wage growth than those in the control group following García-Gómez et al. [24] find that health shocks have larger a health shock. One potential explanation for this could be that they are either not able to perform the same tasks or All results in Tables  8 and 9 are obtained using the drop in health take longer to complete the same tasks as compared to prior status as the health shock. Similar to the previous section, the results to the onset of the health shock. The remaining columns of remain consistent when using the onset of a health condition as the Table 9 examines several proxies for labor productivity that health shock. These additional results are not shown in the paper, but are available upon request. 1 3 O. Lenhart Table 10 heterogeneous effects of health shocks on earnings (health status) Total labor income (£ per year) 3-year 5-year 7-year 9-year Panel A: gender  Male − 2535.31*** (615.30) − 6576.00*** (608.28) − 5552.71*** (721.11) − 3248.65*** (517.66)  Female − 615.02 (584.40) − 1351.57*** (506.31) − 1310.93*** (446.87) − 2131.27*** (649.38) Panel B: education  Advanced degree − 2157.03*** (523.83) − 3166.42*** (572.38) − 3255.57*** (555.75) − 3151.08*** (863.72)  Basic degree/low education − 935.79** (418.38) − 1253.72*** (367.72) − 2592.76*** (497.77) − 2771.53*** (404.12) Panel C: job classification  Managerial/professional job − 1966.34*** (722.16) − 3411.18*** (714.28) − 3250.00*** (965.94) − 6507.92*** (698.68)  Skilled labor 150.67 (386.22) − 2066.42*** (447.94) − 433.09 (554.37) − 2436.33*** (680.92)  Unskilled labor 289.68 (844.24) − 1349.32*** (459.43) − 91.13 (1142.81) − 2079.77*** (475.26) Panel D: age  Below 40 years − 1928.86*** (419.86) − 3845.26*** (599.12) − 2435.64*** (545.65) − 3582.65*** (878.89)  At least 40 − 1110.15* (650.25) − 3836.11*** (566.79) − 2944.72*** (885.39) − 2415.16*** (839.80) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 effects on the income of men, which they relate to the fact that Discussion and conclusions males are accounting for greater shares of household earnings. Using longitudinal data from the USA, Charles [11] further- The findings in this study provide evidence that health more provides evidence that the effects of health shocks on shocks significantly affect the labor market outcomes of earnings are increasing with age. He provides two explana- individuals in the UK for several years after the decline tions for this: (1) older persons have accumulated more human in health. García-Gómez et al. [24] suggest that negative capital that can be destroyed by negative health events; (2) effects of health shocks on labor markets can exist either any subsequent recovery in earnings will be weaker for older due to incentives created by disability benefits or due to individuals. labor market institutions constraining the responsiveness My findings in Panel A confirm the results by García- of wages to reduced productivity. Given that the disability Gómez et al. [24]. For all four sample periods, the effects of benefit scheme in the UK provides benefits at a flat rate, it health status declines on earnings are substantially larger for creates very little incentives for individuals to voluntary male individuals. In the 5-year sample period, a health shock reduce their employment compared to other countries, is shown to reduce labor earnings of men by £6576.00, com- which provide disability benefits that are closely tied to pared to a reduction of earnings of £1351.57 for women (both previous earnings [58]. This suggests that the observed p < 0.01). Similar to García-Gómez et al. [24], I find that men reductions in labor market participation following health have substantially higher pre-shock earnings than women, shocks are not driven by incentives provides by disability which could explain the different effects to some extent. Panels benefits. B and C additionally provide evidence that health shocks have This paper shows that the declines are not entirely stronger effects on labor market outcomes of individuals with driven by changes in employment status, but are also higher education levels and for those who work in managerial observable for individuals who remained employed. Addi- or professional jobs. Again, differences, in income prior to the tionally, the study provides first evidence that changes in health shock can potentially explain the larger effects for these work productivity is a mechanism through which health two groups. Finally, the results in Panel D do not indicate that shocks lead to lower labor earnings. Individuals who suffer the effects differ largely across age groups. 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data at the Southern Economics Annual Conference in Tampa, Florida in sudden health declines are shown to be limited in work- November 2017. related activities and to have difficulties concentrating in the following years, suggesting lower levels of work pro- Compliance with ethical standards ductivity and the inability to complete the same tasks the were able to perform before the health shock. Given that Conflict of interest The author declares that he has no conflict of inter - my results suggest that the negative effects on work pro- est. ductivity are still observable several years after the health shock, policymakers and employers should think about Open Access This article is distributed under the terms of the Crea- tive Commons Attribution 4.0 International License (http://creat iveco ways how the reintegration of employees can be improved mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- and significant productivity loss can be avoided. tion, and reproduction in any medium, provided you give appropriate Additionally, despite the provision of universal health credit to the original author(s) and the source, provide a link to the care through the NHS in the UK, I find significant increases Creative Commons license, and indicate if changes were made. in the likelihood with which individuals pay for health care services following the onset of a health shock. A likely explanation for this might be that individuals want to forego References long waiting times before receiving treatment and decide to pay for private care. The NHS has been dealing with 1. Abadie, A., Imbens, G.W.: Large sample properties of match- the issue of long waiting times for several decades. Using ing estimators for average treatment effects. Econometrica. 74, 235–267 (2006) official data from the NHS, Murray [ 42] shows that 2015 2. Abadie, A., Imbens, G.: On the failure of the bootstrap for match- marked the first year since the introduction of the NHS in ing estimators. Econometrica 76(6), 1537–1557 (2008) which the standard that at least 92% of patients receive their 3. Abadie, A., Imbens, G.: Matching on the estimated propensity treatment within 18 weeks was not achieved, suggesting that score. Econometrica 84(2), 781–807 (2016) 4. Adda, J., Banks, J., von Gaudecker, H.-M.: The Impact of income the issue is becoming worse over time. Thus, policymakers shocks on health: evidence from cohort data. J. Eur. Assoc. 7(6), should discuss ways on how to stop the trend of increasing 1361–1399 (2009) waiting times since they can cause further harm to individu- 5. Antikainen, R., Lönnqvist, A.: Knowledge Work Productivity als suffering health shock by affecting household incomes, Assessment. Institute of Industrial Management, Tampere Uni- versity of Technology (2006) financial stress and overall well-being in the following years. 6. Arie, S.: Is the UK government still serious about reducing smok- Previous data on the overall level of health for the UK ing? BMJ Clin. Res. 356, j1426 (2017) population is mixed. A report of the Economist Intelligence 7. Au, D., Crossley, T., Schellhorn, M.: The effects of health shocks Unit that compares healthcare inputs and outcomes across and long-term health on the work of older Canadians. Health Econ. 14, 999–1018 (2005) 166 countries ranked the UK 23rd in terms of performance 8. Becker, S., Ichino, A.: Simple and bias-corrected matching estima- [19]. A report by the Commonwealth Fund in 2014 places the tors for average treatment effects in STATA. Stata J. 1 (1), 1–18 UK second to last in the ‘healthy lives’ category, which uses (2002) indicators of population health outcomes, including mortality, 9. Bound, J., Schoenbaum, M., Stinebrickner, T.R., Waidmann, T.: The dynamic effects of health on the labor force transitions of infant mortality and life expectancy [17]. Evidence from the older workers. Lab. Econ. 6, 179–202 (1999) Global Burden of Disease Study shows that the leading risk 10. Case, A., Lubotsky, D., Paxson, C.: Economic status and health factors for premature death in the UK are linked to lifestyle, in in childhood: the origins of the gradient. Am. Econ. Rev. 92(5), particular to dietary risk, tobacco smoking, high blood pres- 1308–1334 (2002) 11. Charles, K.K.: The longitudinal structure of earnings losses sure, and alcohol consumption [30]. Given that these causes among work-limited disabled workers. J. Hum. Resour. 