Trends in women’s labor force participation in Australia: 1984–2002

Available online at www.sciencedirect.com

Social Science Research 37 (2008) 287–310 www.elsevier.com/locate/ssresearch

Trends in women’s labor force participation in Australia: 1984–2002 M.D.R. Evans a

a,*

, Jonathan Kelley

a,b

Departments of Sociology and Resource Economics, University of Nevada-Reno, MS 200 Reno, NV 89557-0042, USA b University of Melbourne, Institute of Applied Economic and Social Research, Australia Available online 27 June 2007

Abstract Women’s workforce participation increased strongly over the 1980s and 1990s, especially among middle aged wives. Multivariate analysis of IsssA data (N = 9412) reveals large compositional changes and a trend for succeeding cohorts of women to work more than their predecessors, but few if any period effects. Among the compositional changes, rising women’s education and falling fertility substantially elevate women’s workforce participation and hours worked. No clear time effects were associated with particular policy initiatives. Importantly, interaction tests suggest that the effects of education and of family situation have not changed over time. Finally, family of origin and religiosity have both direct and indirect effects.  2007 Elsevier Inc. All rights reserved. Keywords: Women’s labor force participation; Female labor supply; Hours worked; Australia; Trends; Family background effects; Cohort effects; Age effects; Price effects; Income effects; Education; Life-course stage; Children; Marital status; Husband’s income

1. Introduction Women’s labor force participation is a fascinating issue for both theoretical and policy reasons. Modernization theory has long predicted the convergence of gender roles, as rapid growth in industrial productivity and consequent strong wage gains have drawn first men and then women out of home and farm production and into work in offices and factories (Blumberg, 1984; Inglehart, 1997). And there is evidence that favorable labor markets do draw women into employment (Cotter et al., 1998). Nonetheless, other evidence suggests that this convergence is more apparent than real: the tempo and intensity of labor force participation are, for most women, still largely governed by family considerations (Hakim, 1998).

*

Corresponding author. E-mail address: [email protected] (M.D.R. Evans).

0049-089X/$ - see front matter  2007 Elsevier Inc. All rights reserved. doi:10.1016/j.ssresearch.2007.01.009

288

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

1.1. Long term trends in Australian women’s workforce engagement 1.1.1. All women In a system familiar from northwestern Europe (Hajnal, 1982; Laslett, 1977), a typical life course for 19th century Australian women was leaving home for a series of posts as servant during adolesence and young adulthood, followed by marriage in their mid or late twenties, whereupon many in turn became employers of young household servants (Larson, 1994). Overall, from the mid 19th century to the mid 20th century, women’s labor force participation held steady at about 30%, the vast majority of participants being unmarried (Jones, 1987). Then after WWII, the higher wages and prestige of office and factory jobs drew Australia’s increasingly highly educated young women into clerical and blue collar work, and domestic service dwindled to a niche market. Since the 1940s, at least 90% of Australian women held some paid employment during their early 20s, with recent cohorts working longer hours and more years (Santow and Bracher, 1994). Over the second half of the twentieth century, Australia’s pattern is in the middle of the developed countries, with women’s employment rates lower than the US, for example, but higher than the Netherlands (Jones, 1993; Kempeneers and Lelievre, 1993). 1.1.2. Marital status and women’s workforce involvement In 1950, only about 10% of Australian wives had paid jobs, but that rose sharply between the mid-1950s and the early 1970s; the increase slowed, or even stalled, in the late 1970s and early 1980s (Santow, 1990). Married women’s labor force participation then climbed very gradually over the 1980s and 1990s, passing 60% by 2000 (Evans, 2003). Fig. 1 shows the trends for 1982–2002. By contrast, single women’s employment held constant or declined in the early postwar period (Bracher, 1990; Santow, 1991). Then, in the 1980s, unmarried women’s labor force participation began to climb again, albeit more slowly than did wives’ participation. Throughout the early 1980s, unmarried women’s participation rates exceeded those of wives by 10 or 11 percentage points, age standardized (Evans, 2003). The participation gap then declined to 6 or 7 percentage points through the late 1980s and early 1990s, shrank to 4 or 5 percentage points in the middle and late 1990s, and finally dwindled to 2 points.1 1.1.3. The institutional setting 1.1.3.1. Legal framework. In Australia, substantial disincentives to wives’ labor force participation were enshrined in law and custom until well after WWII. For example, job contracts in government and large private companies often specified lower pay to women than to men in the same job and termination of employment upon marriage. Then, in the late 1960s to early 1970s, a series of judicial initiatives and legislative decisions steadily removed the legal disincentives to women’s employment, culminating in a framework mandating equal employment opportunity and equal pay for equal work.2 Good estimates indicate that, by the 1980s, the reality was nearing the legal ideal: only a small wage gap separated men and women net of productivity-related factors (Marks and Fleming, 1998), despite persistent occupational segregation (Hayes, 1991). 1.1.3.2. Educational system. The postwar period has also been a time of erratically rapid educational expansion in Australia. Early in the postwar period most girls left school after year 8 (Kelley, 2001), so that wives’ potential wages were low relative to the value of their homemaking, and the prestige of the jobs they could get was, on average, lower than that of being a housewife. Subsequent expansion of the educational system and education’s continuing close connection with job quality and pay (Broom et al., 1980; Evans and Kelley, 2002a), make being a full-time homemaker ‘‘cost’’ today’s young women much more than in prior cohorts.

1

The trends would be more distinct if cohabiting women were classed with single women, but the Australian Government data group them this way in keeping with policies to minimize welfare payments. 2 See (Young, 1989) for a detailed assessment of policies affecting women’s employment in the postwar period in Australia; compare (Dex and Shaw, 1986) on Britain and the USA.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Married or co-habitating

289

Not married

70

Percent in labour force

65 60 55 50 45 40 35

02

00

20

98

20

96

19

94

19

92

19

90

19

19

88 19

86 19

84 19

19

82

30

Year Fig. 1. Women’s labour force participation by marital status, adjusted by direct standardization to a uniform age distribution, for women aged 25–64, 1982–2002. Source: Evans 2003 for 1982–2001; for 2002 Australian Bureau of Statistics, 2003. ‘‘Table 01. Labour force status by social marital status, age and sex, series A92221V, A92257W, A92293F, A92329W, A92365F, A92233C,A92269F, A92305C, A92341L, A92377R.’’ 6291.0.55.001 Labour Force, Australia. 1900.5 Free Sample Data—Labour Force Excel Spreadsheets. Available electronically at www.abs.gov.au. Note: 2002 are for January–November as the December data are not yet released; other years are annual averages.

