HEALTH ECONOMICS Health Econ. 24: 659–671 (2015) Published online 8 April 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/hec.3052

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES† DANIEL I. REESa,* and JOSEPH J. SABIAb a

Department of Economics, University of Colorado Denver, Denver, CO, USA Department of Economics, San Diego State University, San Diego, CA, USA

b

SUMMARY While migraine headache can be physically debilitating, no study has attempted to estimate its effects on labor market outcomes. Using data drawn from the National Longitudinal Study of Adolescent Health, we estimate the effect of being diagnosed with migraine headache on labor force participation, hours worked, and wages. Ordinary least squares (OLS) estimates suggest that migraines are associated with reduced labor force participation and lower wages among females. A negative association between migraine headache and the wages of female respondents is also obtained using an instrumental variables (IV) approach, although the IV estimates are imprecise relative to the OLS estimates. Copyright © 2014 John Wiley & Sons, Ltd. Received 19 December 2012; Revised 20 February 2014; Accepted 25 February 2014 KEY WORDS:

migraine headache; wages; labor force participation; productivity

1. INTRODUCTION On the order of one out of every 10 adults in North America and Western Europe suffers from migraine headaches (Breslau and Rasmussen, 2001). Despite the fact that migraine attacks are often accompanied by debilitating pain, sight impairment, and concentration lapses (Headache Classification Subcommittee of the International Headache Society, 2004), and may lead to diminished human capital acquisition (Rees and Sabia, 2011; Sabia and Rees, 2011), no previous study has attempted to assess the impact of migraine headache on labor market outcomes such as wages. To the extent that migraine headaches reduce productivity, and therefore wages, previous estimates of their cost to society should be adjusted upwards. Using ordinary least squares (OLS) and data from the National Longitudinal Study of Adolescent Health, we find that female migraineurs are less likely to participate in the labor market than their non-migraineur counterparts. Moreover, conditional on participation, female migraineurs earn less than their non-migraineur counterparts. In contrast, there is no evidence of an association between migraine headache and male labor market outcomes, nor is there evidence that migraine headache affects female earnings through occupational choice. Because we cannot rule out the possibility that migraine status is correlated with unobserved determinants of labor market outcomes such as on-the-job stress, we pursue an alternative estimation strategy using mother’s self-reported migraine status as a first-stage instrument. This strategy also produces evidence of a negative association between migraine headache and the wages of females. However, second-stage estimates of the association between migraine status and labor force participation, although consistently negative, are never statistically significant at conventional levels. We conclude that migraine headache may reduce wages, most likely through its impact on productivity as opposed to occupation choice or hours worked.

*Correspondence to: Department of Economics, University of Colorado Denver, CB 181, P.O. Box 173364, Denver, CO 80217-3364, USA. E-mail: [email protected] † Supporting information may be found in the online version of this article.

Copyright © 2014 John Wiley & Sons, Ltd.

660

D. I. REES AND J. J. SABIA

2. BACKGROUND Migraine headaches are triggered by a variety of stimuli, including stress, fatigue, irregular sleep patterns, loud noises, strong smells, smoke, and flickering lights (Adelman and Adelman, 2001; Kaniecki, 2002; Lambert and Zagami, 2008; Andress-Rothrock et al., 2010). Approximately 20% of adult women in the USA report having had at least one migraine headache in the past year, while 5% to 6% of adult men in the USA report having suffered an attack in the past year (Stewart et al., 1992; Lipton et al., 2001; Lipton et al., 2007). Attacks last from 4 to 72 h and may involve moderate to severe pain as well as nausea and sensitivity to bright light and loud noises (Headache Classification Subcommittee of the International Headache Society, 2004).1 To our knowledge, no previous study has attempted to estimate the impact of migraine headache on wages or other employment outcomes such as hours or part-time status. However, when asked, migraineurs clearly indicate that they often experience reduced productivity while at work (Clarke et al., 1996; Lipton et al., 2001; Waldie and Poulton, 2002; Pradalier et al., 2004; Lipton et al., 2007; Burton et al., 2009). In fact, migraine sufferers in Western Europe report an average of 2–9 work days lost per year as a result of attacks (Berg, 2004), and 25% of migraine sufferers in the USA report losing at least 1 day work in the past 3 months (Lipton et al., 2007). Based on figures such as these, it has been estimated that the indirect costs to society of migraine headache are much larger than the costs associated with treatment (Berg, 2004).2 Using simple cross-tabulation, medical researchers have found that migraine headache is negatively related to years of schooling and other measures of socioeconomic status (Breslau et al., 1991; Stewart et al., 1992; Waldie et al., 2002; Waldie and Poulton, 2002; Lipton et al., 2007). Nevertheless, there is a perception among the general public that migraine sufferers tend to be overachievers, prone to feelings of anxiety, and driven to succeed.3 This perception is perhaps fueled by the fact that stress is an important trigger of both migraine and tension headaches (Kaniecki, 2002) and is buttressed by a growing number of medical studies documenting a link between migraine headache and neuroticism (Brandt et al., 1990; Merikangas et al., 1993; Breslau and Andreski, 1995; Kentle, 1997; Cao et al., 2002).4