38(3), are preventable, campaigns to promote healthier lifestyles and 618–646 (2003) make people more aware of health risks could be success- 12. Contoyannis, P., Rice, N.: The impact of health on wages: evi- ful in improving health outcomes and in preventing the onset dence from the British household panel survey. Empir. Econ. 26, 599–622 (2001) of sudden health shocks for people in the workforce. Since 13. Currie, J., Stabile, M.: Socioeconomic status and child health: 1998, the UK government has had success in reducing smok- why is the relationship stronger for older children? Am. Econ. ing of the population from 28 to 18% in 2015. It has done Rev. 93(5), 1813–1823 (2003) so by changing taxation, by increasing public awareness of 14. Currie, A., Shield, M.A., Price, S.W.: The child health/family income gradient: evidence from England. J. Health Econ. 26, the harm caused by smoking and by helping people quit [6]. 212–232 (2007) These policy interventions should be continued and expanded 15. Currie, J., Duque, V., Garfinkel, I.: The great recession and moth- to other lifestyle choices that cause harm to people’s health. er’s health. Econ. J. 125(588), F311–F346 (2015) 16. Dano, A.M.: Road injuries and long-run effects on income and Acknowledgements I would like to thank the participants at the Col- employment. Health Econ. 14(9), 955–970 (2005) lege of Business Brown Bag Seminar Series at the University of West 17. Davis, K., Stremikis, K., Squires, D., Schoen, C.: Mirror, Mirror Florida as well as the participants of the session “Labor Economics” on the Wall: How the Performance of the U.S. Health Care System 1 3 O. Lenhart Compares Internationally. New York: The Commonwealth Fund: 40. Miller, D.L., Page, M.E., Stevens, A.H., Filipski, M.: Why are 2014. (2014). http://www.commo nweal thfun d.org/~/media /files recessions good for your health? Am. Econ. Rev. 99(2), 122–127 /publi catio ns/fund-r epor t/2014/jun/1755_davis _mir r o r_mir r o (2009) r_2014.pdf. Accessed 15 May 2017 41. Mora, R., Reggio, I.: Didq: A Command for treatment-effect 18. Disney, R., Emmerson, C., Wakefield, M.: Ill-health and retire- estimation under alternative assumption. Stata J. 15(3), 796–808 ment in britain: a panel data-based analysis. J. Health Econ. 25, (2015) 621–649 (2006) 42. Murray, R.: What is Happening to NHS Waiting Times? The 19. Economist Intelligence Unit. http://stateofr efor m.com/wp-content/ King’s Fund. (2016). https ://www.kings fund.org/blog/2016/-2/ uploads/2015/11/Healt hcar e-outco mes-inde x-2014.pdf . Accessed nhs-waiti ng times . Accessed 16 May 2017 16 May 2017 (2014) 43. Myde, B., Pelletier, B., Lynch, W.: The relationship between 20. Eliason, M., Storrie, D.: Does job loss shorten life? J. Hum. health risks and work productivity. J. Occup. Environ. Med. 46(7), Resour 44(2), 277–302 (2009) 737–745 (2004) 21. Eliason, M., Storrie, D.: Job loss is bad for your health: Swedish 44. Neumayer, E.: recessions lower (some) mortality rates: evidence evidence on cause-specific hospitalization following involuntary from Germany. Soc. Sci. Med. 58(6), 1037–1047 (2004) job loss. Soc. Sci. Med. 68(8), 1396–1406 (2009) 45. Nielsen, T.H.: The relationship between self-rated health and hos- 22. García-Gómez, P., López-Nicolás, A.: health shocks, employ- pital records. Health Econ. 25, 497–512 (2016) ment and income in the spanish labour market. Health Econ. 15, 46. Pelkowski, J., Berger, M.C.: The impact of health on employment, 997–1009 (2006) wages, and hours worked over the life cycle. Q. Rev. Econ. Fin. 23. García-Gómez, P.: Institutions, health shocks and labour outcomes 44, 102–121 (2004) across Europe. J. Health Econ. 30(1), 200–213 (2011) 47. Propper, C., Rigg, J., Burgess, S.: Child health: evidence on the 24. García-Gómez, P., van Kippersluis, H., O’Donnell, O., van roles of family income and maternal mental health from a UK Doorslaer, E.: Long term and spillover effects of health shocks on birth cohort. Health Econ. 16, 1245–1269 (2007) employment and income. J. Hum. Resour. 48(4), 873–909 (2013) 48. Reinhold, S., Jürges., H.: Parental income and child health in Ger- 25. Grossman, M.: On the concept of health capital and the demand many. Health Econ. 21(5), 562–579 (2012) for health. J. Polit. Econ. 80(2), 223–255 (1972) 49. Riphahn, R.: Income and employment effects of health shocks. 26. Hagan, R., Jones, A.M., Rice, N.: Health shocks and the hazard a test case for the German welfare state. J. Popul. Econ. 12(3), rate of early retirement in the ECHP. Swiss J. Econ. Stat. 144, 363–389 (1999) 323–335 (2008) 50. Rosenbaum, P.R., Rubin, D.B.: The central role of the propen- 27. Halkos, G., Bousinakis, D.: The effect of stress and satisfaction on sity score in observational studies for causal effects. Biometrika. productivity. Int. J. Prod. Perform. Manag. 59(5), 415–431 (2010) 70(1), 41–55 (1983) 28. Heckman, J.J., Ichimura, H., Todd, P.: Matching as an econometric 51. Ruhm, C.J.: Are recessions good for your health? Q. J. Econ. evaluation estimator. Rev. Econ. Stud. 65, 261–294 (1998) 115(2), 617–650 (2000) 29. Idler, E.L., Benyamini, Y.: Self-rated health and mortality: a 52. Schaller, J., Stevens, A.H.: Short-run effects of job loss on health review of twenty-seven community studies. J. Health Soc. Behav. conditions, health insurance, and health care utilization. J. Health 38(March), 21–37 (1997) Econ. 43, 190–203 (2015) 30. Institute for Health Metrics and Evaluation.: Global Burden of 53. Schmitz, H.: Why are the unemployed in worse health? The causal Disease Profile: United Kingdom. Seattle: Institute for Health effect of unemployment on health. Lab. Econ. 18, 71–78 (2011) metrics and Evaluation: 2015. (2015). http://www.health data. org/ 54. Smith, J.P.: Unraveling the SES: Health Connection. Population unite d-kingd om. Accessed 16 May 2017 and Development Review 30; Supplement: Aging, Health, and 31. Jiménez-Martín, S., Labeaga, J.M., Prieto, V.C.: A sequential Public Policy (pp. 108–132)(2004) model of older workers’ labor force transitions after a health 55. StataCorp.: Stata Treatment-Effects Reference Manual: Potential shock. Health Econ. 15, 1033–1054 (2006) Outcomes/Counterfactual Outcomes: Release 13. Statistical Soft- 32. Jones, A.M., Rice, N., Roberts, J.: Sick of work or too sick to ware. College Station, TX: StataCorp LP (2013) work? Evidence on self-reported health shocks and early retire- 56. Sullivan, D., von Wachter, T.: (2009). Job displacement and ment from the BHPS. Econ. Model. 27(4), 866–880 (2010) mortality: an analysis using administrative data. Q. J. Econ. 33. Johnston, D.W., Propper, C., Shields, M.A.: Comparing subjective 1265–1306 and objective measures of health: evidence for the income/health 57. Van Doorslaer, E., Wagstaff, A., van der Burg, H., Christiansen, gradient. J Health Econ. 28, 540–552 (2009) T., De Graeve, D., Duchesne, I., Gerdtham, U.-G., Gerfin, M., 34. Khanam, R., Nghiem, H.S., Connelly, L.B.: Child health and the Geurts, J., Gross, L., Häkkinen, U., John, J., Klavus, J., Leu, R.E., income gradient: evidence from Australia. J. Health Econ. 28(4), Nolan, B., O’Donnell, O., Propper, C., Puffer, F., Schellhorn, M., 805–817 (2009) Sundberg, G., Winkelhake, O.: Equity in the delivery of health 35. Lechner, M., Vazquez-Alvarez, R.: The effect of disability on care in Europe and the US. J. Health Econ. 19, 553–583 (2000) labour market outcomes in Germany. Appl. Econ. 43(4), 389–412 58. Van Doorslaer, E., Koolman, X.: Explaining the differences in (2011) income-related health inequalities across European countries. 36. Lindeboom, M., Llena-Nozal, A., van der Klaauw: Health shock, Health Econ. 13, 609–628 (2004) disability and work. Lab. Econ. 43, 186–200 (2016) 59. Vizard, P., Obolenskaya, P.: Labour’s Record on Health: Policy 37. Marcus, J.: The effect of unemployment on the mental health of Spending and Outcomes. Social Policy in a Cold Climate Working spouses—evidence from plant closures in Germany. J. Health Paper WP02. London: CASE (2013) Econ. 32, 546–558 (2013) 60. Zucchelli, E., Harris, A., Jones, A.M., Rice, N.: The effects of 38. McInerney, M., Mellor, J.M.: Recessions and seniors’ health, health shocks on labour market exits: evidence from the HILDA health behaviors, and healthcare use: analysis of the medicare survey. Aust. J. Lab. Econ. 13(2), 191–218 (2010) current beneficiary survey. J. Health Econ. 31, 744–751 (2012) 39. McInerney, M., Mellor, J.M., Nicholas, L.H.: Recession depres- sion: mental health effects of the 2008 stock market crash. J. Health Econ. 32, 1090–1104 (2013) 1 3 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The European Journal of Health Economics Springer Journals

The effects of health shocks on labor market outcomes: evidence from UK panel data

Free
16 pages

Loading next page...
 
/lp/springer_journal/the-effects-of-health-shocks-on-labor-market-outcomes-evidence-from-uk-FbR6rxkj9n
Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Medicine & Public Health; Public Health; Health Economics; Public Finance; Pharmacoeconomics and Health Outcomes; Political Economy/Economic Policy; Health Care Management
ISSN
1618-7598
eISSN
1618-7601
D.O.I.
