Health and adverse selection into marriage

Public Health (1998) 112, 309±311 ß R.I.P.H.H. 1998 http://www.stockton-press.co.uk/ph

Health and adverse selection into marriage: evidence from a study of the 1958 British Birth Cohort YB Cheung1 and A Sloggett2 Institute for Human Services Research, P.O. Box 73815, Kowloon Central Post Of®ce, Hong Kong and 2Centre for Population Studies, London School of Hygiene and Tropical Medicine, London, UK 1

Despite much research on the relationship between marital status and health, the confounding effects of marital selection are not well understood. Even less is known about `adverse selection', the phenomenon that people with poor health or health related attributes have a higher chance of marriage. Using data from the National Child Development Study, a longitudinal study of the 1958 British birth cohort, this paper examines the effects of factors that can select single people into early or later marriages. The selection factors are from three domains, namely, health status, socioeconomic status, and risk-taking behaviour. It is found that, from age 16 ± 23 y, adverse selection is prevalent. People from a lower socio-economic background and smokers are more likely to be married. This has the potential to suppress any association between marriage and health. Meanwhile, men with medical problems are less likely to marry. At ages from 23 ± 33 y, adverse selection reduces and those who have better life chances are more likely to get married. This has the potential of creating a spurious relation between marriage and health. These ®ndings shed light on the understanding of the confounding effects of marital selection according to different life stages. Keywords: marital status; adverse selection; health; Britain

Introduction Until recently prevalent opinions were that marital selection had a unidirectional effect of denying disadvantaged people access to marriage and making them more prone to divorce. This can be referred to as `positive selection'. The possibility of `adverse selection', that is people with health problems or higher exposure to health risks being more likely to get and remain married, is often neglected. In a recent paper published in Public Health,1 the impact of marital selection on the health differentials between married and divorced people in the 1958 British birth cohort was examined. It discussed the hypothesis that people with poor health may have a higher incentive to maintain a marriage.2 It showed that for some disadvantaged people the odds ratio of divorce was lower. Nevertheless, positive selection seemed to be more important in the context of transition from marriage to divorce. As a follow-up, this paper examines marital selection from the single state into marriage and pays particular attention to adverse selection. In a review of British social studies, Coleman and Salt3 summarised that people from a lower socio-economic background tended to marry at younger ages. Additionally an American study showed an association between DSMIII-R psychiatric disorders during adolescence and early marriage.4 It attributed this to young people's attempt to escape stressful environments by marrying early. Two American studies observed associations between drug abuse, juvenile delinquency and early marriage.5,6 The authors suspected that pseudo-maturity, that is a tendency among adolescents with behavioural problems to assume adult roles, may promote premature marriage. Meanwhile, the review by Coleman and Salt3 found that a larger proportion of people with lower socio-economic status remained unmarried.3 Studies looking at marriage in general Correspondence: Mr YB Cheung. Accepted 7 May 1998

also indicated a lower likelihood of marriage for people with increased psychological problems and risk-taking behaviours.4,7 In summary, marital selection with relation to poor life chances and risk-taking behaviour may have a temporal effect of promoting early marriage, as well as a quantum effect of reducing the overall likelihood of marriage. Methods The National Child Development Study is a cohort study of about 17 000 British people born in 1958. It originated in the Perinatal Mortality Survey which obtained information of almost all births during a week in 1958 in England, Wales and Scotland. Its scope has been widened in ®ve sweeps of follow-up: at ages 7, 11, 16, 23 and 33 y (NCDS 1 ± 5).8 This paper examines separately marital transition in the age stages from 16 ± 23 y and from 23 ± 33 y. The subjects being those who were single (never married) at ages 16 or 23 y, and either single or married at age 23 or 33 y. Marital status (single vs married) at age 23 y was regressed on selection factors from three areas, namely, health status, socio-economic status, and risk-taking behaviour, measured at age 16 y. Similar regression analyses were repeated using marital outcome at 33 y and selection factors measured at age 23 y. Government statistics indicate that people who have not married by their early 30s are likely to remain unmarried.9 Therefore the two-stage analyses shed light on the temporal and quantum effects of marital selection. Details about methods and sample representativeness have been reported elsewhere.1,8,10,11 The independent variables at age 23 y have been explained in the previous paper.1 Independent variables collected at age 16 y are as follows. The respondents received a medical examination as part of the NCDS3 study. In this paper `medical abnormality' refers to the presence of any general motor abnormality, physical abnormality, or mental retardation recorded. There was a

