MENTAL ILLNESS, NATIVITY, GENDER AND LABOR SUPPLY

SUMMARY

We analyzed the impacts of nativity and mental health (MH) on work by gender for non-elderly adults using the 2002 National Survey on Drug Use and Health. We employed two indicators of MH – the K6 scale of Mental Illness (MI) and an indicator for symptoms of Mania or Delusions (M/D). Instrumental variable (IV) models used measures of social support as instruments for MI. Unadjusted work rates were higher for immigrants (vs US-born adults). Regressions show that MI is associated with lower rates of work among US-born males but not immigrant males and females; M/D is associated lower rates of work among US-born males and females, and among immigrant males. Results did not change using IV models for MI. Most persons with MI work, yet symptom severity reduces labor supply among natives especially. Immigrants’ labor supply is less affected by MI.

1. INTRODUCTION

Because of continuing immigration, the foreign-born represent an important sector of United State’s workforce. Demographic data suggest that, in aggregate, immigrants differ from natives in their human capital, a condition likely shaping their work experiences in the US. One feature of human capital is health. The relationships between physical and mental health (MH) and labor market outcomes have been shown to be strong and robust (Currie and Madrian, 1999). Clinical and population-based data show that mental illness (MI) is associated with negative outcomes for indicators such as earnings, number of hours worked, and retirement decisions. The impact of MI may differ between native-born and immigrants for two reasons. First, the underlying rates of illness might vary. Epidemiologic studies find that MI is less prevalent among immigrants compared with natives, a finding that contrasts with the disadvantages (e.g. socioeconomic, migration related, or structural) faced by many immigrants. Second, worker or potential worker reaction to illness can also differ across groups. Individuals’ labor market activities are thus not solely a function of their human capital, but also of how they choose to use that capital. This paper addresses this second issue.

Building on earlier research, this paper examines the relationship between nativity, MH, and labor supply. This paper makes a new contribution by studying the differential impact of psychiatric disorders on labor market outcomes by immigrant status. We also take into account the possible effects of income transfers on labor market outcomes. We implement two measures of MH, and we employ an instrumental variable (IV) approach, in an effort to address possible reverse causality from labor market conditions to MH status. This study analyzes a recent nationally representative database with the purpose of detailing the experiences of immigrants and natives with MI in relation to their employment activities. We report on men and women separately since it is well established that labor market trajectories, incomes, and employment characteristics differ by gender (Goldin, 1990) and this approach is supported by prior research (Ettner, 2000; Marcotte et al., 2000).

This paper is organized as follows: in the second section, we review the relevant literature including theoretical underpinnings for this research. Next, descriptions of the analytic framework and data are provided. The fourth section presents results. Policy implications are discussed in the fifth and final section of the paper.

2. PREVIOUS LITERATURE

The US is home to more than 33 million immigrants; Latin Americans and Asians have dominated recent migratory flows (Larsen, 2004). Today, one of every seven workers in the US (~20 million people) is an immigrant (Alsalam and Smith, 2005). Although the background and circumstances of immigrants vary widely, they differ from the native-born population in age, educational attainment, and other socioeconomic characteristics. Immigrants are generally younger and of working age, they have on average completed fewer years of schooling, and have lower incomes than natives (Larsen, 2004). Recent figures indicate that while traditional receiving states such as New York, California, Texas, and Illinois remain attractive to foreign-born persons, immigrants are now settling in ‘new-growth’ states such as Maryland, North Carolina, and Georgia (Passel and Zimmermann, 2001). Therefore, the work activities and characteristics of immigrant populations are having broader impacts on the nation’s labor markets and economy.

Work behaviors are shaped by numerous factors including physical health and MH. Interest in the relationship between health and labor market outcomes is not new and persists not only because poor health may profoundly affect an individual’s quality of life and economic position, but also because morbidity has additional social and economic costs for society as a whole. Part of understanding the potential payoffs to treatment of mental disorder relates to the range of impacts mental disorders have on the total population.

2.1. Psychiatric epidemiology

In the US, MIs are among the most common chronic conditions, though they vary within the population according to demographic and socioeconomic characteristics (Kessler et al., 2005). The National Comorbidity Survey – Replication (NCS-R), a national psychiatric epidemiologic study, finds that 26% of adults in the US meet clinical criteria for at least one MI in the past year; 18% of adults are affected by anxiety disorders, 10% by mood disorders such as depression, 9% by impulse control disorders, and 4% by any substance disorder (Kessler et al., 2005). The early onset of a psychiatric disorder (Kessler et al., 2005) (a health capital shock) can affect the accumulation of human capital (e.g. educational attainment or acquisition of work experience and skills) and may also affect future investments in health capital.

Some research suggests that psychiatric disorders vary by nativity, with foreign-born status appearing to have a protective effect. An epidemiologic study in Central California found that 12-month prevalence rates of DSM-III-R diagnoses were lower among immigrant Mexican-origin non-elderly adults than in their US-born Mexican American peers (Vega et al., 2004). Overall, 13.3% of immigrants had any disorder compared with 27.5%of US-born Mexican Americans. Natives’ rates of disorders were more than twice those of immigrants: for instance, 4.4% of immigrants met criteria for any affective disorder (vs 11% of natives), 7.3% had any anxiety disorder (vs 14.7% of natives), 4.2%had any substance abuse/dependence (vs 12.3% of natives) (Vega et al., 2004). In a study of a diverse, low-income Black population, the Primary Care Evaluation of Mental Disorders was used to assess the MH of women receiving services through county health and welfare services; the study found that the odds of probable current depression was lower for immigrants than US-born black women (OR = 0.34 for African women and 0.42 for Caribbean women) (Miranda et al., 2005). The Chinese American Psychiatric Epidemiological Survey conducted in Los Angeles between 1993–1994 assessed the prevalence of major depression and dysthymia in non-elderly Chinese Americans; 3.4% of adults had an episode of major depression in the prior year and about 1% experienced dysthymia in the prior 12 months (Takeuchi et al., 1998).

Data on the impact of proxy measures of acculturation for MH reveal inconsistent findings; the number of years of living in the US and English proficiency are two measures commonly analyzed. Vega and colleagues (Vega et al., 2004) report that Mexican immigrants with a greater duration (i.e. ≥ 13 years) of residency in the US exhibited higher rates of any psychiatric disorder in the prior year than recent immigrants (<13 years) (18.4 vs 9.2%, respectively). Among African or Caribbean-born black women, report of probable depression is higher with 10 additional years in the US (OR = 1.35) (Miranda et al., 2005). In contrast, for Chinese Americans, neither time in the US nor language are significantly related to 12-month episodes of major depression or dysthymia (Takeuchi et al., 1998). A person’s age at the time of migration may also play an important role in MH status and other health behaviors, though evidence is inconclusive. Younger age of entry into the US was associated with higher rates of substance use disorders and non-substance use disorders in Mexican-origin adults (Vega et al., 2004), whereas age-at-migration was not significantly associated with 12-month depression among Chinese immigrants (Takeuchi et al., 1998). Researchers have also identified an apparent protective effect of a foreign-born status in other areas (e.g. lower rates of infant mortality, low birthweight, preterm births, cigarette smoking, alcohol consumption) (Rumbaut, 1999) beyond MH, suggesting robustness in spite of adversity.

2.2. MH and labor market outcomes

Studies of the impact of MH on labor market outcomes generally show that poor MH adversely affects employment indicators and the estimated impacts differ by gender and disorder. Research on MH and work relies on diverse data, ranging from clinical studies to population-based community surveys. A pioneering study using clinical data and self-report for a population of white male twins born between 1917 and 1927 identified negative and long-term effects of psychoses and neuroses on wages, employment and hours worked (Bartel and Taubman 1979). A follow-up study identified additional effects on the families’ labor supply: women’s work increased among those whose husbands had a diagnosis of neuroses, and was higher especially among those with a recent diagnosis (Bartel and Taubman 1986). In another study, longitudinal data collected in a 30-year follow-up survey of white males who were youths during the 1920s in St. Louis, showed that weekly earnings of psychotic men were about 43% lower than those of healthy men, and persons with psychosis or sociopathy were less likely to be employed (Benham and Benham, 1982).

Community surveys such as the Epidemiologic Catchment Area Survey (ECA) or the National Comorbidity Survey (NCS) elicit symptom information to determine a clinical psychiatric diagnosis, and link diagnoses to labor force outcomes. NCS data showed that MI reduces men and women’s employment by 11%; diagnoses of major depression, agoraphobia, or drug dependence reduced women’s work and reductions in incomes were observed among working women with agoraphobia or schizophrenia. For men, major depression and alcohol dependence negatively impact men’s work (Ettner et al., 1997). Analyses of ECA data indicate that MH status and alcohol abuse shapes older men’s retirement decisions and MH plays a modest role in older women’s retirement decisions (Mitchell and Anderson, 1989). Another analysis of ECA data finds that for women and men, full-time (FT) work is significantly and positively associated with self-rated MH; however, specific diagnoses are important in understanding men’s work decisions; ties between MH and work are also mediated by physical health (Mullahy and Sindelar, 1990).

In addition to the breadth of labor market outcomes analyzed, Ettner (Ettner, 2000) made an important methodological contribution to the field by also estimating two-stage IV models to address the issue of endogeneity of MH status and employment. Using data from the 1995 Midlife in the United States study, Ettner (Ettner, 2000) examined employment status and among workers, nine job characteristics (e.g. occupational category, self-employment, job demands, skills, authority, etc.), as functions of five classes of health-measures (i.e. self-rated health, physical health/MH, functional limitations, medical conditions, and indicators for MH and substance abuse). The study also controlled for respondents’ sociodemographic and family characteristics, spouse’s physical health and MH characteristics, and state unemployment rates. Instruments were respondent’s and parents’ health. In general, better health is associated with employment. However, for both men and women, functional limitations reduce work. Furthermore, for men, a greater number of medical conditions also negatively impacts labor supply. Significantly, results were qualitatively unchanged in the IV analyses. Thus, Ettner concludes that investigators lacking suitable instruments may continue controlling for health status in models, since an IV approach may not produce substantively different findings.

Marcotte and colleagues (Marcotte et al., 2000) analyzed the NCS and adopted a two-stage IVs approach to study the effects of affective disorders on income losses and employment for men and women separately. Instruments include family history of affective disorders and predicted probabilities of psychiatric disorders. Results from standard ordinary least squares (OLS) regressions and IV analyses differed at times. In OLS analyses, non-affective MI and dysthymia reduces women’s annual income; dysthymia reduces men’s incomes. IV analyses show that depression reduces women’s income by ~$6036 annually; for men, neither dysthymia nor other affective disorders are significantly associated with reduced income, yet the number of years since the onset of dysthymia predicts income loss (Marcotte et al., 2000). In OLS and IV analyses depression reduces women’s employment, whereas affective disorders do not reduce men’s employment. In sum, the data showed that depression is associated with women’s lost income and dysthymia is less consistently associated with men’s reduced incomes. Depression negatively impacts women’s employment while affective disorders were not associated with changes in men’s work. The authors remark that chronic nature of affective disorders can be presumed to have important consequences for men’s and women’s labor market outcomes but improvements may occur with time.

Alternatively, analyses of the Health and Retirement Survey identify depression and comorbid pain as a significant determinant of pre-retirement aged (55–65 years) adults’ labor supply, resulting in greater retirement and limitations from work as well as lower annual household incomes (Tian et al., 2005). Using various measures for MI, significant income effects have been identified for males’ contributions to the household income (Frank and Gertler, 1991). Lastly, there is evidence that MI negatively affects workplace behaviors; affective disorders like depression are related to increased number of absences from work (Kessler et al., 1999).

Race/ethnicity are often included as covariates, and to our knowledge, only one study by Chatterji and colleagues (Chatterji et al., 2007) has explicitly focused on the labor market outcomes of ethnic minorities affected by MI. The study relied on data from National Latino and Asian American Study and researchers reported on three outcomes for Latino and Asian men and women separately: (1) employment vs unemployment; (2) among workers, the number of weeks worked; and (3) among workers, report of at least one day’s absence from work in the prior month. Independent variables were various measures of psychiatric disorder (e.g. current psychiatric disorder, current anxiety disorder, current substance use disorder, K10 score, any prior disorder). The analytic plan included standard probit or OLS models, bivariate probits, and an IV approach; identifying variables were (1) number of psychiatric disorders before age 18, (2) attendance of weekly religious services, and (3) use of religious means to deal with life’s problems (Chatterji et al., 2007). Measures of religiosity were considered to represent social capital. Results showed that MI and psychiatric distress both reduced Latino males’ and females’ employment, though the number of weeks worked was not significantly impacted; furthermore, psychiatric distress was also associated with a higher probability of absenteeism. Findings for Asians were mostly non-siginificant. A psychiatric disorder was significantly associated with a reduced probability of employment for men only. Chatterji et al did not observe any gains by applying an IV approach. The investigators concluded that the impacts of MI among Latinos are comparable to those observed in non-Latino whites, while results are inconsistent among Asians; possible explanations include variations in educational attainment, occupational distributions and access to institutional supports via sick-days, and cultural expressions of MI and reactions to it.

The outcomes studied and results from this body of research are diverse, yet findings establish the importance of MI as a factor affecting labor market outcomes. We know of no study that has yet examined the joint contributions of nativity and psychiatric disorders on labor supply. Consequently, it is not yet possible to state whether there may be differential effects of MI on immigrants’ work supply compared with that of natives; this is focus of this investigation.

3. EMPIRICAL SPECIFICATION

This paper investigates whether MI in working-age adults differentially affects the labor supply of immigrant workers as compared with US-born workers. Prior research has documented that the labor market experiences of men and women differ across various domains (e.g. employment rates, earnings, hours worked, hourly wages, occupational segregation, etc.) (Goldin, 1990; Waldfogel and Mayer, 2000). We build on prior gender-focused analyses of MI and labor market outcomes (Ettner, 2000; Marcotte et al., 2000) through this contribution. These analyses rely on two indicators of psychiatric distress. The first is an indicator of MI, which refers to diagnosable MI in the prior 12 months. A second indicator is composed of symptoms of mania and delusions (M/D) and does not necessarily reflect a medical diagnosis. Both measures are described fully below. We also adopted an IV approach. We estimate multivariate logistic regression models of labor supply on indicators of psychiatric morbidity and nativity holding constant other demographic and human capital characteristics of individuals. We concentrate on two dichotomous outcomes: (1) any work for pay activities (i.e. full-time work, part-time work, and other work) vs no work, and among workers: (2) full-time work vs part-time/other work. Separate models were estimated for men and women. Two IV models are employed to test for endogeneity of MI, as described below. Models were estimated via maximum likelihood and goodness of fit was assessed using a modified Hosmer–Lemeshow test (Hosmer and Lemeshow, 1989). A description of the data and the measures included in these analyses is provided below.

3.1. Data and sample

This study makes use of recently collected data that offers a large sample, various indicators of MH status, and information on nativity and labor market experiences, including respondents’ sources of income. However, the survey does not provide information on family history of MI, age-of-onset of illness, timing of illness, or citizenship status.

We examined the labor market experiences of adults ages 18 through 64, excluding students and homemakers from the sample. Data from the 2002 National Survey on Drug Use and Health (NSDUH) were analyzed (US Department of Health and Human Services, 2002). The survey is sponsored annually by the Substance Abuse and Mental Health Services Administration (SAMHSA) and fielded by the Research Triangle Institute of North Carolina. Data were analyzed using SUDAAN version 9.0.1 to account for the complex sample design. STATA was used to develop bivariate probit models (STATA Corporation, 2007). The unweighted sample size is 30 223 persons, which when weighted represent 161.1 million working-age persons ages 18–64 (excluding students and homemakers) nationwide. Our study includes 26 843 US-born (weighted N = 138.6 million adults) and 3380 immigrant non-elderly adults (weighted N = 22.5 million adults). All estimates are weighted using population weights provided by the NSDUH.

The NSDUH’s sample is based on a 50-State design that includes an independent multistage area probability sample for each state and the District of Columbia. Unfortunately, state-identifiers are unavailable in public use files (PUF) to protect respondents’ confidentiality. Furthermore, the survey does not collect any identifying information (e.g. name, social security, etc.) of respondents. The survey employs in-person computer-assisted personal interviewing and audio computer-assisted interviewing, methods that increase respondents’ privacy, confidentiality of sensitive behaviors, and data quality. Beginning in 2002, respondents providing a fully completed interview receive a cash incentive (US Department of Health and Human Services, 2002).

The NSDUH is designed to provide national- and state-level estimates of the prevalence and correlates of drug, alcohol, and tobacco use by the non-institutionalized civilian population age 12 and older residing in the US (including civilians residing on military bases). Certain subgroups are not represented (e.g. active military personnel, persons in residential drug treatment facilities, prisons, or long-term care facilities, or homeless persons not residing in shelters). All persons in the sampling frame are eligible for participation in the survey, regardless of birthplace or citizenship. The survey is well suited for this study since it also collects sociodemographic and employment data, MH status, and information on use of MH and substance use services.

3.2. Indications of MI

MH status can be measured in numerous ways. Our analyses are based on two indicators, which represent different dimensions of illness. Our first measure of MH status relied on a respondent’s score resulting from answers to a six-item validated scale, the K6, which was developed specifically as a screening tool to assess MI in community surveys including the NSDUH (refer to Table I for K6 items). Starting in 2004 wave of the NSDUH, the K6 has been described as an indicator of serious psychological distress rather than serious MI (as listed in the 2002 NSDUH used in this study) (Aldworth et al., 2005). An individual was considered to have a MI if, at some time in the prior year, he or she had one of a set of mental disorders from the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria (excluding substance use disorder) that resulted in functional impairment or significant interference with/limitation of at least one major life activity. More specifically, the K6’s measure of MI reflects a 12-month diagnosis of an anxiety disorder, or mood disorder, or non-affective psychosis (Kessler et al., 2002). To be classified as having a MI, respondents must score 13 or more points (out of a possible 24 points). Persons scoring fewer than 13 points are classified as not having a MI (Substance Abuse and Mental Health Services Administration, 2002). We use a dichotomous measure (0/1) reflecting the presence or absence of MI, although we recognize that the K6 score itself does not represent a specific DSM diagnosis.

Table I.

NSDUH survey items used for mental health status indicator

Survey category	Mental illness (MI)	Mania or delusions (MD)
Lead-in question	‘Most people have periods when they are not at their best emotionally. Think of one month in the past 12 months when you were the most depressed, anxious, or emotionally stressed. If there was no month like this, think of a typical month’.	‘The next questions are about unusual experiences that some people have. When you answer these questions, please do not include times when you were dreaming or half asleep, had a high fever, or were under the influence of illegal drugs or alcohol’.
Survey questions	‘During that month, how often did you feel nervous?’	VOICES: ‘During the past 12 months, have you heard voices – that is, voices that other people said did not exist, voices coming from inside your head, or voices coming out of the air when there was no one around?’
	‘During that same month when you were at your worst emotionally… how often did you feel hopeless?’	FORCE TAKING OVER MIND: ‘During the past 12 months, have you felt that a force was taking over your mind and trying to make you do things you didn’t want to do?’
	‘During that same month when you were at your worst emotionally… how often did you feel restless or fidgety?’	FORCE INSERTING THOUGHTS: ‘During the past 12 months, have you felt that some force was inserting thoughts directly into your head by means of x-rays or laser beams or other methods?’
	‘During that same month when you were at your worst emotionally… how often did you feel so sad or depressed that nothing could cheer you up?’	THOUGHTS STOLEN: ‘During the past 12 months, have you felt that your own thoughts were being stolen out of your mind by someone or something you did not have control over?’
	‘During that same month when you were at your worst emotionally… how often did you feel that everything was an effort?’	PLOT TO HARM: ‘During the past 12 months, have you believed that there was an unfair plot going on to harm you or to have people follow you – when your family and friends did not believe that this was happening?’
	During that same month when you were at your worst emotionally… how often did you feel down on yourself, no good, or worthless?’	SEE VISION: ‘During the past 12 months, have you seen a vision – that is, something that other people could not see?’
		HYPER/MANIC: Respondents were asked ‘During the past 12 months, were there at least four days in a row when you were so excited or hyper that you either got into trouble or people worried about your being so excited, or a doctor said you were manic?’
Response set	(1) All of the time, (2) Most of the time, (3) Some of the time, (4) A little of the time, and (5) None of the time; Don’t know or Refuse to answer are also options.	(1) Yes, (2) No; Don’t know or Refuse to answer are also options.

Note: Items are listed verbatim from the National Survey on Drug Use and Health.

A potential limitation of the K6 is its emphasis on symptoms of depression and anxiety, and the lack of data regarding the age-of-onset of illness and timing of illness with respect to work experiences. As a result, there is potential for reverse causation from labor market outcomes to MH status (i.e. the loss of employment may trigger symptoms of depression and/or anxiety). To address this issue, we constructed a second indicator of MH status based on symptoms of Mania and Delusions (MD) (Table I provides questionnaire items used to construct the MD indicator). To be classified as having symptoms of mania or delusions, respondents must have answered in the affirmative to at least one of the eight questions. We constructed this indicator with the belief that these symptoms are suggestive of more severe and persistent MIs. For example, the symptoms described in our indicator of MD are also features ofserious MIs including schizophrenia and other psychotic disorders as well as bipolar disorder, as described by the DSM-IV (American Psychiatric Association, 1994). These illnesses may be less affected by immediate life circumstances (e.g. employment or other life stressors).

3.3. Dependent variables

The NSDUH provides information on respondents’ labor market activities in the week prior to the survey (refer to Table II). The study focuses on two labor market outcomes: (1) any employment (i.e. full-time, part-time, or other work), with a reference group composed of persons outside the labor market (i.e. unemployed, disabled, retired, no job-reason unspecified, and among workers (2) full-time work vs a reference group of other workers and part-time workers. Both dependent variables are dichotomous (0/1). It is worth mentioning that data on disabled workers reflects self-reported disability from work, meaning that the NSDUH does not request additional documentation to substantiate this status.

Table II.

Selected sociodemographic characteristics by nativity, adults ages 18–64 US

Characterstic	US-born % (se)	Foreign-born % (se)
Population distribution
Mental illness (MI), Past 12 months	9.67 (0.27)	6.19** (0.57)
Mania delusions (MD), Past 12 months	6.69 (0.23)	7.14 (0.70)
Gender
Male	51.26 (0.48)	56.14** (1.34)
Female	48.74 (0.048)	43.86** (1.34)
Race/ethnicity
Non-Latino White	77.65 (0.49)	19.36** (1.14)
Non-Latino African American	12.67 (0.43)	6.58** (0.74)
Latino	6.54 (0.28)	50.38** (1.59)
Other (AIAN/PI/multiple race)	3.14 (0.18)	23.67** (1.36)
Age
18–34	34.60 (0.46)	41.48** (1.25)
35–49	37.64 (0.48)	37.99 (1.39)
50–64	27.76 (0.59)	20.53** (1.33)
Marital status
Married	55.26 (0.59)	62.19** (1.21)
Not married	44.79 (0.59)	37.81** (1.21)
Educational attainment
< 12th grade	11.88 (0.34)	30.00** (1.42)
12th grade graduate	33.94 (0.48)	23.08** (1.32)
College+	54.18 (0.54)	46.92** (1.41)
Survey language
English	99.74 (0.08)	69.88** (1.57)
Spanish	0.26 (0.08)	30.12** (1.57)
Self-rated health
Excellent/very good/good	90.71 (0.32)	88.49* (1.03)
Fair/poor	9.29 (0.32)	11.51* (1.03)
Illicit drug abuser	1.12 (0.10)	0.78 (0.23)
Number of children in household
None	56.31 (0.55)	44.94** (1.46)
One or more	43.69 (0.55)	55.06** (146)
SSI beneficiary	2.67 (0.17)	1.34** (0.30)
MSA status
MSA ≥1 million people	41.49 (0.61)	70.60** (1.45)
MSA <1 million people	58.51 (0.61)	29.40** (1.45)
Work status
Full-time (FT) work	66.48 (0.50)	70.24** (1.34)
Part-time (PT) work	13.52 (0.33)	11.88 (0.89)
Other work	5.43 (0.23)	4.43 (0.53)
Unemployed/looking for work	3.46 (0.16)	5.38** (0.67)
Disabled	4.46 (0.22)	2.74** (0.50)
Retired	4.06 (0.28)	2.03** (0.56)
No job, other reason	2.57 (0.17)	3.32 (0.56)

Data source: Original tabulations of data from the 2002 National Survey on Drug Use and Health (NSDUH). Students and homemakers are excluded. Sample sizes: Unweighted US-born N = 26 843; Weighted US-born N = 138.6 million; Unweighted foreign-born N = 3380; Weighted foreign-born N = 22.5 million. All estimates are based on weighted data. Tests of significance compare US and foreign-born adults; US-born adults are the reference group.

^*p<0.05

^**p≤0.01.

3.4. Independent variables

Table II provides detailed descriptions of the independent variables and also summarizes characteristics of the study population. Respondents’ nativity is determined by an individual’s answer to the question ‘Were you born in the United States?’ Unfortunately, we cannot further distinguish among groups of immigrants since the NSDUH does not collect data on citizenship. Multivariate analyses also include the following demographic characteristics: age and self-reported race/ethnicity. Labor market behaviors may also be shaped by respondent’s marital status or the number of dependent children younger than18. Individuals with disabilities may opt out of the labor market if income transfer programs, such as the Social Security Disability Income (SSDI) or the Supplemental Security Income (SSI), are available. Both programs provide economic assistance to disabled persons; however, SSI is a means-tested program, whereas SSDI is provided to persons who have met other Social Security requirements (e.g. minimum number of quarters worked and payment of Social Security taxes). We could not include a separate indicator for SSDI as it was unavailable in the survey; staff at SAMHSA could not elaborate on whether the SSI variable also included SSDI income (Personal Discussion with Pradip Muhuri, PhD, 7/10/06). We have nevertheless included an indicator for SSI to account for access to income transfer programs.

Human capital characteristics included educational attainment, respondent’s preferred language for participation in the survey (i.e. English or Spanish), self-rated health status, and substance abuse status. A contextual measure of the respondent’s community was included and measured by the region’s metropolitan statistical area (MSA) status, which indicates the number of persons residing in the respondent’s home region.

3.5. Instrumental variables

As discussed above, use of the K6 scale as an indicator for MI introduces the potential for reverse causation from labor market outcomes to MH status. There is also a risk that there exists some unobserved factor affecting both MI and labor market outcomes. A common approach for addressing the resulting endogeneity is IV analysis. IV requires the use of an instrument: a covariate that is predictive of the potentially endogenous variable (in this case, MI) but that is otherwise independent of the dependent variable of interest (in this case, labor market outcome). Bhattacharya et al. (2006) demonstrate that the bivariate probit is an appropriate method for estimating an IV model when both the dependent and the potentially endogenous variable of interest are binary.

Chatterji et al. (2007) review the literature on IV analysis applied to estimating the effect of MI on labor market outcomes. Instruments in their empirical models are the presence of childhood psychiatric disorders and two measures of religiosity as proxies for social capital. Previous investigations have also implemented measures of social support as identifying variables (Hamilton et al., 1997; Alexandre and French, 2001). Following this literature, we employ four identifying instruments that we believe are proxies for social capital – one measure of social support provided by friends and three measures of spousal support: (1) How many friends do you have that you spend time with on shared interests and activities (0–3 vs 4+); (2) How many times during the past 12 months were you and your spouse or partner angry with each other (0–2 vs a few times/no spouse vs many times); (3) How often is your spouse or partner critical of you (always vs sometimes/no spouse vs seldom/never); (4) How often does your spouse or partner show concern for your feelings and problems (always/sometimes/no spouse vs seldom/never). Following Chatterji et al. (2007), we also investigated religiosity, but did not find it to be predictive of MI in our sample.

As with Chatterji et al. (2007), a limitation of this approach is that our instruments are personal characteristics. In addition, we have some residual concern that our instruments may not be exogenous. Although epidemiological studies have demonstrated a positive association between social capital and MH (McKenzie et al., 2002), there remains the possibility that our measures of social capital, focusing on support of spouses and friends, may be affected by MI. Therefore, we include another IV specification where we employ the questionnaire items for mania and delusion symptoms as instruments for the K6 scale. This approach requires the assumption that these variables are only related to employment through their association with the K6 MI indicator. We investigate the exogeneity of each of our instruments using tests of overidentifying restrictions.

4. RESULTS

4.1. Sociodemographic profile

Table II summarizes the sociodemographic characteristics of the study population, stratified by nativity. Immigrants and natives differ in several ways. Foreign-born adults are less likely to meet criteria for MI than US-born adults (6.19% vs 9.67%, respectively) using the K6 scale. However, foreign-born and US-born adults are equally likely to report symptoms of mania/delusions. Immigrants are more likely to be males. Differences in the racial/ethnic distribution by nativity are evident; US-born adults are more likely to be non-Latino white (70.65%), whereas immigrants are more likely to be Latino (50.38%), followed by other groups (23.67%), and non-Latino whites (19.36%). As compared with US-born adults, immigrants are more likely to be ages 18–34, less likely to be ages 50–64, and more likely to be married (62.19%) and to have dependent children (55.06%). Educational attainment differs by nativity: immigrants are more likely than natives to have completed fewer than 12 years of schooling (30.00% vs 11.88%, respectively), less likely have college training (46.92% vs 54.18%, respectively), and are more likely to have completed the survey in Spanish (30.12%).

Assessments of health status varied by nativity, with a greater proportion of immigrants reporting fair/poor health (11.51% vs 9.29%). Immigrants are less likely than US-born adults to receive SSI payments (1.34% vs 2.67%). Immigrants are significantly more likely to reside in MSAs with at least one million persons (70.60% vs 41.49% of natives). Several important differences in labor market activities are observed. Overall, immigrants are more likely than US-born adults to report working full-time (70.24 vs 66.48%, respectively) or to be unemployed (5.38 vs 3.46%, respectively), and less likely to report being disabled (2.74 vs 4.46%, respectively) or retired (2.03 vs 4.06%, respectively).

4.2. Labor market activities

Table III reports on unadjusted work activities during the week prior to administration of the survey. We tested differences between immigrants with MI/MD against all other groups within each work category, while a second set of tests compared natives with MI/MD against all other groups within each work category. Significantly, immigrants with MI are more likely than US-born adults with MI to work full-time (64.16 vs 52.17%, respectively). Similarly, immigrants with MD are more likely than US-born adults with MD to work full-time (60.87 vs 49.48%, respectively). Furthermore, immigrants with MI do not differ in their full-time work activities as compared with natives or immigrants without MI, while immigrants with MD do not differ in their full-time work activities as compared with natives without MD; immigrants with MD are less likely to work than immigrants without MD (60.87 vs 70.96%). Immigrants with MI are substantially less likely than US-born adults with MI to report being disabled (1.66 vs 13.18% for MI; and 5.73 vs 16.23% for MD). US-born adults with MI are less likely to work full-time than adults without MI, regardless of their birthplace. Natives with MI are more likely to report being disabled than US-born adults without MI (13.18 vs 3.53% for MI; and 16.23 vs 3.62% for MD).

Table III.

Unadjusted estimates of work activities during previous week, distributed by nativity and mental health status, ages 18–64, US

	Work activities, Previous week
	Full-time (FT) work % (se)	Part-time (PT) work % (se)	Other work % (se)	Unemployed/ looking for work % (se)	Disabled % (se)	Retired % (se)	No Job, other reason % (se)
Mental illness (MI) by nativity
MI, immigrant	64.16B (4.59)	15.83 (3.05)	7.50 (3.55)	3.82 (1.09)	1.66B (0.80)	NA	5.36 (2.06)
MI, US-born	52.17A (1.51)	15.21 (0.91)	5.71 (0.62)	6.50 (0.77)	13.18A (1.14)	2.96 (0.79)	4.27 (0.62)
No MI, immigrant	70.64B (1.43)	11.62B (0.94)	4.20 (0.511)	5.48 (0.71)	2.81B (0.53)	2.05 (0.59)	3.18 (0.59)
No MI, US-born	68.01B (0.52)	13.34 (0.35)	5.40 (0.24)	3.14B (0.16)	3.53A,B (0.22)	4.18 (0.29)	2.39A,B (0.17)
Mania delusions (MD) by nativity
MD, immigrant	60.87B (4.91)	11.94 (2.54)	6.14 (3.18)	7.52 (1.99)	5.73B (2.57)	NA	3.35 (0.94)
MD, US-born	49.48A (1.75)	14.55 (1.11)	5.91 (0.76)	7.10 (0.84)	16.23A (1.56)	2.52 (0.85)	4.20 (0.63)
No MD, immigrant	70.96A,B (1.35)	11.88 (9.21)	4.27 (0.50)	5.22 (0.66)	2.51B (0.50)	1.85 (0.56)	3.32 (0.58)
No MD, US-born	67.70B (0.49)	13.45 (0.33)	5.40 (0.23)	3.20A (0.15)	3.62B (0.21)	4.17 (0.28)	2.46B (0.16)

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). NA refers to cells where the unweighted sample was less than 5. Sample sizes: Unweighted US-born N = 26 843; Weighted US-born N = 138.6 million. Unweighted foreign-born N = 3380; Weighted foreign-born N = 22.5 million.

^ATest of significance p≤0.05 comparing immigrant with MI/MD to all others, within each work status category (i.e. within FT, etc.).

^BTest of significance p≤0.05 comparing native with MI/MD to all others, within each work status category (i.e. within FT, etc.).

4.3. Multivariate analyses

4.3.1. MH and nativity effects on any work activities.

We further tested the relationship observed in unadjusted analyses between nativity, MH, and work through the specification of multivariate statistical models, separately for men and women. Table IV presents results of multivariate models examining predictors of any work for pay (i.e. full-time, part-time or other work vs unemployed, retired, or disabled) as a function of MI by nativity controlling for the full set of independent variables. Significantly, US-born adults with MI are less likely to be employed than US-born adults without MI; this is true equally for men and women. Notably, in this specification, immigrants with MI are equally likely to work as healthy immigrants and natives. Other factors associated with lower rates of men’s work include being African American or from another racial/ethnic group. For both men and women, being 50 years or older, being unmarried, having completed through a high school education, or being in fair/poor health are all associated with lower rates of work. Factors associated with higher rates of work for men and women include having dependent children and for males, preferring Spanish. Even after excluding SSI beneficiaries, that is, beneficiaries of a cash transfer program, the relationship between nativity and MI and any work was unchanged (Appendix Table AI).

Table IV.

Estimated logistic regression coefficients predicting any work activities by mental illness and nativity among men and women ages 18–64, excluding students and homemakers, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mental illness (MI) by nativity
No MI, US-born	REF	—	—	REF	—	—
No MI, immigrant	−0.01	0.18	0.97	−0.06	0.19	0.76
MI, US-born	−0.71	0.13	0.00	−0.57	0.10	0.00
MI, immigrant	0.92	0.53	0.08	0.36	0.46	0.44
Race/ethnicity
Non-Latino White	REF	—	—	REF	—	—
Non-Latino African American	−0.47	0.13	0.00	−0.20	0.11	0.08
All other race/ethnic groups	−0.47	0.21	0.02	−0.18	0.21	0.38
Latinos	−0.03	0.17	0.84	0.03	0.19	0.86
Age
17–34	0.04	0.09	0.62	0.01	0.09	0.88
35–49	REF	—	—	REF	—	—
50–64	−1.02	0.12	0.00	−1.14	0.11	0.00
Marital status
Married	REF	—	—	REF	—	—
Not married	−0.66	0.11	0.00	−0.24	0.09	0.01
Children < 18 years in household
No children <18 years	REF	—	—	REF	—	—
1 or more Children	0.28	0.10	0.01	0.32	0.09	0.00
Educational attainment
<12 grade	−1.21	0.12	0.00	−1.11	0.12	0.00
High school graduate	−0.55	0.10	0.00	−0.60	0.10	0.00
College+	REF	—	—	REF	—	—
Survey language
English	REF	—	—	REF	—	—
Spanish	1.01	0.33	0.00	−0.53	0.32	0.09
Self-rated health
Excellent/very good/good	REF	—	—	REF	—	—
Fair/poor	−1.44	0.12	0.00	−1.40	0.11	0.00
Substance abuse
Yes	0.16	0.32	0.61	−0.24	0.52	0.64
No	REF	—	—	REF	—	—
MSA size
MSA 1million +people	REF	—	—	REF	—	—
MSA <1 million people	0.03	0.09	0.78	−0.01	0.09	0.92
Constant	3.55	0.61	0.00	2.29	1.03	0.03

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Any work is defined as full-time, part-time or other/volunteer work. The reference group is adults who are unemployed, retired, or disabled.

Table V presents results of multivariate models predicting any work for pay using the mania and delusions (MD) specification of MH. The results from this regression are mostly unchanged from results of analyses using the MI (K6) indicator, with the notable exception that immigrant men with MD are similarly less likely to work as compared with native men with MD. We tested an additional specification using a combination of the MI and MD indicators (data not shown). These results were similar to the two previous specifications, with the point estimate for immigrant men with MI or MD falling between the two previous estimates (−0.035, SE = 0.21) and not statistically significantly different from zero (p = 0.060).

Table V.

Estimated logistic regression coefficients predicting any work activities by mania/delusions and nativity among men and women ages 18–64, excluding students and homemakers, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mania delusions (MD) by nativity
No MD, US-born	REF	—	—	REF	—	—
No MD, immigrant	0.15	0.19	0.42	−0.06	0.28	0.78
MD, US-born	−0.77	0.14	0.00	−0.79	0.13	0.00
MD, immigrant	−0.76	0.33	0.02	0.41	0.54	0.44
Race/ethnicity
Non-Latino White	REF	—	—	REF	—	—
Non-Latino African American	−0.43	0.13	0.01	−0.14	0.12	0.22
All other race/ethnic groups	−0.48	0.20	0.02	−0.19	0.24	0.44
Latinos	−0.04	0.16	0.83	0.05	0.20	0.86
Age
17–34	0.06	0.10	0.51	0.03	0.09	0.78
35–49	REF	—	—	REF	—	—
50–64	−1.02	0.13	0.00	−1.13	0.11	0.00
Marital status
Married	REF	—	—	REF	—	—
Not married	−0.65	0.11	0.00	−0.25	0.09	0.01
Children < 18 years in household
No children < 18 years	REF	—	—	REF	—	—
1 or more children	0.29	0.10	0.01	0.32	0.09	0.00
Educational attainment
<12 grade	−1.18	0.12	0.00	−1.10	0.12	0.00
High school graduate	−0.55	0.11	0.00	−0.62	0.10	0.00
College+	REF	—	—	REF	—	—
Survey language
English	REF	—	—	REF	—	—
Spanish	0.94	0.32	0.00	−0.53	0.31	0.09
Self-rated health
Excellent/very good/good	REF	—	—	REF	—	—
Fair/poor	−1.47	0.12	0.00	−1.42	0.12	0.00
Substance abuse
Yes	−.018	0.28	0.53	−0.29	0.52	0.57
No	REF	—	—	REF	—	—
MSA size
MSA 1million +people	REF	—	—	REF	—	—
MSA < 1 million people	0.04	0.09	0.68	−0.01	0.09	0.90
Constant	3.55	0.60	0.000	2.18	1.04	0.04

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Any work is defined as full-time, part-time or other/volunteer work. The reference group is adults who are unemployed, retired, or disabled.

We present predicted probabilities of reporting any work by MI and nativity for men and women (based on models estimated in Tables IV and V) standardized to the underlying population characteristics in Table VI. For example, the standardized predicted probability of any work among native-born males without MI was calculated as the mean predicted probability of work across all males as if each was native-born without MI. Results based on the MI specification indicate that mentally ill US-born men were significantly less likely than all other men to be employed, while mentally ill male immigrants were more likely to be employed than all other men. Among women, mentally ill US-born women were less likely to be employed than women without MI; however, mentally ill immigrant women were more likely to be employed than mentally ill native-born women. Results based on the MD specification differ for men. Using self-reported symptoms of MD as a proxy, both mentally ill US-born and immigrant males were less likely to work than non-mentally ill males. Results for women were unchanged.

Table VI.

Predicted probabilities of any work ages 18–64, Excluding students and homemakers, US, 2002

	MEN			WOMEN
	Predicted probability	SE	Pair-wise tests	Predicted probability	SE	Pair-wise tests
Mental illness by nativity
No MI, US-born	0.88	0.01		0.84	0.01
No MI, immigrant	0.88	0.02		0.84	0.02
MI, US-born	0.80	0.02	A, B	0.77	0.02	A, B
MI, immigrant	0.94	0.03	A, B, C	0.88	0.04	C
Mania/delusions by nativity
No MD, US-born	0.89	0.01		0.87	0.01
No MD, immigrant	0.91	0.01		0.87	0.02
MD, US-born	0.81	0.02	A, B	0.77	0.02	A, B
MD, immigrant	0.81	0.01	A, B	0.91	0.04	C

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Any work is defined as full-time, part-time or other/volunteer work. Models also control for race/ethnicity, age, marital status, dependent children, educational attainment, survey language, self-rated health, substance abuse status, and metropolitan statistical area size (MSA). A: p<0.05 comparing No MI/MD, US-born vs other categories. B: p<0.05 comparing No MI/MD, Immigrant vs other categories. C: p<0.05 comparing MI/MD US-born vs MI/MD Immigrant.

4.3.2. MH and nativity effects on full-time work activities.

We next examined predictors of full-time work in the population of workers, controlling for all other sociodemographic, health status, and geographic covariates; the reference group includes adults working part-time and employed in other activities. Results from the MI specification are presented in Table VII, with results from the MD specification in Table VIII. No differences in full-time work are observed by MI and nativity for either men or women (Table VII). However, in the MD specification, both US-born men and women with MD were less likely to work full-time than US-born men or women without MD (Table VIII). No significant differences were found for FT work between immigrant men and women with vs without MD. The third specification of MI as MI or MD found native men (but not native women) with MI to be significantly less likely to work full-time (data not shown).

Table VII.

Estimated logistic regression coefficients predicting full-time work by mental illness and nativity among workers ages 18–64 excluding students and homemakers, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mental illness (MI) by nativity
No MI, US-born	REF	—	—	REF	—	—
No MI, immigrant	0.20	0.15	0.18	0.00	0.13	0.97
MI, US-born	−0.17	0.14	0.21	−0.06	0.09	0.50
MI, immigrant	0.18	0.41	0.65	−0.15	0.29	0.59
Race/ethnicity
Non-Latino White	REF	—	—	REF	—	—
Non-Latino African American	−0.09	0.11	0.41	0.46	0.11	0.00
All other race/ethnic groups	−0.23	0.17	0.16	0.23	0.15	0.13
Latinos	−0.10	0.15	0.51	0.36	0.12	0.00
Age
17–34	−0.70	0.09	0.00	−0.61	0.07	0.00
35–49	REF	—	—	REF	—	—
50+	−0.75	0.14	0.00	−0.34	0.11	0.00
Marital status
Married	REF	—	—	REF	—	—
Not married	−0.93	0.10	0.00	0.25	0.07	0.00
Children < 18 years in household
No children < 18 years	REF	—	—	REF	—	—
1 or more children	0.15	0.09	0.11	−0.40	0.07	0.00
Educational attainment
<12 grade	−0.28	0.11	0.01	−0.15	0.10	0.13
High school graduate	0.14	0.09	0.10	−0.02	0.07	0.78
College+	REF	—	—	REF	—	—
Survey language
English	REF	—	—	REF	—	—
Spanish	0.82	0.32	0.01	−0.21	0.23	0.36
Self-rated health
Excellent/very good/good	REF	—	—	REF	—	—
Fair/poor	−0.29	0.18	0.11	−0.54	0.14	0.00
Substance abuse
Yes	−0.46	0.22	0.03	−0.86	0.43	0.04
No	REF	—	—	REF	—	—
MSA size
MSA 1million +people	REF	—	—	REF	—	—
MSA < 1 million people	−0.06	0.08	0.51	0.03	0.07	0.62
Constant	1.82	0.48	0.00	−.053	0.76	0.48

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Workers are defined as full-time, part-time or those doing any other/volunteer work. The reference group is adults who are working part-time or are doing other/volunteer work.

Table VIII.

Estimated logistic regression coefficients predicting full-time work by mania/delusions and nativity among workers ages 18–64 excluding students and homemakers, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mania delusions (MD) by nativity
No MD, US-born	REF	—	—	REF	—	—
No MD, immigrant	0.22	0.17	0.20	0.03	0.13	0.84
MD, US-born	−0.28	0.14	0.04	−0.27	0.13	0.04
MD, immigrant	0.08	0.34	0.81	−0.04	0.37	0.91
Race/ethnicity
Non-Latino White	REF	—	—	REF	—	—
Non-Latino African American	−0.08	0.12	0.53	0.47	0.10	0.00
All other race/ethnic groups	−0.23	0.19	0.23	0.23	0.14	0.11
Latinos	−0.10	0.15	0.50	0.37	0.12	0.00
Age
17–34	−0.70	0.09	0.00	−0.61	0.07	0.00
35–49	REF	—	—	REF	—	—
50+	−0.75	0.14	0.00	−0.34	0.11	0.00
Marital status
Married	REF	—	—	REF	—	—
Not married	−0.93	0.09	0.00	0.26	0.07	0.00
Children < 18 years in household
No children < 18 years	REF	—	—	REF	—	—
1 or more children	0.15	0.09	0.09	−0.41	0.07	0.00
Educational attainment
<12 grade	−0.27	0.12	0.02	−0.15	0.10	0.15
High school graduate	0.14	0.09	0.11	−0.02	0.07	0.75
College+	REF	—	—	REF	—	—
Survey language
English	REF	—	—	REF	—	—
Spanish	0.81	0.33	0.02	−0.21	0.27	0.43
Self-rated health
Excellent/very good/good	REF	—	—	REF	—	—
Fair/poor	−0.29	0.19	0.12	−0.54	0.14	0.00
Substance abuse
Yes	−0.45	0.24	0.06	−0.87	0.38	0.02
No	REF	—	—	REF	—	—
MSA size
MSA 1million +people	REF	—	—	REF	—	—
MSA < 1 million people	−0.05	0.08	0.52	0.03	0.07	0.62
Constant	1.82	0.50	0.00	−0.54	0.76	0.48

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Workers are defined as full-time, part-time or those doing any other/volunteer work. The reference group is adults who are working part-time or are doing other/volunteer work.

Other factors associated with lower rates of full-time work for men and women include being young (ages 17–34) or older (50–64 years) and meeting criteria for being a substance abuser. Gender-based differences in full-time work are observed for other characteristics. For example, men’s lower rates of full-time work are associated with being unmarried or having completed less than a high school education; conversely, completing a Spanish language survey is positively associated with full-time work. Women’s employment in full-time work is positively associated with being African American or Latina and being unmarried. Lower rates for women are associated with having dependent children or being in fair/poor health. The relationship between nativity/MI and full-time work was unchanged in models excluding SSI beneficiaries (Appendix Table AII).

The predicted probabilities of reporting full-time work only for men and women are shown in Table IX. Results based on the MI specification show no significant differences among subgroups by MH status and nativity, for either men or women. Results based on the MD specification show that both male and female US-born workers with MI are less likely to work full-time than their US-born peers without MI.

Table IX.

Predicted probabilities of full-time work, workers ages 18–64 excluding students and homemakers, US, 2002

	MEN			WOMEN
	Predicted probability	SE	Pair-wise tests	Predicted probability	SE	Pair-wise tests
Mental illness (MI) by nativity
No MI, US-born	0.86	0.01		0.70	0.01
No MI, immigrant	0.88	0.01		0.70	0.02
MI, US-born	0.83	0.02		0.69	0.02
MI, immigrant	0.88	0.04		0.67	0.06
Mania/delusions (MD) by nativity
No MD, US-born	0.82	0.01		0.68	0.01
No MD, immigrant	0.85	0.02		0.68	0.03
MD, US-born	0.78	0.02	A, B	0.62	0.03	A
MD, immigrant	0.81	0.05		0.67	0.08

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Workers are defined as full-time, part-time or those doing any other/volunteer work. Models also control for race/ethnicity, age, marital status, dependent children, educational attainment, survey language, self-rated health, substance abuse status, and metropolitan statistical area size (MSA). A: p<0.05 comparing No MI, US-born vs other categories. B: p<0.05 comparing No MI, Immigrant vs other categories. C: p<0.05 comparing MI US-born vs MI immigrant.

4.3.3. IVAnalysis of MH and nativity effects on any work activities.

We present IV estimates of predicted probabilities of any work calculated using the bivariate probit in Table X. As our instrument set applied to MI (not MI interacted with nativity), we respecified the models to include both genders and stratified by nativity. We present three specifications of bivariate probit models: the first specification includes measures of social support (SS) as instruments, the second model includes self-reported symptoms of mania and delusions (MD) as instruments, and the third model includes both measures of social support and mania/delusions. We present results from a standard probit model for comparison.

Table X.

IV Analysis: predicted probabilities of any work ages 18–64, excluding students and homemakers, US, 2002

	US-born			Immigrant
	Predicted probability	SE	Pair-wise tests	Predicted probability	SE	Pair-wise tests
Mental illness
Probit model (standard)
No MI	0.89	0.01	A	0.88	0.01
MI	0.82	0.01		0.91	0.02
Bivariate probit model 1
No MI	0.89	0.01		0.86	0.02
MI	0.77	0.07		0.93	0.09
Bivariate probit model 2
No MI	0.90	0.01	A	0.88	0.01
MI	0.63	0.04		0.81	0.10
Bivariate probit model 3
No MI	0.90	0.01	A	0.88	0.01
MI	0.65	0.04		0.85	0.10

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Any work is defined as full-time, part-time or other/volunteer work. Models also control for race/ethnicity, age, marital status, dependent children, educational attainment, survey language, self-rated health, substance abuse status, and metropolitan statistical area size (MSA). Instruments included in models were: Model 1: How many friends do you have that you spend time with on shared interests and activities; How many times during the past 12 months were you and your spouse or partner angry with each other; How often is your spouse or partner critical of you; How often does your spouse or partner show concern for your feelings and problems. Note: the selection effect was non-significant in both US-born and immigrant models. Model 2: During the past 12 months: have you heard voices – that is, voices that other people said did not exist, voices coming from inside your head, or voices coming out of the air when there was no one around? have you felt that a force was taking over your mind and trying to make you do things you didn’t want to do? have you felt that some force was inserting thoughts directly into your head by means of x-rays or laser beams or other methods? have you felt that your own thoughts were being stolen out of your mind by someone or something you did not have control over? have you believed that there was an unfair plot going on to harm you or to have people follow you - when your family and friends did not believe that this was happening? have you seen a vision – that is, something that other people could not see? were there at least four days in a row when you were so excited or hyper that you either got into trouble or people worried about your being so excited, or a doctor said you were manic? Note: the selection effect was significant in the US-born model and non-significant in the immigrant model. Model 3: Includes instrument sets from both models 1 and 2. A: p<0.05 comparing NoMI to MI.

The IVs were highly predictive of MI with F statistics of 93 and 295 for SS and MD, respectively. We employed a modified test of overidentifying restrictions for the bivariate probit by sequentially excluding each IV from the set of IVs and including them in the set of independent variables predicting either any work or full-time work. Although each IV proved to be otherwise unassociated with any work, all of our potential IVs were related to the decision between full-time and part-time work among workers. Thus, we present IV estimates for the model predicting any work only.

The standard probit model shows results similar to the previous analysis of MI among males: MI reduces the probability of work among those who are native-born, but MI does not affect work status among immigrants. The first bivariate probit specification (SS) showed no significant differences in work status between US-born and immigrants with MI compared with those without MI. However, the selection effect was not statistically significant in either model, and the results from the IV analysis are largely similar to the standard analysis, although the standard errors of the MI estimates are larger. The second bivariate probit specification (MD) shows significant declines in the probability of work among US-born with MI, and no difference among immigrants with MI. The selection effect was statistically significant in the US-born model but not the immigrant model. The standard error of the probability of work is much smaller in this specification of US-born model. Results from the third specification (SS+MD) are nearly identical to the MD specification. Overall, the results from this analysis support our earlier finding that MI negatively impacts labor market participation for US-born non-elderly adults only.

5. DISCUSSION

This study uses recent nationally representative data that, with large numbers of immigrants in the sample, allows us to assess the effects of MI on nativity groups by gender. We analyzed two indicators of MI, thereby providing a more nuanced view of the effects of MI on labor supply. In sensitivity analyses, we controlled for a possible selection bias based on self-reported disability and access to cash transfer programs by estimating models that excluded beneficiaries of the federal SSI program. Lastly, we build on other studies (Ettner, 2000; Marcotte et al., 2000; Chatterji et al., 2007) that have applied an IV approach in order to account for possible endogeneity between MH status and labor market outcomes. As with Chatterji (Chatterji et al., 2007) and Ettner (Ettner, 2000), we did not observe significant gains in the estimation of models using an IV approach. We concur with Ettner (Ettner, 2000) that using health status indicators may be sufficient. In separate analyses, we also tested the use of the K6 as a continuous indicator. Our results were qualitatively unchanged suggesting that a dichotomous variable is sufficient to detect the effect of MI on work. The K6 is not intended to measure severity of illness, thus limiting us from commenting further on the relationship between severity and labor market outcomes.

Our findings corroborate results from earlier studies, affirming that MH affects labor supply, but not uniformly across immigration and gender groups. Using the K6 scale to assess MI, a classification that reflects anxiety or mood disorders or non-affective psychosis, we found that male and female immigrants with MI are just as likely as healthy adults to be active in the labor market, while US-born adults were significantly less likely to be employed. This finding was supported in both multivariate and IV analyses. In contrast, when using questions that assess symptoms of mania and delusions, we found that both US-born and immigrant males with MD were less likely to work than US-born and immigrant males without MD. With respect to full-time vs part-time work among workers, we found no effect of MI, but a negative effect of MD among US-born males and females.

The inconsistent effects of MI for part-time vs full-time work suggest that our findings may be affected by the type of work considered by native vs foreign-born persons. It may be that these groups face different options with respect to work. The choice set for natives may be more dependent on MH than the jobs being considered by immigrants, for whom the labor market opportunities are mediated more immediately by their immigration status (for which we do not have a measure since it is not collected by the NSDUH). Alternatively, it may be that immigrants with MI have fewer alternatives for income replacement and therefore have to work, despite poor MH, leading to further negative consequences.

Our differences in findings between the MI and MD specifications likely reflect the fact that we are measuring different aspects of MI, and that these aspects have differential effects on work by nativity. International rates of depression and anxiety vary across countries and cultures (Demyttenaere et al., 2004). These disorders are embedded in the environment and cultural surroundings (Salgado de Snyder et al., 2000). Therefore, the impact of MI on labor market outcomes may be mediated by the person’s response to their illness. Psychiatric conditions with symptoms of mania and delusions, such as schizophrenia and bipolar disorder, are less likely to be affected by concurrent life events; the fact that someone is laid off is not likely to cause them to hallucinate. As MI and MD measure different aspects of MI, it is not surprising that they have different impacts among immigrants than among US-born populations. One possible interpretation is that immigrants are more robust, at least in their labor supply response to symptoms of depression and anxiety than US-born workers. Mania and delusions have a similar deleterious effect on the probability of employment among US born and immigrant males, perhaps because the conditions represented by these symptoms are more severe and are less likely to be mediated by culture.

Conceptualization of MI is understood to vary by cultural background and nativity. Somatization, or the presentation of physical symptoms in response to stressors, has been identified in foreign-born populations residing in the US and abroad, while labeling of physical or emotional symptoms may also vary among ethnic subgroups (Salgado de Snyder et al., 2000; Leong and Lau, 2001; Vega et al., 2006). Thus, majority culture-based community surveys, such as the NSDUH, may under-estimate psychiatric disorders in culturally diverse populations, especially if symptoms experienced by immigrants have a cultural manifestation.

It is not entirely clear what resources immigrants draw upon to mitigate post-migration mental conditions. One small convenience and community-based study found that stress-coping strategies differ among Mexican-origin immigrants, their native-born peers, and non-Latino whites. Compared with other groups, Mexican immigrants most frequently used positive reframing, denial, and religion as coping strategies, whereas whites employed substance abuse more often (Farley et al., 2005). Additional research could elucidate immigrants’ conceptualizations of the source or symptoms of psychological distress as well as the perceived impact of distress on life domains, including work.

The NSDUH data provide another opportunity to examine the relationship between MH and labor supply. However, the outcomes observed here may be shaped by unmeasured factors. The NSDUH is neither a labor market or immigration/immigrant survey. Consequently, some of our comparisons were limited by a smaller sample of immigrants relative to the number of US-born persons in the study. In addition, it is unknown whether immigrants surveyed by the NSDUH are separated from their families or whether having a spouse or dependents in the sending country motivates work activities despite poor MH. Prior literature has demonstrated an important relationship between physical health and MH and labor market outcomes. Our data revealed important differences in self-reported disability by nativity. Yet, lack of data on comorbid physical conditions or activity limitations due to physical or emotional conditions in the NSDUH limited us from determining more precisely the impact of comorbid physical health and chronic medical conditions (with psychiatric conditions) on work outcomes. Similarly, the lack of identifying data (due to the sensitive nature of the topics addressed by the survey) did not permit us to link the NSDUH data with other databases that may have allowed us to examine the influence of other comorbid conditions. The current survey is lengthy, nevertheless, the SAMHSA may consider including additional items on physical health status in future versions of the survey instrument; such items would enhance studies such as this by addressing the issue of comorbidities.

We conducted sensitivity analyses in an effort to account for selection bias in the sample and incentives to withdraw from the labor market based on the availability of cash transfers, we estimated models that excluded any persons receiving SSI. Our results were qualitatively unchanged, suggesting that other factors beyond selection or income substitution are accountable for the observed differences in labor market participation between immigrants and natives. The NSDUH lacks structured diagnostic interviews that would permit identification of the type and severity of MI. Thus, we are unable to test whether immigrants and natives differ on type or severity of their disorder, and doing so would certainly inform our understanding of the nature of psychiatric distress among population subgroups in relation to their labor market activities. Nevertheless, we did control for substance abuse, which could serve as another indicator of illness severity.

Possible explanations for our findings include another form of sample selection bias among immigrants. The ‘Hispanic paradox’ indicates that immigrants display favorable health and mortality outcomes relative to non-immigrants despite poor socioeconomic profiles (Abraido-Lanza et al., 1999; Franzini et al., 2002; Crimmins et al., 2007; Alegria et al., 2008). Immigration of the healthiest individuals may partially contribute to these outcomes (Franzini et al., 2002). Alegría and colleagues (Alegria et al., 2008) recently examined the prevalence of MI in Latino populations by nativity and Latino subgroup. Psychiatric illness was more prevalent among non-Latino whites as compared with Latinos and lifetime rates of MI were more prevalent among the US-born as compared with immigrant Latinos. Yet, the paradox was consistently observable among Mexicans, especially for depressive and anxiety disorders. Several possible interpretations were offered, including that the geographic proximity of Mexico, older age of migration, and greater numbers of Mexican-origin individuals, may maintain the culture and its social networks thus buffering against demoralization in a challenging post-migration environment.

One important limitation in testing hypotheses such as the ‘Hispanic paradox’ has been the lack of data collected in both sending and receiving countries. Studies such as the World Health Organization’s World Mental Health Surveys (Wang et al., 2007) may facilitate examination of the ‘healthy migrant’ theory, which proposes that the healthiest persons migrate. Finally, longitudinal studies in both sending and receiving communities are needed to fully understand the relationship between health status and migration decisions, including decisions by migrants to return to the country of origin (i.e. the salmon hypothesis) (Abraido-Lanza et al., 1999) since it has been proposed that estimates of illness among immigrants in the US are downwardly biased by ill persons who return to their country of origin and thus are not counted in US mortality and other statistics. These issues remain unresolved and require better data collected in multiple sites to more clearly understand the manner in which health impacts migration and labor market opportunities in sending and receiving communities.

5.1. Implications

Our results suggest that the impact of MI on overall work activities differs by nativity. The economic contributions of immigrants and natives with MH concerns are impressive, and suggestive of resiliency, though this cannot be assumed. A next step in this line of research is to examine the conditions under which immigrants with psychiatric symptoms work. Such research would consider variation by psychiatric disorder as well as other occupational characteristics (e.g. occupation, industry, firm size, etc.). At work, psychiatric illness may also result in spill-over effects for affected persons such as increased absenteeism or reduced functioning (Kessler and Frank, 1997; Kessler et al., 1999), and this merits further investigation as well.

MI often presents during a person’s early years. The National Comorbidity Survey Replication showed that three-quarters of lifetime cases of DSM-IV psychiatric disorders emerged by age 24 (Kessler et al., 2005), potentially disrupting opportunities to amass human and health capital. A longitudinal study is needed to assess the temporal course of illness vis-à-vis labor market outcomes, migration, or fluctuations in labor supply. Data on childhood family socioeconomic conditions, parents’ MH status, and family history of MI would permit further contextualization of the nature of MI.

An extensive body of literature has identified gaps in MH service utilization by nativity, gender, and race/ethnicity (Vega et al., 1999; Wells et al., 2001; Alegria et al., 2002; Ojeda and McGuire, 2006). Results from this investigation reinforce the need to preserve and improve MH. This remains a priority issue within the nation’s health policy agenda and is an issue of great importance for employers. Firms may examine the adequacy of institutional policies and programs pertaining to MH and health-care coverage and services covered as well as workplace wellness programs, keeping in mind the particular needs of the workforce (e.g. language, literacy skills, etc.). Lastly, the research community must make additional efforts to ensure that population-based instruments can more precisely assess MI and variations in its expression among ethnic subgroups; such data are critical for structuring programs that meet the needs of populations that have consistently been underserved and unserved.

APPENDIX

Logistic regression analyses predicting any work and full time work are given in Tables AI and AII. Bivariate probit analyses predicting any work is given in Table AIII.

Table AI.

Logistic regression analyses predicting any work, by gender, ages 18–64, excluding students/homemakers and supplemental security income beneficiaries, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mental illness by nativity
No MI, US-born	REF	0	—	REF	0	—
No MI, immigrant	−0.10	0.18	0.58	−0.09	0.20	0.66
MI, US-born	−0.75	0.13	0.00	−0.52	0.11	0.00
MI, immigrant	0.85	0.52	0.11	0.25	0.46	0.59
Race/ethnicity
Non-Latino White	REF	0	—	REF	0	—
Non-Latino African American	−0.45	0.14	0.00	−0.16	0.12	0.18
All other race/ethnic groups	−0.42	0.21	0.05	−0.13	0.21	0.54
Latinos	0.03	0.18	0.89	0.15	0.18	0.41
Age
18–34	−0.09	0.10	0.35	−0.07	0.10	0.48
35–49	REF	0	—	REF	0	—
50–64	−1.16	0.13	0.00	−1.17	0.11	0.00
Marital status
Married	REF	0	—	REF	0	—
Not married	−0.66	0.11	0.00	−0.09	0.10	0.33
Children < 18 years in household
No children < 18 years	REF	0	—	REF	0	—
1 or more Children	0.26	0.10	0.01	0.35	0.09	0.00
Educational attainment
<12 grade	−1.13	0.13	0.00	−1.05	0.13	0.00
High school graduate	−0.54	0.11	0.00	−0.53	0.10	0.00
College+	REF	0	—	REF	0	—
Survey language
English	REF	0	—	REF	0	—
Spanish	0.88	0.34	0.01	−0.68	0.33	0.04
Self-rated health
Excellent/very good/good		0	—	REF	0	—
Fair/poor	−1.23	0.14	0.00	−1.30	0.12	0.00
Substance abuse
Yes	0.09	0.33	0.78	−0.37	0.52	0.48
No	REF	0	—	REF	0	—
MSA size
MSA 1million +people	REF	0	—	REF	0	—
MSA < 1 million people	0.01	0.10	0.88	0.01	0.09	0.94

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Any work is defined as full-time, part-time or other/volunteer work. The reference group is adults who are unemployed, retired, or disabled.

Table AII.

Logistic regression analyses predicting full-time work only, by gender, ages 18–64, excluding students/homemakers and supplemental security income beneficiaries, US, 2002

	MEN			WOMEN
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
Mental illness by nativity
No MI, US-born	REF	0	—	REF	0	—
No MI, immigrant	0.19	0.16	0.24	−0.02	0.13	0.89
MI, US-born	−0.15	0.14	0.29	−0.04	0.10	0.65
MI, immigrant	0.16	0.40	0.69	−0.16	0.29	0.57
Race/ethnicity
Non-Latino White	REF	0	—	REF	0	—
Non-Latino African American	−0.05	0.12	0.64	0.46	0.11	0.00
All other race/ethnic groups	−0.22	0.17	0.20	0.22	0.15	0.14
Latinos	−0.09	0.15	0.55	0.36	0.12	0.00
Age
18–34	−0.72	0.09	0.00	−0.60	0.07	0.00
35–49	REF	0	—	REF	0	—
50–64	−0.75	0.15	0.00	−0.34	0.11	0.00
Marital status
Married	REF	0	—	REF	0	—
Not married	−0.95	0.10	0.00	0.25	0.07	0.00
Children < 18 years in household
No Children < 18 years	REF	0	—	REF	0	—
1 or more Children	0.14	0.09	0.14	−0.41	0.07	0.00
Educational attainment
<12 grade	−0.24	0.12	0.04	−0.11	0.10	0.30
High school graduate	0.14	0.09	0.12	−0.01	0.07	0.85
College+	REF	0	—	REF	0	—
Survey language
English	REF	0	—	REF	0	—
Spanish	0.77	0.33	0.02	−0.22	0.23	0.33
Self-rated health
Excellent/very good/good	REF	0	—	REF	0	—
Fair/poor	−0.33	0.18	0.07	−0.53	0.15	0.00
Substance abuse
Yes	−0.43	0.22	0.06	−0.87	0.43	0.04
No	REF	0	—	REF	0	—
MSA Size
MSA 1million +people	REF	0	—	REF	0	—
MSA < 1 million people	−0.09	0.08	0.25	0.02	0.07	0.74

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Workers are defined as full-time, part-time or those doing any other/volunteer work. The reference group is adults who are working part-time or are doing other/volunteer work.

Table AIII.

Bivariate probit analyses predicting any work, by US and immigrant status, ages 18–64, excluding students/homemakers, US, 2002

	US-born			Immigrant
	Est. beta coeff.	SE beta	p-Value T-test	Est. beta coeff.	SE beta	p-Value T-test
EQUATION 1: prob (any work)
Mental illness	−0.56	0.31	0.067	0.76	0.52	0.15
Male	0.11	0.04	0.01	0.39	0.10	0.00
Race/ethnicity
Non-Latino White	REF	0	—	REF	0	—
Non-Latino African American	−0.21	0.05	0.00	0.24	0.18	0.19
All other race/ethnic groups	−0.15	0.11	0.16	−0.06	0.15	0.67
Latinos	−0.08	0.07	0.28	0.24	0.16	0.13
Age
18–34	0.01	0.04	0.88	−0.10	0.11	0.38
35–49	REF	0	—	REF	0	—
50–64	−0.60	0.05	0.00	−0.60	0.16	0.00
Marital status
Married	REF	0	—	REF	0	—
Not married	−0.25	0.04	0.00	−0.16	0.11	0.12
Children < 18 years in household
No children < 18 years	REF	0	—	REF	0	—
1 or more children	0.19	0.04	0.00	0.09	0.10	0.37
Educational attainment
<12 grade	−0.68	0.05	0.00	−0.36	0.14	0.01
High school graduate	−0.32	0.04	0.00	−0.281	0.12	0.02
College+	REF	0	—	REF	0	—
Survey language
English	REF	0	—	REF	0	—
Spanish	−0.19	0.41	0.64	−0.12	0.15	0.44
Self-rated health
Excellent/very good/good	REF	0	—	REF	0	—
Fair/poor	−0.82	0.08	0.00	−0.59	0.14	0.00
Substance abuse
Yes	−0.03	0.15	0.85	0.21	0.44	0.64
No	REF	0	—	REF	0	—
MSA size
MSA 1million +people	REF	0	—	REF	0	—
MSA < 1 million people	0.00	0.04	0.99	0.02	0.09	0.84
EQUATION 2: prob(MI)
Male	−0.37	0.03	0.00	−0.30	0.11	0.01
Race/ethnicity
Non-Latino White	REF	0	—	REF	0	—
Non-Latino African American	−0.29	0.06	0.00	−0.17	0.18	0.36
All other race/ethnic groups	−0.05	0.09	0.63	−0.05	0.15	0.74
Latinos	−0.14	0.07	0.05	−0.40	0.15	0.01
Age
18–34	0.12	0.04	0.001	0.36	0.11	0.01
35–49	REF	0	—	REF	0	—
50–64	−0.31	0.06	0.00	−0.30	0.20	0.128
Marital status
Married	REF	0	—	REF	0	—
Not married	0.29	0.05	0.00	0.19	0.11	0.09
Children < 18 years in household
No children < 18 years	REF	0	—	REF	0	—
1 or more children	−0.10	0.04	0.01	0.13	0.11	0.26
Educational attainment
<12 grade	0.13	0.05	0.02	0.13	0.13	0.30
High school graduate	0.03	0.04	0.42	0.00	0.12	0.972
College+	REF	0	—	REF	0	—
Survey language
English	REF	0	—	REF	0	—
Spanish	−0.08	0.58	0.90	−0.22	0.16	0.17
Self-rated health
Excellent/very good/good	REF	0	—	REF	0	—
Fair/poor	0.78	0.06	0.00	0.48	0.14	0.001
Substance abuse
Yes	−0.43	0.13	0.001	0.46	0.32	0.14
No	REF	0	—	REF	0	—
MSA size
MSA 1million +people	REF	0	—	REF	0	—
MSA < 1 million people	0.01	0.04	0.77	0.02	0.10	0.87
Friends with shared interests/activities
0–3	0.28	0.04	0.00	0.19	0.12	0.11
4+	REF	0		REF	0
Times in the past 12 months angry with spouse/partner
0–2	−0.16	0.06	0.004	−0.01	0.16	0.93
A few times/no spouse	REF	0		REF	0
Many times	0.24	0.06	0.00	0.69	0.18	0.00
How often is spouse/partner critical
Always	0.13	0.09	0.15	0.28	0.27	0.29
Sometimes/no spouse	REF	0		REF	0
Seldom/never	−0.27	0.05	0.00	−0.52	0.14	0.00
How often does your spouse/partner show concern for feelings
Always/sometimes/no spouse	REF	0		REF	0
Seldom/never	0.35	0.09	0.00	0.17	0.22	0.44

Data source: Original tabulations of the 2002 National Survey on Drug Use and Health (NSDUH). Workers are defined as full-time, part-time or those doing any other/volunteer work. The reference group is adults who are working part-time or are doing other/volunteer work.