1.2. Prior research 1.2.1. Time This paper focuses on time and women’s labor force engagement. In particular, we investigate (1) to what degree the rise in women’s labor force participation reflects period influences affecting women of all ages to the same extent at particular times, (2) to what degree on-going changes in socialization experiences have continually ratcheted up workforce engagement by launching successive cohorts of young women on increasingly job-oriented ways of life, and (3) to what degree there are mixed effects. To better understand the socioeconomic processes generating cohort and period differences one should measure the posited processes directly rather than relying on catchall proxies such as date of birth and year of survey (Firebaugh, 1989), so where possible, we include substantive measures. For example, family background characteristics are clearly cohort variables, since they do not change after one has grown up, and education is substantially a cohort characteristic (although there is some adult education). By contrast, marital status and fertility involve period, cohort and age components, so it is simpler to interpret them as ‘‘mixed’’ influences (Ono, 1999). After netting out these measured influences, we also examine the residual effects of age, year of survey (period effects), and year of birth (cohort effects). Our models involving these three terms achieve identification by using broad categories of age—in effect, imposing equality constraints on year-by-year coefficients within the categories. Prior research shows that broad dummy variables for age capture its effects on women’s labor force participation best (Evans and Kelley, 2001). 1.2.2. Family background Prior research showing family background to have very important influences in labor-market-related domains such as education (Crook, 1995) and occupational status (Evans and Kelley, 2002a; Hayes, 1991), encourages us to explore its effects on women’s workforce involvement. Accordingly, we investigate the effects of mother’s education and work history, father’s occupational status, immigrant parents, parental divorce,

290

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

having a step parent, childhood urban residence, parents’ age at respondent’s birth, parents’ religious denomination, and religiosity. 1.2.3. Effects of education Education encourages married women’s labor force participation in Australia (Evans, 1984; Kelley and Evans, 2002; Santow, 1991) as in many other countries (Kerckhoff, 2001; Rosenfeld, 1996).3 Educational attainments have risen substantially in Australia in the postwar period (Kelley, 2001), thereby shifting more women into the part of the educational distribution where labor force participation is higher. Education draws women into the workforce both by inculcating more career-oriented attitudes and by enhancing potential wages (Evans, 1988), but we will not attempt a decomposition here, focusing instead on the magnitude of the effect, and on whether it is changing over time. 1.2.4. Effects of marital status Marrying per se does not much affect women’s labor force participation in the US: newlyweds build assets before having children, so women’s workforce participation is a standard part of a family strategy (Rosenfeld, 1996). This tends to be true in Australia as well, at least since the middle 1980s (Bracher and Santow, 1990). Accordingly, we anticipate no main effect of being married, and, contrary to the postmaterialist hypothesis that marriage has changed its meaning in ways that encourage female employment (Inglehart, 1997), we anticipate no significant interaction of marriage with time. 1.2.5. Effects of family context Women still substantially curtail their labor force participation while there are babies and toddlers at home in Australia (Bracher and Santow, 1990; Daly, 1990; Evans, 2000), as in other developed countries such as the US (Budig and England, 2001; Rosenfeld, 1996). Accordingly, we expect to find that small children reduce women’s labor force participation, even net of a wide range of socioeconomic variables. School age children also reduce labor force participation, but the effect is smaller (Daly, 1990; Evans, 2000). 2. Data 2.1. Data source The data are from IsssA-Pool (the pooled cross-sections of the International Social Science Surveys/ Australia).4 We mainly use the pooled cross-sections of primary respondents selected at random from the electoral rolls5, together with panel data to assess the reliability of retrospective measures (given in Table 1). The population sampled by the IsssA consists of citizens of Australia who reside at the address which they have provided to the Electoral Office, who can read enough English to answer a self-completion questionnaire, and who are not cognitively impaired.6 For simplicity, we refer to this population as ‘‘Australians’’. The IsssA surveys are sent by post, individually addressed by name, to simple random samples of Australian citizens drawn by the Electoral Commission from the compulsory electoral rolls (which are public documents) using a minor modification of Dillman’s Total Response Method (Dillman, 1993). Details on the survey‘s fieldwork and data preparation procedures are in (Evans and Kelley, 2002b; Kelley et al., 1999). Comparison with the census shows that samples are representative (Bean, 1991; Sikora, 1997). Some respondents are re-contacted over time, so we use robust standard errors to correct for clustering.

3

Student status temporarily depresses labour force involvement, but the effect seems to be entirely transitory (Blossfeld and Huinink, 1991; Hoem, 1986; Santow and Bracher, 1994). 4 The IsssA is conducted by the International Survey Centre under the auspices of the Melbourne Institute of Applied Economic and Social Research at the University of Melbourne. 5 The IsssA also collects data on primary respondents’ siblings, but they are not used in this paper. 6 The selection on citizenship should have little effect: prior research shows that non-citizen immigrants differ from citizen immigrants principally in their duration of residence, with few or no differences in variables relevant to this paper, namely marital status and stratification characteristics (Evans, 1988).

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

291

Table 1 Measurement Concept (mnemonic, if different) Dependent variables Labor force participation (LFP) Hours worked Family background Father’s occupational status Mother’s education Parents’age Parents divorced

Step parent Number of siblings Immigrant Second generation Mother worked (MomWorked) Parents’church going (InParentsChurchGo) Catholic childhood Anglican childhood Christian belief (Transcendental Belief) Size of place at age 14 (UrbanChildhood)

SES and life cyde Size of place now (UrbanNow) Education

Single mother [2] (SoloMom) Young married, no children (WedNoKids) Married, children under 5 (HasPreSchooler)

Details

Test–retest reliability (over 5 years; N > 1040)

Mean (N > 9432)

LFP: 1 = engaged in any paidwork or unemployed and looking for work, 0 = everyone else. Timeframe: week of the survey # hours respondent normally works weekly in her current job (even if temporarily away at survey date). Range: 0–80

[1]

0.58

[1]

17.8

Worldwide status scores (Evans and Kelley, 2002b: 305–337). Range: 0–100

r = .81

40

# years of education completed by respondents mother. Range: 0–20. Some immigrants’ parents had little or no schooling Mean of direct questions about mother’s and father’s ages when respondent was born A direct question. The reference category is non-divorced parents—mostly intact married couples, but including widowed and never married parents (the latter have increased recently but were rare among our sample’s parents) A direct question about who was in the household: 1 = step parent present, 0 = all other A direct question. Range: 0–8 A direct question on birthplace: 1 = born overseas, 0 = born in Australia Coded from dired questions on birthplace of self and parents. 1 = Australian born and at least one parent born overseas, 0 = all other Scale averaging three direct questions: whether respondents mother worked full-time (1), parttime (0.5) or not at all (0) when respondent was a preschooler, age 6–10, and about 14 Log of number of services attended per year (0 set to 0.5 before logging)

r = .73

8.8

r = .99

30

r = .74

0.07

r = .68

0.07

r = .93 (N = 837) r = .89 r = .93

3.0 0.21 0.06

r = .81

0.26

r = .84

1.77

From a direct question on parents’ denomination (in same surveys, by mother’s denomination): 1 = catholic, 0 = everyone else From the same direct question as ParertsCathdic. Anglicans are similar to American Episcopalians, 1 = Anglican, 0 = everyone else 4-Item scale covering belief in God, heaven, hell, and the afterlife (Kelley, Evans, and Headey, 1993). Range: 100 = Strondy believe all, 0 = reject all

r = .87

0.21

r = .78

0.23

r = .85 (N = 826; 7 years)

60

Natural log of the population of the place where respondent lived at age 14: ‘‘Farm or property’’, a ‘‘Village (under 1000)’’; ‘‘Town (to 20,000)’’; ‘‘Midsized city (to 100,000)’’; ‘‘City (to 500,000)’’; Metropolitan (500,000+) Farm is scored as 10, the others are coded to the midpoint the interval [3]

r = .81

56,000 (anti-log)

From a direct question, is measured and scored in the same way as UrbanChiIdhood [3] From a series of questions on respondent’s years of primary and secondary schooling and details on highest educational qualification. These were coded into the Australian Bureau of Statistics’ 3 digit educational code and then recoded into usual years of Schooling Derived from questions on marriage and number and ages of children: 1 = non-married mother, 0 = all others Same source as SoloMom 1 = married, under age 40, no children yet; 0 = all others (Note: Some could be pregnant; the data lacks information on this.) Same source as SoloMom 1 = married, has at least 1 child under age 5; 0 = all others

r = .78

101,000 (anti-log) 11.3

r = .87

Marital status reliability r = .66 [4]

0.06

[see above]

0.13

0.05

(continued on next page)