HealthÐadverse selection into marriage YB Cheung and A Sloggett

310

question `How many cigarettes do you usually smoke in a week?' in the NCDS3 survey. The answers were categorised as `non-smoker (none)', `occasional smoker (1 ± 19)' and `frequent smoker (520)'. The categorisations of social class, education and housing tenure are selfexplanatory. Social class was classi®ed according to the occupation of a respondent's father or father ®gure (Table 1). Note that education was ascertained at NCDS4 (age 23 y) but was considered a potential selection factor in the stage 16 ± 23 y, since most respondents completed their education before 23 y. Multivariate logistic regression with backward elimination of insigni®cant variables based on likelihood ratio test was employed. Models were ®tted for males and females separately. Results

standing illness (OR ˆ 0.45) and males who rated their own health as fair or poor (OR ˆ 0.7) had lower odds of being married at age 33 y. Males with high malaise level also had lower chance of marriage (OR ˆ 0.58). Education was no longer associated with marriage at this stage. Men of classes IV and V had about 40% lower odds of getting married. Female owner-occupiers had a better chance of getting married than all other tenure groups. Council tenants had the poorest chance (OR ˆ 0.37 in relation to owner-occupiers). Unemployment was signi®cant for both sexes: the longer one has been unemployed, the lower the odds of marriage. Drinking and smoking were not signi®cantly associated with a man's chance of getting married. However, female heavy smokers were less likely to get married. Discussion

The life stage 16 ± 23 y Results of the ®nal multivariate logistic regression models are shown in Table 1. Initial examination showed that education was a strong confounder of the relation between most of the selection factors and marital status. This is understandable since a longer time spent in education is commonly associated with delayed marriage. Males suffering a medical abnormality at age 16 y were less likely to be married at 23 y (OR ˆ 0.43). For both sexes, growing up in an owner-occupier's family was associated with lower odds of early marriage (that is being married at age 23 y). Father's social class was insigni®cant among females but signi®cant for males: the lower the class status, the higher the odds of early marriage. Smoking was associated with higher odds of early marriage regardless of sex. The life stage 23 ± 33 y The ®nal multivariate logistic regression models are shown in Table 2. Health status at age 23 y remained an important aspect of positive selection. Females with limiting long-

The ®ndings show clear patterns of marital selection at different stages of life. At young ages, lower socioeconomic status and smoking had a temporal effect of increasing the chances of getting married. At the stage from 23 ± 33 y, however, this impact was reversed. Both were associated with lower odds of marriage. Finally, those who suffered poor health had lower odds throughout. It seems reasonable to assume a unidirectional effect for selection into marriage during middle adulthood (age 23 ± 33 y). This can exaggerate the health differences between single and married people. Nonetheless, marital selection in the stage from adolescence to early adulthood (age 16 ± 23 y) is more complex. The protective effects of marriage can be suppressed by adverse selection. Future investigation into marriage and health during young adulthood should take both positive and adverse selection into account. Also the mechanisms of adverse selection are largely unknown, though the wish to escape from stressful environments and pseudo-maturity have been suggested.4±6 Detailed investigation of this issue may open up a research area relating to the health and coping behaviour of adolescents.

Table 1 Relative odds of early marriage vs being single at age 23 y: ®nal multivariate logistic regression models for females and males.a,b Relative odds (95% CI) Area variable

Value

Education

Below O level O level or equivalent A level or above

Health status Medical abnormality Socio-econ. status Social class Housing tenure Risk-taking behaviour Smoking

Females (n ˆ 2434)

Males (n ˆ 2501)

1.00 0.82 (0.69,0.97)* 0.31 (0.26,0.38)*

1.00 0.85 (0.70,1.04) 0.35 (0.26,0.46)*

Condition absent Present

1.00 0.43 (0.26,0.70)*

I and II III IV and V Owner Private renter Council renter

1.00 1.40 (1.20,1.63)* 1.37 (0.97,1.94)

1.00 1.50 (1.21,1.85)* 1.81 (1.39,2.35)* 1.00 1.15 (0.79,1.69) 1.36 (1.14,1.62)*

Non-smoker Occasional smoker Frequent smoker

1.00 1.12 (0.92,1.36) 1.24 (1.02,1.50)*

1.00 1.24 (0.96,1.59) 1.36 (1.12,1.65)*

Independent variables measured at age 16. Empty cells indicate that the variables were excluded from the statistical models because likelihood ratio tests showed that including them did not improve the goodness-of-®t (P > 0.05). *P < 0.05.

a

b

HealthÐadverse selection into marriage YB Cheung and A Sloggett

311

Table 2 Relative odds of being married vs staying single at age 33 y, conditional upon being single at age 23 y: ®nal multivariate logistic regression models for females and malesa, b Relative odds (95% CI) Area variable

Value

Health status Limiting long-standing illness Self-rated health Malaise level Socio-econ. status Social class Housing tenure

Unemployment Risk-taking behaviour Smoking

No Yes Excellent or good fair or poor Low (0 ± 7) High ( > ˆ 8)

Females (n ˆ 1724) 1.00 0.45 (0.28,0.75)*

I and II III IV and V Owner-occupier Private renter Council renter Other tenant < 1 month 1 ± 12 months > 12 months

1.00 0.58 (0.36,0.95)* 0.37 (0.22,0.63)* 0.57 (0.37,0.87)* 1.00 0.79 (0.64,0.98)* 0.52 (0.37,0.74)*

Never-smoker Ex-/light smoker Heavy smoker

1.00 0.87 (0.69,1.10) 0.64 (0.49,0.82)*

Males (n ˆ 1698)

1.00 0.70 (0.49,0.99)* 1.00 0.58 (0.36,0.95)* 1.00 0.94 (0.74,1.19) 0.58 (0.36,0.95)*

1.00 0.70 (0.57,0.86)* 0.39 (0.29,0.54)*

Independent variables measured at age 23 y. Empty cells indicate that the variables were excluded from the statistical models because likelihood ratio tests showed that including them did not improve the goodness-of-®t (P > 0.05). Education and alcohol consumption were found insigni®cant for both sexes. So they are not shown here. *P < 0.05.

a

b

Acknowledgements Thanks are due to the ESRC Data Archive and the Social Statistics Research Unit at City University for access to the data of the National Child Development Study. References 1 Cheung YB. Can marital selection explain the differences in health between married and divorced people? Public Health 1998; 112: 113 ± 117. 2 Lillard LA, Panis CWA. Marital status and mortality: the role of health. Demography 1996 33: 313 ± 327. 3 Coleman D, Salt J. The British Population: Patterns, Trends and Processes: Oxford University Press: Oxford, England, 1992. 4 Forthofer MS, Kessler RC, Story AL, Gotlib IH. The effects of psychiatric disorder on the probability and timing of ®rst marriage. J Health Soc Behav 1996; 37: 121 ± 132.

5 Newcomb MD, Bentler PM. Consequences of Adolescent Drug Use. Sage: Newbury Park, England, 1989. 6 Fu H, Goldman N. Incorporating health into models of marriage choice: demographic and sociological perspectives. J of Marriage and the Family 1996; 58: 740 ± 758. 7 Mastekaasa A. Marriage and psychological well-being: some evidence on selection into marriage. J of Marriage and the Family 1992, 54: 901 ± 911. 8 Power C. A review of child health in the 1958 birth cohort: National Development Study. Paediatr Perinat Epidemiol 1992; 6: 81 ± 110. 9 OPCS. 1993 Marriage and Divorce Statistics. Series FM 2, No. 21. HMSO: London, 1995. 10 Ferri E. Introduction, In: Ferri E (eds). Life at 33: The Fifth Follow-up of the National Child Development Study. National Children Bureau: London, 1993, 1 ± 15. 11 Power C, Manor O. Fox J. Health and Class: The Early Years. Chapman and Hall: London, 1991.