3. DATA AND MEASURES Our data are drawn from the National Longitudinal Study of Adolescent Health (Add Health), a nationally representative school-based survey conducted by the University of North Carolina at Chapel Hill, now made available to researchers by the University of Michigan. The Wave I in-home survey was administered to 20,745 seventh through 12th graders between April and December of 1995 and includes oversample of black students with college-educated parents as well as students of Chinese, Cuban, and Puerto Rican descent (see Harris et al., 2008, for a detailed discussion of the Add Health sampling strategy.) According to the Headache Classification Subcommittee of the International Headache Society (2004, p. 150–151), moderate pain ‘inhibits, but does not wholly prevent usual activities’, whereas severe pain ‘prevents all activities’. 2 After reviewing a number of studies using data from Western Europe, Berg (2004) concluded that the ‘vast majority of the total costs, nearly 90%, are due to indirect costs in the form of absenteeism and reduced effectiveness’. Direct medical costs were, on average, €39 per patient per year in Western Europe in Berg (2004). 3 Writing for the New York Times, Brody (1988) described migraines as occurring ‘with disproportionate frequency among high achievers who expect a lot from themselves and accomplish a lot’. The overachieving-migraineur hypothesis seems to be common even among medical professionals. In an interview with Vibrant Life, Dr. Seymour Diamond, the Executive Chairman of the National Headache Foundation, was asked if there ‘is such a thing as a migraine personality’. He replied, ‘[m]any migraine patients are perfectionists. They tend to be intelligent, doers, and achievers.’ The full text of this interview is available at http://findarticles.com/p/articles/mi_m0826/is_n4_v9/ ai_14022967/?tag=content. 4 It should be noted, however, that the link between migraine and personality is still being investigated and debated. Nylander et al. (1996) found no evidence that migraine sufferers had different ‘temperaments or characters’ than non-sufferers; Mattsson and Ekselius (2002) found no relationship between migraines and personality type among women ages 40–74 years; and Stronks et al. (1999) found that neither migraineurs nor subjects prone to tension headaches showed higher anxiety levels than members of a healthy control group when confronted with an arithmetic problem. 1

Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

661

Three follow-up surveys to the baseline Add Health survey were conducted. The first was administered in 1996, 1 year after the baseline survey; the second was administered in 2001, when respondents were ages 18 to 26 years; and the third was conducted in 2007–2008, when respondents were ages 24 to 32 years. Migraine status and labor market outcomes were constructed using data from the third follow-up to the baseline Add Health survey. After eliminating observations with missing information on labor force participation, migraine status, and the controls, our sample consists of 14,487 respondents. The sample is reduced to 11,929 after deleting respondents with missing information on biological mother’s migraine history. It is reduced to 11,069 when we restrict our attention to respondents who reported positive earnings in the previous year. A number of labor market outcomes are considered. We created a dichotomous variable, Any Earnings, equal to one if the respondent reported positive personal earnings in the last year and equal to zero if the respondent reported no earnings. In addition, we created the variables Hours Worked and Hourly Wage, both of which are logged in the regression analysis.5 Hourly Wage was based on the answers to the following question: Now think about your personal earnings. In [the previous year], how much income did you receive from personal earnings before taxes, that is, wages or salaries, including tips, bonuses, and overtime pay, and income from self-employment? If a respondent replied ‘do not know’ to the earnings question, he or she was prompted with seven categories of earnings; the midpoints of each were used to code missing values. To calculate the hourly wage in dollars, we divided the answer to the yearly earnings question by usual hours worked per week times 50.6 The variable Hourly Wage is bottom coded at $2.13 per hour (the tipped Federal minimum wage in 2007) and top coded at $500 per hour.7 Approximately 5% of employed respondents in our sample earned less than $2.13 per hour, and 0.04% earned more than $500 per hour. The independent variable of interest comes from the third follow-up survey. Specifically, respondents were asked, ‘Has a doctor, nurse or other health care provider ever told you that you have or had migraine headaches?’ Migraine is equal to one if a respondent answered this question in the affirmative and is equal to zero otherwise. Consistent with research showing that women often experience their first migraine headache after adolescence (O’Brien et al., 1994), fully 20% of female Add Health respondents had been told by a health professional that they suffered from migraines by the third follow-up survey (Table I). In contrast, only 8% of male respondents had been told by a health professional that they suffered from migraine headaches. Respondents with zero earnings in the past year were more likely to be migraineurs. Specifically, 22.3% of respondents with zero earnings had been told by a health professional that they suffered from migraine headaches, while 13.7% of respondents with positive earnings had been told by a health professional that they suffered from migraine headaches (Table I). Conditional on having positive earnings in the past year, respondents who worked fewer than 40 h per week were more likely to suffer from migraine headache (Table I). Likewise, respondents who earned less than $15 per hour were more likely to suffer from migraine headaches (Table I).

4. ESTIMATION AND RESULTS 4.1. OLS estimation and results While the descriptive statistics reported in Table I are suggestive, previous research has shown that observables such as race and family income are important correlates of migraine status (Bigal et al., Hours Worked was based on answers to the question, ‘[h]ow many hours a week (do/did) you usually work at this job?’ This question was only asked if a respondent reported working at least 10 h per week for their current employer. The Add Health survey does not ask respondents about usual weeks worked. See Sabia and Rees (2012) for more information on how the hourly wage was calculated. 7 Indicator variables are included on the right-hand side of the wage equation for wages above the top-coded value and below the bottom-coded value. 5 6

Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

662

D. I. REES AND J. J. SABIA

Table I. Means of selected variables

Any Earnings Hours Hourly Wage (in dollars) Migraine Any Earnings = 1 Any Earnings = 0 Hours < 40 Hours ≥ 40 Hourly Wage ≤ $15 $15 < Hourly Wage ≤ $25 Hourly Wage > $25 Mother Migraine

Pooled sample

Males

Females

0.933 (0.251) [14,487] 41.5 (11.2) [13,431] 18.17 (21.7) [13,431] 0.143 (0.350) [14,487] 0.137 (0.344) [13,510] 0.223 (0.417) [977] 0.158 (0.365) [3102] 0.131 (0.337) [10,329] 0.155 (0.362) [6369] 0.127 (0.333) [4525] 0.110 (0.313) [2531] 0.273 (0.454) [11,069]

0.968 (0.175) [6879] 44.1 (11.4) [6622] 19.75 (23.7) [6622] 0.080 (0.271) [6879] 0.079 (0.270) [6662] 0.106 (0.309) [217] 0.084 (0.278) [1045] 0.077 (0.267) [5577] 0.087 (0.282) [2828] 0.074 (0.262) [2315] 0.069 (0.254) [1475] 0.268 (0.443) [5486]

0.900 (0.300) [7608] 39.1 (10.3) [6809] 16.62 (19.5) [6809] 0.200 (0.400) [7608] 0.194 (0.396) [6848] 0.257 (0.437) [760] 0.196 (0.397) [2057] 0.193 (0.395) [4752] 0.208 (0.406) [3541] 0.182 (0.386) [2210] 0.768 (0.374) [1056] 0.277 (0.448) [5583]

Unweighted means are generated using data drawn from Waves I and IV of the National Longitudinal Study of Adolescent Health. Standard deviations are in parentheses, and sample sizes are in brackets.

2007). In an effort to account for the influence of observables such as these, we estimate the following equation using OLS: yi ¼ β0 þ β′1 Xi þ β2 Migrainei þ εi ;

(1)

where yi represents one of the three dependent variables described earlier and X includes the following: the respondent’s age, age squared, race, ethnicity, urbanicity, the respondent’s Peabody Picture Vocabulary Test (PPVT) score (a measure of cognitive ability),8 educational attainment, current school enrollment,9 marital status, number of children, the educational attainment of the parent who answered the parental questionnaire, household income during adolescence, parents’ marital status during adolescence, height, BMI, number of biological siblings, and whether the respondent had an older sibling. In addition, we experiment with including a broader set of (potentially endogenous) controls in an effort to capture the influence of individual characteristics that have been found to be associated with both migraine headache and labor market outcomes. These controls include the following: the Center for Epidemiological Studies Depression (CES-D) score and the Rosenberg Self-Esteem score (Breslau et al., 1991; Zwart et al., 2003), whether the respondent suffered from hypertension (Kelman and Rains, 2005), whether the respondent reported usually being unable to get enough sleep (Scher et al., 2005), indicators for number of times the respondent was drunk in the past month, a neuroticism score (Brandt et al., 1990; Merikangas et al., 1993; Breslau and Andreski, 1995; Kentle, 1997; Cao et al., 2002), and an anxiety score (Kentle, 1997; Cao et al.,

8

The PPVT measures verbal comprehension and vocabulary. The test is conducted by an interviewer who reads a word to a respondent and then has the respondent choose among four illustrations to determine the picture that best fits the word. The PPVT consists of 78 items (Harris and Thomas, 2002). Adolescent Health respondents were administered 39 of these 78 items. 9 Controlling for educational attainment and current school enrollment is important because there is evidence that migraineurs are less likely to graduate high school and less likely to attend college than non-migraineurs (Rees and Sabia, 2011; Sabia and Rees, 2011). Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

663

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

2002).10 Finally, because there is qualitative evidence that migraineurs often change jobs and even career paths in an effort to manage their condition (Peters et al., 2004; Belam et al., 2005), we experiment with adding controls for occupation and job tenure to the right-hand side of Eqn (1).11 Table II reports OLS estimates of the effect of migraine status on labor force participation (as measured by Any Earnings) and the natural log of current hours. Standard errors are corrected for clustering at the school level, as per the Add Health research design. The top panel of Table II presents estimates of estimates of β2 for the pooled sample, which is composed of both male and female respondents. The bottom panels present OLS estimates of β2 by gender. In the pooled sample, migraine headache is associated with a 0.019 to 0.044 reduction in the probability of participating in the labor market. When the sample is restricted to male respondents, the estimates of β2, although negative, are not statistical significant at conventional levels. The relationship between migraine status and labor force participation is stronger among female respondents. In the fully specified model, being diagnosed with migraines is associated with a 0.020 reduction in the probability of reporting positive earnings. There is little evidence that migraine status affects hours worked conditional on participating in the labor market. In the pooled sample with no controls, migraine headache is associated with a 2.9% decrease in hours worked (e0.029 – 1 = 0.029). However, when the demographic and family background controls are included, this estimate is reduced in absolute magnitude and becomes statistically insignificant at conventional levels. Restricting the sample by gender produces insignificant and often positive estimates of the relationship between migraine status and hours worked. Turning our attention to the estimates presented in Table III, migraine headache is associated with a 5.0% decrease in the wages of male respondents after conditioning on personal characteristics and family background. However, this association becomes statistically insignificant and is substantially reduced in absolute magnitude when the education controls and measures of personality/comorbidities are included. Controlling for occupational choice and job tenure has only a modest impact on the estimate of β2. Table III provides stronger evidence that female migraineurs are less productive in the workplace than their non-migraineur counterparts. Migraine headache is associated with a 5.1% reduction in the wages of females controlling personal characteristics and family background. This association is reduced in absolute magnitude but is not eliminated, when the education controls and measures of personality/comorbidities are included. Specifically, migraine headache is associated with a 3.3% reduction in the wages of females. Once again, including 10

Previous studies have found that diagnosis with a neurosis or an anxiety disorder can lead to sharply lower wages (Bartel and Taubman, 1986; Ettner et al., 1997; Marcotte and Wilcox-Gök, 2003). The Add Health contains a series of questions designed to measure neuroticism. At the third follow-up, respondents were asked whether they ‘strongly agree’ (=1), ‘agree’ (=2), ‘neither agree nor disagree’ (=3), ‘disagree’ (=4), or ‘strongly disagree’ (=5), with the each of the following statements: 1 2

I have frequent mood swings. I am relaxed most of the time.

3

I get upset easily.

4

4. I seldom feel blue. Each respondent’s neuroticism score was calculated by summing their responses (reverse coding their responses to the first and third statements). In addition, we calculated an anxiety score for each respondent based on their answers to a similar set of questions. Specifically, at the third follow-up, respondents were asked whether they ‘strongly agree’ (=1), ‘agree’ (=2), ‘neither agree nor disagree’ (=3), ‘disagree’ (=4), or ‘strongly disagree’ (=5) with the following statements:

11

1 2

I worry about things. I am not easily bothered by things.

3

I get stressed out easily.

4

I don’t worry about things that have already happened.Each respondent’s anxiety score was calculated by summing their responses (reverse coding their responses to the first and third statements).

Table A1 (available online) presents descriptive statistics for selected control variables by migraine status.

Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

664

D. I. REES AND J. J. SABIA

Table II. Ordinary least squares (OLS) estimates of the relationship between migraine diagnosis and any earnings, hours of work Any Earnings Estimated bias (4)

(5)

0.044*** 0.022*** 0.019*** (0.007) (0.007) (0.007) [14,487] [14,487] [14,487]

0.382*** (0.052) [14,487]

0.029*** (0.007) [13,431]

0.011 (0.007) [6879]

0.008 (0.008) [6879]

0.063 (0.037) [6879]

0.020** (0.009) [7608] Yes

0.175*** (0.032) [7608] Yes

Yes

Yes

(1) Pooled sample Migraine Males Migraine Females Migraine Personal characteristics and family background Education, personality, and comorbidities

Ln(Hours)

(2)

0.012 (0.007) [6879]

0.035*** 0.021** (0.009) (0.009) [7608] [7608] No Yes No

No

(3)

(6)

(7)

0.001 0.009 (0.007) (0.007) [13,431] [13,431]

Estimated bias (8) 0.435*** (0.042) [13,431]

0.010 (0.013) [6622]

0.004 (0.013) [6622]

0.011 (0.013) [6662]

0.007 (0.030) [6662]

0.003 z(0.009) [6809] No

0.004 (0.008) [6809] Yes

0.013 (0.009) [6809] Yes

0.231*** (0.004) [6809] Yes

No

Yes

Yes

No

Unweighted OLS coefficients are reported. Controls include age, age squared, race, ethnicity, gender (for the pooled sample), measured BMI, measured height, marital status, number of children, whether pregnant (for females), urbanicity, family structure during adolescence, household income in adolescence, parental educational attainment, Peabody Picture Vocabulary Test score, years of schooling completed, and whether currently attending school. Standard errors corrected for clustering at the school level are in parentheses. ***Statistically significant at the 1% level; **at the 5% level; *at the 10% level.

job tenure and indicators for the respondent’s two-digit SIC occupational code on the right-hand side of the estimating equation has only a modest impact on the estimate of β2, suggesting that the relationship between migraine headache female earnings is not driven by occupational choice.12 Although a rich set of controls were included in the vector X, we cannot rule out the possibility that migraine status is correlated with the unobserved determinants of labor market outcomes, potentially biasing OLS estimates of β2. One method of addressing this concern is to assume that the correlation between migraine status and the error term of Eqn (1), ε, is of the same sign and magnitude as the correlation between migraine status and β′1X (Altonji et al., 2005).13 Estimated biases under this assumption are reported in columns (4) and (8) of Table II and column (5) of Table III. Because respondents who suffered from migraine headaches tended to be heavier, scored higher on the CES-D, were less likely to have graduated high school, and were more likely to have suffered from hypertension than their non-migraineur counterparts (see Table A1, available online), the estimated biases are,

12

Table A2 (available online) presents estimates of Eqn (1) by age of diagnosis. They provide some evidence that the relationship between migraine status and labor force participation is more pronounced among female respondents who were first diagnosed at a young age. The relationship between migraine status and the hourly wage appears to be strongest among female respondents who were diagnosed as an adolescent. Table A3 (available online) provides OLS estimates controlling for school and grade fixed effects. They are similar to those presented in Tables 2 and 3, suggesting that environment/contextual factors do not drive the estimates between migraine status and the outcomes under study. 13 Following Krauth (2011), the bias under this assumption is equal to corr Migraine; β’1 X

Copyright © 2014 John Wiley & Sons, Ltd.





1=2 var ðεÞ : var ðMigraineÞ Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

665

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

Table III. Ordinary least squares (OLS) estimates of relationship between migraine diagnosis and wages Ln(Hourly Wage)

Pooled sample Migraine Males Migraine Females Migraine Personal characteristics and family background Education, personality, and comorbidities Occupation indicators and job tenure

(1)

(2)

(3)

(4)

Estimated bias (5)

0.107*** (0.015) [13,431]

0.057*** (0.014) [13,431]

0.034*** (0.013) [13,431]

0.039*** (0.012) [13,431]

0.138*** (0.005) [13,431]

0.073** (0.033) [6622]

0.051* (0.030) [6622]

0.027 (0.030) [6622]

0.036 (0.029) [6622]

0.084*** (0.007) [6622]

0.078*** (0.017) [6809] No No No

0.052*** (0.017) [6809] Yes No No

0.034** (0.017) [6809] Yes Yes No

0.040** (0.018) [6809] Yes Yes Yes

0.082*** (0.005) [6809] Yes Yes Yes

Unweighted OLS coefficients are reported. Controls include age, age squared, race, ethnicity, gender (for the pooled sample), measured BMI, measured height, marital status, number of children, whether pregnant (for females), urbanicity, family structure during adolescence, household income in adolescence, parental educational attainment, Peabody Picture Vocabulary Test score, years of schooling completed, and whether currently attending school. Standard errors corrected for clustering at the school level are in parentheses. ***Statistically significant at the 1% level; **at the 5% level; *at the 10% level.

with only one exception, negative.14 Moreover, they are much larger (in absolute magnitude) than the OLS estimates of β2, which, if taken at face value, suggests that migraineurs are actually more likely to participate in the labor market and are more productive than their non-migraineur counterparts. However, there is reason to view this interpretation of the results in Tables II and III with some degree of skepticism, as we discuss in the next section. 4.2. Instrumental variables estimation and results The assumption that the correlation between migraine status and the error term of Eqn (1) is of the same sign and magnitude as the correlation between migraine status and β′1X is, of course, arbitrary (Krauth, 2011), and in the context of estimating the effect of migraine headache on labor market outcomes, there is reason to view it as implausible. Work-induced stress, irregular hours, extreme temperatures, loud noises, strong smells, and flickering/bright lights are all potential triggers of migraine headaches (Adelman and Adelman, 2001; Kaniecki, 2002; Lambert and Zagami, 2008; Andress-Rothrock et al., 2010). Because the labor market compensates workers for stress and disamenities such as irregular hours (McNabb, 1989; Kostiuk, 1990; Schumacher and Hirsch, 1997; Daniel and Sofer, 1998; French and Dunlap, 1998; Groot and Maassen van den Brink, 1999; Glomb et al., 2004; Cole et al., 2009; Bryson et al., 2010), the actual reduction in wages due to migraine headache could be larger than that suggested by the OLS estimates presented in Table III (in other words, the bias could be positive).15

14

OLS estimates of Eqn (1) provide evidence that BMI, the symptoms of depression (as measured by the respondent’s CES-D score), and hypertension are negatively related to labor force participation and wages. Graduating from high school (relative to completing 9–11 years of education) is associated with a 4.8% reduction in the probability of labor force participation and a 7.1% increase in wages. 15 To the extent that migraineurs tend to be ‘high achievers who expect a lot from themselves and accomplish a lot’ (Brody, 1988), the expected direction of the bias would also be positive. Although we include measures of personality in the vector of controls, these measures are not designed to capture motivation or effort. Whether health insurance was provided by the respondent’s employer is another unmeasured job characteristic that could be positively correlated with both migraine headache diagnoses and wages. Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

666

D. I. REES AND J. J. SABIA

One method of addressing the issue of unobservables is through instrumental variables estimation. A number of medical studies have concluded that migraine headache is inheritable. For instance, Russell and Olesen (1995) found that having a first-degree relative who suffered from migraine headache with aura was associated with a substantial increase in the likelihood of suffering from migraine headache with aura, and specific gene mutations have been linked to familial hemiplegic migraine and sporadic hemiplegic migraine, rare and particularly severe subtypes of migraine headache (De Fusco et al., 2003; de Vries et al., 2007; Anttila et al., 2008).16 Relying on this evidence, we estimate the following first-stage equation in an effort to identify exogenous variation in the respondent’s migraine status: Migrainei ¼ α0 þ α′1 Xi þ α2 Mother Migrainei þ ηi ;

(2)

where Mother Migraine is equal to 1 if the respondent’s biological mother was reported to have suffered from migraine headache and is equal to 0 otherwise. This estimation strategy will produce consistent second-stage estimates of β2 provided that Mother Migraine is uncorrelated with the error term of Eqn (1). However, it is possible that mothers who suffered from migraine headache had other inheritable health problems. It is also possible that mothers with migraine headache were unable to devote as much time and resources to their children’s development as their non-migraineur counterparts. In an effort to address these concerns, we augment the vector X with indicators of parental health, whether the respondent’s family moved to the neighborhood because of the quality of the school system, whether the parent who filled out the Add Health parental questionnaire was a member of the local parent teacher association, whether the respondent’s mother worked outside of the home during his or her adolescence, whether the respondent and his or her parents made decisions together, the intensity of the parent’s desire that their child go to college, and the degree to which the respondent’s parents valued scholastic brilliance.17 Estimates of the relationship between mother’s migraine status and whether the respondent was diagnosed with a migraine headache by a health professional are presented in Table IV. Consistent with the results of medical studies (Russell and Olesen, 1995; De Fusco et al., 2003; de Vries et al., 2007; Anttila et al., 2008), we find that having a biological mother who suffered from migraine headaches is associated with a statistically significant 0.046 increase in the probability that male respondents suffered from migraine headache, and an 0.08 to 0.085 increase in the probability that female respondents suffered from migraine headache. The F-statistics in the first-stage easily meet the Staiger and Stock (1997) standard, with values ranging from 19.8 to 55.0. Table V presents two-stage least squares (2SLS) estimates of the relationship between Migraine and the labor market outcomes under study. Although consistently negative, the 2SLS estimates of the relationship between migraine headache and labor force participation are not statistically significant. In contrast, the 2SLS estimate of the relationship between Migraine and wages in the pooled sample is statistically significant and substantially more

Migraines with aura typically begin with visual disturbances such as flashes of light or flickering spots, but otherwise, they exhibit the same symptoms as migraine headache without aura, the most common type of migraine headache (The Headache Classification Subcommittee of the International Headache Society, 2004). 17 Indicators of parental health were based on responses to the question, [h]ow is your general physical health? The possible responses were ‘excellent’, ‘very good’, ‘good’, ‘fair’, and ‘poor’. A dichotomous measure of whether the respondent moved to the area because of school quality was based on responses to the question, ‘[y]ou live here because the schools here are better than they are in other neighborhoods?’ Parental aspirations with regard to their children’s scholastic performance were based on responses to the question, ‘[i]f {NAME} could be one of the following in high school, which would be most important to you?’ Possible responses were ‘a brilliant student’, ‘a leader in school activities’, ‘an athletic star’, ‘the most popular’, and ‘(He/she) has already graduated from high school.’ Parental expectations of college attendance were based on responses to the question, ‘[h]ow disappointed would you be if {NAME} did not graduate from college?’ Possible responses were ‘very disappointed’, ‘somewhat disappointed’, and ‘not disappointed’. Finally, we assessed involvement in official school organizations using responses to the following question, ‘[p]lease tell me whether you are a member of any of the following: Parent/teacher organization.’ Parental employment was measured by parental responses to the question, ‘Do you work outside the home?’ The quality of the parental-child relationship was measured by the parents’ responses on the frequency with which they reported the following: ‘You get along well with him/her’ and ‘You make decisions about life together with him/her.’ Responses to each item were ‘always’, ‘often’, ‘seldom’, ‘sometimes’, and ‘never’. 16

Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

667

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

Table IV. First-stage estimates of relationship between mother’s migraine status and the whether the respondent was diagnosed with migraines by a health professional Pooled sample (1) Mother Migraine First-stage F-stat N Labor Force Participation sample Wages and hours sample

0.066*** (0.009) 55.0 11,929 Yes No

(2) 0.067*** (0.009) 51.1 11,069 No Yes

Males (3)

Females (4)

0.046*** (0.010) 21.9 5696 Yes No

0.046*** (0.010) 19.8 5486 No Yes

(5)

(6)

0.080*** (0.014) 31.7 6233 Yes No

0.085*** (0.015) 31.4 5583 No Yes

Unweighted ordinary least squares coefficients are reported. Controls include age, age squared, race, ethnicity, gender (for the pooled sample), measured BMI, measured height, marital status, number of children, whether pregnant (for females), urbanicity, family structure during adolescence, household income in adolescence, parental educational attainment, Peabody Picture Vocabulary Test score, years of schooling completed, and whether currently attending school. Standard errors corrected for clustering at the school level are in parentheses. ***Statistically significant at the 1% level; **at the 5% level; *at the 10% level.

Table V. Two-stage least squares estimates of the relationship between migraine diagnosis and labor market outcomes Pooled sample Migraine First-stage F-stat N Migraine First-stage F-stat N Migraine First-stage F-stat N Parental health/involvement

0.099 (0.071) 55.0 11,929

0.092 (0.075) 53.0 11,929

0.013 (0.104) 51.5 11,069

0.003 (0.107) 51.0 11,069

0.340* (0.175) 51.1 11,069 No

0.328* (0.180) 50.4 11,069 Yes

Males Panel I: Any Earnings 0.066 0.050 (0.119) (0.124) 21.9 19.6 5696 5696 Panel II: Ln(Hours) 0.162 0.138 (0.202) (0.215) 20.5 18.6 5486 5486 Panel III: Ln(Hourly Wage) 0.434 0.393 (0.366) (0.393) 19.8 18.0 5486 5486 No Yes

Females 0.115 (0.092) 31.7 6233

0.103 (0.095) 31.0 6233

0.052 (0.103) 29.5 5583

0.065 (0.101) 30.0 5583

0.309 (0.191) 31.4 5583 No

0.309* (0.187) 31.8 5583 Yes

Unweighted ordinary least squares coefficients are reported. Controls include age, age squared, race, ethnicity, gender (for the pooled sample), measured BMI, measured height, marital status, number of children, whether pregnant (for females), urbanicity, family structure during adolescence, household income in adolescence, parental educational attainment, Peabody Picture Vocabulary Test score, years of schooling completed, and whether currently attending school. Wage models include additional controls for occupation and job tenure. Standard errors corrected for clustering at the school are in parentheses. ***Statistically significant at the 1% level; **at the 5% level; *at the 10% level.

negative than the corresponding OLS estimate. Specifically, without controls for parental health and involvement, being diagnosed with a migraine headache is associated with a 28.8% decrease in wages. When the sample is restricted to males, migraine headache is associated with a (statistically insignificant) 35.2% decrease in wages; when the sample is restricted to females, migraine headache is associated with a (statistically insignificant) 26.6% decrease in wages. This pattern of results is consistent with the hypothesis that migraine triggers such as work-induced stress, bad lighting, and loud noises are compensated in the labor market. The second-stage estimates in the odd-numbered columns of Table V could be biased if mothers who suffered from migraine headache had other inheritable health problems or were unable to devote as much time and resources to their children’s development as their non-migraineur counterparts. In an effort to explore this possibility, we include controls for parental health and involvement in the even-numbered columns of Table V. Adding these controls has only a modest impact on the pooled estimate of β2 and absolutely no impact on the second-stage estimate Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

668

D. I. REES AND J. J. SABIA

of the relationship between migraine status and the wages of females. In fact, controlling for parental health and involvement in school-related activities actually improves the precision of the latter estimate. Although still imprecise (the standard error is roughly 10 times the standard error produced using OLS), migraine diagnosis is associated with a 26.6% decrease in the wages of females even controlling for parental health and involvement. However, it should be noted that we cannot formally reject the hypothesis that this estimate is larger than the corresponding OLS estimate presented in column (4) of Table III.

5. CONCLUSION Using data drawn from the National Longitudinal Study of Adolescent Health (Add Health), this study is the first to investigate the effect of migraine headache on labor market outcomes. OLS estimates suggest that migraine headache is associated with a 2 percentage-point reduction in the probability that young adult females participate in the labor market; conditional on working, migraine headache is associated with a 3% reduction in their wages. OLS estimates of the relationship between migraine headache and male labor market outcomes are smaller and imprecise. Although we observe and control for individual and family background characteristics, the possibility that migraine status is correlated with the unobserved determinants of labor market outcomes cannot be ruled out. We employ two estimation strategies in an effort to account for the influence of unobservables. Following Altonji et al. (2005), we assume that the correlation between migraine status and the error term of estimating equation is of the same sign and magnitude as the correlation between migraine status and the observables. However, because work-induced stress and irregular hours are potential triggers of migraine headaches, and because the labor market compensates workers for stress and disamenities such as irregular hours, we argue that this assumption is implausible. Indeed, estimates based on this assumption suggest that migraineurs are actually more likely to participate in the labor market and are more productive than their non-migraineur counterparts. Our second method of accounting for the influence of unobservables relies on evidence from medical studies that migraines are inheritable. Specifically, we instrument for migraine status using an indicator of whether the respondent’s mother suffered from migraines. We find strong evidence that migraine headache leads to reduced productivity among females, but second-stage estimates of the relationship between migraine headache and the wages of males are not statistically significant at conventional levels. In an effort to test whether maternal migraine status could have affected adult children’s labor market outcomes through non-genetic routes, we control for several observable measures of parental time and preferences. Adding these controls has only a modest impact on the second-stage estimates of the relationship between migraine status and the wages of females. Recently, a number of new prescription migraine medications from a class of drugs called triptans have come onto the market and are now in wide use. Although no medication is 100% effective, the introduction of these more effective drugs may have increased the workplace productivity of migraine sufferers. To the extent that this is the case, our estimates should be viewed as lower bounds. It should be noted, however, that many sufferers delay taking prescription medications when they sense a headache coming on, and some choose to avoid taking them altogether. For instance, one study found that approximately two thirds of migraineurs delayed or avoided taking prescription drugs because of concerns about their side effects, which include difficulty concentrating, dizziness, nausea, and sleepiness (Gallagher and Kunkel, 2003). To put the magnitude of our estimates in perspective, it is useful to compare them to estimated effects of other health conditions on wages. For instance, Currie and Hyson (1999) found that low birth weight is associated with a 2–4% decrease in wages; Fletcher (2009) found that adolescent depression is associated with a 15% reduction in labor market earnings; and Fletcher (2010) found that ADHD is associated with a 33% reduction in earnings. Thus, migraine headaches do not appear to be as debilitating as ADHD, but their effect on productivity appears to be at least as large as weighing less than 2500 g at birth, the cut-off for low birth weight typically used in the medical profession. Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

669

Approximately 10% of adults in North America and Western Europe suffer from migraine headaches (Breslau and Rasmussen, 2001). Yet, no previous study has examined the effects of this debilitating condition on labor market outcomes such as hours and wages. Our results suggest that female migraine sufferers are less productive than their counterparts who do not struggle with the pain, sight impairment, and concentration lapses that come with migraines. Future estimates of the cost of migraine headache to society should be adjusted upwards, perhaps substantially.

ACKNOWLEDGEMENT All authors contributed equally to the data collection effort, the analysis of the data, the interpretation of the results, and the writing of the manuscript. Because this study used secondary data, IRB approval was not needed or obtained.

CONFLICT OF INTEREST

The authors report no conflicts of interest or outside funding.

REFERENCES

Adelman JU, Adelman RD. 2001. Current options for the prevention and treatment of migraine. Clinical Therapeutics 23(6): 772–788. Altonji J, Conley T, Elder T, Taber C. 2005. Using selection on observed variables to assess bias from unobservables: assessing the effectiveness of catholic schools. Journal of Political Economy 113(1): 151–184. Andress-Rothrock D, King W, Rothrock J. 2010. An analysis of migraine triggers in a clinic-based population. Headache: The Journal of Head and Face Pain 50(8): 1366–1370. Anttila V, Nyholt DR, Kallela M, Artto V, Vepsalainen S, Jakkula E, Wennerstro A, Tikka-Kleemola P, Kaunisto MA, Hamalainen E, Widen E, Terwilliger J, Merikangas K, Montgomery GW, Martin NG, Daly M, Kaprio J, Peltonen L, Farkkila M, Wessman M, Palotie A. 2008. Consistently replicating locus linked to migraine on 10q22–q23. American Journal of Human Genetics 82(5): 1051–1063. Bartel A, Taubman P. 1986. Some economic and demographic consequences of mental illness. Journal of Labor Economics 4: 243–256. Belam J, Harris G, Kernick D, Kline F, Lindley K, McWatt J, Mitchell A, Reinhold D. 2005. A qualitative study of migraine involving patient researchers. British Journal of General Practice 55(511): 87–93. Berg J. 2004. Economic evidence in migraine and other headaches: a review. European Journal of Health Economics 5(1): S43–S54. Bigal EM, Lipton RB, Winner P, Reed ML, Diamond S, Stewart WF. 2007. Migraine in adolescents: association with socioeconomic status and family history. Neurology 69(1): 16–25. Brandt J, Celentano D, Stewart W, Linet M, Folstein MF. 1990. Personality and emotional disorder in a community sample of migraine headache sufferers. American Journal of Psychiatry 147(3): 303–308. Breslau N, Andreski P. 1995. Migraine, personality, and psychiatric comorbidity. Headache 35(7): 382–386. Breslau N, Rasmussen BK. 2001. The impact of migraine epidemiology, risk factors, and co-morbidities. Neurology 56(1): S4–S12. Breslau N, Davis GC, Andreski P. 1991. Migraine, psychiatric disorders, and suicide attempts: an epidemiologic study of young adults. Psychiatry Research 37: 11–23. Brody JE. 1988. Personal health. New York Times, Thursday October 13. Available at: http://www.nytimes.com/1988/10/ 13/us/health-personal-health.html Bryson A, Barth E, Dale-Olsen H. 2010. Do higher wages come at a price? NIESR Discussion Paper No. 371. Available at: http://www.niesr.ac.uk/pdf/231110_173447.pdf Burton WN, Landy SH, Downs KE, Runken MC. 2009. The impact of migraine and the effect of migraine treatment on workplace productivity in the United States and suggestions for future research. Mayo Clinic Proceedings 84(5): 436–445. Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

670

D. I. REES AND J. J. SABIA

Cao M, Zhang S, Wang K, Wang Y, Wang W. 2002. Personality traits in migraine and tension-type headaches: a five-factor model study. Psychopathology 35(4): 254–258. Clarke CE, MacMillan L, Sondhi S, Wells NE. 1996. Economic and social impact of migraine. Q J M 89(1): 77–84. Cole MA, Elliott RJR, Lindley JK. 2009. Dirty money: is there a wage premium for working in a pollution intensive industry? Journal of Risk and Uncertainty 39(2): 161–180. Currie J, Hyson R. 1999. Is the impact of health shocks cushioned by socioeconomic status? The case of birth weight. American Economic Review 89(2): 245–250. de Vries B, Freilinger T, Vanmolkot KRJ, Koenderink JB, Stam AH, Terwindt GM, Babini E, van den Boogerd EH, van den Heuvel JJMW, Frants RR, Haan J, Pusch M, van den Maagdenberg AMJM, Ferrari MD, Dichgans M. 2007. Systematic analysis of three FHM genes in 39 sporadic patients with hemiplegic migraine. Neurology 69(23): 2170–2176. Daniel C, Sofer C. 1998. Bargaining compensating wage differentials, and dualism of the labor market: theory and evidence for France. Journal of Labor Economics 16(3): 546–575. De Fusco M, Marconi R, Silvestri L, Atorino L, Rampoldi L, Morgante L, Ballabio A, Aridon P, Casari G. 2003. Haploinsufficiency of ATP1A2 encoding the Na+/K+ pump α2 subunit associated with familial hemiplegic migraine type 2. Nature Genetics 333(2): 192–196. Ettner SL, Frank RG, Kessler RC. 1997. The impact of psychiatric disorders on labor market outcomes. Industrial and Labor Relations Review 51(1): 64–81. Fletcher JM. 2009. Adolescent depression and adult labor market outcome. Yale University Working Paper. Fletcher JM. 2010. The effects of childhood ADHD on adult labor market outcomes. Yale University Working Paper. French MT, Dunlap LJ. 1998. Compensating wage differentials for job stress. Applied Economics 30(8): 1067–1075. Gallagher RM, Kunkel R. 2003. Migraine medication attributes important for patient compliance: concerns about side effects may delay treatment. Headache: The Journal of Head and Face Pain 43(1): 36–43. Glomb TM, Kammeyer-Mueller JD, Rotundo M. 2004. Emotional labor demands and compensating wage differentials. Journal of Applied Psychology 89(4): 700–714. Groot W, Maassen van den Brink HM. 1999. The price of stress. Journal of Economic Psychology 20(1): 83–103. Harris DR, Thomas JL. 2002. The educational costs of being multiracial: evidence from a national survey of adolescents, Research Report No. 02-521, Population Studies Center at the Institute of Social Research, University of Michigan. Harris KM, Halpern CT, Entzel P, Tabor J, Bearman PS, Udry JR. 2008. The National Longitudinal Study of Adolescent Health: research design [WWW document]. URL: http://www.cpc.unc.edu/projects/addhealth/design Headache Classification Subcommittee of the International Headache Society. 2004. International classification of headache disorders, 2nd edition. Cephalalgia 24(S1): 8–160. Kaniecki RG. 2002. Migraine and tension-type headache: an assessment of challenges in diagnosis. Neurology 58(9): S15–S20. Kelman L, Rains JC. 2005. Headache and sleep: examination of sleep patterns and complaints in a large clinical sample of migraineurs. Headache: The Journal of Head and Face Pain 45(7): 904–910. Kentle RL. 1997. Elements of neuroticism in relation to headache symptomatology. Psychological Reports 80(1): 227–235. Kostiuk PF. 1990. Compensating differentials for shift work. Journal of Political Economy 98(5): 1054–1075. Krauth B. 2011. Bounding a linear causal effect using relative correlation restrictions. Working Paper, Department of Economics Simon Fraser University. Available at: http://www.sfu.ca/~bkrauth/research.htm. Lambert GA, Zagami AS. 2008. The mode of action of migraine triggers: a hypothesis. Headache: The Journal of Head and Face Pain 49(2): 253–275. Lipton RB, Stewart WF, Diamond S, Diamond ML, Reed ML. 2001. Prevalence and burden of migraine in the United States: data from the American migraine study II. Headache 41(7): 646–57. Lipton RB, Bigal ME, Diamond M, Freitag F, Reed ML, Stewart WF. 2007. Migraine prevalence, disease burden, and the need for preventive therapy. Neurology 68(5): 343–349. Marcotte DE, Wilcox-Gök V. 2003. Estimating earnings losses due to mental illness: a quantile regression approach. The Journal of Mental Health Policy and Economics 6(3): 123–134. Mattsson P, Ekselius L. 2002. Migraine, major depression, panic disorder, and personality traits in women aged 40–74 years: A population-based study. Cephalalgia 22(7): 543–551. McNabb R. 1989. Compensating wage differentials: some evidence for Britain. Source: Oxford Economic Papers 41(2): 327–338. Merikangas KR, Stevens DE, Angst J. 1993. Headache and personality: results of a community sample of young adults. Journal of Psychiatric Research 27(2): 187–196. Nylander P-O, Schlette P, Brändström S, Nilsson M, Forsgren T, Forsgren L, Adolfsson R. 1996. Migraine: temperament and character. Journal of Psychiatric Research 30: 359–368. O’Brien B, Goeree R, Streiner D. 1994. Prevalence of migraine headache in Canada: a population-based survey. International Journal of Epidemiology 23(5): 1020–1026. Peters M, Abu-Saad HH, Vydelingum V, Dowson A, Murphy M. 2004. Migraine and chronic daily headache management: a qualitative study of patients’ perceptions. Scandinavian Journal of Caring Sciences 18(3): 294–303. Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

MIGRAINE HEADACHE AND LABOR MARKET OUTCOMES

671

Pradalier A, Auray J-P, El Hasnaoui A, Alzahouri K, Dartigues J-F, Duru G, Henry P, Lanteri-Minet M, Lucas C, Chazot G, Gaudin A-F. 2004. Economic impact of migraine and other episodic headaches in France: data from the GRIM2000 study. PharmacoEconomics 22(15): 985–999. Rees DI, Sabia JJ. 2011. The effect of migraine headache on educational attainment. The Journal of Human Resources 46(2): 317–332. Russell MB, Olesen J. 1995. Increased familial risk and evidence of genetic factor in migraine. British Medical Journal 311(7004): 541–544. Sabia JJ, Rees DI. 2011. Individual heterogeneity and reverse causality in the relationship between migraine headache and educational attainment. Economics of Education Review 30(5): 913–923. Sabia JJ, Rees DI. 2012. Body weight and wages: evidence from Add Health. Economics and Human Biology 10(1): 14–19. Scher AI, Terwindt GM, Picavet SJ, Verschuren MM, Ferrari MD, Launer LJ. 2005. Cardiovascular risk factors and migraine: the GEM population-based study. Neurology 64(4): 614–620. Schumacher EJ, Hirsch BT. 1997. Compensating differentials and unmeasured ability in the labor market for nurses: why do hospitals pay more? Industrial and Labor Relations Review 50(4): 557–579. Staiger D, Stock JH. 1997. Instrumental variables regression with weak instruments. Econometrica 65(3): 557–586. Stewart WF, Lipton RB, Celentano DD, Reed ML. 1992. Prevalence of migraine headache in the United States. Relation to age, income, race, and other sociodemographic factors. The Journal of the American Medical Association 267(1): 64–69. Stronks DL, Tulen JH, Pepplinkhuizen L, Verheij R, Mantel GW, Spinhoven P, Passchier J. 1999. Personality traits and psychological reactions to mental stress of female migraine patients. Cephalalgia 19(6): 566–574. Waldie KE, Poulton R. 2002. The burden of illness associated with headache disorders among young adults in a representative cohort study. Headache: The Journal of Head and Face Pain 42(7): 612–619. Waldie KE, Hausmann M, Milne BJ, Poulton R. 2002. Migraine and cognitive function: a life-course study. Neurology 59(6): 904–908. Zwart JA, Dyb G, Hagen K, Ødegård KJ, Dahl AA, Bovim G, Stovner LJ. 2003. Depression and anxiety disorders associated with headache frequency. The Nord-Trøndelag Health Study. European Journal of Neurology 10(2): 147–152.

Copyright © 2014 John Wiley & Sons, Ltd.

Health Econ. 24: 659–671 (2015) DOI: 10.1002/hec

Migraine headache and labor market outcomes.

While migraine headache can be physically debilitating, no study has attempted to estimate its effects on labor market outcomes. Using data drawn from...
129KB Sizes 0 Downloads 3 Views