292

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Table 1 (continued) Concept (mnemonic, if different)

Details

Test–retest reliability (over 5 years; N > 1040)

Mean (N > 9432)

Married, school age children (HasFledgling) Married, children grown (EmptyNest) Husband’s workforce participation(SoLFP) Husband’s income (Sp$1k)

Same source as SoloMom 1 = married, has at least 1 child age 5–18; 0 = all others Same source as SoloMom 1 = married, over age 40, all children over age 18, 0 = all others A direct question. 1 = husband working full-time, part-time, or unemployed; 0 = all others A direct question. Husband’s income from all sources last year in thousands of dollars

[see above]

0.20

[see above]

0.28

[1]

0.67

[1]

29

A direct question on date of birth. The surveys were conducted from 1984 to 2002, so age and cohort are not colinear, as they would be in a single survey A direct question on date of birth. The surveys were conducted from 1984 to 2002, so age and cohort are not colinear, as they would be in a single survey Same source as age. 1 = age 45–54, 0 = all other Same source as Age. 1 = age 55–64, 0 = all other A series of dummy variables representing survey year. Identification issues involving year born, age, and year of survey are dealt within text

r = .998

1948

r = .998

43

[see above] [see above] r = 1.0

0.23 0.18 1992

Age and time Cohort (YearBorn)

Age, linear

Age 45–54 Age 55–64 Year of survey

[1] Varies month-by-month, so our 5 year latest-retest is not appropriate. [2] Prior research suggests that cohabitor behave more like singles than like married people in Australia (Khoo, 1987) as in the US. (Rindfuss and Vanden Heuvel, 1990), so we treat them as unmarried in the analysis. To assess the sensitivity of the specification, we estimated an alternative model grouping them with the married Cohabitors are only 5% of the sample and categorizing them with the single or the married does not influence the slope coefficients (all differences ns at p < .05, Appendix Table A, at www.internationalsurvey.org), so their location is not a particularly consequential decision. For consistency with the behavioral research, we group them with the singles. [3] We use the In functional form because it has a stronger correlation with the dependent variables than does raw size. [4] This relatively low reliability is found in many surveys. Some of the apparent unreliability is genuine change. Source: International Social Science Survey/Australia, pooled file 1984–2002.

2.2. Sample restrictions This analysis is based on the IsssA-Pool’s 9412 women aged 25–64. The restriction to women under age 65 is because labor force participation rates at ages 65–69 have never exceeded 5% (Kelley and Evans, 2002), so that labor force involvement is no longer an issue.7 We omitted women under age 25 to ensure that those in the analysis have had time to complete their education. It is well known in the status attainment literature that including 18- to 24-year-olds could seriously bias estimates of education’s causes and consequences, which are important here.8 2.3. Measurement Table 1 gives the information on variable measurement.

7 There is an argument for including older women nonetheless, since ‘‘0 hours worked’’ is a perfectly good answer and the larger sample size would give greater precision in estimates. The offsetting disadvantage is due to social change: older cohorts have rather different attitudes, values, and life histories in a number of relevant ways. Those would need to be modeled. The uncertainty introduced by that would, we think, more than offset any gains in precision from the larger sample size. 8 There is still some residual bias for the youngest groups because some complete tertiary qualifications as adults, after age 25.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

293

3. Methods and models 3.1. Statistical significance Since the IsssA-Pool database provides a large sample of women—over 9000 cases for most analyses—we use the conservative .001 level as the criterion of statistical significance. 3.2. Basic models This paper uses recursive models. The basic model (Table 2, below)) has the total effects of family background including those that may operate indirectly through respondent’s SES and life-cycle variables

Table 2 Basic model of women’s workforce involvement Panel A Hours worked (OLS)

Family background (Eq. (1)) Father’s occupational status Mother’s education Parents’ age when R born Parents divorced Lived with step parent Number of siblings Migrant: first generation Second generation migrant Mother worked when R young Parents church attendance (In) Catholic Anglican Christian belief Urban at age 14 (In)

Panel B Labor force participation (Logistic regression)

b

SE b

Std.

t

b

SE b

z

Exp(b)

.04 .56 ns ns ns ns ns ns 4.75 ns ns ns .04 ns

.01 .10 ns ns ns ns ns ns .82 ns ns ns .01 ns

.05 .08 ns ns ns ns ns ns .08 ns ns ns .06 ns

3.32 5.68 ns ns ns ns ns ns 5.78 ns ns ns 4.55 ns

.01 .07 ns ns ns ns ns ns .55 ns .24 ns 0.004 ns

.00 .01 ns ns ns ns ns ns .09 ns .08 ns .00 ns

4.6 6.2 ns ns ns ns ns ns 5.9 ns 3.2 ns 3.6 ns

1.01 1.07 ns ns ns ns ns ns 1.73 ns 1.28 ns 0.996 ns

SES end life-cycle (Eq. (2)) Urban resident (In) Education Never married (reference) Single mother Young married, no children Married, children under 5 Married, school age children Married, children grown Husband employed Husband’s income

.30 1.30

.07 .10

.06 .19

4.17 13.50

.04 .16

.01 .01

4.8 12.2

1.05 1.18

7.53 ns 16.84 5.60 2.69 3.61 .05

1.05 ns .79 .75 .70 .58 .01

.09 ns .30 .12 .06 .09 .09

7.19 ns 21.32 7.43 3.85 6.25 6.08

.79 ns 2.12 .60 .44 .74 .01

.13 ns .12 .11 .10 .08 .00

6.2 ns 18.0 5.5 4.4 9.3 4.9

.45 ns .12 .55 .65 2.10 .99

Age and time Year born (cohort or vintage) Age: 45–54 (dummy) Age: 55–64 (dummy) (R-squared; psuedo-R-squared)

.18 .04 .11 ns ns ns 9.41 1.13 .19 (2% for family background; 20% for

4.76 ns 8.30 full model)

.03 .01 6.3 ns ns ns 1.11 .15 7.42 (3% for family background; 23% for

1.03 ns .33 full model)

OLS regression on hours worked for pay; logistic regression on labor force participation. Total effects with Huber-White robust standard errorsa Women, age 25–64. Australia 1984–2002. N = 9412. ns, not statistically significant at p < .001, two-tailed. a Family background is assured to be causally prior, so total effects include both direct effects and indirect effects through other variables in the model. For the other variables there are no indirect effects, so total effects are the same as direct effects. Further details are available on our web site. Huber-White robust standard errors adjust for non-independence among panel respondents, who make up 32% of the cases. Source: International Social Science Survey/Australia, pooled file 1984–2002.

