Printer Friendly

Does labor market status influence self-assessed health?

Introduction

The empirical literature offers conflicting views on the relationship between labor market status and health. In a study of US workers, for example, Strully (2009) finds that unemployment leads to a deterioration in the health and possibly mortality of those laid-off, a result paralleling a number of earlier studies (e.g., Sullivan and von Wachter 2009; Clark 2003; Linn et al. 1985, or a literature summary by Berkman and Kawachi 2000). On the other hand, a study of Dutch workers (van der Horst et al. 1992) finds little difference in the health of those working and those laid-off. In a study of layoffs at Boeing Corporation, Greenberg et al. (2010) find that laid-off workers had fewer health problems, and were happier and less stressed than those keeping their jobs (who were working harder to maintain production and were worrying about losing their employment).(1) These latter results provide support for the conclusions of earlier studies (summarized by Bose and Bohle 2002) of the adverse health effects of downsizing on workers who retain their jobs.

When mortality is used as the measure of health, there is also debate over the link between labor market status and health. Ruhm (2000) found that periods of high unemployment are associated with lower mortality rates. He suggested that possible mechanisms behind this effect were cyclical changes in risky behaviors (e.g. cigarette smoking, alcohol consumption, obesity) and in the time cost of beneficial behaviors (e.g. exercise, preventive medical care, diet). Miller et al. (2009), however, question the plausibility of these mechanisms because the deaths that occur over the business cycle are concentrated in age groups in the population that have relatively low labor force attachment, namely, those under age 25 and above 64. As a result, Miller et al. (2009) conclude that it is unlikely that individuals' labor market status is a key determinant of aggregate mortality changes over the business cycle.

This study focuses on morbidity and addresses two questions. First, conditional on self-assessed health in the past, does labor market status at that time predict current self-assessed health? Second, does past self-assessed health predict current self-assessed health? We are motivated by the current incomplete understanding of the mechanism linking labor market status and health, regardless of whether health is measured in terms of mortality or morbidity. In certain respects this paper parallels previous studies on the impact of current and past unemployment on life satisfaction (e.g., Clark et al. 2001) and previous studies on the impact of past health on current health (e.g., Halliday 2008) but with a shift in focus to consideration of the impact of a wider variety of possible past labor market statuses on current self-assessments of physical and mental health. We consider alternative and indirect channels that could plausibly link labor market status and health outcomes. Our focus on self-perceptions of health is based on the view that perception of what is true can have a significant influence on what turns out to be true (Rabin and Schrag 1999).

The two questions we seek to answer are also relevant for public policy. Individuals often initiate their interaction with the health care system. Presumably, that initiation is a result of the perception that health care is needed. In many industrialized countries, the total financial value of many individual interactions with the health care system amounts to a significant share of national income in a given year. If specific current reasons for an individual's labor market status were found to influence future self-assessed health, then, in principle, it might be possible to design targeted policies that would change negative perceptions that might be manifested as negative actual health outcomes in the future.

The organization of the analysis is as follows. In The Data, we summarize our data set. In Specification and Estimation, we describe our empirical model, estimation methods, and hypotheses to be tested. In Results and Interpretation, we present the econometric results and their interpretation. In Conclusion, we draw together our major conclusions and provide suggestions for further research.

The Data

The Nationwide Sample

Our data come from the Medical Expenditures Panel Survey (MEPS) that is carried out for a given group of people in five rounds over two years by the Agency for Healthcare Research and Quality (2008). Our sample draws from the longitudinal weight file covering all age groups in the United States in 2005 and 2006 (Panel 10; HC-106). Obviously, the breadth of coverage gives rise to some problems for studying the relationship between labor market status and health in the long term. Schooling often interrupts the employment of those less than 25 years of age, while those over 62 years of age may be retiring from work. For these reasons, our sample includes only those aged 25 to 62 years at the time of the first round of the survey. We also eliminate soldiers, whose health status may have little to do with their employment per se, but rather with their participation in war. This leaves us with an initial sample of approximately 7,000 people, whose employment and health status are investigated for the five survey rounds in which they were followed by MEPS. Over the course of the two years of the survey, an average of 13 people died in each round. They remain in the sample until the round of their death. We find no evidence that omitting them in the round of their death biased our results.

