Parenthood, gender and sickness absence

Parenthood, gender and sickness absence

Social Science & Medicine 50 (2000) 1827±1842 www.elsevier.com/locate/socscimed Parenthood, gender and sickness absence Arne Mastekaasa a,b,* a Dep...

331KB Sizes 0 Downloads 81 Views

Social Science & Medicine 50 (2000) 1827±1842

www.elsevier.com/locate/socscimed

Parenthood, gender and sickness absence Arne Mastekaasa a,b,* a

Department of Sociology & Human Geography, Faculty of the Social Sciences, University of Oslo, P.O. Box 1096, 0317 Oslo, Norway b Institute for Social Research, Oslo, Norway

Abstract It is well documented that women have generally higher morbidity rates than men. In line with this women are also more absent from work due to sickness. This paper considers one popular explanation of the morbidity di€erence in general and of the di€erence in sickness absence in particular, viz. that women to a greater extent than men are exposed to the `double burden' of combining paid work with family obligations. We discuss theories of role overload and role con¯ict, which both assume that the combination of multiple roles may have negative health e€ects, as well theories of role enhancement, which assume positive health e€ects of multiple roles. Using two large Norwegian data sets, the relationship between the number of and the age of children on the one hand and sickness absence on the other is examined separately for men and women and for a number of theoretically interesting subpopulations of women de®ned in terms of marital status (also taking account of unmarried cohabitation), level of education, and working hours. Generally speaking the association between children and sickness absence is weak, particularly for married people of both genders. To the extent that married persons with children are more absent than married persons without children, this is largely due to respiratory conditions. The relationship between children and sickness absence is somewhat stronger for single, never married mothers, but not for single mothers who have been previously married or for women living in unmarried cohabitation. The ®ndings thus provide little support for either role overload/con¯ict or role enhancement theories. The possibility that these e€ects are both present and counterbalancing each other or that they are confounded with uncontrolled selection e€ects can not, however, be ruled out. 7 2000 Elsevier Science Ltd. All rights reserved. Keywords: Parenthood; Children; Multiple roles; Sickness absence

Introduction A large body of health statistics and empirical research shows that women have higher morbidity rates than men. In line with this overall pattern, most studies also show large gender di€erence in absence from work due to sickness (see, e.g., the

* Fax: +47-2285-5253. E-mail address: [email protected] Mastekaasa).

(A.

review in Mastekaasa and Olsen, 1998). In Norway the number of sick leave days compensated under the national health insurance system per employee per year is 1.65 as high for women as for men (National Insurance Administration, 1998). A popular explanation of the morbidity di€erence in general and of the di€erence in sickness absence in particular is that women to a greater extent than men are exposed to the `double burden' of combining paid work with family obligations. The double burden may consist of both role overload (having too much to do) and role con¯ict (being subject to

0277-9536/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 7 - 9 5 3 6 ( 9 9 ) 0 0 4 2 0 - 7

1828

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

con¯icting expectations or demands). The view that multiple roles are detrimental to health is not unanimous, however. There is an equally large literature stressing the positive health e€ects of multiple roles (the role enhancement hypothesis). A large number of studies have examined the relationship between the combination of employment and parental roles and various morbidity measures. On the whole this literature provides more support for the role enhancement than for the role overload/con¯ict hypothesis (Weatherall et al., 1994). A positive relationship between parenthood and health, however, is less well established than that between employment and health. In a review of research during the 1980s Ross et al. (1990, p. 1066) conclude that ``children at home have small, inconsistent, or insigni®cant e€ects on parent's physical health . . .''. A similar conclusion is reached by AÊkerlind et al. (1996), who also review a large number of studies from the 1970s and 1980s1. This impression is con®rmed in more recent studies, for instance Hibbard and Pope (1991), and in studies focusing on psychological problems or psychological well-being (see e.g., Roxburgh, 1996, and the literature cited there). Several studies have examined the relationship between parenthood and mortality. Some of these report women with children to have lower mortality whereas others ®nd no mortality di€erence between parents and nonparents. Using both number of children and a dichotomous child/no child variable, Hibbard and Pope (1991, 1992, 1993) found no relationship with mortality, and this was true for employed as well as nonemployed women (and for men). With regard to employed women and men, similar results are reported by Kotler and Wingard (1989), although their data provided some indications that children could be associated with higher mortality among single employed women. Two other studies report a negative association between children and mortality. Weatherall et al. (1994) found a lower mortality rate among women with children below 10. Martikainen (1995) reports a clearly lower mortality rate for women with children below 16, and this held for employed as well as nonemployed and for married as well as not married women. Martikainen (1995) found the negative relationship between parenthood and mortality to be particularly strong for deaths due to accidents and violence. Consistent with this, several studies have found married women with children to have a considerably lower suicide risk than both married women without 1 Their own empirical ®ndings on sickness absence do not con®rm to this pattern, however. See below.

children and unmarried women (Veevers, 1973; Kozak and Gibbs, 1979; Fawcett et al., 1987; Hùyer and Lund, 1993). Previous research on the relationship between children and sickness absence has yielded mixed results. In the American literature the most common ®nding has been that women with children are more absent than women without children (Naylor and Vincent, 1959; Nicholson and Goodge, 1976; Garrison and Muchinsky, 1977; Leigh, 1983; Scott and McClellan, 1990; Vistnes, 1997), but the opposite relationship has also been reported (Fitzgibbons and Moch, 1980; Paringer, 1983). A number of Swedish studies have shown that women with children have less short-term sickness absence (de®ned by a duration of one week or less) than women without children (or, in some cases, that women with more children have less sickness absence than those with fewer children) (BjoÈrklund, 1991; Vogel et al., 1992; Svensson, 1995). With regard to absence spells of longer duration the ®ndings in the Swedish studies are less consistent. A positive relationship between children and sickness absence is reported by Vogel et al. (1992), BjoÈrklund (1991) and AÊkerlind et al. (1996), while others report negative or zero relationships (Puranen, 1991; Svensson, 1995). With regard to Norway, Mastekaasa (1990) found that married or cohabiting women with small children were somewhat less likely to have an absence spell (of at least one week's duration) than married/cohabiting women without children. To sum up, there is little evidence that children have any strong impact on most measures of morbidity or mortality. A notable exception is suicide mortality, which seems to be clearly lower among parents. Mortality due to other particular causes and speci®c types of morbidity may also to a greater or lesser extent deviate from the overall pattern. With regard to sickness absence in particular, there seems to be a weak preponderance of studies showing children to be associated with a somewhat higher level of sickness absence. The purpose of this paper is to provide a detailed picture of how sickness absence is related to both the number of children and their age. The study adds to the literature in several ways. Many previous studies have not distinguished between single and married women or between di€erent marital statuses more generally, either because this information has been missing or because of inadequate sample sizes. The present study utilises two very large samples, one of which is drawn from the population of all employees in Norway. Another important advantage of the present data is that information on diagnoses is available. Third, we are able to take into account that a considerable part of the gender di€erence in sickness absence may be due to pregnancy-related absences (Alexanderson et

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

al., 1996); in many previous studies the e€ects of being a mother has been confounded with the e€ects of child bearing. The rest of the paper is organised as follows. In the next section we present the role enhancement, role con¯ict and role overload arguments in more detail and develop hypotheses about the relationships between children and sickness absence. The third section presents data and methods. We then present a series of empirical analyses. Although most analyses are limited to married people, we also include some results on single and cohabiting women.