294

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

HoursWorked or logitðLFP Þ ¼ b0 þ b1 FathersOccupationalStatus þ b2 MothersEducation þ b3 ParentsAge þ b4 ParentsDivorced þ b5 StepParent þ b6 NumberOfSiblings þ b7 Immigrant þ b8 SecondGeneration þ b9 MomWorked þ b10 lnParentsChurchAttendance þ b11 ParentsCatholic þ b12 ParentsAnglican þ b13 TranscendentalBelief þ b14 UrbanChildhood þ e1 The effects of SES and life-cycle variables are

ð1Þ

9

HoursWorked or logitðLFP Þ ¼ Eq:ð1Þ þ b15 UrbanNow þ b16 Education þ b17 SoloMum þ b18 WedNoKids þ b19 HasPreSchooler þ b20 HasFledgling þ b21 EmptyNest þ b22 SpLFP þ b23 Sp$1k þ b24 YearBorn þ b25 Age4554 þ b26 Age5564 þ e2

ð2Þ

Where e1 and e2 are random error terms with the usual properties. Estimation. For labor force participation, a dichotomous variable, estimates are by logistic regression. Hours worked, a continuous variable, is heavily heaped on zero, which might create statistical problems. However Tobit results, which cater for such possibilities, are virtually indistinguishable from OLS (Appendix Table B, available at www.international-survey.org). We therefore prefer OLS on the grounds of simplicity and robustness. 3.3. Age, period and cohort effects: identification issues The model of Fig. 3, below, extends the age and cohort (year born) variables by adding dummy variables for each survey year, so capturing all unmeasured period effects HoursWorked or logitðLFP Þ ¼ Eq:ð2Þ þ dummy variables for year of survey þ e3

ð3Þ

The age–period–cohort model is identified because of the restrictive age specification with two broad age dummies (suggested by prior research on age patterns of labor force participation). The dummies effectively constrain most cohort-by-cohort coefficients to be equal, an identification strategy recommended by Pullum (1978), who also makes the key point (1978: 116–117) that the age–period–cohort problem is equivalent to stratification’s ‘‘mobility effects’’ (for example, Kelley, 1992). The logic of our approach follows Kelley (1992); see Appendix Table D (at www.international-survey.org). The model is clearly over-identified. 3.4. Time interactions We also extended Eq. (2) by including multiplicative interaction terms with year of survey to assess whether the impacts of key variables vary in different times. For example, for education HoursWorked or logitðLFP Þ ¼ f ðvariables of Eq:ð2Þ;Education  Year of Survey; e4 Þ

ð4Þ

This allows the effect of education to be larger, the same, or smaller in recent years than it was in the past. We also test other interactions in the same way. Results are in Appendix Table G (available at www.international-survey.org). None is statistically significant. 3.5. Model for projection forward in time To draw out the substantive implications of the model, we use it in projections forward 20 years. To allow for the possibility that the future could bring different opportunities and constraints for subgroups of women 9

Since there are no intervening variables here, total effects and direct effects are the same.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

295

in different life course situations, we re-estimated the model separately for (1) young, single, childless women, (2) young childless wives, (3) young wives with preschool children, (4) wives with children of school age, (5) early middle aged wives with grown children, (6) wives about age 50 with grown children and (7) wives around age 60 with grown children. For married women, the husband’s employment situation is set to currently working and his income is $25,000 annually. The projection for young unmarried women (‘‘singles’’), for example, is based on the analogue of Eq. (1) estimated for single women only. We dropped variables that did not apply to them (for example, husband’s income). (1) To get adjusted figures for 2002, we then applied the equation to everyone, single and married, to get predicted values for the entire population. This whole population standardization (Kelley and Evans, 1995) adjusts for compositional differences between single and other women in family background, SES, and age.10 (2) To get adjusted figures for 20 years in the future, in 2022, we assume that this same equation holds but adjust year born forward by 20 years. We also model on-going changes in education and mother working, projecting them forward by 20 years on the basis of simple OLS regressions.11 We then apply the equation to everyone in this adjusted population. The projections for other groups, for example mothers with young children, are obtained analogously from Eq. (1) estimated for them. Note that predicted values from these seven models differ somewhat from the pooled model of Eq. (2) because they allow the effects of the independent variables to vary for different life course situations, rather than imposing a common slope for all. This has advantages (especially flexibility) and disadvantages (especially due to the smaller samples). Reassuringly, in practice the differences are rarely significant (see Appendix Table H, available at www.international-survey.org). On balance, we believe the best estimates for the effects of most independent variables are from the pooled model of Eq. (2) but the best projections forward in time come from the seven separate models. 4. Results This section presents the estimates of influences on women’s workforce engagement—both participation (lfp) and hours worked. The general expectation is that they will show broadly the same results, but with more significant results evident in the more powerful OLS regression analysis of hours worked. Table 2 presents the results of our basic model. These estimates assume no reciprocal effects. For many of the variables here, temporal ordering and basic logic make clear that there could be no reciprocal causation (for example, it is nonsense to think of an adult’s hours worked as potentially causing her father’s occupational status decades before). For variables potentially subject to reciprocal causation, the estimates shown here provide an upper bound on the true effect.12 4.1. Family background Family background has some enduring effects on women’s hours worked for pay (left panel of Table 2 and Fig. 2) and on labor force participation (right panel). Daughters of well-educated mothers work substantially more. For example, a college graduate’s daughter of could be expected to work over 2 h more weekly than a high school graduate’s daughter. Similarly, daughters of high status fathers also work more. For example, a professional’s daughter is likely to work about 4 h more weekly than a farm laborer’s daughter. Almost all of these family background effects come about indirectly because they lead the daughter to get more education herself and that, in turn, leads her to work more. Maternal role models also matter. The daughters of persistently working mothers will themselves work about 5 h more than homemakers’ daughters. About half of this comes about indirectly because they get more 10 We used Stata 9’s predict function. Because maximum likelihood estimates are sometimes unstable when there are many irrelevant variables, in the probit analysis of labor force participation we restricted the independent variables to those with statistically significant effects (but including YearBorn whether or not significant). 11 Specifically, we replace education by = 0.088313 * YearBorn + 7.1016 and mother working by 0.006016 * YearBorn  .0281. 12 This assumes that reciprocal effect would have the same sign, as seems reasonable in this case.

296

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310 FAMILY BACKGROUND: Father's occupational status .00 .05 Mother's education .01

.07

Parents' age when R born More hours direct

Parents divorced

More hours, indirect

Lived with step parent

Fewer hours, direct

Number of siblings

Fewer hours, indirect

Migrant: First generation Second generation migrant Mother worked when R young

.04

.04

Parents church attendance (ln) Catholic Anglican Christian belief .02 .03 Urban at age 14 (ln) .02 .04 SES AND LIFE-CYCLE: Urban resident (ln)

.06 .19

Education Never married (reference) Single mother

.09

Young married, no children Married, children under 5

.30

Married, school age children Married, children grown

.12 .06

Husband employed

.09

Husband's income

.09

AGE AND TIME: Year born

.11

Age: 45 to 54 (dummy) Age: 55 to 64 (dummy) 0.00

.19 0.10

0.20

0.30

Importance: Standardized effect .

Fig. 2. Influences on hours worked: Standardized effects.

education and somewhat different life course patterns. Half comes about directly, probably through socialization into different skills and attitudes. The only other family background effects are from Christian belief (staunch atheists work about 4 h more a week than strong believers) and urban origins (its direct effect increases work but its indirect effects decrease it). These patterns are essentially identical for hours worked and for labor force participation. This suggests that the decision about hours of work is not a separate step from the decision about participation. Instead of a sequence of decisions, women appear to simultaneously consider whether to participate and how much to work. Further analysis shows that these effects are the same for women’s work at different ages, except that religiosity only reduces work involvement for younger women (Appendix Table E, at www.internationalsurvey.org).