Measures of Health

A major focus of this investigation is on self-assessed health. In each of five rounds of the survey, respondents were asked by MEPS to report both their physical and mental health status on a five point scale, with good health at the high end of the scale. These subjective measures raise the problem of "justification bias" (Currie and Madrian 1999; or McGarry 2004), which allegedly occurs, for instance, when poorer health can be linked to higher government benefits or can be used to assuage the personal sting of being laid off or fired. By taking into account a person's self-assessed health both before and after the change in labor market status, we gain a quantitative perspective on the extent of such bias in a change in labor market status, which turns out to be minuscule.

In addition to the two self-assessed health indicators, we examine three other measures of health.(2) The first is the sum of the number of work days missed because of sickness plus non-work days also spent in bed because of illness. This information was gathered each round. The second and third indicators are composite measures of physical and mental health, which we label "quasi-objective." They are only available for rounds two and four. The physical component summary is an index that weighs 35 indicators: objective (e.g., ability to climb steps or carry out various activities of daily living) and subjective (e.g., reported pain or energy level). The mental component summary is an index weighing 35 indicators in a much different manner, placing more emphasis on measures of reported depression, nervousness, and emotional state.(3)

Using data from only rounds two and four. Table 1 shows the correlation matrix of the five indicators. As expected, the self-assessed and also the quasi-objective physical and mental health indices are correlated with each other. Since the higher values of these measures indicate better health, they are negatively correlated with days sick. All of the correlations are highly significantly different from zero. Presumably, the common factor behind all of these indicators is "health." A more formal way to examine this hypothesis is to carry out a factor analysis of the five health indicators and then to examine the correlation of these five individual indicators to the derived factor. We do this using data on the five health indictors from rounds two and four. Such a procedure yields two relevant results: 1) there is only one major factor; 2) all five of the health indicators are related to the factor, with the self-assessed physical health and self-assessed mental health indicators having the strongest association; and sick days, the weakest association. Such an exercise gives us confidence that the self-assessed health measures have some "objectivity." In the Part B of the Statistical Appendix (Jefferson and Pryor 2013), we report the results of the factor analysis in greater detail.

Table 1 Correlation matrix of five different indicators of physical and mental health

                        Quasi-objectivc measures

                        Physical health     Mental health

Quasi-objective measures

Physical health         1.000

Mental health           0.124               1.000

Days sick              -0.3 19             -0.221

Sclf-assc'Sscd measures

Physical health         0.589               0.357

Menial health           0.347               0.489

 Days sick    Self-asscsscd measurcs

              Physical health         Mental health







 1.000



-0.270         1.000

-0.224         0.633                   1.000

Cell entries arc pair-wise correlation coefficients. Excepl for days sick, the health correlations are reported for the case where higher values of the measure denote better health. All coefficients arc statistically significant at the 0.01 level. The sample is for the entire population between 25 and 62 years old. Sample covers survey rounds two and four. Sample size is 12752 observations


Since our goal is to examine health perception dynamics at the individual level, we focus, as noted above, on the measures of self-assessed physical and mental health. The other raw health data have limited use for us. The quasi-objective measures were calculated only for the second and fourth rounds which would severely limit our dynamic analysis. The data on sick days per period, although collected four to five months apart, raise a different difficulty: in a number of cases the number of days sick recorded exceeded the number of days in a five month period. Our conjecture is that this occurred because, in some instances, the survey period in practice took additional months so that six or seven months elapsed between survey responses.

To simplify the interpretation and estimation of the dynamic econometric model, we transform the health measures used in the empirical analysis. For both self-assessed physical and mental health measures, we define an indicator variable equal to one if the respondent's health is rated good, very good, or excellent and equal to zero otherwise. Table 2 Panel A provides summary statistics on the transformed health measures. This table covers data for all five rounds. Respondents reported that their physical health was good or better approximately 84 % of the time. They reported that their mental health was good or better approximately 92 % of the time.