10.1007/s10198-018-0985-z
Publisher site
See Article on Publisher Site

Abstract

This study examines the link between health shocks and labor market outcomes in the United Kingdom. For sample periods of up to 9 years, I use longitudinal data from the British Household Panel Survey to test how sudden health shocks affect a number of labor market outcomes, such as labor and household income, employment status, and hours worked. Addition- ally, the study examines potential mechanisms underlying the link between health declines and labor market outcomes. By estimating propensity score matching difference-in-differences models, the study shows that sudden health declines lead to significant and persistent reductions in earnings. The effects are strongest for individuals experiencing severe health shocks, males, individuals with higher education and those working in managerial jobs. When examining potential channels, I provide evidence that increased health care expenditures and health care usage as well as reduced work productivity can explain the observed effects on labor market outcomes. Keywords Health shocks · Labor market · Mechanisms · United Kingdom JEL Classification C23 · I10 · I12 · J60 Introduction employment and earnings, policymakers should discuss ways to provide improved assistance to workers experienc- Despite being previously examined by several research- ing sudden declines in health in dealing with the situation ers, there is still uncertainty about the relationship between and allowing them to return to work. Additionally, there labor market and health outcomes. While the majority of is little evidence about the potential mechanisms through existing studies has focused on examining health effects which health shocks could affect immediate and long-term following changes in income and employment, there has labor market outcomes. Using longitudinal data from the been growing interest in learning more about the effects British Household Panel Survey (BHPS) for up to 9 years, of health on labor market outcomes in recent years. Given this study examines the effects of health shocks on several the close link between employment status and both health labor market outcomes and explores potential channels and health insurance coverage, an improved understand- through which labor market outcomes are affected follow - ing of how labor market outcomes are affected following ing health shocks. health shocks is relevant for policymakers. If health shocks The analysis of this paper adds to the existing literature have significant long-term negative spillover effects on in two ways. First, the study uses longitudinal data from the UK. I am following the individuals over four different sam- ple periods: 3, 5, 7 and 9 years. Besides testing for the imme- Electronic supplementary material The online version of this diate ee ff cts of health shocks on labor market outcomes, this article (https ://doi.org/10.1007/s1019 8-018-0985-z) contains supplementary material, which is available to authorized users. also allows me to provide evidence whether people adapt to health shocks over time or whether the effects are persistent. * Otto Lenhart These findings could indicate whether there are potential ottolenhart@gmail.com labor market institutions in place that make it challenging to Department of Economics, Duncan Wing, Strathclyde reintegrate workers into the workforce following the health Business School, University of Strathclyde, 199 Cathedral issues. Street, Glasgow, UK Vol.:(0123456789) 1 3 O. Lenhart Second, besides testing for the presence of a causal link conditions and health outcomes. These findings are mixed: from health to labor market outcomes, this study examines earlier work provides evidence that economic downturns potential mechanisms through which health shocks can actually improve health outcomes [40, 44, 51], while more affect labor market outcomes. An examination of potential recent work suggest that health declines along with the channels is important from a policy and a social welfare economy [15, 38, 39]. perspective since it provides evidence for how policymak- A large number of studies have examined the relation- ers can potentially help prevent substantial losses in pro- ship between income and health. Following the pioneer- ductivity following health shocks. This study examines the ing study by Case et al. [10], several papers have also role of three mechanisms: (1) changes in the frequency of provided evidence for the presence of a strong positive health care usage. This can impact labor market outcomes association between household income and health [4, 13, by taking time away from work and work-related activi- 14, 34, 47, 48]. This phenomenon has become known ties; (2) changes in the likelihood of paying for health care. as the income gradient in health. In more recent year, While universal health coverage is provided in the UK by researchers have expressed the need to test for the causal the National Health Service (NHS), individuals have the nature between income and health by pointing out that option to purchase additional private care to forego long the presence of a positive association could be the result waiting times and receive potentially lower quality care; (3) of third factors that inf luence both health and income or changes in the worker’s productivity. Observing changes due to reverse causality, which exists if health outcomes is the productivity of workers who previously experienced influence people’s employment status and, therefore, negative health shocks could suggest that employees should their income. find ways on how to allow workers to be better reintegrated Several previous studies have examined the effects of to the workforce, while policymakers should create an envi- health shocks on labor market outcomes. The majority of ronment that mitigates the risks to employees experiencing early work on the topic has focused on elderly groups of sudden health shocks. the population and the effects of health on retirement [ 7, 9, The study finds that health shocks, captured by declines 18, 26, 31, 32, 49, 54, 60]. These studies established that in self-reported health status and the onset of health condi- older adults are significantly less likely to be employed and tions, significantly affect labor market outcomes. Negative more likely to retire following the occurrence of a health health events are shown to reduce labor earnings, household shock. Other studies additionally show that health shocks income, and the likelihood of being employed. The negative have negative labor market effects for younger individuals effects are found for all four variations of the sample length, by examining several types of health shocks. These include which suggests that health shocks have lasting impacts on the presence of permanent health conditions [46], reduced labor market outcomes. When examining the effects across psychological health [12], injuries from road accidents different subgroups of the population, I find larger effects [16], the onset of disability [11, 36], reduced physical for males, higher educated individuals, and those working health [22, 23] and sudden illness [24]. managerial and professional jobs. When examining poten- Van Doorslaer and Koolman [58] find that income- tial mechanisms, the study provides evidence that increased related health inequalities in the UK are larger than in health care usage and health care expenditures as well as most other European countries, while García-Gómez et al. reduced work productivity can explain the observed per- [24] suggest that differences in the provision of disability sistent negative effects of health shocks on labor market benefits could explain these differences across nations. outcomes. The authors argue that relating the size of benefits to previous earnings, as done in the Netherlands, reduces the average income loss from health shocks compared to Labor market outcomes and health: when benefits are paid at a flat rate like in the UK. Besides previous evidence examining the effects of negative health events on labor market outcomes, this study tests for additional potential A number of previous studies have examined health effects mechanisms that can explain health-related inequalities in of negative employment shocks on people’s health. It has the UK (health care usage, health care expenditures, and been established that negative employment events such as work productivity). mass layoffs, plant closings and job loss have significant negative effects on health outcomes of affected individuals [20, 21, 37, 52, 53, 56]. Furthermore, other studies have examined the association between worsened economic 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 1 DD model setup Treatment group Control group Health Employment Health Employment Panel A 3-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (shock) Fair/poor/very poor Working Excellent/very good Working  2002 (post) Fair/poor/very poor Excellent/very good Panel B 5-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (shock) Fair/poor/very poor Working Excellent/very good Working  2003 (post) Fair/poor/very poor Excellent/very good  2004 (post) Excellent/very good Panel C 7-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (pre) Excellent/very good Working Excellent/very good Working  2003 (shock) Fair/poor/very poor Working Excellent/very good Working  2004 (post) Fair/poor/very poor Excellent/very good  2005 (post) Excellent/very good  2006 (post) Excellent/very good Panel D 9-year  2000 (pre) Excellent/very good Working Excellent/very good Working  2001 (pre) Excellent/very good Working Excellent/very good Working  2002 (pre) Excellent/very good Working Excellent/very good Working  2003 (pre) Excellent/very good Working Excellent/very good Working  2004 (shock) Fair/poor/very poor Working Excellent/very good Working  2005 (post) Fair/poor/very poor Excellent/very good  2006 (post) Excellent/very good  2007 (post) Excellent/very good  2008 (post) Excellent/very good the self-reported health measure is reduced since each Data individual’s health is only compared to his or her own prior assessment. This allows controlling for the fact that This study uses data from waves 10–18 (2000–2008) of each respondent may have their own scales in ranking their the British Household Panel Survey (BHPS), a nationally health (reference bias). Furthermore, in comparison to the representative panel survey of private households in Great two other commonly used UK datasets with detailed infor- Britain that started interviewing 10,300 individuals from mation on labor market outcomes (Labor Force Survey and 5,500 families in 1991. The use of the BHPS provides New Earnings Survey), the BHPS also provides informa- several advantages for the purpose of this study. Due to tion on several health outcomes. Finally, the BHPS gives it longitudinal nature, the dataset allows accounting for a complete representation of incomes across the pay dis- time-invariant unobserved heterogeneity and composi- tribution since it questions all individuals above 15 years tional selection. The potential for measurement error in of age who live in the household at the time of the inter- view. Given that individuals in the UK become eligible to receive state pensions at the age of 65, the sample is Taylor (1998) provides a full description of the sampling strategy restricted to all individuals between the ages 18 and 64 in applied in the initial wave in order to design a nationally representa- tive sample of the British population. 