Theoretical background and hypotheses The relationship between sickness absence and health or morbidity Use of sickness absence as a health indicator has been suggested by several authors (e.g., Bourbonnais et al., 1992; Marmot et al., 1995; Messing, 1997). Nevertheless, as these authors also note, sickness absence is clearly not a simple re¯ection of morbidity. It is more appropriately regarded as an illness behaviour, the latter concept being de®ned as ``the manner in which persons monitor their bodies, de®ne and interpret their symptoms, take remedial action, and utilise various sources of help as well as the more formal health care system'' (Mechanic, 1986, p. 101). To the extent that sickness absence requires certi®cation from a physician, it is also directly in¯uenced by the physician's behaviour. Moreover, the need to be absent from work does not only depend on the seriousness of the medical condition, but also on the characteristics of the job; in most cases it will be easier, for instance, to do oce work while sick than to work as a manual labourer or a stewardess. In addition, depending on the degree of control, absence from work formally reported as due to sickness may also be due to other factors, like interpersonal con¯icts in the workplace or low work motivation. Thus sickness absence is not ideal as a general measure of health or morbidity. But other commonly used morbidity measures like those based on selfreports in surveys or on medical help-seeking also suffer from biases, and it is not clear what type of data has the highest validity. The important point, however, is that di€erent types of data often su€er from di€erent biases; use of di€erent types of data thus may contribute to a more valid and complete overall picture (cf. Gijsbers van Wijk et al., 1992). Moreover, the purpose of this article is not to provide a general picture of the health status of the Norwegian workforce, but only to examine the degree to which responsibility for children

1829

is associated with health problems. Thus we need only be concerned about sources of error that covary systematically with the individual's family composition. Moreover, we limit the analyses to sickness absences certi®ed by a physician, and in some cases even to absences (`spells') of relatively long duration (more than two weeks). Marmot et al. (1995) found strong relationships between self-reported health problems and long term absences; for shorter absences the relationships were weaker. Theories and hypotheses Most discussions of the e€ects of parenthood and employment on health draw on the role concept. More speci®cally, the combination of the roles of parent and employee is discussed as a special case of the more general phenomenon of multiple roles. There are two main streams of thought, one stressing the negative e€ects of multiple roles, the other the positive ones. The idea that multiple roles may have negative health e€ects has been associated in particular with Goode's (1960) scarcity hypothesis. Goode assumes that human energy or role resources are ®xed and limited. Therefore, the more roles a person occupies, the more strain he or she experiences, and the greater the likelihood of negative e€ects on health and well-being. Later discussions of these e€ects have often distinguished between role overload and role con¯ict. The role overload hypothesis is very closely related to Goode's scarcity hypothesis, focussing on time constraints and the amount of work involved in simultaneously performing several demanding roles. Role con¯ict refers to the idea that di€erent roles may involve contradictory expectations. Although often closely associated in practice, role overload and role con¯ict refer to conceptually distinct mechanisms. Role overload may be mitigated, for instance, by placing small children in a kindergarten, but this does not solve the role con¯ict if staying at home with the children is part of the expectations associated with the mother role. The idea that multiple roles may have positive e€ects on health and well-being has been referred to variously as the role accumulation or role enhancement hypothesis (Sieber, 1974; Marks, 1977). More roles means ``more potential sources of self-esteem, stimulation, privileges, social status, and social identity'' (Baruch et al., 1987). Among other things, a greater number of roles reduce the individual's vulnerability to negative experiences associated with any particular role. Although multiple role theories have been strongly dominant in discussions of the impact of children and employment, other perspectives have also been suggested. Although perhaps more famous for his the-

1830

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

ory of marriage as a protection against suicide, Durkheim (1970) argued that children were in fact a more important protective factor. This was so because marriage protected only men whereas children were important even for women. In discussing why children had such a bene®cial e€ect, Durkheim stressed the importance of the share amount of interaction in a family. The larger the family, the more intense the social life within it, and the lower the probability of suicide. Another possible health-promoting e€ect of responsibility for children is through increased social control. Parents are likely to limit health risks like dangerous leisure time activities or alcohol, tobacco or substance abuse to a greater extent than nonparents (cf. Martikainen's (1995) ®nding mentioned above that parenthood was most strongly associated with mortality due to accidents or violence). This may happen because of internalised norms or because of negative reactions from friends or relatives. Negative e€ects of parenthood on health may also come about by means of other mechanisms than role overload or role con¯ict. One factor of importance could be that parents are exposed to the infectious diseases (especially coughs and colds) caught by their children. With regard to the e€ects of parenthood and employment on sickness absence, the basic hypotheses implied by the various theories of multiple roles are simple. According to the role overload/con¯ict theory, one would expect the level of sickness absence to increase monotonically with the number of children. Since small children are generally more demanding than older children and since the expectations associated with the mother role are also probably strongest for small children, one would further expect the level of sickness absence to decline with the age of the youngest child. According to the role enhancement hypothesis the main di€erence should be between parents and nonparents; the number of children and their age seem less crucial to the theory. The hypothesis is therefore that parents have lower sickness absence than nonparents. Implicitly or explicitly most theoretical discussions of the e€ects of combining parenthood and employment have focussed on women and particularly married women2. In principle, at least, the various mechanisms discussed above also seem to apply to a greater or lesser extent to men and to single parents. Considering single women with children ®rst, they 2 The di€erent e€ects of children on the psychological wellbeing of married and nonmarried women has, however, been the subject of many studies in the family and divorce literature (e.g., Umberson and Gove, 1989).

obviously lack the role of wife. Automatic application of the scarcity hypothesis suggests that one role less should contribute to less overload and con¯ict. However, a single parent may also be considered as one who has to perform the tasks and ful®l the expectations that are otherwise shared by two parents. The single mother role may therefore be considerably more demanding than the married mother role. Research on marriage and psychological well-being also support the idea that parenting is more stressful for single than for married parents (Gar®nkel and McLanahan, 1986). It seems reasonable to expect the changes of the mother role to be more important than the loss of the wife role and to hypothesise that the above stated e€ects of the number of children and the age of the youngest child to be stronger for single mothers than for married ones. Additional factors contributing to more negative health e€ects among single women is that this group will consist of many formerly married or cohabiting persons who may experience the transition to single parenthood as stressful. Role enhancement theories do not necessarily imply that the e€ects of parenthood should di€er between single and married mothers. It seems reasonable to expect, however, that role accumulation is subject to marginal utility e€ects. That is, an additional role is less important the more roles a person occupies to begin with. It follows from this that the bene®cial e€ects of parenthood should be stronger for single than for married women. Unmarried cohabitation has grown increasingly common in many countries in the last couple of decades, not the least in Norway. In 1994, unmarried cohabitants made up more than 22% of all women 20±44 years of age (Noack, 1996, p. 22). Although less institutionalised than marriage, cohabitation probably has many of the same e€ects (Mastekaasa, 1995). We therefore expect the results for cohabitants to be quite similar to those for the married. Theories of multiple roles are not explicitly gender speci®c, and should apply to men as well as women. Thus children should increase the probability of overload even for men. The demands of the father role may, however, be less than those of the mother role. The overload e€ects can therefore be expected to be weaker for men than for women. Role con¯ict depends on the extent to which di€erent roles imply contradictory expectations. At this point there is a clear gender di€erence. For men the combination of paid employment and parenthood has been the dominant pattern for a long time and the normative expectations associated with these roles have had plenty of time to adjust to one another. On the other hand, the content of both the father role and the work role have changed in recent decades, and at least in some ways both roles may have become more