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

297

4.2. Education Education substantially increases women’s work involvement, as expected: the standardized regression coefficient is .19 (Table 2). Each additional year of education increases a woman’s probability of participating in the labor market by 18% and increases her hours worked by 1.3 h a week, on average and all else equal. So, for example, a university honors graduate (16 years of education) can be expected to work about 5 h a week longer than an otherwise comparable woman who only finished high school. Here again, we find parallel results for labor force participation and for hours worked. Further analysis shows that the effects of education do not differ significantly between younger and older women (Appendix Table E, at www.international-survey.org). 4.3. Husbands Wives of husbands who are in the workforce are more likely to participate in the workforce themselves, and work longer hours, to boot. The model provides only upper bound estimates on these effects, because some of the causal influence is likely to flow in the other direction. Specifically, everything else being equal wives of employed men are twice as likely to be in the work force (Table 2, panel B) and they work for pay 4 h more per week (Table 2, panel A). Further analysis suggests that this effect is restricted to older women: it amounts to 6 h a week for women over 45 and is not significant for their juniors (Appendix Table E, at www.international-survey.org). Note, moreover, that 72% of the husbands who are not working are retired. So the effect probably largely represents a joint retirement decision. By contrast, husband’s income is associated with very slightly lower levels of wife’s labor force participation and hours worked, all else equal. Specifically, each extra $1000 the husband earns reduces the probability of his wife’s being in the labor force by a fraction of 1%13, and reduces his wife’s weekly hours of employment by only 3 min14, ceteris paribus. This effect is the same for wives of all ages. The direction of this effect is consistent with economists’ standard arguments about ‘‘income effects’’ but is much smaller than their arguments suggest.15 Overall, although statistically significant, husband’s income and employment are of only of modest importance to wives’ labor force engagement, with standardized regression coefficients a little under .10. 4.4. Stage in the life-cycle Life-cycle stage is influential in some respects, but not in others, with young children the crucial distinction. For these estimates, on balance we prefer the more flexible model of Table 3, based on separate estimates for each life-cycle stage, to the more parsimonious model of Table 2 based on data pooled over life-cycle stages. Details are in the methods section. The comparison (or reference) group for these effects is never married, childless women. According to these estimates, unmarried, childless women would have a labor force participation rate of 78% in 2002. If prevailing trends continue, we project that this would rise sharply to 95% in 2022 (Table 3, column 1). Hours worked weekly would also rise sharply, from 25 to 34, according to these same assumptions. Single mothers work rather less (column 2).16 However, if present trends continue, we project that their work involvement will not change greatly over the next few decades, in sharp contrast to single women’s. So by 2022 single mothers will lag well behind. 13

Recall that these are upper bounds, because of the possibility of causal influence flowing in the opposite direction. More exactly, .06 * 60 = 3.6 min. 15 Because well-educated women tend to marry prosperous men, education’s direct and indirect effects are in opposite directions: welleducated women tend to work more but their husband’s prosperity reduces that to some degree. Our rough calculations suggest this suppressor effect is quantitatively small: each year of wife’s education implies an extra $2400 in husband’s income. That income in turn leads to a reduction of wife’s hours worked of 4 min per week for each $1000 of husband’s income. So a college education (versus only high school) leads to a reduction of 15 min worked per week (an ‘‘income’’ effect, as economists put it). This is small compared to a college education’s ‘‘price’’ effect (as economists would call it) of 5 h and 12 min. The same is true for the US lately (Cohen and Suzanne, 1999). 16 Even in this large sample, there were not enough single mothers to enable reliable estimation of differences according to the ages of the children. 14

298

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Table 3 Predicted values for women’s labor force participation and hours worked per week by life-cycle stage, for women average in family background, religion, place of residence and other respectsa Age and life-cycle stage Single Single mother Panel A: now in 2002 Adjusted labor .78 force participation Adjusted hours 25 worked

Young married, Young married, no children children under 5

Married, children in school

Married, children past school

Age 55+, married, children past school

.68

.70

.39

.64

.71

.59

21

26

11

18

21

15

.76

.44

.72

.94

.89

27

12

20

34

24

Panel B: projected for year 2022 Predicted labor .95 .72 force participation Predicted hours 34 19 worked

Source: International Social Science Survey/Australia, pooled file 1984–2002. a Assumes that after marriage the husband works and earns $25,000 per year. Young are those under 40. Current trends in education and in mother’s employment are projected forward. Other variables are assumed to remain unchanged through 2020. Details on methods are given in the text.

Young women who are married but as yet childless do not now greatly differ from unmarried women in their chances of being in the workforce or in how many hours they supply to the labor market each week, all else equal (Table 2). Thus, there is no evidence that marriage per se has any effect on women’s workforce engagement. However, if present trends continue, we estimate that young childless wives’ work involvement will not change much in the next two decades, unlike single women’s (Table 3). Thus by 2022, they will be working much less than single women. Married women with preschool children are much less likely to be in the workforce and work 14 or 15 fewer hours than do childless women, ceteris paribus (Table 3). This is the largest single influence in the model: its standardized regression coefficient, .30, is half again as large as the next most important (.20 for education; see the pooled model in Fig. 2). According to our projections, their labor force involvement will hardly change in the next few decades. So by 2022, they will be even more distinctive than they are now. Older children make less difference to their mother’s work involvement. The reduction in hours worked by mothers of school age children (compared to childless women), 7 h, is only about half of the reduction induced by preschoolers (Table 3). This is a moderately important effect in the pooled model, with a standardized regression coefficient larger than those for husband’s employment and income, but smaller than the effects of education or young children (Fig. 2). According to our projections, their work force involvement will increase only slightly in the next few decades, leaving them even further behind single women. Grown children have little effect on women’s work involvement. Married women under age 55 who have grown children now work almost as much as single women, on average and all else being equal. According to our projections, their work involvement—like that of single women—can be expected to increase sharply in the next few decades (Table 3). By 2022, 94% would be in the labor force, working 34 h a week—fully as much as single women. As aging continues, work involvement declines. Mature age women over 55 start to retire, working an average of only 15 h a week, less than middle aged mothers and far less than single women. But this will change dramatically in the next few decades. According to our projections, hours worked will increase more than half by 2022. That will make mature age one of the more active stages in the life-cycle. 4.5. Age The life-course stage differences just described account for most apparent age differences, but not quite all. Compared to younger women, those aged 55–64 work 9 h less on average, all else equal (see the pooled model

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

299

in Table 2), as expected from prior research. The standardized regression coefficient of .19 makes it the 2nd or 3rd most important determinant of hours worked: less important than being married with preschool age children; about as important as education; more important than year of birth; more important than husband’s income; more important than husband’s employment; more important than having school age children; more important than having grown children; more important than being married or not; more important than urban residence; and more important than family background or religion (Fig. 2). 4.6. Cohort (year of birth) Different currents seem to be pulling time effects in different directions (Table 2). Having just discussed the age effects, we now focus on birth cohort or ‘‘vintage’’ effects. Note that these are residual effects of time, because several influential cohort effects—mother’s education and work history—are already controlled in the model, as are rising education (largely a cohort effect), and marriage and children (blends of period and cohort effects). A key finding in the pooled model of Table 2 is the strong positive effect of year of birth on labor force participation and on hours worked.17 The coefficient suggests that each year later in the 20th century that a women was born raises her weekly hours worked by about 11 min18 weekly, on average and all else equal, with a moderately large standardized regression coefficient of .11. Thus, For example, women born in 1980 would be expected to work about 5.5 h a week more than the prior generation born about 1950. Colinearity. Age and year of birth are not perfectly correlated (because the surveys were done over an 18 year period) but the correlation is high, r = .91. So there is some uncertainty due to colinearity (Appendix Table C at www.international-survey.org). A baseline model has all our standard variables except for age and year born. Adding year born to the baseline gives a statistically highly significant effect, substantially increasing the percent of variance explained. Alternatively, adding age (both the two dummies) also shows a highly significant effect, mainly due to the dummies. Comparing models shows that the year of birth effect is roughly halved by the age control, but remains large and statistically significant. Age effects are little changed by controlling for year born. Importantly, the effects of family background and SES variables are much the same so long as either age or year of birth, or both, are controlled. This is evidence that our main conclusions are robust. 4.7. Period (Year of survey) In addition to age and cohort effects, there might be unmeasured period effects—notably the changing policy climate. We modeled these by a full set of dummy variables for each survey year, with 1984 being the reference category (Eq. (3)). These age–period–cohort effects are identified for the reasons set out in the methods section. The results are shown in Fig. 3 (details are in Appendix Table F available at www.internationalsurvey.org). There is no monotonic trend in hours worked weekly, but period effects are not entirely absent (F = 9.25, 11 df, p < .0001). Compared to 1984, there are only two individually significant effects, a high in 1989 and a low in 1990. Many policy initiatives were intended to encourage women’s workforce engagement over this period, but the evidence does not reveal any changes at the times that might have been expected.19