1 3 O. Lenhart the surveyed households for whom information on health off from work to help their family member with the doc- and labor market outcomes is available. tor appointments. Given that the analysis uses four different The analysis uses two different types of health shocks. In sample lengths, it is able to provide evidence whether any the main specification, which is presented in Table  1, health potential changes in health care usage only occur immedi- shocks are defined as a decline in self-reported health sta - ately after the health shock or whether these changes persist tus. Self-assessed health is categorized from 1 (= excellent) for several years. to 5 (= very poor) in the BHPS. This measure of health While health shocks can reduce the labor force partici- has been shown to be a good predictor of other health out- pation of people who can no longer work, another channel comes, including mortality [29], future health care usage through which adverse health events can affect earnings is [57] and hospitalizations [45]. To remove concerns about by reducing labor productivity. Workers might not be able potential reporting heterogeneity of health status, Johnston to perform the same tasks or might need longer to complete et al. [33] suggest the additional use of more objective health the same tasks compared to before the health shock. Without outcomes. In an additional specification, health shocks are empirically testing for the presence of this channel, García- defined as the onset of a new health condition. In each wave, Gómez and López-Nicolás [22] point out that productiv- the BHPS asks respondents whether they suffer from any ity losses could either be absorbed by the employer or by of the following 15 health conditions: body pain, migraine, the inability to work extra time. Using a sample of 2264 skin issues/allergy, asthma/chest pain, anxiety, heart or workers, Myde et al. [43] provide evidence for a strong link blood pressure, hearing problems, stomach/liver/kidney between health risks and self-reported work productivity. pain, seeing problems, epilepsy, diabetes, alcohol or drug To capture whether changes in work productivity are a problems, stroke, cancer or other conditions. As indicated by potential mechanisms underlying the relationship between García-Gómez [23], using the onset of health conditions as a health shocks and labor market outcomes, this analysis health shock could provide evidence regarding any potential examines four proxies for work productivity: (1) hourly anticipation effects [23]. wages of workers, (2) the likelihood of reporting that cur- Besides examining the short- and long-run effects of rent work is limited by one’s health, (3) the likelihood of health shocks on earnings and employment, this study addi- having difficulty to concentrate, and (4) the likelihood of tionally tests for the role of potential mechanisms underlying constantly feeling under strain. While hourly wages is the the link between sudden declines in health and labor market most direct way of measuring productivity, the other three outcomes. According to the Grossman [25], health can be outcomes should provide additional evidence for poten- viewed as both a consumption and investment good since it tial changes in labor productivity following the onset of not only makes people feel better, but it also increases the health shocks. Antikainen and Lönnqvist [5] suggest that, number of healthy days to work and to earn income. Gross- especially in “knowledge-intensive” organizations, where man [25] states that to keep certain levels of health capital, knowledge has more importance than other inputs, work individuals invest into their health through channels such as performance can be negatively affected by health problems market inputs of health care, diet and exercise. or other personal issues because they are highly dependent The first mechanism examined is changes in health care on the ability to concentrate. Using a factor analysis model, usage, which is captured by three outcomes: (1) the likeli- Halkos and Bousinakis [27] provide empirical evidence that hood of having more than five annual doctor visits, (2) the increased stress leads to reduced work productivity. Using likelihood of having spent a night at a hospital in the previ- the four proxies of work productivity listed above, my study ous year, (3) the likelihood of having used a number of other tests whether individuals who suffered health shocks are not health services, and (4) changes in the likelihood of having able to perform the same tasks compared to prior to the paid for health care services in the previous year. While the experiencing the sudden health decline. NHS provides universal coverage to all individuals in the UK, two serious issues that the program has been dealing with are the quality of care and long waiting times [59]. To Econometric methods avoid these problems, individuals have the option to pur- chase additional private care to forego long waiting times DD matching models before seeing a doctor in some cases. Persistent increases in out-of-pocket expenditures on health can be associated with Similar to García-Gómez and López-Nicolás [22], this labor market outcomes in two ways: (1) earned income of study estimates propensity score matching combined with individuals recovering from health shocks could be reduced difference-in-differences (DD) models to estimate the Aver - due to time spent away from work for doctor visits, and (2) age Treatment Effect on the Treated (ATET). This empirical household income could be affected to a larger extent than strategy allows me to compare the distributions of outcomes individual income if other household members take time for treated individuals (who suffer the health shock) with 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data the distributions of outcomes of matched individuals in propensity score, and (2) kernel matching on the propensity the control group, without having to make any functional score. Since there are no reasons to expect one of the meth- form assumptions. As pointed out by García-Gómez and ods to be preferable to the other, the use of both methods López-Nicolás [22], matching frameworks are often criti- allows the analysis to test for the robustness of the observed cized for assuming away potential biases that might exist effects. When examining the ATET of health shocks on due to unobserved heterogeneity. The authors argue that one labor market outcomes, the analysis examines four different solution to remove concerns about such biases is the use of outcomes: (1) total annual labor income, (2) total annual longitudinal data that includes data from before and after the household income, (3) the probability of being employed, health shock. Using longitudinal data from the BHPS, this and (4) hours worked per week. Standard errors are obtained study is able to first difference the outcomes of the treated following recent findings by Abadie and Imbens [3 ], who and the controls to eliminate any unobservable fixed effects established how to take into account that propensity scores that influence the selection into the groups as well as the are estimated in the first stage. They show that ignoring outcomes of interest [22]. Thus, the estimated ATET’s are this fact when estimating average treatment effects on the weighted averages of the differences in differences between treated in the second stage may lead to confidence inter - each of the treated individuals and his/her matched control. vals that are either too large or small. By showing that the The study uses estimated propensity scores, which cal- propensity matching estimator have approximately Normal culate the probability of treatment given a vector of observ- distributions, Abadie and Imbens [3] provide evidence that able variables, to match individuals who receive a health the matching on estimated propensity score is more efficient shock to individuals that are similar but do not experience than matching on the true propensity score in large samples. the health shock. The propensity scores are based on pre- treatment variables and are estimated using probit models. Assignment of treatment and control groups Observable characteristics that are included to obtain the propensity scores are age, gender, household size, educa- This assignment of individuals into treatment and control tional attainment, and area. In additional specifications, I group used in this study is similar to previous work by furthermore include information on lagged health status as García-Gómez and López-Nicolás [22] and García-Gómez a covariate when estimating the propensity scores. Follow- [23] as well as Lechner and Vázquez Álvarez [35]. A crucial ing Rosenbaum and Rubin [50], the use of a function of X, challenge when estimating the effects of health shocks on called the propensity scores P(X), rather than a potentially labor market outcomes is the fact that many health and labor high-dimensional vector of covariates implies that: market outcomes are potentially jointly determined by many people. The use of propensity score matching DD model can E Y  D = 1, P(X) = E Y  D = 0, P(X) , (1) 0 0 overcome this concern by identifying arguably exogenous where Y denotes the untreated state, D = 1 indicates treat- 0 health shocks that are independent of employment status. ment and D = 0 indicates non-treatment. The analysis of this Table 1 shows the setup for the two groups used in the DD study follows Heckman et al. [28] difference-in-difference models estimated in this study, which analyzes four varia- matching methods, which uses both comparisons between tions of sample length to test for both immediate and the treated and non-treated, and differencing over time. Thus, long-term effects of adverse health events on labor market the conditions needed to identify the ATET using the differ - outcomes. Despite different sample periods, all three models ence-in-difference matching estimator is: share the following characteristics: E Y − Y  D = 1, P(X) = E Y − Y  D = 0, P(X) , � � 0,t 0,t 0,t 0,t 1. Individuals from both treatment and control group are (2) in excellent or very good health and are working in the where t and t′ represent the post- and pre-treatment peri- pre-treatment period (Pre). ods, respectively. Thus, the ATET provides a weighted aver- 2. Individuals forming the treatment group experience a age of the differences in differences between individuals in health shock in the treatment period (Shock), meaning the treatment and the control group and it is obtained by their health status drops to fair, poor or very poor. Indi- estimating the following equation: ATET = E Y D = 1, P(X) − E Y D = 0, P(X) The analysis is conducted using the “teffects psmatch” command in DID 1t 0t Stata, which incorporates the findings by Abadie and Imbens [3] and − E Y D = 1, P(X) − E Y D = 0, P(X) � � 1t 0t takes into account potential estimation errors in the propensity score (3) due to the fact that the propensity scores were estimated in the first stage. In an earlier paper, Abadie and Imbens [2] show that bootstrap- In the empirical analysis, the study uses two alternative ping, a technique that had previously often been used to obtain stand- methods when matching treated individuals with those in ard errors for propensity score matching estimators, is generally not the control group [8]: (1) nearest neighbor matching on the valid for matching estimators. 