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

demanding (see e.g., Hochschild, 1997). The existence of role con¯ict even for men should therefore not be ruled out, but such e€ects should clearly be weaker than for women. It seems reasonable to expect that even role enhancement e€ects may be weaker for men than for women, but again these e€ects should not be expected to be zero. Weaker e€ects follow if one assumes that family and children is a more important life domain for women than for men. In our analyses we distinguish among absence spells with di€erent diagnoses. The various theories of multiple roles are not very speci®c, however, and often the same theory is applied to general feelings of well-being and to speci®c types of morbidity like coronary heart disease. Therefore the diagnosis speci®c analyses are mainly exploratory. A number of authors have criticised the multiple roles literature for treating women as a homogeneous group and not taking into account for instance class di€erences or di€erences in working hours (Arber et al., 1985; Arber, 1991). It seems reasonable to expect both role overload and role con¯ict e€ects to be stronger for full-time than for part-time workers. With regard to role enhancement, a marginal utility assumption (see above) leads to a hypothesis of stronger e€ects of parenthood for part-time workers. Although several authors have stressed the importance of taking class di€erences into account, exactly how class should a€ect the relationship between the combination of parenthood and employment on the one hand and health on the other has rarely been speci®ed. In general, however, the various multiple role e€ects may perhaps be expected to be stronger for lower or middle class women. Lower class women may be expected to have more physically demanding work. According to Karasek's theory, low decision latitude combined with monotonous, high-pressure work may also make many lower class (and lower middle class) jobs more psychologically demanding. On the other hand, (upper) middle class jobs are often associated with more responsibility. With regard to role enhancement, one could argue that the role of parent becomes less important for people who have a greater number of other valued roles, which middle class women are more likely to have. Role enhancement e€ects may therefore be stronger among lower class people. Since our most important data set lacks information about occupation, our ability to take class di€erences into account is limited. It is, however, possible to use level of education as a proxy for class. Selection e€ects and other confounding factors The statistical association between (the number and

1831

age of) children and sickness absence (or morbidity more generally) does not simply re¯ect the causal impact of the former variable on the latter. Since both responsibility for small children and morbidity vary over the life cycle, there is an obvious need to control for age. In most of our analyses we also control for education, earnings and working hours, since these variables may be expected both to correlate with the number of children and the children's age (e.g., women with many children are more likely to work part-time and thereby to have lower earnings) and to a€ect sickness absence. In addition to the potential e€ects of parenthood per se, part of women's higher morbidity is associated with their reproductive functions (Gijsbers van Wijk et al., 1992). Several Norwegian and Swedish studies show that periods of sickness absence are very common in connection with pregnancy (Alexanderson et al., 1996; Strand et al., 1997; Sydsjù et al., 1997). We take child bearing into account by modelling it as part of the e€ect of the age of the youngest child (see below). With regard to causal inferences, the possibility of selection e€ects is a major problem. Selection e€ects are dicult to handle even with longitudinal data, but particularly with cross-sectional data. The multiple roles literature includes a number of longitudinal studies which have examined the reciprocal e€ects of employment and health (e.g., Waldron and Jacobs, 1988; Elstad, 1995), but studies of the importance of parenthood (the present one included) are largely based on cross-sectional data. In the present context, two types of selection e€ects may be of importance. First, those who experience problems combining parenthood and employment may withdraw from the labour force. The labour force (and our samples, which are drawn from this population) will therefore contain a disproportionate number of relatively healthy parents. Since there is no corresponding selection of more healthy nonparents, a comparison of parents and nonparents may give a misleading picture of the e€ects of children; overload and con¯ict e€ects will be underestimated and role enhancement e€ects overestimated. The other type of selection e€ect is that people with health problems may refrain from having children or at least limit the number of children. The consequences with regard to the detection of overload, con¯ict and enhancement e€ects are similar: parents will be a relatively healthy subpopulation, and a comparison with nonparents will underestimate overload and con¯ict e€ects and overestimate enhancement e€ects. The level of labour force participation among women in general and among mothers of small children in particular varies considerably from one country

1832

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

to another3. This makes it likely that the importance of the type of selection e€ects discussed above may also vary. This could be part of the explanation for the somewhat divergent ®ndings in previous research, as discussed above. The level of female labour force participation could also be important in other respects. It seems reasonable to assume, for instance, that normative expectations adapt to some extent to people's actual behaviour, and that role con¯icts will therefore be weaker the more common it is for mothers to be occupationally active.

Data and methods

sampled. We include in our analyses all nonstate female employees 18 to 59 years of age who had been employed (4 h per week or more) for at least part of the year …N ˆ 99,742). Self-employed and state employees are excluded, the latter due to lack of information on sickness absence. We also include only women, since information about children is not available for nonmarried men. (Analysis of married men are limited to the 1995 data to limit the size of the article, but can be obtained upon request.) For brevity we refer to the data sets as the 1995 data and the 1990 data, respectively (although the speci®c years of data collection are not central to the issues taken up in this article).

Samples We use two Norwegian data sets in our analyses. The main data set is based on a national register of all employment relationships with four or more hours of work per week. A sample of 10% of all employment relationships active on 18 May 1995 was drawn. The sample consists of 190,947 persons. In the present study we limit the analyses to married persons between 18 and 59 years of age. The sample analysed includes 94,869 persons. The nonmarried are excluded since the register data only contain information about formal marital status. In Norway, unmarried cohabitation is very common and an analysis of the impact of children that does not take this into account is likely to be highly misleading. Nonmarried women are, however, analysed in our second data set. The second data set is based on the Norwegian 1990 Census, which has been supplemented by data from various public administrative registers. The 1990 Census was conducted as a sample survey of about 10% of the population. People in rural areas were over3

In Norway, the percentage of women with at least one child less than 7 years of age who were occupationally active was 71 in 1990 and 76 in 1995 (Statistics Norway, 1991, 1996). 4 Some information about the Norwegian sickness absence bene®t scheme is given at this website: http://www.trygdeetaten.no/pub/nsss.html. 5 We tried, e.g., to analyze the number of spells using Poisson regression. This works well for the total number of spells (all diagnoses) in the 1995 data, but is not appropriate for the diagnosis speci®c analyses or for the 1990 data, where the number of zeros will be much higher than expected in the Poisson distribution. Extensions of the Poisson model (socalled hurdle or zip-models) could probably be used, but the pay-o€ in terms of more ecient estimates would be small. For the 1990 data we have also made some analyses of days lost using tobit regression. Again, the results were similar to those obtained with logistic regression.