Table 2 Summary statistics

Panel A: health status measures



Mean

Sid. deviation

Panel B: labor force categories

Employment status

   Average number: currently employed

   Average number: job to return to

   Average number: got job dunng period

   Average number: not employed

Reasons for employment change(a)

   Persons not changing jobs or work mg part time

   Average number: job ended or business dissolved

   Average number: quit voluntarily

   Average number laid-0ff or fired

   Average number: not working because of illness or injury

   Average number: quit to take another Job

   Other reasons

Reasons for not working

   lnapplicable(b)

   Average number: inability to find work

   Average number: voluntarily stayed at home

   Average number: ill or disabled

   Average number: temporary laid-off

   Average number: waiting to start a new Job

   Other reasons

Self-assessed    Self-assessed

Physical health  Mental health

0.842            0.915

0.365            0.278

Percent          Observations

72.7 %           25,377

0.2              82

2.7              949

23.4             8,186

92.4             25,805

1.2              326

1.1              294

0.8              222

0.5              140

2.8              772

1.3              369

84.0             29.323

1.8              616

7.2              2,527

6.5              2.275

0.0              15

0.1              29

0.3              125

The self-assessed physical and mental measures have been redefined as
an indicator variable equal to one if a respondent's health is good or
better and equal to zero otherwise. That is, these indicators will
equal one if the reported physical or mental health is excellent, very
good, or good
(a) There are five rounds but only four chances for job changes between
rounds, so this percentage breakdown is based only on four sets of
changes
(b) The inapplicable consist of those who were employed and those who
were never employed; both were excluded from the regression


Labor Market Variables: Three Groupings

The MEPS asked respondents their employment status for each round.(4) Their responses were coded according four mutually exclusive possibilities: currently employed; not employed now but has a job to return to ("job to return to"); not employed at beginning of the survey round but got job since then ("got job during period"); and not employed at beginning of period, not employed now, and has no job to return to ("not employed"). Such a classification makes intuitive sense for this project: clearly being unemployed but having a job to return to might have a different impact on health than being unemployed with no future job prospects.

Respondents were also asked if they lost their job or changed their job during the survey round. These responses were coded by MEPS into five mutually exclusive categories: the job ended or the business dissolved; the respondent voluntarily quit; illness or injury led to a job loss; a lay off (or. implicitly, a firing) occurred; or the respondent left his or her original place of employment during the survey round to take another job.

Finally, respondents who were not working at the time of the survey round were asked whether this was because they could not find work, voluntarily remained home, were ill or disabled, had a temporary job layoff, or were waiting to start a new job.

Table 2 Panel B presents an overview of the labor force variables used in the regressions reported below. Since some people have poor health as a result of their being unemployed, while others arc unemployed because of their poor health, it is necessary to avoid confusing these two reasons for poor health. To distinguish these two situations, we can take advantage of the MEPS questions about job loss because of sickness or injury and whether people are not working because of poor health or injury.

Control Variables

The MEPS also contains a large number of possible control variables that other studies of health have employed. These include age, gender, race, ethnicity, family size, marital status, education, family income, health insurance status, region, and urban residency (whether the respondent lives in a metropolitan statistical area, MSA). It seems likely that most of these are related to self-assessed health status, although in the case of health insurance, the resulting sign is ambiguous. We also include dummies for the rounds (time dummies) and also those control variables that vary over time in our empirical analysis.(5) Very few of these are significantly related to health status, once other variables are taken into account. The full set of results is available at Jefferson and Pryor (2013).

Specification and Estimation

The two effects that we are most interested in estimating are the impact of current labor market status on health self-assessments in the future and the impact of current health self-assessments on future health self-assessments, that is, the presence of state dependence. To do this, we estimate autoregressive distributed lag forecasting models of health self-assessments at the individual level. These effects arc identified by controlling for unobserved heterogeneity and the autocorrelation structure of the idiosyncratic error term; the imposition of a restriction on the contemporaneous correlation between the idiosyncratic error and measures of labor market status; and the use of valid instrumental variables. We also include a number of standard individual control variables, including whether the respondent has health insurance. Therefore, we specify the following empirical model:

[h.sub.it] = [[gamma].sub.hit-1] + [s'.sub.it-1][delta] + [x'.sub.it-1][beta] + r'.sub.t[theta] + [[alpha].sub.i]+ [u.sub.it] (1)

where i = 1,2,......., N denotes the cross section unit (individuals) and t = 1,2, ..., T represents time (round). In Eq. 1, a measure of health is denoted by h; a row vector of labor market status variables (discussed above) is denoted by s; a row vector of control variables is denoted by X; and a row vector of survey round dummies is denoted by r; unobserved heterogeneity is represented by [alpha], the idiosyncratic error is denoted by u [Tilde] IID(0,[[sigma].sub.u.sup.2]). By assumption, [alpha] and u are independent. The coefficient on the lagged dependent variable, [gamma], will capture the presence of state dependence, if any. The coefficient column vectors, [delta],[beta] [theta], are conformable with the variable row vectors to which they are adjacent. Apart from the round (time) dummies, all of the variables on the right hand side are lagged one period. This eliminates any potential contemporaneous correlation between the idiosyncratic error and the right hand side variables.

We estimate Eq. 1 using the Arellano and Bover (1995) and Blundell and Bond (1998) system generalized method of moments estimator. To implement this estimator, we add the assumption that the initial conditions process for health is stationary. This permits the use of lagged differences as instruments in Eq. 1 as suggested by Blundell and Bond (1998). At the same time, we use lagged levels as instruments in the first differenced version of Eq. 1 as suggested by Arellano and Bond (1991). In some of the models estimated below, the Arellano and Bond (1991) test for zero autocorrelation of order two in [DELTA]u is rejected. When this is the case, we present estimates based on the assumption that u follows a moving average process of order one, MA(1). (For the details of sensitivity tests and 95 % confidence intervals for key parameter estimates, see the Statistical Appendix Part C in Jefferson and Pryor 2013.)

With respect to Eq. 1. we formulate the following two hypotheses for empirical testing: first, conditional on current self-assessed health, current labor market status does not help to predict future self-assessed health, [[delta].sub.j]=0 for each of the j parameters in the vector, [delta]; second, there is no state dependence in self-assessments of physical and mental health, [gamma]=0. We test these hypotheses using separate samples for women and men.

Results and Interpretation

Tables 3, 4 and 5 present our estimates for Eq. 1 for each of the three labor market status groupings (employment status, reasons for employment change, and reasons for not working). Since the dependent variable is an indicator variable, we interpret our results within the context of the linear probability model.

Table 3 Determinants of employment status

                                   Self-assessed physical health (t)

                                   A.1       A.2

                                   Women      Men

Self-assessed physical health (t-1) 0.103***   0.959**

                                   (0.02)     (0.42)

Self-assessed mental health(I- 1)   -          -



Temp. unemployed but job to

return to(t- 1)                   - 0.101     0.208*

                                  (0.07)      (0. 11)

Got job during period (t-1 )      0.033*      0.041

                                  (0.02)      (0.04)

Not employed (t-1)                0.036**     0.014

                                  (0.02)      (0.04)

Sample size                       14400       12415

Chi square                        88.007      47.087

p-value                           0.000       0.000

p-value of AR(2) test             0.847

Self-assessed mental health (t)

A.3           A.4

Women         Men

-             -

0.064**       -0.023

(0.03)        (0.62)

- 0.062       0.121*

(0.05)        (0.07)

- 0.020       - 003

(0.02)        (0.02)

0.042***      -0.024

(0.02)        (0.03)

14402         12406

31.809        19.798

0.016         0.285

0.1 79

The reference group is those employed. t denotes time period. * p<0.10;
 ** p<0.05; *** p < 0.001 where p denotes probability value. Estimated
using the Arellano and Bover (1995) and Blundell and Bond (1998) system
generalized method of moments estimator. The self-assessed physical and
self-assessed mental measures have been redefined as an indicator
variable equal to one if a respondent's health is good or better
and equal to zero otherwise. That is, these indicators will equal one
if the reported physical or mental health is excellent, very good, or
 good. All regressions in Tables 3. 4 and 5 include controls for
income. age. age squared, marital status (widow(er), divorced,
separated from spouse, never married), family size, live in
metropolitan statistical area, survey round (time), and whether
 the individual had health insurance
Table 4 Determinants of employment change

                                    Self-assessed physical health (t)

                                    B.1        B.2

                                    Women      Men

Self-assessed physical health (t-1) 0.085***   0.010*

                                    (0.03)     (0.04)

Self-assessed mental health (t-1 )  -          -



Job ended or business dissolved(t-1)-0.051     -0.047

                                    (0.03)     (0.04)