1 3 O. Lenhart Table 2 Sample sizes for Health shock: drop in health status Health shock: onset of health condi- treatment and control group tion Treated Control Total Treated Control Total 3-year sample 591 9720 10,311 1620 5034 6654 5-year sample 585 11,155 11,740 1525 5190 6715 7-year sample 504 11,760 12,264 1001 5061 6062 9-year sample 315 11,268 11,583 432 4635 5067 viduals in the control group remain in excellent or very that are present in all waves of the corresponding sample good health. All members of the treatment and the con- periods. Given that the onset of new health conditions occur trol group are working during the period in which the less frequently than declines in health status, the sample treatment occurs. sizes are smaller when using health condition as the health 3. Self-reported health status of individuals in the treat- shock. ment group remains in fair, poor or very poor health in Given that the analysis in this study only includes indi- the first year after the health shock, while individuals in viduals who are present in all survey waves for each sample the control group remain in excellent or very good health length period (either 3, 5, 7 or 9 years), attrition could pose a throughout the post-treatment period (Post). potential issue. The obtained treatment effect estimates could potentially be biased if people drop out of the survey due to Additionally, using the same setup as shown in Table 1, health-related problems. Given the longitudinal data set of I use the onset of a health condition as an alternative health the BHPS, I am able to identify individuals who drop out of shock. Individuals forming the treatment group report the the survey and compare them to those who remain in it and onset of a health condition in the treatment period, while are used in the analysis. Online Appendix Table A1 shows those in the control group report no health conditions comparisons of descriptive statistics for the two groups for throughout the study period. Again, all individuals work in the sample periods of 5, 7, and 9 years. It is noticeable that both the pre-treatment and the treatment period. Given that there are only very small differences in health status between information regarding the presence of health conditions are people remaining in the survey and those who drop out at potentially less subjective than self-assessed health status, some point. Individuals who stay in the BHPS for all the the findings from this additional health shock can remove years in each of the three period analyzed are shown to be concerns about potential reporting heterogeneity of health slightly more likely to be employed and have higher labor status [33]. and household incomes than those who drop out of the sur- By examining a sample of individuals who are employed vey. Table A1 shows that the attrition rates were between 13 during both the pre-treatment period and the year of the and 15% for the sample periods shown in Online Table A1. health shock, the potentially simultaneous determination One of the main assumptions of estimating propensity of health and labor market outcomes is accounted for and score matching DD estimates is that the overlap assump- allows testing for the effects of experiencing health declines tion, which is satisfied when there is a chance of seeing on labor market outcomes in the post-treatment period. This observations in both control and treatment groups at each framework ensures that the observed effects are not the result combination of covariate values. As highlighted in the refer- of reverse causality, which would exist if changes in labor ence manual for Stata Treatment Ee ff cts, the analysis cannot market outcomes lead to the health shock. One assumption predict or otherwise account for the unobserved outcomes of this framework is that there are no anticipation effects, of some individuals if the assumption is violated [55]. Fig- meaning that people report declines in health because they ure  1a–d provide plots of the estimated densities of the expect negative employment shocks to occur in the future. propensity scores for both treatment and control groups for all four sample periods. All the graphs present clear evi- Descriptive statistics dence that the overlap assumption is satisfied since there are chances of seeing observations in both groups at each Table 2 presents sample sizes for the four different sample combination of covariate values. length for each of the two health shocks that examined in Table 3 provides results from covariate balance tests for the study. For the health status shock in the shortest period the matching conducted in the analysis. In well specified of study (2000–2002), the control group consists of 9720 matching models, the covariates should be balanced, which observations and the analysis includes 591 observations for allows for the outcomes to be conditionally independent of the treatment group. The analysis includes only individuals the treatment when conditioning on covariates [55]. The left 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Fig. 1 Density of propensity scores, a 3-year sample, b 5-year sample, c 7-year sample, d 9-year sample side of Table 3 shows the balance test results for the main individuals. The nearest neighbor matching estimates find analysis, while the right side shows differences between reductions in annual labor income in the range of £1181.40 raw and matched data when lagged health information is for the year after the shock in the 3-year sample (p < 0.01) included as a covariate. Overall, the balancing results indi- to £4432.32 for the 9-year sample (p < 0.01), which exam- cate that the matching succeeds in balancing the covariates ines the effects for up to 4 years after the health shock. and reducing the standardized differences between the two The immediate effects on earnings in the year following the groups. drop in health status is slightly smaller in magnitude than estimated by García-Gómez and López-Nicolás [22], who find a decline in income of €1118 (measured in 2001 Euros, Results which corresponds to £1763.31 using the 2001 €/£ conver- sion rate) using Spanish data and a 3-year sample. The effects of health status shocks While the kernel matching estimates also provide evi- dence for declines in labor income in all four periods, two of Table 4 presents propensity score matching DD effects of the effects are statistically insignificant. Overall, given that sudden declines in health status on four labor market out- losses in labor income are larger in magnitude for the longer comes and four different sample lengths. The estimates in the sample periods, the results do not suggest that individuals first two columns provide evidence that health shocks have adapt to the health shock. On the contrary, it appears that substantial negative effects on labor earnings of affected individuals struggle to be reintegrated into the labor force following the declines in health. The next two columns show Table  3 shows balancing test results for the Nearest Neighbor matching analysis. The results are consistent for the kernel matching Income results are adjusted for inflation using the U.K. Consumer analysis. These results are not shown, but are available upon request. Price Index and 2000 as the base year. 1 3 O. Lenhart Table 3 Covariate balance tests Main analysis Analysis with controls for lagged health Standardized differ - Variance ratio Standardized differ - Variance ratio ences ences Raw Matched Raw Matched Raw Matched Raw Matched 3-year  HH size − 0.0175 0.0024 0.9420 0.9163  Education 0.1734 − 0.0018 0.9906 0.9687  Age 0.1071 − 0.0186 1.0413 1.0315  Gender − 0.0022 0.0275 1.0014 1.0016  Area − 0.0335 − 0.0127 0.9285 0.9490 5-year  HH size − 0.0672 0.0417 1.0414 1.1571 − 0.0672 − 0.0310 1.0414 1.1079  Education 0.3427 0.0149 0.9780 0.9559 0.3427 − 0.0757 0.9780 0.9161  Age 0.1533 0.0432 0.9215 0.8796 0.1533 0.0243 0.9215 0.8692  Gender 0.1038 − 0.0715 0.9996 0.9896 0.1038 − 0.0017 0.9996 0.9999  Area − 0.0769 − 0.0320 1.0018 0.9967 − 0.0769 − 0.1255 1.0018 0.9490  Lagged health status – – – – 0.6766 0.0187 0.6010 0.9708 7-year  HH size − 0.1542 0.0182 1.1199 1.2669 − 0.1542 − 0.2428 1.1199 0.9570  Education 0.1278 − 0.0738 0.9103 0.8231 0.1278 0.4209 0.9103 1.3985  Age − 0.0831 − 0.0123 1.0280 1.0092 − 0.0831 0.1321 1.0280 1.1516  Gender − 0.0428 − 0.0892 1.3490 1.9217 − 0.0428 − 0.2305 1.3490 1.1363  Area 0.2269 0.0432 0.7737 0.8212 0.2269 − 0.0232 0.7737 0.7864  Lagged health status – – – – 0.7242 − 0.0782 0.5197 1.0818 9-year  HH size 0.0455 0.0744 0.8414 0.7221 0.0455 0.0440 0.8414 0.9859  Education 0.2734 − 0.0498 1.0289 1.1293 0.2734 0.0232 1.0289 0.8383  Age 0.1377 0.0050 0.8731 0.8611 0.1377 − 0.1599 0.8731 1.0662  Gender − 0.0170 − 0.0145 1.4092 2.0770 − 0.0170 0.0592 1.4092 0.9511  Area 0.2399 0.0667 0.8979 0.9703 0.2399 0.2130 0.8979 0.8791  Lagged health status – – – – 0.7109 − 0.0410 0.6144 0.8386 Table 4 Effects of health shocks on labor market outcomes (health status) Total labor income (£ per year) Total HH income (£ per year) Employed Weekly work hours NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel ing ing ing matching 3 year sample − 1181.40*** − 769.08 − 2834.63*** − 3355.70*** − 0.0068 − 0.0186* − 0.06 (0.57) − 1.14 (0.72) (430.54) (621.24) (756.41) (1065.85) (0.0073) (0.0109) 5-year sample − 3041.75*** − 3948.23*** − 4362.41*** − 4255.36*** − 0.0356*** − 0.0370*** − 1.17** − 0.57 (0.65) (462.60) (752.80) (777.40) (1063.76) (0.0103) (0.0149) (0.58) 7-year sample − 2097.46*** − 671.85 − 3025.02*** − 4677.16*** − 0.0378*** − 0.0268* 0.93 (0.57) 0.39 (0.78) (437.16) (720.86) (715.66) (1345.02) (0.0128) (0.0149) 9-year sample − 4432.32*** − 3345.17** − 5005.84*** − 4871.36*** − 0.0052 − 0.0159*** 0.35 (1.07) 0.51 (0.88) (810.96) (1583.87) (842.24) (1683.60) (0.0071) (0.0052) Robust standard errors, based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 5 Annual treatment Total labor income (£ per year) Employed effects 5-year sample (health shock in 2002)  Treat*2000 456.45 (571.33) 0.0089 (0.0182)  Treat*2001 29.13 (548.56) 0.0056 (0.0152)  Treat*2003 − 496.32 (526.55) 0.0017 (0.0192)  Treat*2004 − 1913.57** (763.25) − 0.0734*** (0.0310) 7-year sample (health shock in 2003)  Treat*2000 695.75 (873.84) − 0.0090 (0.0236)  Treat*2001 124.97 (762.76) − 0.0083 (0.0194)  Treat*2002 1336.91 (846.87) 0.0063 (0.0213)  Treat*2004 − 1358.27* (781.94) 0.0568 (0.0367)  Treat*2005 − 2078.62** (923.18) − 0.0572 (0.0361)  Treat*2006 − 2498.40** (1022.74) − 0.1212*** (0.0431) 9-year sample (health shock in 2004)  Treat*2000 592.78 (2527.47) − 0.0005 (0.0077)  Treat*2001 882.40 (2567.48) 0.0004 (0.0048)  Treat*2002 2161.92 (2325.20) 0.0021 (0.0054)  Treat*2003 128.95 (2301.63) − 0.0009 (0.0057)  Treat*2005 385.63 (2226.96) 0.0145 (0.0259)  Treat*2006 − 948.35 (2033.88) − 0.0198*** (0.0063)  Treat*2007 − 1551.61 (2298.16) − 0.0319*** (0.0066)  Treat*2008 − 1625.28 (2229.07) − 0.0421*** (0.0071) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in paren- theses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 the effects on health declines on annual household income. evidence that there are any changes on the intensive margin For all sample periods and both matching techniques, I find of employment. While four of the eight estimates for the even larger reductions in household income than for labor likelihood of being employed are statistically significant at income. For the 9-year sample, the results suggest that the 1% level, the immediate effects are substantially smaller household incomes are reduced by £5005.84 and £4871.36 than those observed by García-Gómez and López-Nicolás following the health shock (both p < 