Variables Sickness absence The 1990 data provide information only on absence spells of more than two weeks duration; the register used for the 1995 data includes all physician certi®ed absences irrespective of duration (detailed information about duration is not given in these data). Absence spells lasting for three days or less, however, generally do not require certi®cation from a physician, and are not included. The compensation rate during sickness absence is 100% of earnings up to a ceiling of NKO 201,000 …1£16,000† in 1990 and NKO 233,000 …1£19,000† in 1995. Employees with higher earnings also in most cases receive full compensation as part of collective tari€ agreements or individual employment contracts4. In the analyses we measure sickness absence by a simple dichotomous variable distinguishing between at least one sickness absence spell on the one hand and no instance of sickness absence on the other during the period of observation (9 months in the 1990 data; 1 year in the 1995 data). Thus we do not take into account either the number of absence spells or the duration of the spells. The main reason for this is that it makes it possible to use the same statistical methods (logistic regression) in all analyses, which facilitates the presentation of the results and comparisons between the data sets. Operationalisations in terms of the number of spells or days lost (the latter available only in the 1990 data) lead to very similar results5. An indication of this is provided by the descriptive statistics presented below (Tables 1 and 2). Diagnostic categories In the 1995 data all absences are coded according to the International Classi®cation of Primary Care

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

1833

Table 1 Sickness absence in 1995 (all physician certi®ed absences) by number of children and gender. Married persons Women

Number of children 0 1 2 3±9

Men

average # of spells

at least one absence (%)

N

average # of spells

at least one absence (%)

N

0.578 0.596 0.549 0.528

36.6 37.2 35.4 34.1

19780 10244 11769 5112

0.386 0.379 0.401 0.402

26.1 25.4 26.7 26.7

17088 10940 13568 6782

(ICPC) (Lamberts and Wood, 1987)6. In addition to analyses of all absences irrespective of diagnosis, separate analyses are carried out for the numerically most important diagnostic categories, musculoskeletal (category L), respiratory (R), pregnancy and female genital (W and X), psychological (P), and digestive system (D). Together these categories include 83% of women's absence spells in 1995 and 79% of men's. Children In analysing the impact of children on sickness absence it is necessary to take both the number of children and their age into consideration. We do this by means of dummy variables for one child, two children, and three or more children, using no child as reference. Only children less than 16 years of age are included. The age of the youngest child is included as a continuous variable. We add, however, two dummy variables for the age of the youngest child being 0 and 1 year, respectively. The age 0 dummy captures sickness absence in connection with pregnancy, whereas the age 1 dummy adjusts for lower sickness absence during a period in which women are on parental leave7. Marital status and cohabitation We make separate analyses of married (both data 6 The ICPC is based on an abridged version of the International Classi®cation of Diseases (ICD) and on a Reason for Encounter Classi®cation developed in primary care (Lamberts and Wood, 1987). 7 All women are entitled to paid parental leave in connection with child birth (or adoption). In 1990 the length of this leave was 28 weeks with 100% pay or 35 weeks with 80%. By 1995 this had been extended to 38 and 48 weeks, respectively. 8 The distinction between the latter two groups was made because preliminary analyses showed that the relationship between children and sickness absence was very di€erent. See below. 9 Information on education is missing for 1 to 1.5% of our samples. In stead of excluding these persons we include a dummy variable for missing information on education.

sets), cohabiting, and never married and previously married single persons (the 1990 data)8. The distinctions between married and nonmarried and between never married and previously married are made on the basis of the public population register, whereas the distinction between cohabiting and single is made on the basis of self-reports. Control variables Age is measured in years. Level of education is measured in years beyond compulsory school. The highest level (Ph.D.) is 12 years9. Earnings per year (sum of wages and salaries received by the individual) is measured in units of the local currency (NOK; NOK11£0:08). In the analyses, a log transformation of this variable is used. Part-time work is measured by two dummy variables, one for 4 to 19 h per week and one for 20 to 29 h, with 30 h or more as reference. We also control for time of observation, measured in years. For most sample members the observation period is slightly in excess of 0.75 years (9 months) in 1990 and exactly one year in 1995. People who change jobs during 1990 or 1995 will be under observation for a shorter period, since we limit the analyses to the main employment relationship of each employee in May 1995 or November 1990.

Statistical methods As noted above, our dependent variable is dichotomous, taking the value 1 if an individual has at least one absence spell during the period of observation and 0 otherwise. Logistic regression is therefore employed. Nonlinear e€ects of age, education and earnings are taken into account by inclusion of second order terms. An interaction term allowing the age of the youngest child to have di€erent e€ects for women with only one and with two or more children is included when signi®cant (at the 0.05 level). For simplicity, the coecients for the control variables education, earnings, and working hours are not

1834

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

Table 2 Sickness absence (absence spells or more than two weeks duration) during the period 1 January to 2 October 1990 by number of children and marital/cohabiting status: women Married

Number of children 0 1 2 3±9

Cohabiting

Single

at least one absence (%)

average # of days lost

N

at least one absence (%)

average # of days lost

N

at least one absence (%)

average # of days lost

N

22.6 22.3 18.0 19.2

14.098 12.402 9.885 10.613

25157 16659 19504 8123

22.5 30.5 27.5 32.3

11.536 16.571 15.417 20.156

5223 3695 1770 344

18.6 24.3 26.2 29.3

10.542 13.218 14.793 18.058

13894 4613 1645 311

presented in the tables, but complete results are available upon request.

the level of sickness absence increases monotonically with the number of children. Married men and women

Results We organize the presentation of results in six subsections. In the ®rst one both data sets are used and in the third one, on ``Cohabiting and single women'', only the 1990 data are used. Otherwise only the 1995 data are used. Descriptive results Tables 1 and 2 present some simple descriptive statistics on the relationship between number of children and sickness absence. Since the 1995 data does not make it possible to distinguish between single and cohabiting persons, Table 1 includes married people only. For married women the percentage with at least one sickness absence is almost the same for those with one child as for the childless. Beyond the ®rst child the percentage declines slightly as the number of children increases. The same pattern emerges if we examine in stead the average number of absence spells. For men the relationship between number of children and either measure of sickness absence is even weaker. Table 2 provides information on children and sickness absence for single and cohabiting as well as married women. The percentage with at least one absence spell exceeding two weeks is slightly lower if the woman has two or more children than if she has one child or less, but the di€erence is small. The results for cohabiting and single women are quite di€erent. For cohabiting women the major di€erence is between the childless on the one hand and those with one child or more on the other, with the level of sickness absence markedly higher among the latter. For single women

Table 3 presents results from two logistic regression models, estimated separately for married women and married men. Both models include several variables describing the number of children and the age of the youngest child, as well as controls for time of observation and age. Model 2 for both genders in addition controls for level of education, earnings and working hours. In model 2 for men, the interaction of the number of children with the age of the youngest child was not signi®cant, and was deleted from the model. According to model 1, only married women with one child have a probability of sickness absence that is signi®cantly di€erent from their childless counterparts, and even this di€erence declines with the child's age. When controlling for education, earnings and working hours, even women with two or three or more children have a signi®cantly higher probability of sickness absence than the childless. More detailed analyses show that the di€erence between the two models is most strongly in¯uenced by the control for level of education: in model 1 the level of sickness absence among women with children is kept down by these women's relatively high levels of education. A striking result in both models for women is the very high level of absence if children are born during the year of observation, and even more the very low level of absence the year after the birth. This is, however, in line with our expectations and previous research. A very high proportion of women have a period of sickness absence in connection with their pregnancy, and for a substantial part of the year after birth the woman will be on paternal leave. Apart from the year a child is born and the year after, the results for men are very similar to the results