Qui1 voluntarily(t-1)               -0.045     -0,031

                                    (0.03)     (0.03)
Laid off or fircd(t-1)              0.051      0.012

                                    (0.04)     (0.03)
Quit for another job(t-1)           0.000      -0.007

                                    (0.02)     (0.02)
Other reasons(t-1)                  -0.027     -0.003

                                    (0.04)     (0.02)
Sample size                         10709      9233

Chi square                          76.762     73.324

p-value                             0.000      0.000

Self-assessed mental health (t)

B.3        B.4

Women      Men

-          -

0.063**    0.005

(0.03)     (0.05)

-0.037     -0.016

(0.03)     (0.02)

-0.002     0.002

(0.02)     (0.02)

0.021      -0.030

(0.04)     (0.03)

-0.001     0.016

(0.02)     (0.02)

-0.004     0.011

(0.02)     (0.02)

10711      9226

23.178     15.017

0.184      0.661

The reference group is those retaining existing jobs. *p<0.10;
** p<0.05; *** p<0.01 For other notes, see Table 3
Table 5 Detenninants of not working

                                    Self-assessed physical health (t)

                                    C. 1            C.2

                                    Women           Men

Self-assessed physical health (t-1) 0.014           0.026

                                    (0.05)          (0.13)

Self-assessed mental health (t-1)   -               -



Couldn't find work(t-1)             -0.011          0.071*

                                    (0.04)          (0.04)

Temporarily la id off(t-1)          0.026           -0.088

                                    (0.03)          (0.15)

Waiting to start new job{t-1)       0.011           -0.107

                                    (0.02)          (0. 11)

Other reasons(t-1)                  0.012           -0.104

                                    (0.08)          (0.09)

Size of sample                      1897            616

Chi square                          41.032          19.131

p-value                             0.002           0.384

p-value of AR(2) test               0.339           0.550

Self-assessed mental health(t)

C.3             C.4

Women           Men

-               -

0.024           0.148

(0.09)          (0. 17)

0.043           0.025

(0.04)          (0.04)

0.003           0.0 16

(0.02)          (0.04)

0.003           -0.009

(0.01)          (0.04)

-0.023          0.036

(0.05)          (0.03)

1897            615

10.471          19.095

0.915           0.386

0.349           0.554

The reference group is those voluntarily staying at home. *p<0.10;
 ** p<0.05; *** p<0.01. For other notes, see Table 3


Tables 3, 4 and 5 show the results for women and men for the three employment status models. The most startling overall result is that none of the independent variables dealing with labor market status have a statistically significant adverse impact on changes in self-assessed physical or mental health once past self-assessed health is controlled for. State dependence is present in self-assessments of physical health for both sexes in two of the three employment status groupings.

Table 3 focuses on employment status. The null hypothesis that there is no state dependence in self-assessments of physical and mental health, [gamma]=0, is rejected at the one percent level for women and at the five percent level for men. As indicated in equations A. 1 and A.2 in the table, self-assessed physical health in the previous survey period is a significant determinant of self-assessed physical health in the current period. An interpretation of the state dependence parameter for the self-assessed physical health regression for women is that if a respondent's self-assessed physical health was good or better in the previous period, then the probability that it is good or better in the current period is increased 10.3 % given last period's employment status and ceteris paribus. For men, that number is 96 %.

Staying with self-assessed physical health, the impact of employment status, relative to those employed, can be interpreted as follows. For women who got a job during the current period, the probability that they will assess their physical health in next period to be good or better increases by 3.3 %. The cumulative impact of this effect over the 2 year sample period is given by the cumulative multiplier, [delta]j/(1 -[gamma])=0.033/(1-0.103)= 3.7%. For men who are currently unemployed but have a job to which they can return, the probability that they will assess their physical health next period to be good or better increases by 20.8 %. The cumulative impact of this effect is calculated to be in excess of 500 %. Thus, almost surely men in this situation will assess their physical health to be good or better next period.

When it comes to self-assessed mental health, many of the effects of employment status are mute as indicated in equations A.3 and A.4. Relative to those employed, women who currently are not employed and men who have a job to which they can return are predicted to raise their self-assessment of mental health next period. There is evidence of state dependence only for women. The null hypothesis, [gamma]=0, is rejected at the five percent level for women.