0.01). A potential expla- [22] for Spain. When re-estimating the analysis with only nation for the difference in magnitudes for total household individuals who remained at work throughout the sample income and individual labor earnings is that other household periods, I find very similar declines in labor earning and members take time away from work to assist the individuals household income. This suggests that changes in employ- recovering from health shocks. ment are not the only driver of the observed income losses. Appendix Table A2 furthermore provides matching DD The later part of the study examines some other potential results for the effects of sudden declines in health status on mechanisms through which health shocks can affect labor the natural log of both total labor and household income. market outcomes. Consistent with the results in Table 4, all estimates show that health shocks negatively affect labor earnings and total Annual treatment effects household incomes of affected individuals. While it should be noted that two of the eight labor income estimates are Table 5 shows annual estimates for the effects of health imprecisely estimated, Online Table A2 confirms that the shocks on total annual labor income and the likelihood of observed treatment effects are robust to the measure of income used in the analysis. Table 4 additionally shows the effects of health shocks on Appendix Table  A3 furthermore shows DD matching estimates the likelihood of being employed and weekly hours worked. when lagged health status is included as a covariate to obtain the pro- My analysis finds that individuals reduce their labor mar - pensity score values. The results are consistent with the main results ket activity on the extensive margin, while there is little from Table  3, providing further evidence that negative health events affect labor market outcomes. 1 3 O. Lenhart Fig. 2 Annual treatment effects on labor income, a 5-year sample. b Annual treatment effects on employment, 5-year sample. c Annual treat- ment effects on labor income, 7-year sample. d Annual treatment effects on employment, 7-year sample being employed, which are obtained by interacting each Mild vs. severe health shocks year with the treatment indicator. Since this test includes effects during pre-shock periods, it provides a test for the The longitudinal nature of the BHPS furthermore allows parallel trends assumption made in the main DD model. me to identify individuals who experienced large changes Given that this analysis is not feasible in the 3-year sam- in self-reported health as well as others whose health status ple, Table 5 only shows treatment effects for sample peri- only slightly declined. Information on self-reported health ods of 5, 7 and 9 years. status in the BHPS is provided on a scale from 1 (= excel- For all the sample periods, no statistically significant lent) to 5 (= very poor). For this analysis, I define the two differences are estimated during the years before the onset types of treatments the following way for all sample lengths: of the health shocks. Furthermore, none of the pre-shock (1) mild health shocks are average declines in self-reported treatment effects that are shown in Table  5 are jointly health by at most one point on the scale between the pre- significant. This provides suggestive evidence that the and post-shock period; (2) severe health shocks are average parallel trends assumption is satisfied. The estimates for declines in self-reported health by more than one point on both labor income employment status become larger in the scale between the pre- and post-shock period. Individu- magnitude several years after the shocks, indicating that als in the control group are those whose average health status the effects of health shocks on labor market outcomes remained the same across both periods. Compared to the are persistent rather than temporary. Figure 2a–d confirm main analysis in Sect. 5.1, this specification allows using this by providing graphical representations of estimates changes in the entire distribution of health status. Consistent presented in Table 5. with the main DD setup shown in Table 1, all individuals 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 6 The effects of health shocks on labor market outcomes (average differences in health status) Total labor income (£ per year) Total HH income (£ per year) Employed Weekly work hours NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- ing ing ing ing 3-year sample  Mild shock − 565.58** − 568.12* − 368.59 − 584.43 − 0.0097** − 0.0117** − 0.03 (0.27) 0.22 (0.38) (239.51) (322.62) (442.20) (514.00) (0.0041) (0.0053)  Severe − 1743.84*** − 1108.57* − 1411.01 − 1495.33 − 0.0261** − 0.0294*** 1.11** (0.49) 1.58** (0.75) shock (462.67) (598.55) (915.97) (1138.41) (0.0107) (0.0111) 5-year sample  Mild shock − 298.32 − 406.03 372.45 610.45 − 0.0079** − 0.0064 0.17 (0.21) 0.12 (0.27) (234.22) (295.50) (341.97) (452.44) (0.0033) (0.0042)  Severe − 1353.91*** − 1178.19** − 79.18 562.42 − 0.0373*** − 0.0326*** 0.76 (0.51) − 0.10 (0.66) shock (395.69) (515.06) (768.42) (918.08) (0.0097) (0.0109) 7-year sample  Mild shock − 777.23*** − 684.35* − 432.31 − 607.45 − 0.0071 − 0.0012 0.79*** 1.16*** (0.36) (254.05) (379.52) (399.68) (578.83) (0.0044) (0.0059) (0.26)  Severe − 3697.61*** − 2483.74*** − 4366.20*** − 2546.32*** − 0.0652*** − 0.0594*** − 1.52** − 1.90*** shock (362.84) (547.43) (587.85) (1006.68) (0.0132) (0.0119) (0.66) (0.71) 9-year sample  Mild shock − 1739.05*** − 1840.00*** − 3908.70*** − 3758.49*** − 0.0205 − 0.0113** 0.25 (0.23) 0.46 (0.31) (269.20) (341.71) (371.18) (490.90) (0.0142) (0.0053)  Severe − 3335.97*** − 3873.31*** − 5716.03*** − 7504.61*** − 0.0723*** − 0.0619*** − 2.66*** − 1.74** shock (578.20) (832.38) (775.68) (1361.91) (0.0130) (0.0140) (0.68) (0.78) Robust standard errors, cluster by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 Table 7 Effects of health shocks on labor market outcomes (health condition) Total labor income (£ per Total HH income (£ per year) Employed Weekly work hours year) NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- NN matching Kernel match- ing ing ing ing 3-year sample − 1049.24*** − 1068.43** − 2107.76*** − 2500.87*** − 0.0066 − 0.0029 − 0.39 (0.37) − 0.83* (0.50) (364.98) (515.82) (580.60) (860.66) (0.0070) (0.0071) 5-year sample − 1653.48*** − 1414.43*** − 2105.80*** − 3490.51*** 0.0020 0.0046 − 0.83** − 1.15** (0.48) (340.84) (521.67) (716.49) (940.21) (0.0022) (0.0029) (0.36) 7-year sample − 3129.55*** − 3292.39*** − 3342.17*** − 4202.84*** − 0.0025 − 0.0035 − 0.30 (0.39) − 0.51 (0.55) (444.22) (698.85) (777.31) (1078.69) (0.0013) (0.0025) 9-year sample − 3482.73*** − 5122.64** − 4097.99*** − 8083.43*** 0.0097 0.0116 0.11 (0.58) 0.45 (0.94) (554.06) (2017.59) (1059.44) (1996.55) (0.0074) (0.0152) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 are still required to be employed throughout the pre-shock for the effects of labor market outcomes following a 5-point period and in the year that the shock occurred. This analysis drop in health satisfaction in the German Socio-Economic is similar to two previous studies that test for employment Panel (GSOEP), which collects self-reported health informa- effects for individuals near retirement with panel data sets. tion on a scale from 0 to 10. Smith [54] data from the Health and Retirement Survey Table 6 presents the results for the two levels of health (HRS) to separately test the effects of experiencing either shocks. As expected, the negative effects on labor income a major or a minor health shock, while Riphahn [49] tests and the likelihood of being employed are substantially larger 1 3 O. Lenhart Table 8 Effects of health shocks on health care usage More than 5 annual doctor visits Spent a night at hospital Used any other health services Paid for any health services NN Kernel NN Kernel NN Kernel NN Kernel 3 years 0.2329*** 0.2555*** 0.0980*** 0.0931*** 0.2403*** 0.2477*** 0.0346*** 0.0261*** (0.0190) (0.0204) (0.0160) (0.0174) (0.0234) (0.0282) (0.0159) (0.0181) 5 years 0.1742*** 0.1906*** 0.0631*** 0.0720*** 0.1397*** 0.1809*** 0.0349*** 0.0570*** (0.0207) (0.0204) (0.0135) (0.0162) (0.0226) (0.0283) (0.0145) (0.0181) 7 years 0.1794*** 0.2104*** 0.0773*** 0.0813*** 0.2058*** 0.2288*** 0.0530*** 0.0489** (0.0202) (0.0218) (0.0167) (0.0161) (0.0261) (0.0308) (0.0183) (0.0197) 9 years 0.1611*** 0.1774*** 0.0345 (0.0219) 0.0556*** 0.1498*** 0.1127*** − 0.0194 − 0.0175 (0.0440) (0.0250) (0.0174) (0.0397) (0.0393) (0.0161) (0.0227) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Examples of health services asked for in the BHPS are usage of a physiotherapist, psychotherapist, health visitor at home and a hospital consultant. Pregnancies are excluded when examining changes in the likelihood of being a hospital in-patient *p < 0.10, **p < 0.05, ***p < 0.01 for individuals who experienced severe health shocks com- the health shock. The observed effects on hours worked are pared to individuals whose average health status declined mixed, with three estimates finding statistically reductions by at most one point. Similar to the previous findings, the in the weekly time spent working following the onset of the results are larger for the two longer sample periods (5 and 7 health condition. Overall, the results in Tables 4 and 7 pro- years), suggesting that the labor market effects are persistent vide consistent evidence that sudden health declines lead to rather than temporary. In the 7-year period, it is observ- substantial and persistent negative effects on labor earning able that individuals with mild health shocks significantly and household income. increase their weekly work hours, while those who experi- enced severe health shocks work significantly fewer hours after the health shock. Overall, the results in Table 6 point Mechanisms out that labor market outcomes are significantly worsened after severe health declines. However, given that labor The effects on health care usage income for those with mild shocks is reduced by £1840 in the 9-year sample (p < 0.01, kernel matching), the results Table 8 presents estimates for the effects of health shocks on also indicate that even relatively small health declines three indicators of health care usage and on the likelihood can negatively affect labor market outcomes of affected with which individuals paid for any health services out of individuals. their own pockets. The first six columns show that, as one could expect, individuals are more likely to have more than The effects of health conditions five annual doctor visits, to spend a night at the hospital and to have used any other services (e.g., physiotherapist, psy- For the results shown in Table 7, I use the onset of a new chotherapist, health visitor at home) over the last 12 months. health condition as an alternative health shock. Given that While the effects are largest in the 3-year sample, where the the presence of health conditions is likely to be more objec- results capture the results in the years immediately after the tive than self-reported health status, these estimates can health shock, the results remain relatively large and statisti- potentially provide additional robustness to the findings cally significant for the longer sample periods. Given that shown in Table 4 by removing concerns about the use of spending a night in the hospital or frequent doctor visits self-assessed health. As shown in Table 2, the number of means lost time at work, the observed changes in health care treated individuals captured with this alternative