17

Eligibility for the government age pension varies (for women) from age 60 for older cohorts to age 65 for younger, with a gradual transition for those born after July 1935 and before January 1949. Unfortunately for the data analyst, age of eligibility is correlated r = .91 with year born. The high correlation between the two variables make it impossible reliably to separate their effects even in this large sample, so age at eligibility had to be omitted from the analysis. 18 More exactly, each successive birth cohort has been working .27 of an hour per week more that the prior cohort throughout their lives, all else equal. 19 For example, in 1996, the government introduced subsidies to defray day care center costs which were not available to housewife mothers taking care of their own children.

300

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Additional hours worked weekly

Hours worked compared to 1984 10

5

0

-5

-10

)

ce

n re

e ef

(R

86 987 989 990 993 994 995 6.5 999 001 002 9 1 1 1 2 1 2 1 1 1 19

19

84

19

Year

Fig. 3. Changes in hours worked, compared to 1984.

4.8. Interactions with time Time could operate more subtly—not as a sweeping effect that shapes the destinies of women in all walks of life in the same ways and to the same degree, but rather as a social force that has differential effects on women with different socioeconomic resources, or in different stages of life. To test for such possibilities, we augmented the model of Eq. (1) and Table 2 with a new set of interaction variables representing the multiplicative interactions of time with many of the focal variables. The model is Eq. (4), results are in Appendix Table G, at international-survey.org. In the event, not one of these interactions has a statistically significant effect at p < .001. But even more strikingly, inclusion of the entire group of them does not significantly improve the fit of the original model (F = 3.27, 7 df, ns). The absence of significant interactions here suggests that time has not had differential effects on different groups of women. Thus the effects of education and life-course stage have not substantially altered over this 18-year period, net of other influences. 5. Discussion 5.1. Summary of effects Family background has some enduring effects on women’s hours worked for pay and on labor force participation. Daughters of well-educated mothers work substantially more, as do daughters of high status fathers. Almost all of these advantages come about indirectly because they lead the daughter to get more education herself and that, in turn, leads her to work more. Maternal role models also matter: the daughters of working mothers themselves work more. About half of this comes about indirectly and half directly. The only other substantial family background effect is that devout Christians work less. Most analyses of work force engagement are based on census and survey data that lack information on family background. But because the effects of family background are almost entirely indirect, their omission will in general not bias the coefficients of education and life-cycle characteristics, but one should be careful in interpreting these coefficients. The only exception is maternal role models which does have a small direct effect. As it is correlated with respondent’s education and SES, that creates a small omitted variable bias in conventional analyses. Education greatly increases women’s labor force participation and hours worked. The steady increase in education is the source of a good deal of the rise in married women’s labor force participation since 1984. Importantly, there is no significant decline in education’s influence over time.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

301

Being married, on its own, does not significantly affect women’s degree of workforce engagement. Husband’s income reduces women’s workforce engagement, as predicted by economists’ standard arguments about ‘‘income effects’’, but the impact is much smaller than economists’ arguments suggest. Because well-educated women tend to marry prosperous men, education’s direct and indirect effects are in opposite directions: welleducated women tend to work substantially more but their husband’s prosperity reduces that by a few minutes a week. By contrast, husband’s employment increases women’s workforce engagement, all else equal. This probably represents a joint retirement decision: three-quarters of the husbands who are not working are retired. When the husband retires, the wife is likely to retire too; conversely, while the husband is still working, the wife is more likely to work too. Even in this large database, there are not enough cases reliably estimate the effect of husband’s unemployment. It should be born in mind is that formal marriage may not be an exogenous process if women and men select marriage partners in ways that are unmeasured by the survey but are relevant to the labor market. If so, then the apparent effects of husband’s employment and income could really reflect the unmeasured preexisting differences rather than having any intrinsic influence of their own.20 The presence of young children greatly reduces women’s labor force participation and work intensity. Moreover, there is no significant decline in children’s effect over time. Other influences—the gradual increase over the generations in workforce engagement, higher levels of education, and the like—mean that the mothers of young children today are more deeply engaged in the workforce than were the mothers of young children two decades ago. But if our models are correct, this is because the levels of the other influences have changed, not because the impact of young children has changed. Moreover, today’s mothers are more likely to have only one child, which also elevates their participation rates compared to mothers in the 1980s and early 1990s. Family variables’ effects do not change over time, a striking finding. There is the illusion of change in the impact of family, if one merely compares workforce engagement of mothers today with mothers of 15 or 20 years ago. But that is misleading, because the change is not limited to the mothers, but is common to the cohorts as a whole. The family effects within cohorts remain. Of course, this could alter in the future, and warrants continued monitoring. Time matters in some ways, but not others. The apparent influence of age is mainly illusory, due to lifecycle differences. The exception is that women over age 55 work less on average. There were no evident period (year of survey) effects associated with particular policy initiatives, nor indeed any clear period trends at all. By contrast, cohort (year of birth) effects are strong and clear. Throughout their working lives, each succeeding cohort has a slightly higher probability of being in the workforce and works longer hours. Finding such strong cohort effects is particularly interesting because individual women’s work force participation is very dynamic over the life course, so one might well have expected period effects to predominate. On the contrary, these findings suggest that underneath the high turnover and frequent shifts in work hours are strong cohort effects. Note that these cohort increases in employment-intensity are above and beyond those due to education and to changes in family background over time. They reflect some other social forces that are not measured in our models: discovering what those are could greatly improve forecasts as well as advancing knowledge.

5.2. Projections for the future Our projections suggest that over the next 20 years there will be two quite different patterns of change in women’s work involvement. When children are not a constraint—that is, for single women and married women whose children are past school age—women’s work involvement will increase substantially. Hours worked, for example, will increase by about 50%. In contrast, when there are young children mother’s work involvement will not change appreciably.21 Hours worked may rise or fall by about 10%. Thus if our projections are cor20 This issue could be addressed in future research via a two-pronged strategy: first using fixed effects models in panel data to attempt to control for unmeasured pre-existing difference; and, second, expanding the range of potentially relevant characteristics that are measured in new survey data. 21 We assume most young women who are married but childless expect soon to have children, and adjust their labor force plans accordingly. The average gap between marriage and first birth is only 2 or 3 years.

302

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

rect, the long historical trend toward greater labor force involvement will end for mothers with young children, while continuing for women before and after that stage in the life-cycle.

Acknowledgments This research was supported by the Australian Commonwealth Government’s Department of Family and Community Services’s (FaCS’s) SPRC Grant #026 with the Melbourne Institute of Applied Economic and Social Research. The interpretations expressed herein are not necessarily endorsed by the sponsor. Combining the IsssA surveys to make the IsssA-Pool dataset was supported by the Australian Research Council’s RIEF Grant R19920093.

Appendix A See Appendix Tables A, B, C, D, E, F, G, H. Appendix Table A Alternative treatment of cohabitation makes no significant difference to the results: OLS predicting hours worked weekly, for women only; direct effects Variable

Marriage and life-cycle Never married (reference) Single mother Young married, no children Married, children under 5 Married, school age children Married, children grown Other variables Father’s occupational status Mother’s education Parents’ age when R born Parents divorced Lived with step parent Number of siblings Migrant: first generation Second generation migrant Mother worked when R young Parents church attendance (In) Catholic Anglican Christian belief Urban at age 14 (In) Urban resident (In) Education Husband employed Husband’s income Age: 45–54 Age: 55–64 Year born Year of survey (Constant)

Panel A: cohabit = married

Panel B: cohabit = single

b

SE b

Beta

b

0.95 0.83 0.68 0.58 0.54

.09 .03 .31 .11 .04

— 7.91 1.02 17.63 5.69 2.25

0.00 0.01 0.10 0.07 0.03 0.03 0.62 0.75 0.20 0.72 0.04 0.10 0.12 0.44 1.30 0.75 2.43 0.58 0.26 0.12 0.19 0.48 0.32 0.45 0.02 0.01 0.21 0.05 0.27 0.05 1.35 0.07 3.03 0.48 0.05 0.01 0.32 0.65 7.74 0.97 0.26 0.04 0.18 0.04 489.30 71.25 Adjusted R-squared

.00 .01 .01 .01 .00 .00 .00 .02 .04 .02 .00 .01 .03 .04 .05 .20 .07 .09 .01 .16 .16 .05

— 8.11 2.25 16.96 4.92 1.70

.20

Difference in b (t statistic)

SE b

Beta

0.83 0.91 0.68 0.58 0.54

.10 .01 .32 .12 .05

0.16 0.99 0.69 0.95 0.72

ns ns ns ns ns

0.00 0.01 0.09 0.07 0.04 0.03 0.72 0.75 0.20 0.72 0.02 0.10 0.09 0.44 1.30 0.75 2.41 0.58 0.28 0.12 0.20 0.48 0.36 0.45 0.02 0.01 0.21 0.05 0.28 0.05 1.34 0.07 3.31 0.48 0.05 0.01 0.13 0.65 7.87 0.97 0.25 0.04 0.18 0.04 481.97 70.64 Adjusted R-squared

.00 .01 .01 .01 .00 .00 .00 .02 .04 .02 .00 .01 .03 .04 .05 .20 .08 .09 .00 .16 .15 .05

0.07 0.05 0.01 0.10 0.00 0.13 0.05 0.00 0.03 0.11 0.01 0.07 0.09 0.01 0.07 0.11 0.41 0.44 0.21 0.10 0.07 0.01

ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns

.20

Source: International Social Science Survey/Australia, pooled file 1984–2002. Panel A: cohabitors are treated as married. Panel B: cohabitors are treated as single. Australia, 1984–2002; N = 9433.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

303

Appendix Table B Tobit and OLS estimates of influences on hours worked; direct effects Tobit

Family background Father’s occupational status Mother’s education Parents’ age when R born Parents divorced Lived with step parent Number of siblings Migrant: first generation Second generation migrant Mother worked when R young Parents church attendance (In) Catholic Anglican Christian belief Urban at age 14 (In)

OLS

b

z

b

t

ns ns ns ns ns ns ns ns 4.09 ns ns ns ns 0.28

ns ns ns ns ns ns ns ns 4.23 ns ns ns ns 3.09

ns ns ns ns ns ns ns ns 2.45 ns ns ns ns 0.20

ns ns ns ns ns ns ns ns 4.24 ns ns ns ns 3.65

SES and life-cycle Urban resident (In) Education Never married (reference) Single mother Young married, no children Married, children under 5 Married, school age children Married, children grown Husband employed Husband’s income

0.51 2.19

5.40 17.69

0.30 1.30

5.38 17.77

10.75 ns 27.90 7.92 5.12 8.26 0.08

7.90 ns 24.24 8.30 5.42 10.09 7.43

7.53 ns 16.84 5.60 2.69 3.61 0.05

9.14 ns 25.16 9.69 4.97 7.59 8.21

Age and time Year born Age (years) Age: 45–54 (dummy) Age: 55–64 (dummy)

0.40 ns ns 20.44

5.85 ns ns 11.09

0.18 ns ns 9.43

4.38 ns ns 8.66

Only effects significant at p < .001 are shown. The correlation between Tobit predicted values and OLS predicted values is r = .98. Australia, 1984–2002; women only. N = 9433. Source: International Social Science Survey/Australia, pooled file 1984–2002.

304

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Appendix Table C Sensitivity of age and cohort specifications

Alternative models predicting number of hour worked weekly. OLS estimates; direct effects. Australia, 1984–2002; women only. N = 9433. *Grayed-out figures are for metric coefficients that are not significantly different from zero at p < .001. Source: International Social Science Survey/Australia, pooled file 1984–2002. a The correlation between age and year born is r = .915.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

305

Appendix Table D Identification of age–period–cohort effects Panel 1 Identify effects of date born and survey year (i.e. cohort and period effects) from all the data in green. This is a standard 2 variable problem. The effects could be linear, or curvilinear, or there could be a separate dummy variable for each row and each column (among other possibilities). Even with row and column variables (12 + 9 = 21 of them), the model is heavily over-identified (there are 12 * 9 = 108 data points, less 10 dropped because of the age 65 restriction, so 98 data points to estimate only 21 parameters). Survey year (period) 1960 1965

1970

1975

1980

1985

1990

1995

2000

Date born: 1920 (cohort) 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975

Panel 2 Now add two dummy variables, one for all the cells in yellow below (one parameter) and another for the cells in blue (another parameter). These are the age effects. They are just shift factors moving the diagonal cells above (or below) the level implied by the row and column effects. The model is still firmly over-identified, with 23 parameters to estimate and 98 data points. Survey year (period) 1960 1965 Date born: (cohort)

1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975

1970

1975

1980

1985

1990

1995

2000

306

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Appendix Table E Models predicting number of hours worked weekly separately for women under age 45 and those over age 45

OLS estimates: direct effects. Australia 1984–2002. *Grayed-out figures are for metric coefficients that are not significantly different from zero at p < 001. Source: International Social Science Survey/Australia, pooled file 1984–2002.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

307

Appendix Table F Changes over time in hours worked Year of survey

Robust b

1984 (reference) .0 s1986 2.1 s1987 2.4 s1989 2.5 s1990 4.7 s1993 .6 s1994 .6 s1995 .5 s1996 .7 s1999 1.4 s2001 .6 s2002 2.8 Joint significance test for differences over time: Increase in R-squared: .008 Controlled but not shown: all variables in the model of Table 1.

Significance SE

t

— .86 .88 .75 1.06 .88 .97 .88 .86 .97 1.04 .99

— — 2.40 ns 2.74 ns 3.38 p < 001 4.49 p < 001 .70 ns .60 ns .61 ns .80 ns 1.43 ns .60 ns 2.89 ns F = 9.25, 11 df, p < .0001

Australian women, age 25–64, in 1984–2002. N = 9412 (The model is that of Table 1, first panel, with year of survey measured by a set of dummy (indicator) variables, with 1984 the reference category Robust Huber-White standard errors. R-squared: .208.). ns, not statistically significant at p < .001, two-tailed. Source: International Social Science Survey/Australia pooled file 1984–2002.

Appendix Table G Changes over time in the impact of key determinants of women’s workforce involvement

Interaction terms Education · year of survey Single mother · year of survey Young married, no children · year Married, children under 5 · year of survey Married, school age children · year of survey Married, children grown · year of survey Age: 55–64 · year of survey Controlled but not shown: all variables in Eq. 2 Year of survey (1930 = 0) Joint significance test: F = 3.27, 7 df, ns Change in R-squared: .002

b

SE

t

ns ns ns ns ns ns ns

ns ns ns ns ns ns ns

0.20 2.92 1.22 2.54 1.52 2.96 0.81

ns

ns

1.78

ns, not statistically significant at p < .001 two-tailed. Source: International Social Science Survey/Australia, pooled file 1964–2002. OLS regression on hours worked for pay with multiplicative interactions. Women, age 25–64. Australia 1984–2002. N = 9412.

308

Appendix Table H Projection equations for predicted hours worked now and in 2003 (used in Table 3)

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

*Grayed-out figures are for metric coefficients that are not significantly different from those for single women at p < .001. Source: International Social Science Survey/Australia, pooled file 1984–2002. OLS for each group separately; direct effects. Coefficients in bold face type are significantly different from zero at p < .05.

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

309

References Bean, C.S., 1991. Comparison of National Social Science Survey with the 1986 Census. NSSS Report 2:12–19. Available at www.international-survey.org. Blossfeld, H.P., Huinink, J., 1991. Human-capital investments or norms of role transition – how Women’s schooling and career affect the process of family formation. American Journal of Sociology 97, 143–168. Blumberg, R.L., 1984. A general theory of gender stratification. Sociological Theory 2, 23–101. Bracher, M., 1990. Explaining first marriage trends in Australia. Journal of the Australian Population Association 7, 128–150. Bracher, M., Santow, M.G., 1990. The family histories of Australian women. European Journal of Population 6, 227–255. Broom, L., Jones, F.L., McDonnell, P., Williams, T., 1980. The Inheritance of Inequality. Routledge & Kegan Paul, London. Budig, M.J., England, P., 2001. The wage penalty for motherhood. American Sociological Review 66, 204–225. Cohen, P.N., Binachi, S.M., 1999. Marriage, children, and women’s employment: what do we know? Monthly Labor Review 122, 22–31. Cotter, D.A., DeFiore, J., Hermsen, J.M., Kowalewski, B.M., Vanneman, R., 1998. The demand for female labor. American Journal of Sociology 103, 1673–1712. Crook, C.J., 1995. The role of mothers in the educational and status attainment of Australian men and women. Australian and New Zealand Journal of Sociology 31, 45–73. Daly, A., 1990. Women in the workforce and family structure in Australia. Journal of the Australian Population Association 7, 25–39. Dex, S., Shaw, L.B., 1986. British and American Women at Work: Do Equal Employment Opportunities Matter? Macmillan, Basingstoke. Dillman, D.A., 1993. The design and administration of mail surveys. Annual Review of Sociology 17, 225–249. Evans, M.D.R., Kelley, J., 2002a. Family background and occupational success in Australia. Australian Social Monitor 5, 73–78. Evans, M.D.R., 1988. Working wives in Australia: influences of the life cycle, education, and feminist ideology. In: Bean, C. (Ed.), Australian Attitudes: Social and Political Analyses from the National Social Science Survey. Allen & Unwin, Melbourne, pp. 147–162. Evans, M.D.R., 1984. Immigrant women in Australia—resources, family, and work. International Migration Review 18, 1063–1090. Evans, M.D.R., 2000. Women’s participation in the labour force: ideals and behaviour. Australian Social Monitor 3, 49–57. Evans, M.D.R., 2003. Women’s employment trends, 1982–2002, by marital status. Australian Social Monitor 6, 16–17. Evans, M.D.R., Kelley, J., 2001. Gender and employment biographies in Australia. Australian Social Monitor 3, 103–105. Evans, M.D.R., Kelley, J., 2002b. Australian Economy and Society 2001: Education, Work, and Welfare. Federation Press, Sydney. Firebaugh, G., 1989. Methods for estimating cohort replacement effects. Sociological Methodology 19, 243–262. Hajnal, J., 1982. Two kinds of preindustrial family formation systems. Population and Development Review 8, 449–494. Hakim, C., 1998. Developing a sociology for the twenty-first century: preference theory. British Journal of Sociology 49, 137–143. Hayes, B.C., 1991. Intergenerational mobility among employed and non-employed women: the Australian case. Australian and New Zealand Journal of Sociology 26, 368–388. Hoem, 1986. The impact of education on modern family-union initiation. European Journal of Population 2, 113–133. Inglehart, R., 1997. Modernization and Postmodernization. Princeton University Press, Princeton. Jones, F.L., 1987. Occupational statistics revisited: the female labour force in early British and Australian censuses. Australian Economic History Review 27, 56–76. Jones, G.W., 1993. Is demographic uniformity inevitable? Journal of the Australian Population Association 10, 1–17. Kelley, J., 1992. Methods and pitfalls in the analysis of social mobility: class of origin, class of destination, and mobility per se. In: Turner, F.C. (Ed.), Social Mobility and Political Attitudes: Comparative Perspectives. Transaction Press, New Brunswick and London, pp. 233–251. Kelley, J., 2001. Changing educational attainments of women and men in twentieth century Australia. Australian Social Monitor 4, 63–66. Kelley, J., Evans, M.D.R., 2002. Women’s employment in middle and old age. Australian Social Monitor 5, 39–51. Kelley, J., Evans, M.D.R., 1995. Class and class conflict in six western nations. American Sociological Review 60, 157–178. Kelley, J., Evans, M.D.R., 1999. Australian and international survey data for multivariate analysis: the IsssA. The Australian Economic Review 32, 298–303. Kempeneers, M., Lelievre, E., 1993. Women’s work in the EC. European Journal of Population 9, 77–92. Kerckhoff, A.C., 2001. Education and social stratification processes in comparative perspective. Sociology of Education 74, 3–18. Khoo, S.-E., 1987. Living together as married: a profile of de facto couples in Australia. Journal of Marriage and the Family 49, 185–191. Larson, A., 1994. Growing Up in Melbourne: Family Life in the Late Nineteenth Century. Family Formation Project Monograph Number 12. Canberra: Department of Demography, Australian National University. Laslett, P., 1977. Characteristics of the western family considered over time. Journal of Family History 2, 89–116. Marks, G. N., Fleming, N., 1998. Youth Earnings in Australia: 1980–1994: A Comparison of Three Youth Cohorts. LSAY Research Report 8. Ono, H., 1999. Historical time and US marital dissolution. Social Forces 77, 969–999. Pullum, Thomas W., 1978. Parametrizing age, period, and cohort effects: an application to U.S. delinquency rates, 1964–1973. Sociological Methodology 9, 116–140. Rindfuss, R.R., Vanden Heuvel, A., 1990. Cohabitation: a precursor to marriage or an alternative to being single? Population and Development Review 16, 703–726. Rosenfeld, R.A., 1996. Women’s work histories. Population and Development Review 22, 199–222. Santow, M.G., 1990. The family histories of Australian women. European Journal of Population 6, 227–255.

310

M.D.R. Evans, J. Kelley / Social Science Research 37 (2008) 287–310

Santow, M.G., 1991. Work and family in the lives of australian women. In: Becker, H.A. (Ed.), Life Histories and Generations. University of Utrecht Press, Utrecht, pp. 89–114. Santow, M.G., Bracher, M., 1994. Change and continuity in the formation of first marital unions in Australia. Population Studies 48, 475– 496. Sikora, J., 1997. International Survey of Economic Attitudes in Australia, Finland and Poland: Comparison with the Census. WwA: Worldwide Attitudes 1997.12.31:1–8. Available at www.international-survey.org. Young, C.M., 1989. Life cycle experience of cohorts in the evolution of female labour force participation in Australia. In: Pope, D., Alston, L. (Eds.), In Australia’s Greatest Asset. The Federation Press, Sydney.