Table 4 focuses on individuals whose employment status changes. Note that those individuals who report that they were fired or laid off because of sickness are omitted from the models in Table 4. As indicated in equations B.1 and B.2, state dependence is present in self-assessments of physical health for both sexes. Self-assessed physical health in the previous survey period is a significant determinant of self-assessed physical health in the current period. The null hypothesis, [gamma]=0, is rejected at the 1 % level for women and the 10 % level for men. When it comes to self-assessed mental health, there is evidence of state dependence only for women, as indicated in models B.3 and B.4. We find that, relative to those retaining their existing jobs, a current employment status change for any of several reasons has no statistically significant predictive power for future self-assessments of physical or mental health.

In Table 5, we look more closely into the relationship between the self-assessed health of those not working and the reasons they are not working. In this sample, we also eliminated those who were not working because of illness or injury, since this variable does not distinguish between those who were laid off because of illness or injury and those who never had employment for this reason. There is no evidence in support of state dependence in these equations. That is, we are unable to reject the null hypothesis for women or men that the coefficient on health in the previous period is zero ([gamma]=0). For these individuals, any persistence in self-assessed health is more likely due to unobserved heterogeneity. Relative to those staying at home voluntarily, the inability to find work currently increases the probability that a man's self-assessed physical health will be good or better by 7.1 %. An interpretation of this finding may be that there is signaling that men in this situation (not working but not at staying home voluntarily) are "able-bodied" but lack employment opportunity. No other current reason for not working has statistically significant predictive power for future self-assessments of physical or mental health.

We close this section with a caveat. The MEPS follows respondents for only a two-year period. Therefore, the context of our analysis is the short- and medium-term effects of labor market status on health. For example, our data do not tell us how long respondents have been unemployed. In the face of long-term unemployment, the lower family income during the period may lead to a decline in health if those who have lost their health insurance along with their job could not take advantage of Medicaid or afford their own medical insurance. As others have noted, it may lead to a lower probability of obtaining future employment because of "scarring" or loss of skills.

Conclusion

This study addressed two questions. First, conditional on current self-assessed health, does current labor market status predict future self-assessed health? Second, does current health self-assessments predict future health self-assessments? To address these questions, we used a random sample of approximately 7.000 individuals between the ages of 25 and 62 drawn from the Medical Expenditures Panel Survey (MEPS). The survey was carried out in five rounds over 2 years in the United States in 2005 and 2006. We used autoregressive distributed lag forecasting models of health self-assessments at the individual level to estimate the impact of current health self-assessments on future health self-assessments, that is, the presence of state dependence; and also the impact of current labor market status on future self-assessed health.

An overview of our main findings is as follows. The key finding of this paper is that labor market status, defined in terms cither of a person's past position in the labor force, or the reasons for changing jobs, or the reasons for not working, has no statistically significant adverse impact on future self-assessed physical or mental health once current self-assessed health is controlled for. Conditional on current health self-assessments, having a job to return to, getting a job, non-employment, and the inability to find work all positively influenced health self-assessments in the future. These conclusions, of course, pertain only for the 2 year period of the survey. We also found evidence that state dependence is more prevalent in self-assessments of physical health than in self-assessments of mental health. Unobserved heterogeneity is the more likely source of intertemporal correlation in self-assessments of mental health in men. For women, the same intertemporal correlation is more likely driven by state dependence.

Future research should focus on a longer time period and also on the specific types of health problems experienced by the unemployed.

Acknowledgments We thank Kathleen McCann of Social and Scientific Systems for assistance with our data; Adrian Lucas and members of the Swarthmore-Byrn Mawr-Haverford Summer Seminar for helpful comments; and Victoria Wilson-Schwartz for editorial assistance. Jefferson's research is supported by a Swarthtmore College Faculty Research Grant.

Published online: 20 December 2013

[c] International Atlantic Economic Society 2013

(1) This result runs counter to the findings of Thcodossiou (1998).

(2) Details on the location of these measures in MEPS are in Part A of the Statistical Appendix (Jefferson and Pryor 2013).

(3) Our measure of mental health differs from that of Theodossiou (1998) who measured psychological well-being and found that unemployment led to "greater strain," "less enjoyment of day-to-day activities." "being less able to face up to problems", "less confidence," "more thinking of being a worthless person," and "feeling less happy." He used the British Household Panel Study for his regression with a sample of roughly 7900 individuals in 1992.