definition usage can potentially explain the earnings losses to some of health shock is larger than for the drop in health status. extent. Table 7 shows that the negative effects on both labor and The final two columns of Table  8 additionally provide household income are consistent with the results from the evidence that treated individuals are more likely to pay for health status shock, with all effects being statistically signifi- any health care services following the health care shock. The cant at the 1% level. The results for employment indicate the nearest neighbor matching results suggest that the effect is onset of a new health condition did not affect employment on largest for the 7-year sample, again indicating that the effects the extensive margin, which again suggests that other factors on health are persistent. Given that only a small share of explain the losses of earnings and household incomes after individuals in my samples report that they have any health 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data Table 9 Effects of health shocks on work productivity Hourly wage (£ per hour) Work limited by health Having difficulty to concentrate Feeling constantly under strain Nearest neigh- Kernel match- Nearest neigh- Kernel match- Nearest neigh- Kernel match- Nearest neigh- Kernel matching bor matching ing bor matching ing bor matching ing bor matching 3 years − 0.2933 − 0.6774 0.1953*** 0.1915*** 0.1806*** 0.1721*** 0.1493*** 0.1570*** (0.3998) (0.4684) (0.0180) (0.0182) (0.0213) (0.0178) (0.0217) (0.0277) 5 years − 1.4467*** − 1.3654*** 0.0837*** 0.0922*** 0.0809*** 0.0789*** 0.1470*** 0.1535*** (0.1870) (0.3897) (0.0144) (0.0167) (0.0182) (0.0239) (0.0271) (0.0279) 7 years − 0.8068** − 0.0947 0.1060*** 0.1007*** 0.1833*** 0.1643*** 0.1956*** 0.1677*** (0.3794) (0.3334) (0.0158) (0.0200) (0.0240) (0.0241) (0.0239) (0.0291) 9 years − 2.0683*** − 2.0709*** 0.0860*** 0.0857*** 0.0230* 0.0362** 0.1032*** 0.0712*** (0.2474) (0.7561) (0.0277) (0.0234) (0.0120) (0.0180) (0.0338) (0.0338) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses *p < 0.10, **p < 0.05, ***p < 0.01 care expenditures, the increase of paying for health care ser- can provide more evidence on how work performance can vices of 5.30% points (p < 0.01) corresponds to an increase be affected by health shocks. of 52.01% compared to prior to the health shock. The next two columns show that treated workers are These observed changes in health care expenditures significantly more likely to report that their health is lim- could furthermore explain the fact that household income iting their work. Similar to changes in health care usage, reductions following health shocks are even larger than the the effects are largest in the year after the health shock. losses in labor earnings, as previously shown in Tables 4 Using the 3-year sample, I observe a 19.53% point increase and 7. Other household members might reduce their work in the likelihood of reporting health-related work limita- time to support the family members with health issues with tions (p < 0.01). While the effects are smaller for the three their doctor visits, which goes along with increased health longer sample periods, they still show statistically signifi- expenditures. While increases in health care expenditures are cant increases (p < 0.01). The other two proxies of work observable for the first three sample periods, no statistically productivity I examine are reporting having difficulties to significant effects are found for the 9-year sample. concentrate [5] and being constantly under strain [27]. The DD matching estimates obtained for these two outcomes The effects on worker’s productivity provide additional evidence that reductions in work pro- ductivity might explain the losses of labor income to some Another potential channel through which health shocks extent. Again, the effects are quite large and remain persis- can affect labor market outcomes are changes in the level tent across the different sample periods. Overall, the results of work productivity. In Table 9, I show the effects on four in Table  9 suggests that individuals who suffered from a proxies for work productivity for the sample of people who sudden health shock are less likely to perform the same tasks work throughout the sample period. compared to prior to the health shock. First, I examine whether health shocks affect the hourly wages of individuals who remain in the workforce. The DD results provide evidence that wage rates declined substan- Heterogeneous effects tially for workers who experienced adverse health events compared to those who did not. While the estimates for the In a number of additional specifications, I examine whether 3-year period are relatively small and imprecisely estimated, the effects of sudden health declines on labor earnings differ I find that hourly wages are reduced by £2.07 (p < 0.01) across subgroups of the population. Table 10 presents nearest when analyzing the 9-year sample. These estimates suggest neighbor DD matching results across gender, education level, that individuals who remain in the workforce experience job classifications, and age. Using data from the Netherlands, less wage growth than those in the control group following García-Gómez et al. [24] find that health shocks have larger a health shock. One potential explanation for this could be that they are either not able to perform the same tasks or All results in Tables  8 and 9 are obtained using the drop in health take longer to complete the same tasks as compared to prior status as the health shock. Similar to the previous section, the results to the onset of the health shock. The remaining columns of remain consistent when using the onset of a health condition as the Table 9 examines several proxies for labor productivity that health shock. These additional results are not shown in the paper, but are available upon request. 1 3 O. Lenhart Table 10 heterogeneous effects of health shocks on earnings (health status) Total labor income (£ per year) 3-year 5-year 7-year 9-year Panel A: gender  Male − 2535.31*** (615.30) − 6576.00*** (608.28) − 5552.71*** (721.11) − 3248.65*** (517.66)  Female − 615.02 (584.40) − 1351.57*** (506.31) − 1310.93*** (446.87) − 2131.27*** (649.38) Panel B: education  Advanced degree − 2157.03*** (523.83) − 3166.42*** (572.38) − 3255.57*** (555.75) − 3151.08*** (863.72)  Basic degree/low education − 935.79** (418.38) − 1253.72*** (367.72) − 2592.76*** (497.77) − 2771.53*** (404.12) Panel C: job classification  Managerial/professional job − 1966.34*** (722.16) − 3411.18*** (714.28) − 3250.00*** (965.94) − 6507.92*** (698.68)  Skilled labor 150.67 (386.22) − 2066.42*** (447.94) − 433.09 (554.37) − 2436.33*** (680.92)  Unskilled labor 289.68 (844.24) − 1349.32*** (459.43) − 91.13 (1142.81) − 2079.77*** (475.26) Panel D: age  Below 40 years − 1928.86*** (419.86) − 3845.26*** (599.12) − 2435.64*** (545.65) − 3582.65*** (878.89)  At least 40 − 1110.15* (650.25) − 3836.11*** (566.79) − 2944.72*** (885.39) − 2415.16*** (839.80) Robust standard errors, clustered by individuals and based on Abadie and Imbens [1], are shown in parentheses. Income is adjusted for inflation, using the UK. Consumer price Index and 2000 as the base year *p < 0.10, **p < 0.05, ***p < 0.01 effects on the income of men, which they relate to the fact that Discussion and conclusions males are accounting for greater shares of household earnings. Using longitudinal data from the USA, Charles [11] further- The findings in this study provide evidence that health more provides evidence that the effects of health shocks on shocks significantly affect the labor market outcomes of earnings are increasing with age. He provides two explana- individuals in the UK for several years after the decline tions for this: (1) older persons have accumulated more human in health. García-Gómez et al. [24] suggest that negative capital that can be destroyed by negative health events; (2) effects of health shocks on labor markets can exist either any subsequent recovery in earnings will be weaker for older due to incentives created by disability benefits or due to individuals. labor market institutions constraining the responsiveness My findings in Panel A confirm the results by García- of wages to reduced productivity. Given that the disability Gómez et al. [24]. For all four sample periods, the effects of benefit scheme in the UK provides benefits at a flat rate, it health status declines on earnings are substantially larger for creates very little incentives for individuals to voluntary male individuals. In the 5-year sample period, a health shock reduce their employment compared to other countries, is shown to reduce labor earnings of men by £6576.00, com- which provide disability benefits that are closely tied to pared to a reduction of earnings of £1351.57 for women (both previous earnings [58]. This suggests that the observed p < 0.01). Similar to García-Gómez et al. [24], I find that men reductions in labor market participation following health have substantially higher pre-shock earnings than women, shocks are not driven by incentives provides by disability which could explain the different effects to some extent. Panels benefits. B and C additionally provide evidence that health shocks have This paper shows that the declines are not entirely stronger effects on labor market outcomes of individuals with driven by changes in employment status, but are also higher education levels and for those who work in managerial observable for individuals who remained employed. Addi- or professional jobs. Again, differences, in income prior to the tionally, the study provides first evidence that changes in health shock can potentially explain the larger effects for these work productivity is a mechanism through which health two groups. Finally, the results in Panel D do not indicate that shocks lead to lower labor earnings. Individuals who suffer the effects differ largely across age groups. 1 3 The effects of health shocks on labor market outcomes: evidence from UK panel data at the Southern Economics Annual Conference in Tampa, Florida in sudden health declines are shown to be limited in work- November 2017. related activities and to have difficulties concentrating in the following years, suggesting lower levels of work pro- Compliance with ethical standards ductivity and the inability to complete the same tasks the were able to perform before the health shock. Given that Conflict of interest The author declares that he has no conflict of inter - my results suggest that the negative effects on work pro- est. ductivity are still observable several years after the health shock, policymakers and employers should think about Open Access This article is distributed under the terms of the Crea- tive Commons Attribution 4.0 International License (http://creat iveco ways how the reintegration of employees can be improved mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- and significant productivity loss can be avoided. tion, and reproduction in any medium, provided you give appropriate Additionally, despite the provision of universal health credit to the original author(s) and the source, provide a link to the care through the NHS in the UK, I find significant increases Creative Commons license, and indicate if changes were made. in the likelihood with which individuals pay for health care services following the onset of a health shock. A likely explanation for this might be that individuals want to forego References long waiting times before receiving treatment and decide to pay for private care. The NHS has been dealing with 1. Abadie, A., Imbens, G.W.: Large sample properties of match- the issue of long waiting times for several decades. Using ing estimators for average treatment effects. Econometrica. 74, 235–267 (2006) official data from the NHS, Murray [ 42] shows that 2015 2. Abadie, A., Imbens, G.: On the failure of the bootstrap for match- marked the first year since the introduction of the NHS in ing estimators. Econometrica 76(6), 1537–1557 (2008) which the standard that at least 92% of patients receive their 3. Abadie, A., Imbens, G.: Matching on the estimated propensity treatment within 18 weeks was not achieved, suggesting that score. Econometrica 84(2), 781–807 (2016) 4. Adda, J., Banks, J., von Gaudecker, H.-M.: The Impact of income the issue is becoming worse over time. Thus, policymakers shocks on health: evidence from cohort data. J. Eur. Assoc. 7(6), should discuss ways on how to stop the trend of increasing 1361–1399 (2009) waiting times since they can cause further harm to individu- 5. Antikainen, R., Lönnqvist, A.: Knowledge Work Productivity als suffering health shock by affecting household incomes, Assessment. Institute of Industrial Management, Tampere Uni- versity of Technology (2006) financial stress and overall well-being in the following years. 6. Arie, S.: Is the UK government still serious about reducing smok- Previous data on the overall level of health for the UK ing? BMJ Clin. Res. 356, j1426 (2017) population is mixed. A report of the Economist Intelligence 7. Au, D., Crossley, T., Schellhorn, M.: The effects of health shocks Unit that compares healthcare inputs and outcomes across and long-term health on the work of older Canadians. Health Econ. 14, 999–1018 (2005) 166 countries ranked the UK 23rd in terms of performance 8. Becker, S., Ichino, A.: Simple and bias-corrected matching estima- [19]. A report by the Commonwealth Fund in 2014 places the tors for average treatment effects in STATA. Stata J. 1 (1), 1–18 UK second to last in the ‘healthy lives’ category, which uses (2002) indicators of population health outcomes, including mortality, 9. Bound, J., Schoenbaum, M., Stinebrickner, T.R., Waidmann, T.: The dynamic effects of health on the labor force transitions of infant mortality and life expectancy [17]. Evidence from the older workers. Lab. Econ. 6, 179–202 (1999) Global Burden of Disease Study shows that the leading risk 10. Case, A., Lubotsky, D., Paxson, C.: Economic status and health factors for premature death in the UK are linked to lifestyle, in in childhood: the origins of the gradient. Am. Econ. Rev. 92(5), particular to dietary risk, tobacco smoking, high blood pres- 1308–1334 (2002) 11. Charles, K.K.: The longitudinal structure of earnings losses sure, and alcohol consumption [30]. Given that these causes among work-limited disabled workers. J. Hum. Resour. 38(3), are preventable, campaigns to promote healthier lifestyles and 618–646 (2003) make people more aware of health risks could be success- 12. Contoyannis, P., Rice, N.: The impact of health on wages: evi- ful in improving health outcomes and in preventing the onset dence from the British household panel survey. Empir. Econ. 26, 599–622 (2001) of sudden health shocks for people in the workforce. Since 13. Currie, J., Stabile, M.: Socioeconomic status and child health: 1998, the UK government has had success in reducing smok- why is the relationship stronger for older children? Am. Econ. ing of the population from 28 to 18% in 2015. It has done Rev. 93(5), 1813–1823 (2003) so by changing taxation, by increasing public awareness of 14. Currie, A., Shield, M.A., Price, S.W.: The child health/family income gradient: evidence from England. J. Health Econ. 26, the harm caused by smoking and by helping people quit [6]. 212–232 (2007) These policy interventions should be continued and expanded 15. Currie, J., Duque, V., Garfinkel, I.: The great recession and moth- to other lifestyle choices that cause harm to people’s health. er’s health. Econ. J. 125(588), F311–F346 (2015) 16. Dano, A.M.: Road injuries and long-run effects on income and Acknowledgements I would like to thank the participants at the Col- employment. Health Econ. 14(9), 955–970 (2005) lege of Business Brown Bag Seminar Series at the University of West 17. Davis, K., Stremikis, K., Squires, D., Schoen, C.: Mirror, Mirror Florida as well as the participants of the session “Labor Economics” on the Wall: How the Performance of the U.S. Health Care System 1 3 O. Lenhart Compares Internationally. New York: The Commonwealth Fund: 40. Miller, D.L., Page, M.E., Stevens, A.H., Filipski, M.: Why are 2014. (2014). http://www.commo nweal thfun d.org/~/media /files recessions good for your health? Am. Econ. Rev. 99(2), 122–127 /publi catio ns/fund-r epor t/2014/jun/1755_davis _mir r o r_mir r o (2009) r_2014.pdf. Accessed 15 May 2017 41. Mora, R., Reggio, I.: Didq: A Command for treatment-effect 18. Disney, R., Emmerson, C., Wakefield, M.: Ill-health and retire- estimation under alternative assumption. Stata J. 15(3), 796–808 ment in britain: a panel data-based analysis. J. Health Econ. 25, (2015) 621–649 (2006) 42. Murray, R.: What is Happening to NHS Waiting Times? The 19. Economist Intelligence Unit. http://stateofr efor m.com/wp-content/ King’s Fund. (2016). https ://www.kings fund.org/blog/2016/-2/ uploads/2015/11/Healt hcar e-outco mes-inde x-2014.pdf . Accessed nhs-waiti ng times . Accessed 16 May 2017 16 May 2017 (2014) 43. Myde, B., Pelletier, B., Lynch, W.: The relationship between 20. Eliason, M., Storrie, D.: Does job loss shorten life? J. Hum. health risks and work productivity. J. Occup. Environ. Med. 46(7), Resour 44(2), 277–302 (2009) 737–745 (2004) 21. Eliason, M., Storrie, D.: Job loss is bad for your health: Swedish 44. Neumayer, E.: recessions lower (some) mortality rates: evidence evidence on cause-specific hospitalization following involuntary from Germany. Soc. Sci. Med. 58(6), 1037–1047 (2004) job loss. Soc. Sci. Med. 68(8), 1396–1406 (2009) 45. Nielsen, T.H.: The relationship between self-rated health and hos- 22. García-Gómez, P., López-Nicolás, A.: health shocks, employ- pital records. Health Econ. 25, 497–512 (2016) ment and income in the spanish labour market. Health Econ. 15, 46. Pelkowski, J., Berger, M.C.: The impact of health on employment, 997–1009 (2006) wages, and hours worked over the life cycle. Q. Rev. Econ. Fin. 23. García-Gómez, P.: Institutions, health shocks and labour outcomes 44, 102–121 (2004) across Europe. J. Health Econ. 30(1), 200–213 (2011) 47. Propper, C., Rigg, J., Burgess, S.: Child health: evidence on the 24. García-Gómez, P., van Kippersluis, H., O’Donnell, O., van roles of family income and maternal mental health from a UK Doorslaer, E.: Long term and spillover effects of health shocks on birth cohort. Health Econ. 16, 1245–1269 (2007) employment and income. J. Hum. Resour. 48(4), 873–909 (2013) 48. Reinhold, S., Jürges., H.: Parental income and child health in Ger- 25. Grossman, M.: On the concept of health capital and the demand many. Health Econ. 21(5), 562–579 (2012) for health. J. Polit. Econ. 80(2), 223–255 (1972) 49. Riphahn, R.: Income and employment effects of health shocks. 26. Hagan, R., Jones, A.M., Rice, N.: Health shocks and the hazard a test case for the German welfare state. J. Popul. Econ. 12(3), rate of early retirement in the ECHP. Swiss J. Econ. Stat. 144, 363–389 (1999) 323–335 (2008) 50. Rosenbaum, P.R., Rubin, D.B.: The central role of the propen- 27. Halkos, G., Bousinakis, D.: The effect of stress and satisfaction on sity score in observational studies for causal effects. Biometrika. productivity. Int. J. Prod. Perform. Manag. 59(5), 415–431 (2010) 70(1), 41–55 (1983) 28. Heckman, J.J., Ichimura, H., Todd, P.: Matching as an econometric 51. Ruhm, C.J.: Are recessions good for your health? Q. J. Econ. evaluation estimator. Rev. Econ. Stud. 65, 261–294 (1998) 115(2), 617–650 (2000) 29. Idler, E.L., Benyamini, Y.: Self-rated health and mortality: a 52. Schaller, J., Stevens, A.H.: Short-run effects of job loss on health review of twenty-seven community studies. J. Health Soc. Behav. conditions, health insurance, and health care utilization. J. Health 38(March), 21–37 (1997) Econ. 43, 190–203 (2015) 30. Institute for Health Metrics and Evaluation.: Global Burden of 53. Schmitz, H.: Why are the unemployed in worse health? The causal Disease Profile: United Kingdom. Seattle: Institute for Health effect of unemployment on health. Lab. Econ. 18, 71–78 (2011) metrics and Evaluation: 2015. (2015). http://www.health data. org/ 54. Smith, J.P.: Unraveling the SES: Health Connection. Population unite d-kingd om. Accessed 16 May 2017 and Development Review 30; Supplement: Aging, Health, and 31. Jiménez-Martín, S., Labeaga, J.M., Prieto, V.C.: A sequential Public Policy (pp. 108–132)(2004) model of older workers’ labor force transitions after a health 55. StataCorp.: Stata Treatment-Effects Reference Manual: Potential shock. Health Econ. 15, 1033–1054 (2006) Outcomes/Counterfactual Outcomes: Release 13. Statistical Soft- 32. Jones, A.M., Rice, N., Roberts, J.: Sick of work or too sick to ware. College Station, TX: StataCorp LP (2013) work? Evidence on self-reported health shocks and early retire- 56. Sullivan, D., von Wachter, T.: (2009). Job displacement and ment from the BHPS. Econ. Model. 27(4), 866–880 (2010) mortality: an analysis using administrative data. Q. J. Econ. 33. Johnston, D.W., Propper, C., Shields, M.A.: Comparing subjective 1265–1306 and objective measures of health: evidence for the income/health 57. Van Doorslaer, E., Wagstaff, A., van der Burg, H., Christiansen, gradient. J Health Econ. 28, 540–552 (2009) T., De Graeve, D., Duchesne, I., Gerdtham, U.-G., Gerfin, M., 34. Khanam, R., Nghiem, H.S., Connelly, L.B.: Child health and the Geurts, J., Gross, L., Häkkinen, U., John, J., Klavus, J., Leu, R.E., income gradient: evidence from Australia. J. Health Econ. 28(4), Nolan, B., O’Donnell, O., Propper, C., Puffer, F., Schellhorn, M., 805–817 (2009) Sundberg, G., Winkelhake, O.: Equity in the delivery of health 35. Lechner, M., Vazquez-Alvarez, R.: The effect of disability on care in Europe and the US. J. Health Econ. 19, 553–583 (2000) labour market outcomes in Germany. Appl. Econ. 43(4), 389–412 58. Van Doorslaer, E., Koolman, X.: Explaining the differences in (2011) income-related health inequalities across European countries. 36. Lindeboom, M., Llena-Nozal, A., van der Klaauw: Health shock, Health Econ. 13, 609–628 (2004) disability and work. Lab. Econ. 43, 186–200 (2016) 59. Vizard, P., Obolenskaya, P.: Labour’s Record on Health: Policy 37. Marcus, J.: The effect of unemployment on the mental health of Spending and Outcomes. Social Policy in a Cold Climate Working spouses—evidence from plant closures in Germany. J. Health Paper WP02. London: CASE (2013) Econ. 32, 546–558 (2013) 60. Zucchelli, E., Harris, A., Jones, A.M., Rice, N.: The effects of 38. McInerney, M., Mellor, J.M.: Recessions and seniors’ health, health shocks on labour market exits: evidence from the HILDA health behaviors, and healthcare use: analysis of the medicare survey. Aust. J. Lab. Econ. 13(2), 191–218 (2010) current beneficiary survey. J. Health Econ. 31, 744–751 (2012) 39. McInerney, M., Mellor, J.M., Nicholas, L.H.: Recession depres- sion: mental health effects of the 2008 stock market crash. J. Health Econ. 32, 1090–1104 (2013) 1 3

Journal

The European Journal of Health EconomicsSpringer Journals

Published: Jun 6, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off