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

1835

Table 3 Logistic regression of sickness absence on the number and age of children and controls. Married women and men. 1995 data. Regression coecients with standard errors in parenthesesa Married women model 1

Married men model 2

model 1

model 2

Intercept Time of observation 1 child 2 children 3 or more children Age of youngest child Age of youngest child 2 or more children Child born in 1994 Child born in 1995 Age/10 Age2/100

ÿ0.4764 0.0043 0.1582 0.0588 ÿ0.0578 ÿ0.0179 ÿ0.0168

(0.0242) (0.0003) (0.0510) (0.0425) (0.0440) (0.0043) (0.0054)

ÿ0.4712 0.0038 0.2279 0.1884 0.1358 ÿ0.0200 ÿ0.0197

(0.0278) (0.0003) (0.0518) (0.0435) (0.0455) (0.0043) (0.0054)

ÿ1.0904 0.0032 0.0906 0.1724 0.1560 ÿ0.0109 ÿ0.0138

(0.0282) (0.0004) (0.0505) (0.0428) (0.0436) (0.0043) (0.0054)

ÿ0.9238 0.0038 0.2075 0.2293 0.2321 ÿ0.0206 ±

(0.0313) (0.0004) (0.0486) (0.0414) (0.0437) (0.0038)

ÿ0.7242 0.3896 ÿ0.1071 0.0391

(0.0625) (0.0555) (0.0156) (0.0129)

ÿ0.7243 0.3412 ÿ0.1359 0.0792

(0.0631) (0.0561) (0.0160) (0.0132)

ÿ0.1482 ÿ0.1585 ÿ0.1229 0.1227

(0.0506) (0.0532) (0.0170) (0.0144)

ÿ0.1158 ÿ0.1225 ÿ0.0690 0.0878

(0.0509) (0.0533) (0.0176) (0.0146)

ÿ2 log likelihood N

60530.54 46680

59692.88 46680

55295.50 48189

53895.69 48189

a

In addition to the variables included in the table, model 2 also controls for the following variables: education, education squared, log(earnings), log(earnings) squared, long part-time, and short part-time work. The reference individual has the following characteristics: no children, 40 years of age, 3 years of secondary school, earnings of NKO200,000, full-time job. Coecients that that are signi®cantly di€erent for men and women at the 0.01 level (two-tailed test) are in bold type. Signi®cance probability for each coecient is given by p < 0.001, p < 0.01 and p < 0.05.

for women. In fact the coecients for the number of children dummies are slightly larger for men, but the di€erences between the genders are not signi®cant. The results for both men and women are illustrated in Fig. 110. The ®gure shows a slightly higher level of absence when the child is small, but the level of absence declines with the age of the child. This decline is most rapid for women who have more than one child, whereas there is no di€erence in this respect between women with one child and men irrespective of how many children they have. The results so far are quite clear. First, married women with very young children have slightly higher sickness absence than married women without children, but as the children grow older women with children tend to have the lower level of sickness absence. Second, the relationship between children and sickness absence is in general very similar for men. Cohabiting and single women As noted above, there are reasons to believe children

10 Since the probability of absence depends on several variables, it is necessary to choose some speci®c values on these variables. This does not, however, a€ect the shape of the relationship between the child variables and sickness absence.

to have stronger e€ects for single women, whereas cohabiting women are expected to di€er little from the married. In Table 4 the 1990 data are used to test these hypotheses. Preliminary analyses showed that the relationship between children and sickness absence was very di€erent for never married and previously married single women. These categories are therefore analysed separately. For purposes of comparison, results for married women are also included. The results for married women in the 1990 data (Table 4) di€er only slightly from the results in the 1995 data (Table 3). The main di€erence is that in the 1990 data women with two children are signi®cantly less absent from work than the childless (no di€erence in the 1995 data). This is probably due to the fact that the 1990 data only include absence spells of more than two weeks duration, whereas the lower limit in the 1995 data is four days. We show in the next section that the results in the two data sets become more similar if absences due to respiratory conditions, which are mainly of short duration, are excluded from the analysis of the 1995 data. Our expectation of small di€erences between cohabiting and married women is con®rmed. None of the coecients for number of children or the age the youngest child are signi®cantly di€erent between these two categories. For never married single women the level of sickness

1836

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

Fig. 1. Probability of sickness absence for married men and women as a function of the number of children and the age of the youngest child.

Table 4 Logistic regression of sickness absence on the number and age of children and controls. Married, cohabiting and single women. 1990 data. Regression coecients with standard errors in parenthesesa Married

Cohabiting

Single never married

Intercept Time of observation 1 child 2 children 3 or more children Age of youngest child Child born in 1989 Child born in 1990 Age/10 Age2/100

ÿ1.1883 0.0042 0.1286 ÿ0.1712 ÿ0.0839 ÿ0.0183 ÿ0.4392 1.2972 ÿ0.1278 0.1623

ÿ2 log likelihood N

67652.58 69443

(0.0335) (0.0003) (0.0475) (0.0417) (0.0447) (0.0036) (0.0630) (0.0481) (0.0151) (0.0119)

ÿ1.0893 0.0056 0.0535 ÿ0.0606 0.1619 0.0018 ÿ0.2445 1.6467 0.1044 0.0741 11591.21 11032

(0.0693) (0.0006) (0.0792) (0.0891) (0.1465) (0.0091) (0.1132) (0.0946) (0.0410) (0.0300)

ÿ1.5622 0.0040 0.3994 0.4836 ÿ0.1011 ÿ0.0143 ÿ0.4094 1.7158 0.0300 ÿ0.0259 9770.421 12227

previously married (0.0679) (0.0007) (0.1038) (0.1534) (0.4325) (0.0115) (0.2177) (0.1561) (0.0355) (0.0270)

ÿ0.9810 0.0034 ÿ0.2610 ÿ0.1620 0.1253 0.0204 0.0977 1.2850 ÿ0.0470 0.0079

(0.0764) (0.0008) (0.1445) (0.1303) (0.1688) (0.0112) (0.3397) (0.3103) (0.0457) (0.0334)

9416.947 8236

a In addition to the variables included in the table the following variables are controlled for: education, education squared, log(earnings), log(earnings) squared, long part-time, and short part-time work. Coecients in the cohabiting and single subsamples that are signi®cantly di€erent from those in the married subsample (0.01 level, two-tailed test) are in bold types. Otherwise, see the note to Table 3.

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

absence is clearly higher if the woman has one child and it increases further for two children. Using the coecients in Table 4, the estimated probability of sickness absence for a 40 year old woman with 3 years of secondary education, average earnings (NKO 200,000 per year) and a full-time job is 0.17 if she has no child. This increases to 0.24 if she has one 2-yearold child and 0.25 if she has two children (and the youngest is 2 years old). The estimated e€ect of two children is also signi®cantly larger for single never married women than for married women. The relationship changes dramatically if she has three or more children. Very few never married women have as many as three or more children, however …N ˆ 51 in our data). This makes the estimate for this category very unreliable (note the large standard error in Table 4). The age of the youngest child is not signi®cantly related to the sickness absence of single never married women, but the size of the coecient is only moderately weaker than for married women (ÿ0.014 versus ÿ0.018). For single previously married women, neither the number of children nor the age of the youngest child is signi®cantly related to the probability of sickness absence. The coecients for one and two children are both negative and may seem to be of some magnitude. Note, however, the estimated positive e€ect of the age of the youngest child, which means that the estimated di€erence between women with and without children declines quite rapidly over time. It is important to note that although single never married mothers of one or two children have considerably higher sickness absence than single never married women without children, they di€er very little from married as well as cohabiting mothers. It is mainly single never married women without children who deviate considerably from the other categories by having a particularly low level of sickness absence. There are very strong e€ects of having a child during the year of observation, particularly for never married single and cohabiting women. Even for married women the e€ect is much stronger than that found in the analysis of the 1995 data. This di€erence could be due to the fact that we now include only relatively long-term absences (more than two weeks); previous research has shown that absence spells in connection with pregnancy are often of quite long duration (Strand et al., 1997). The most important reason, however, may well be that the rules for use of parental leave were slightly di€erent in the two years: in 1995 11 This does not imply, of course, that sickness absence among childless married women is mainly due to morbidity in the reproductive organs. Even for this group the probability of at least one absence spell during a year is much larger for the musculoskeletal and respiratory categories.

1837

three weeks of the leave period had to be taken before delivery, whereas this was not required in 1990. This change probably reduced the need for sickness absence during the late part of the pregnancy. For the single the relationship between age and sickness absence is very weak. Among the married the probability of sickness absence is relatively high among the young, it reaches a minimum for women in the mid-forties, but then increases somewhat again. Major diagnostic categories Separate analyses were carried out for the most important diagnostic categories, musculoskeletal (category L), respiratory (R), pregnancy and female genital (W and X), psychological (P), and digestive system (D). To conserve space, the main ®ndings are only summarised brie¯y (but tables are available upon request). For all categories except R and W and X, the results are very simple: Neither the number of children nor the age of the youngest child has any signi®cant impact on the probability of sickness absence. On the other hand, sickness absence with respiratory diagnoses is much higher among parents of both genders than among the childless. This is shown more clearly in Fig. 2. As in the analyses above, and probably due to parental leave, the level of absence is low the year after a child is born, particularly for women. Contrary to what was found for the overall level of sickness absence, however, women also have a very low level of absence due to respiratory conditions during the year they give birth to a child. Apart from the time immediately preceding and following childbirth, the level of absence is high when the youngest child is small and decreases with the age of the child, and this holds for men as well as women. When the youngest child is two years old, a woman with one or two children has a 43% higher probability of absence than a childless woman. For a woman with three or more children, the level of sickness absence is somewhat lower, but still considerably higher than for a childless woman. A man with one two-year old child likewise has a 37% higher sickness absence probability than a childless man. Women's sickness absence with diagnoses related to pregnancy and genital organs is very high if a child is born during the year of observation. Otherwise, this type of absence is much lower for women with children than for the childless. This is probably largely due to a selection e€ect: the population of childless married women includes a considerable number of women who are childless precisely because of morbidity in the sexual organs11. The results so far indicate that the weak tendency to

1838

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

Fig. 2. Probability of sickness absence due to respiratory conditions for married men and women as a function of the number of children and the age of the youngest child.

higher sickness absence among persons with children documented in Table 3 could mainly be a result of the association between children and absence due to respiratory conditions. Table 5 provides results from an analysis of all sickness absence except that classi®ed as respiratory12. Women with one small child still have a signi®cantly higher probability of absence than childless married women, but the coecient is very small. The positive e€ects of having two or three or more children disappear completely when the respiratory diagnoses are excluded. Even for married men, the exclusion of these diagnoses removes most of the relationship between children and sickness absence. The only coecient remaining signi®cant is for three or more children. To the extent that having small children increases the probability of sickness absence at all, this e€ect seems to be largely con®ned to absence due to respiratory conditions.

12 An individual receives a score of 1 if she has at least one absence spell within any diagnostic category except respiratory and 0 if she has no absence spell or one or more absence spells classi®ed as respiratory. 13 Full results are available from the author.

Full-time and part-time work We tested the hypothesis that the relationship between children and sickness absence is stronger for full-time than for part-time employees by carrying out separate analyses for women with short part-time, long part-time and full-time work13. None of the coecients for the number of children dummies had coecients that varied signi®cantly with working hours (if a 0.01 level of signi®cance and a two-sided test are chosen). E€ects by level of education As noted above, it has been suggested that the combination of children and paid employment may be more problematic for lower class women. Separate analyses for married women with low, medium and high levels of education, however, revealed no clear tendency for the association between children and sickness absence to either decline or increase with the level of education13. None of the regression coecients vary signi®cantly among the educational categories. There is no evidence that the combination of employment and children is more problematic for women with low socio-economic status.

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

1839

Table 5 Logistic regression of sickness absence on the number and age of children and controls. Absence due to respiratory conditions excluded. Married women and men. 1995 data. Regression coecients with standard errors in parenthesesa Married women

Married men

Intercept Time of observation 1 child 2 children 3 or more children Age of youngest child Child born in 1994 Child born in 1995 Age/10 Age2/100

ÿ0.8430 0.0035 0.1013 ÿ0.0888 ÿ0.0411 ÿ0.0118 ÿ0.6191 0.8523 ÿ0.1320 0.0602

ÿ1.2198 0.0032 0.0551 0.0675 0.1349 ÿ0.0057 ÿ0.1194 ÿ0.0705 ÿ0.0446 0.0792

ÿ2 log likelihood N

5411.721 46680

a

(0.0289) (0.0003) (0.0512) (0.0436) (0.0477) (0.0039) (0.0711) (0.0569) (0.0169) (0.0140)

(0.0338) (0.0004) (0.0537) (0.0456) (0.0479) (0.0042) (0.0575) (0.0595) (0.0192) (0.0159)

46801.51 48189

See note to Table 3.

Discussion The above analyses show that the relationship between children and sickness absence is generally weak for married and cohabiting women as well as for previously married women living without a new partner. It seems very unlikely that the low association between these variables is due to measurement or sampling errors. We have reason to believe that the absence registration system is highly reliable, particularly since it is partly used to administer the paid sick leave. Moreover the samples are very large, and sampling error should be of minor importance. Although the weak association between children and sickness absence is not in line with either the role enhancement or the con¯ict/overload hypotheses, it is, as noted above, consistent with empirical research using several measures of morbidity. A weak association does not necessarily mean that parenthood has no e€ect on health. It is also possible that parenthood have both negative and positive e€ects, for instance both role enhancement and role overload e€ects. In the aggregate these e€ects may cancel each other out, if positive e€ects are dominant in one part of the population and negative e€ects in another. However, positive and negative e€ects may also coexist in the same individual. The contribution of our analysis is to show that the net e€ect of these possibly opposing processes is close to zero. We have found the relationship between children and sickness absence to be very similar for married men and married women. On the face of it this suggests that the underlying causal processes are also similar. Both mothers and fathers of small children

have a considerably higher level of sickness absence due to respiratory conditions, but otherwise the relationship between children and sickness absence is very weak. There seems to be no strong theoretical reason to expect role con¯ict or role overload to a€ect speci®cally the probability of respiratory diseases. It is more likely that the high level of respiratory conditions among parents is due to the contagiousness of many such diseases, and to parents being infected through their children who attend schools or kindergartens. As far as causal interpretations are concerned, an important limitation of the analyses must be considered: like most previous research we have not been able to take selection e€ects into account. If those who have the greatest problems combining child care and employment leave the labour force or refrain from having children (or have fewer children), a positive e€ect of children on sickness absence will be underestimated (and a negative e€ect will be exaggerated). There are some indications in the results that selection e€ects could be present. Most important, the probability of sickness absence tends to decrease when the number of children increases beyond one. This would happen if women who have problems combining children and employment either refrained from having a second child, or, if they had a second child, left the labour force. However, the lower level of sickness absence among married women with two or more children could also be due to learning or adaptation. One piece of evidence against the hypothesis of selection e€ects is that the relationship between children and sickness absence is about the same at both high and low levels of education. Since women with a high level of education are more likely to remain in the

1840

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

labour force when they have children, the selection e€ects should be weaker here than among the low educated. A counter argument would be that the selection e€ects are really stronger among persons with low level of education, but that the e€ects of combining childcare and employment are also stronger here (since jobs for people with low education are more dicult to combine with children). This being the case, these e€ects could counterbalance one another. The ®ndings for single women are quite complex. For previously married single women the relationship between children and the probability of sickness absence is weak and not signi®cantly di€erent from the relationship between these variables in the married subpopulation. For single never married women the probability of sickness absence increases markedly from zero to one or two children. (The estimated probability of sickness absence is very low for single never married women with three or more children, but this category is small and the results are very imprecise.) This could be interpreted as evidence that children increase the probability of sickness absence in this group of women. It is worth noting, however, that single mothers with one or two children do not have a higher level of sickness absence than married or cohabiting women with the same number of children. Thus it is not so much that children increases the probability of absence among the single never married, but rather that the absence of children among the single never married is associated with a particularly low absence probability. One interpretation of this is that any sort of family role (spouse, cohabitant or mother) is associated with a somewhat higher probability of absence. Whether the woman has one or two of these family roles (e.g., only spouse or both spouse and mother) does not, however, seem to be important. To some extent, this seems consistent with a role overload or role con¯ict hypothesis. A slightly di€erent interpretation is that the work role may become especially important for a woman who is neither spouse, cohabitant nor mother, and this may reduce her probability of absence. Before concluding that marital/cohabiting status roles a€ect sickness absence, however, the possibility of di€erential selection must be considered even here. Such selection could be health based, but it could also be based on more psychological characteristics, like work commitment. Health related selection does not seem to provide a very convincing account of the di€erences among the marital/cohabiting status categories. To judge from the literature on marriage selection, the never married should be a negatively rather than positively selected group (Goldman, 1993). It seems more likely that never married childless women are positively selected in terms of work commitment. Some support for a selection argument is also provided

by the very high level of sickness absence among young married women. It seems intuitively reasonable that women who marry in their early twenties are particularly family oriented and correspondingly less work oriented. From the point of view of the selection argument it is also important to note that the absence of family roles does not necessarily decrease the probability of sickness absence: in most age groups single previously married women without children have at least as high a level of sickness absence as the married. We noted earlier that the relationship between parenthood and sickness absence might well di€er considerably from one country to another. One factor behind the weak relationship between these variables in our data could be the high female labour force participation in Norway. An examination of this issue must, however, await the availability of appropriate cross-national data.

Conclusion Generally speaking the association between the number of children as well as their age and sickness absence seem to be weak. This is particularly true for married women. Also, the relationship between children and sickness absence is quite similar for both married men and married women. Apart from absence due to respiratory conditions married people with children do not have higher levels of sickness absence than married people without children. Neither among women cohabiting without marriage nor among single women who have been previously married is there any notable relationship between children and sickness absence. Among single women who have not been married previously, on the other hand, there is a clear positive association between children and sickness absence. However, this is due to a low level of sickness absence among single never married women without children and not to a particularly high level of sickness absence among single never married women with children. In general the ®ndings thus provide little support for either role overload/con¯ict or role enhancement theories. The possibility that the positive and negative e€ects of children assumed by these theories are both present and cancel each other out can not, however, be ruled out. The same applies to selection e€ects. An important ®nding in our research is that marital status is at least as strongly related to sickness absence as is the presence of children in the household. A methodological implication of this is that it is very important to control for marital/cohabiting status in studies of the impact of children on sickness absence (or to perform separate analyses in marital/cohabiting sta-

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

tus groups as has been done in this article). If, for instance, all single women are included in the same regression and no control for marital status is performed, the generally high probability of sickness absence among the previously married (who have a larger number of children) will create a spurious positive relationship between the number of children and sickness absence. Among women, we have found the lowest level of sickness absence among single persons without children. This can be interpreted as evidence that family roles contribute to increased sickness absence. Both the low level of sickness absence in this group and the general lack of even a modest association between children and sickness absence could, however, also be the result of various selection e€ects. This selection could be health based, meaning, for instance, that women with health problems limit their labour force participation and/or their childbearing. It could also be related to psychological characteristics, like work commitment. One interpretation of the particularly low probability of sickness absence among never married women who live alone and have no children is that this group is positively selected in terms of work commitment. A further study of such selection e€ects requires longitudinal data, and should be an important priority in further work on the relationship between family roles and sickness absence.

Acknowledgements This research was supported by a grant from the Norwegian Ministry of Health and Social A€airs. The excellent research assistance of Harald Dale-Olsen is gratefully acknowledged. I am grateful for comments by Hege Torp and the journal's reviewers. Responsibility for errors and interpretations rests solely with the author.

References AÊkerlind, I., Alexanderson, K., Hensing, G., Leijon, M., Bjurulf, P., 1996. Sex-di€erences in sickness absence in relation to parental status. Scandinavian Journal of Social Medicine 24, 27±35. Alexanderson, K., SydsjoÈ, A., Hensing, G., SydsjoÈ, G., Carstensen, J., 1996. Impact of pregnancy on gender di€erences in sickness absence. Scandinavian Journal of Social Medicine 24, 169±176. Arber, S., Gilbert, G.N., Dale, A., 1985. Paid employment and women's health: a bene®t or a source of role strain? Sociology of Health and Illness 7, 375±400. Arber, S., 1991. Class, paid employment and family roles:

1841

Making sense of structural disadvantage, gender and health status. Social Science & Medicine 32, 425±436. Baruch, G.K., Biener, L., Barnett, R.C., 1987. Women and gender in research on work and family stress. American Psychologist 42, 130±136. BjoÈrklund, A., 1991. Vem faÊr sjukpenning? En empirisk analys av sjukfraÊnvarons bestaÈmningsfaktorer. In: Arbetskraft, arbetsmarknad och produktivitet, Expertrapport No. 4 til Produktivitetsdelegationen, Stockholm, pp. 285±299. Bourbonnais, R., Vinet, A., VeÂzina, M., Gingras, S., 1992. Certi®ed sick leave as nonspeci®c morbidity indicator: a case-referent study among nurses. British Journal of Industrial Medicine 49, 673±678. Durkheim, E., 1970. Suicide. Routledge and Kegan Paul, London. Elstad, J.I., 1995. Employment status and women's health: exploring the dynamics. Acta Sociologica 38, 231±249. Fawcett, J., Scheftner, W.A., Clark, D.C., Hedeker, D., Gibbons, R.D., Coryell, W., 1987. Clinical predictors of suicide in parents with major a€ective disorders: a controlled prospective study. American Journal of Psychiatry 144, 35±40. Fitzgibbons, D., Moch, M., 1980. Employee absenteeism: a multivariate analysis with replication. Organizational Behavior and Human Performance 26, 349±372. Gar®nkel, I., McLanahan, S.S., 1986. Single Mothers and their Children. Urban Institute Press, Washington, DC. Garrison, K.R., Muchinsky, P.M., 1977. Attitudinal and biographical predictors of incidental absenteeism. Journal of Vocational Behavior 10, 221±230. Gijsbers van Wijk, C.M.T., Kolk, A.M., van den Bosch, W.J.H.M., van den Hoogen, H.J.M., 1992. Male and female morbidity in general practice: the nature of sex di€erences. Social Science & Medicine 35, 665±678. Goldman, N., 1993. Marriage selection and mortality patterns: inferences and fallacies. Demography 30, 1993. Goode, W.J., 1960. A theory of role strain. American Sociological Review 25, 483±496. Hibbard, J.H., Pope, C.R., 1991. E€ect of domestic and occupational roles on morbidity and mortality. Social Science & Medicine 32, 805±811. Hibbard, J.H., Pope, C.R., 1992. Women's employment, social support and mortality. Women and Health 18, 119± 133. Hibbard, J.H., Pope, C.R., 1993. The quality of social roles as predictors of morbidity and mortality. Social Science & Medicine 36, 217±225. Hochschild, A.R., 1997. The Time Bind. Metropolitan Books, New York. Hùyer, G., Lund, E., 1993. Suicide among women related to number of children in marriage. Archives of General Psychiatry 50, 134±137. Kotler, P., Wingard, D.L., 1989. The e€ect of occupational, marital, and parental roles on mortality: the Alameda County Study. American Journal of Public Health 79, 607±612. Kozak, C.M., Gibbs, J.O., 1979. Dependent children and suicide of married parents. Suicide and Life Threatening Behavior 9, 67±75. Lamberts, H., Wood, M. (Eds.), 1987. International

1842

A. Mastekaasa / Social Science & Medicine 50 (2000) 1827±1842

Classi®cation of Primary Care. Oxford University Press, Oxford. Leigh, J.P., 1983. Sex di€erences in absenteeism. Industrial Relations 22, 349±361. Marks, S.R., 1977. Multiple roles and role strain: some notes on human energy, time and commitment. American Sociological Review 42, 921±936. Marmot, M.G., Feeney, A., Shipley, M., North, F., Syme, S.L., 1995. Sickness absence as a measure of health status: from the UK Whitehall II study. Journal of Epidemiology and Community Health 49, 124±130. Martikainen, P., 1995. Women's employment, marriage, motherhood and mortality: a test of the multiple role and role accumulation hypotheses. Social Science & Medicine 40, 199±212. Mastekaasa, A., 1990. Kjùnnsforskjeller i sykefravñr: betydningen av omsorgsoppgaver og jobbegenskaper. Tidsskrift for Samfunnsforskning 31, 531±554. Mastekaasa, A., 1995. Age variations in the suicide rates and subjective well-being of married and never married persons. Journal of Community and Applied Social Psychology 5, 21±39. Mastekaasa, A., Olsen, K.M., 1998. Gender, absenteeism and job characteristics: a ®xed e€ects approach. Work and Occupations 25, 195±228. Mechanic, D. 1986. Illness behaviour: an overview. In: McHugh, S., Vallis, T.M. (Eds.), Illness Behavior: A Multidisciplinary Model. Plenum Press, New York, pp. 101±110. Messing, K., 1997. Women's occupational health: a critical review and discussion of current issues. Women and Health 25, 39±68. National Insurance Administration (Rikstrygdeverket), 1998. Trygdestatistisk aÊrbok 1997. National Insurance Administration, Oslo. Naylor, J.C., Vincent, N.L., 1959. Predicting female absenteeism. Personnel Psychology 12, 81±84. Nicholson, N., Goodge, P.M., 1976. The in¯uence of social, organizational and biographical factors on female absence. Journal of Management Studies 13, 234±254. Noack, T. 1996. Familieutvikling i demogra®sk perspektiv. In: Brandth, B., Moxnes, K. (Eds.), Familie for tiden. Stabilitet og forandring. Tano Aschehoug, Oslo, pp. 11± 29. Paringer, L., 1983. Women and Absenteeism: Health or Economics? American Economic Review 73, 123±127.

Puranen, B., 1991. Kvinnors ohaÈlsa Ð beror den paÊ dubbelarbete? Socialmedisinsk Tidsskrift 68, 387±399. Ross, C.E., Mirowsky, J., Goldsteen, K., 1990. The impact of the family on health: The decade in review. Journal of Marriage and the Family 52, 1059±1078. Roxburgh, S., 1996. Gender di€erences in work and wellbeing: e€ects of exposure and vulnerability. Journal of Health and Social Behavior 37, 265±277. Scott, D.K., McClellan, E.L., 1990. Gender di€erences in absenteeism. Public Personnel Management 19, 229±252. Sieber, S.D., 1974. Toward a theory of role accumulation. American Sociological Review 39, 567±578. Statistics Norway, 1991. Labour Force Statistics 1990. Statistics Norway, Oslo. Statistics Norway, 1996. Labour Force Statistics 1995. Statistics Norway, Oslo. Strand, K., Wergeland, E., Bjerkedal, T., 1997. Work load, job control and risk of leaving work by sickness certi®cation before delivery, Norway 1989. Scandinavian Journal of Social Medicine 25, 193±201. Svensson, D., 1995. Om koÈnsskillnader i sjukfraÊnvaro. Paper presented at the Meetings of the Scandinavian Sociological Association, Helsinki, June 9±11. Sydsjù, A., Sydsjù, G., Kjessler, B., 1997. Sick leave and social bene®ts during pregnancy Ð a Swedish±Norwegian comparison. Acta Obstetricia et Gynecologica Scandinavica 76, 748±754. Umberson, D., Gove, W.R., 1989. Parenthood and psychological well-being: theory, measurement and stage in the family life course. Journal of Family Issues 10, 440±462. Veevers, J.E., 1973. Parenthood and suicide: an examination of a neglected variable. Social Science & Medicine 7, 135± 144. Vistnes, J.P., 1997. Gender di€erences in days lost from work due to illness. Industrial and Labor Relations Review 50, 304±323. Vogel, J., Kindlund, H., Diderichsen, F., 1992. ArbetsfoÈrhollanden, ohaÈlsa och sjukfraÊnvaro 1975±1989. Rapport No. 78. Statistics Sweden, Stockholm. Waldron, I., Jacobs, J., 1988. E€ects of labor force participation on women's health: new evidence from a longitudinal study. Journal of Occupational Medicine 30, 977±983. Weatherall, R., Joshi, H., Macran, S., 1994. Double burden or double blessing? Employment, motherhood and mortality in the longitudinal study of England and Wales. Social Science & Medicine 38, 285±297.