(4) Details on the location of the labor market status measures in MEPS are discussed in Part A of the Statistical Appendix (Jefferson and Pryor 2013).

(5) In the MEPS, the education variable is defined as number of years of education when first entered MEPS. This variable docs not change over time. The region variable is defined as region of residency at the end of the year. There is essentially no time variation in this variable.

References

Agency for Healthcare Research and Quality. (2008). Medical Expenditures Panel Survey Household Component (MEPS HC - 106), 2006 Full Year, Consolidated Data File, Panel 10, (www.meps.ahrq. gov/mepsweb).

Arcllano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58, 277-297.

Arcllano, M., & Bover. O. (1995). Another look at the instrumental variable estimation of error-components models. Journal of Econometrics, 68(1), 29-51.

Berkman, L., & Kawachi, I. (Eds.). (2000). Social Epidemiology. New York: Oxford University Press.

Blundell. R., & Bond. S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics. 87(1), 115-143.

Bose. L., & Bohle, P. (2002). Health and social effects of downsizing: A review. The Economic and Labor Relations Review, 13(2), 270-287.

Clark, A. E. (2003). Unemployment as a Social Norm: Psychological Evidence from Panel Data. Journal of Labor Economics. 2/(2), 323 351.

Clark, A. E., et al. (2001). Scarring: The psychological impact of past unemployment. Economica, 68, 221 241.

Curric. J., & Madrian, B. (1999). Health, health insurance and the labor market. In O. Ashenfelter & D. Card (Eds.), Handbook of labor economics (Vol. 3C, pp. 3309-3416). New York: Elsevier.

Greenberg, E., et al. (2010). Turbulence: Boeing and the State of American workers and managers. New Haven: Yale University Press.

Halliday, T. (2008). Heterogeneity, state dependence, and health. Econometrics Journal. II, 499-516.

Jcfferson. P., & Pryor F. (2013). "Statistical Appendix: Labor/Health." At http://www.swarthmore.edu/acadcmics/economics/faculty-and-staff/philip-jefferson.xml.

Linn, M., et al. (1985). Effects of unemployment on mental and physical health. American Journal of Public Health. 75(5), 502 506.

McGarry, K. (2004). Health and retirement: do changes in health affect retirement expectations? Journal of Human Resources, 39(3), 624-648.

Miller, D., et al. (2009). Why arc recessions good for your health? American Economic Review; Papers and Proceedings, 99(2), 122 127.

Rabin, M., & Schrag, J. (1999). First impressions matter: a model of confirmatory bias. The Quarterly Journal of Economics. 114(1), 37-82.

Ruhm, C. J. (2000). Are recessions good tor your health? Quarterly Journal of Economics, 115(2), 617 650.

Strully, K. (2009). Job loss and health in the U.S. labor market. Demography, 48(2), 221 246.

Sullivan, D., & von Wachter, T. (2009). Job displacement and mortality: an analysis using administrative data. The Quarterly Journal of Economics. 124(3), 1265-1306.

Thcodossiou, I. (1998). The effects of low-pay and unemployment on psychological well-being: a logistic regression approach. Journal of Health Economics. I7(1), 85-104.

van der Horst, F. G. E. M. et al. (1992). "Causality in the Relation between Health and Long-term Unemployment." In C. H. A. Verhaar and Lammert Gosse Jansman, (eds). On the Mysteries of Unemployment, (pp. 225 53). Boston: Kluwer.

DOI 10.1007/s l 1294-01 3-945 1-y

P. N Jeffcrson (*) * F. L. Pryor

Deparonent of Economics, Swarthmore College, Swanhmorc, PA 19081. USA

e-mail: pjefferl@swarthmorc.edu
COPYRIGHT 2014 Atlantic Economic Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:N. Jefferson, Philip; L. Pryor, Frederic
Publication:International Advances in Economic Research
Article Type:Report
Geographic Code:1USA
Date:Feb 1, 2014
Words:5697
Previous Article:Online discussion and learning outcomes.
Next Article:How much do technological gap, firm size, and regional characteristics matter for the absorptive capacity of ltalian enterprises?
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters