Trajectories of Unpaid Labour and the Probability of Employment Precarity and Labour Force Detachment Among Prime Working-Age Australian Women

Worldwide, women are over-represented in precarious and insecure employment arrangements. Importantly, the high unpaid labour demands women experience over the life course compromise paid labour force participation for women. This study explores the way different trajectories of time spent in unpaid labour throughout women’s prime working and child-rearing years (from baseline age of 25–35 yrs to 42–52 yrs) are associated with indicators of precarious employment and labour force detachment later in life. We applied group-based trajectory modelling to 17 waves (2002–2018) of data from the Household Income and Labour Dynamics in Australia survey to identify trajectories in unpaid labour. We then examined associations between these estimated trajectories and employment outcomes in wave 19 (2019). Our study shows that chronic exposure to high amounts of unpaid labour over prime working-age years (compared to lower exposure levels) increases women’s probability of precarious employment and labour force detachment later in prime working life. This provides evidence that ongoing inequity in the division of unpaid labour has considerable long-term implications for gender inequality in the paid labour force, and underscores the importance of urgently addressing how men and women share and prioritise time across both paid and unpaid labour domains.


Background
Employment precarity has been on the rise in high-income countries since the turn of the millennium (Kalleberg & Vallas, 2018).Importantly, women are over-represented when it comes to precarious employment (Vosko et al., 2009).This is partly attributable to "non-standard"1 work becoming increasingly normative across many female-dominated employment sectors such as care, retail, hospitality, education, and health (Menendez et al., 2007).However, it is also likely that women's disproportionate responsibility for unpaid labour plays a significant part (Organisation for Economic Cooperation and Development (OECD), 2017).High unpaid labour demands over the life course create and perpetuate paid labour force participation compromises and penalties for women (Charmes, 2019).One such penalty is women's increased exposure to greater employment precarity and labour force detachment (Ferrant et al., 2014).Importantly, these exposures not only have lasting economic consequences, but are also linked to poorer health and wellbeing outcomes (Benach et al., 2014;McKee-Ryan et al., 2005).
The purpose of this paper is to focus on this under-researched topic and, using Australian data, examine how chronic exposure to unpaid labour over women's prime working years predicts precarious employment and labour force detachment later in working life.
There is no universal definition or measurement of precarious employment (PE) (Kreshpaj et al., 2020).Nonetheless, PE is generally accepted to be a multidimensional and complex concept, involving elements of insecurity, powerlessness, lack of rights and protection, and opportunity loss (Kreshpaj et al., 2020;Vosko et al., 2009).In conceptualising what constitutes PE, it is important to consider the standard employment relationship, defined as a "stable, socially protected, full-time job" (Vosko et al., 2009).This is important as it represents the employment norm (historically underpinned by the malebreadwinner model and informing the concept of the "ideal worker"), established to protect against forms of precariousness, and is taken by many to be the benchmark for assessing PE (Vosko et al., 2009).Thus, with full-time permanent (secure) work (with all its associated protections and benefits) as the yardstick, possible indicators of PE include part-time work, casual work, insecure work, and work lacking social benefits (Kreshpaj et al., 2020;Vosko et al., 2009).Low income and low job control have also been rationalised as indicators of PE (Benach et al., 2014).Despite uniform acceptance that PE is a multidimensional construct, a key limitation hindering much of the extant evidence is the use of singular or unidimensional measures of employment precarity (Benach et al., 2016;Utzet et al., 2020).Examining one-dimensional measures of PE in isolation not only ignores other important PE considerations, but also potentially misclassifies individuals whilst simultaneously stymieing any holistic or true understanding of the breadth of how PE is experienced.Whilst prior constructs of multi-dimensional PE do exist (Lewchuk, 2017;Milner et al., 2018;Vives et al., 2015), work to date has lacked a gender lens.This study aims to begin addressing this gap, recognising that utility of multiple indicators and levels of analysis is strongly encouraged for future PE research, as is accounting for emerging perspectives on how the dimensions of PE may be measured in a gender sensitive way (Valero et al., 2020;Vosko et al., 2009;Young, 2010).Lastly, we also consider that labour force detachment (LFD) to be a more extreme (particularly gendered) form of PE, where people of prime working-age (usually women) are temporarily or permanently detached from the paid work force due to unpaid care and domestic responsibilities.
Unpaid labour (UL) exerts enormous influence over women's lives across the globe.Variation exists in how UL (or unpaid domestic work) is conceptualised and measured.Nonetheless, it is broadly acknowledged to be inclusive of all responsibilities and tasks

Conceptual Frameworks Linking Unpaid Labour with Employment Precarity
The gendered division of labour is a dynamic construct that, whilst operationalised at the couple and household level, has complex individual, community, workplace, and national levers (Jung & O'Brien, 2017).In brief, three main theoretical perspectives have been historically recognised.Described in detail elsewhere (Baxter & Tai, 2015;Brines, 1994;Deutsch & Gaunt, 2020), these are; time availability (allocation based on each partners hours spent in paid work within household), economic exchange/bargaining (related to power and relative resources/income) and gender display (also known as "doing gender" where women and men conform to performing socially constructed gendered behaviours).More recently, however, these theoretical approaches have been challenged (Geist & Ruppanner, 2018).Given the assumptions about gender and families upon which they are based, contemporary sociologists have described them as insufficient to account for the division of UL in modern-day households (Geist & Ruppanner, 2018).Of relevance to our study is the critique of the time availability perspective and its assumptions, whereby the likely reciprocal relationship (or reverse causality) in the relationship between UL and relative hours of paid work has largely been ignored.Likewise, economic exchange/bargaining power in households also hinges on time availability for paid work, which we argue is restricted by unpaid work in the first instance.
Whilst the dualisation of the workforce, and the gendered nature of part-time work has been documented extensively in the sociological literature, synthesising this discourse with theories of PE is relatively nascent (Rubery et al., 2022;Samtleben & Müller, 2021).This is an oversight in the existing PE research because PE is inherently gendered (Ferrant et al., 2014;Gray et al., 2021;Vosko et al., 2009), and is underpinned by many of the same gendered processes found in the dualisation research, but these processes have largely been ignored.Patriarchal structures, power constructs and how these shape divisions of household labour heavily influence women's susceptibility to PE (Menendez et al., 2007).As such, acknowledging that PE is inextricably interwoven with inequities in the gendered division of UL is key, as is appreciating that male-breadwinner ideology continues to channel women into under/unemployment and non-standard work (Menendez et al., 2007;Vosko et al., 2009).Notably, whilst the male-breadwinner model has been largely replaced by the "one and a half earner" variant (full-time male worker and part-time female) in Australia, the theoretical underpinning of the key drivers of gendered PE remain the same (Pocock et al., 2013;Vosko et al., 2009).Whilst much has been written regarding the pervasive influence of UL (as a construct) on labour force participation and elements of PE (such as part-time work and casualisation) (Baxter & Tai, 2015;Vosko et al., 2009), scant empirical research exists drawing these two gender equality dimensions together.Our study is novel in empirically examining these highly gendered and inter-related theoretical constructs in combination, interrogating the effects of women's UL on their future employment precarity and attachment.

Prior Evidence
To our knowledge, no studies have empirically examined the association between UL and PE specifically, nevertheless a small body of research has examined unpaid care/ work and labour market attachment or employment more broadly.For example, a longitudinal Finnish study examined labour market trajectories in cohabiting couples with children (Peutere et al., 2017).Their results indicated that high responsibility for housework (but not childcare) was related to weaker labour force attachment trajectories for both men and women.Moreover, an earlier US longitudinal study demonstrated positive labour force participation effects for women whose male partners perform relatively more of the housework (Cunningham, 2008).Although, neither of these studies captured total UL either conceptually (inclusive of all elements) or quantitively (as a measure of time burden versus division of responsibilities).More recently, a 2021 study of German couples did consider total UL time, reporting that time spent in UL (as well unequal division of UL) had adverse effects on women's labour market participation and working hours (Samtleben & Müller, 2021).
In addition to time caring for children, the provision of informal care to older or disabled/unwell adults also contributes to total UL and is highly gendered.Compared to other dimensions of UL, the association between informal unpaid caregiving and labour force participation has been more extensively examined.A 2007 systematic review gathered the international research (noting that 2/3 of papers pertained specifically to elder care), and reported that caregivers were equally as likely to be in the labour force as non-caregivers, but worked fewer hours (Lilly et al., 2007).This aligns with a more recent Canadian study examining the gendered impact of informal care to elderly relatives on labour force participation, reporting women were > 5 times more likely to work part-time and 73% more likely to leave the labour market due to informal care (Smith et al., 2020).Indeed, it is the pressures and requirements of care provision (children and other) that are most frequently cited as the predominant drivers for reduced labour force participation, or for not being in the labour force at all (Baum & Mitchell, 2010;Lilly et al., 2007;Smith et al., 2020).

Current Study
Importantly, the time burden that UL imposes on women's lives changes over time.Women's UL time is often low in early adulthood (and usually not too disparate from men of a similar age), before typically increasing upon transition to cohabitation with a partner, beforethen rising steeply upon birth of first child.Whilst the average age of motherhood has been incrementally rising over time in Australia, most mothers (nearly 2/3) were aged between 25 and 34 years in 2020 [Australian Institute of Health and Welfare (AIHW), 2022].Thus, by an average age of 30 years, a vast proportion of Australian women experience a significant and often sudden increase to their UL in line with this major life transition.It is well recognised that parenthood induces short term LFD for many new mothers (Baxter, 2013;Kuitto et al., 2019), a phenomenon keenly felt by Australian women, due to weak political will and limited supportive policy.Until 2010, Australia was still one of only two countries in the Organisation for Economic Cooperation and Development (OECD) (alongside the United States) without a national paid parental leave (Baird et al., 2021).Whilst there have been welcome advancements in Australian parental leave policy in recent years, its gendered nature continues to place family care responsibilities firmly with women (Baird et al., 2021).Moreover, upon returning to the workforce, further challenges present themselves.For example, weak and ineffective childcare policy in Australia means that quality childcare is not only unaffordable, but is often difficult to access or of limited availability (Pennington, 2020;Wood, 2020).Such barriers not only disincentivise women returning to work (over 20% of Australian women between 25 and 54 years are not participating in the labour force), but also perpetuate a trend of Australian women not returning to full time employment (almost 50% of employed women worked part-time in 2018), leading to a much lower female labour force participation rate compared to similar high-income countries (Pennington, 2020).Coined the "workforce disincentive rate", it is reported that the proportion of income lost through higher taxes, lost family benefits, and higher childcare costs, financially disadvantages many women working more than three days a week (Wood, 2020).It is also important to note that the heightened UL load a growing family demands does not diminish with re-entry to the workplace, but remains highly gendered, and persists through women's prime working years.Importantly, as per the focus of this paper, persistent exposure to increased UL creates ongoing issues of time poverty, resulting in women commonly trading off paid working hours to meet their high UL demands, consequentially inducing both employment precarity and attachment implications, even in the longer term (Vosko, 2000).Whilst the motherhood penalty for lifetime earnings and income is well documented (Budig & England, 2001;Weeden et al., 2016), the long-term effects of exposure to UL on PE and LFD has not been investigated.This study addresses this gap by utilising group-based trajectory modelling to explore how different trajectories of time spent in UL for women throughout their prime working and child-rearing years (25-54 years) (Organisation for Economic Co-operation and Development (OECD), 2022), are associated with indicators of PE and LFD later in working life.We hypothesise that women with greater exposure to UL over time will have greater odds of PE and LFD later in working life.

Research Aims
Utilising Australian data, the specific aims of this research were to: 1. Identify trajectories of UL for women throughout their prime working years 2. Describe the characteristics of the women most likely to be within each trajectory of UL 3. Examine the association between trajectories of UL for women and future indicators of PE and LFD

Data Source and Sample
Data source was the Household Income and Labour Dynamics in Australia (HILDA) Survey.HILDA is a nationally representative longitudinal study of Australian households that commenced in 2001 with a sample of 13,969 persons from 7682 Australian households (Watson & Wooden, 2021).A top-up sample was added in 2011.Administered in the form of a self-completion questionnaire, HILDA is an omnibus survey that collects data annually on a wide range of topics including family, economic and subjective well-being, and labour market dynamics.This study used data from wave 2 through to wave 19 (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019).
Wave 1 (2001) of HILDA could not be utilised given it does not capture the key unpaid labour exposure variables of interest.Our analytic sample was restricted to women who were between 25 and 35 years of age at wave 2 in 2002 (n = 1381).Given our research focus, this age cohort was selected to capture one of the key UL transition points (parenthood), noting that the average age of all mothers in Australia in 2002 was 29.4 years and 27.6 years for a first birth (Laws PJ, 2004).Seventeen years of annual data (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) from baseline age of 25-35yrs to 42-52yrs informed the UL trajectories.PE and LFD were then measured in 2019 (at wave 19) when the cohort were aged 43-53yrs (final n = 862).While wave 20 was available for this study, this data was not included as it was collected in 2020 during the COVID pandemic and associated lock downs.

Unpaid Labour
The group-based trajectories were based on a measure of total UL at each wave (waves 2-18).This was derived by summing the number of hours and minutes in a typical week spent doing the following unpaid labour tasks: ( 1) housework (such as preparing meals, washing dishes, cleaning house, washing clothes, ironing, and sewing), (2) household errands (such as shopping, banking, paying bills and keeping financial records (but not driving children to school and other activities), (3) caring for own children (playing with own children, helping them with personal care, teaching, coaching or actively supervising them, or getting them to childcare, school, or other activities), (4) caring for other children (looking after other people's children (aged under 12 years) on a regular, unpaid basis), ( 5) caring for a disabled spouse or disabled adult relative or caring for elderly parents or parents-in-law, and (6) outdoor tasks, including home maintenance (repairs, improvements, painting, etc.), car maintenance or repairs, and gardening.Given that time-use data are particularly vulnerable to over-estimation bias, all unpaid labour variables were capped (rounded off) at the 99th percentile (where applicable) to contain extreme outliers.This measure was operationalized as a continuous variable (hours/week).

Precarious Employment and Labour Force Detachment
Our analysis examined five dependent variables (outlined below), each capturing different dimensions of PE and LFD.All were measured at wave 19 and were operationalised as binary variables.Notably, variables b, c, d, and e were only asked of women who were employed/had a job in 2019.a) Labour force attachment (not employed vs employed) Labour status is ascertained annually in the HILDA survey from a labour force status question, classifying participants as employed, unemployed, or not in the labour force/ retired for the year preceding the present interview year.Our sample were aged 42-52 years at outcome wave 19 and therefore some years from official retirement age.Resultantly, if they were classified as not in the labour force (NILF), rather than being retired, they were more likely willing to work but encountering barriers precluding them from starting imminently, or part of the highly gendered "hidden unemployed" group (Metcalf, 2013).For the purposes of examining LFD amongst our cohort, labour status was dichotomised into either employed or not employed (unemployed/hidden unemployed).
b) Part-time employment (part-time vs full-time hours) Part-time versus full-time employment was determined using a derived variable collating the number of hours usually worked per week in all jobs.As per the Australian Bureau of Statistics (ABS) definition (Australian Bureau of Statistics (ABS), 2021a), women who worked 35 hours or more per week (in all jobs) were classified as full-time, while those who work less than 35 hours were classified part-time.
c) Casual employment (casual vs permanent) Casual workers are determined in the HILDA survey using the ABS definition of casual employees, predominantly ascertained via lack of any paid leave entitlements including paid sick leave and paid holiday leave.(Australian Bureau of Statistics (ABS), 2021b).In Australia, casual work also invariably means the worker is offered no firm advance commitment to ongoing work, nor an agreed pattern of work.It was operationalised in our model as a binary variable, casual worker or non-casual (permanent) worker.d) Job security (low vs medium/high) Our measure of job security was a scale derived from 3 items ("secure future in job", "worry about future of job", "company will still exist 5 years from now"), all assessed using a 7-point Likert scale (from strongly agree to strongly disagree).The second item was reverse-coded and therefore higher scores indicated greater job security (Leach et al., 2010).This measure has been validated (Leach et al., 2010), and in line with previous research, scores were summed and divided into tertiles (Karasek et al., 1998;Taouk et al., 2020).For the final dichotomous measure, those in the lowest tertile of job security were considered to have low job security, and medium and high job security were combined as the reference category.
e) Job control (low vs medium/high) Our measure of job control was based on 11 items (assessed using Likert scales) pertaining to skill discretion and decision authority (such as "use many of my skills and abilities" and "freedom to decide how to work" (with full list available in supplementary Table S1) (Leach et al., 2010).High scores represented higher control, and as previously done, scores were summed and divided into tertiles (Karasek et al., 1998;Taouk et al., 2020).Like job security, those in the lowest tertile of the distribution were considered to have the lowest job control, and the remaining two tertiles were combined to create the reference category medium/high job control.

Covariates
We controlled for a range of covariates considered to be plausible common causes (i.e., potential confounders) of UL trajectories and paid employment outcomes amongst prime working-age Australian women.All were measured in Wave 2. These included: country of birth (Australia, elsewhere-English speaking, elsewhere-non-English speaking), education (did not complete high school, completed high school, certificate/trade, diploma, bachelor's degree or higher), physical health (limiting long term health condition/not), household disposable income in quintiles (calculated by summing the income components for the previous financial year for all adults in the household and equivalized using the modified OECD scale) (Australian Bureau of Statistics, 2022), place of residence (city/regional/ remote), and household structure (couple, lone person, with and without children < 15 yrs).Age was also included in the models as a confounder, given the potential ten-year age difference between participants, which is considered significant in prime child-rearing and prime working-age cohorts in the context of UL and employment.

Statistical Analysis
Analysis proceeded in three main stages.Firstly, we used group-based trajectory modelling to identify the different trajectories of UL experienced by women.This was followed by a multinomial logistic regression identifying the different characteristics associated with membership in each trajectory group.In the third analytic stage, we conducted logistic regressions examining the associations between the UL trajectories for women and our five indicators of PE and LFD.

Group-Based Trajectory Modelling
Group-based trajectory modelling (GBTM) was utilised to classify distinct groupings of women with similar UL time/demands over time.As a specialised application of finite mixture modelling, GBTM allows identification of groups of individuals who follow a similar average progression (or trajectory) on a variable of interest over time (Nagin, 2009).GBTM employs maximum likelihood estimation methods, whereby resultant effect estimates remain unbiased in the presence of incomplete data, on the assumption that data is missing at random (Nagin, 2009).The resultant trajectory groups are conceived to be discrete clusters of individuals following similar trajectories that may be present within the population (Nagin, 2009).
Trajectories of UL were identified using the traj user-written command (Jones & Nagin, 2013), in Stata SE version 17 (StataCorp, 2021) and were based on 17 waves of annual data (Waves 2-18) with participant's age at each year/wave used as the underlying time scale (Nagin & Odgers, 2010).The censored normal (cnorm) model was used to estimate our unpaid labour trajectories (given total unpaid labour was a continuous variable).To identify the optimal model, we followed Nagin's recommended approach (Nagin, 2009;Nagin & Odgers, 2010).Firstly, a series of one to five groups were tested to identify the ideal number of latent trajectory groups.Determination of optimal group number was based on the Bayesian Information Criterion (BIC), values of group membership probabilities, trajectory group sizes, and expert opinion of the authors.We then specified different polynomial functions across the resultant groups (testing each possible combination) to determine the optimal shape (guided by BIC) of each trajectory.After ascertaining the optimal trajectory model, women were then assigned to trajectory groups based on their maximum posterior probability of group membership.To further assess model fit, we calculated the average posterior probability of group membership and odds of correct classification (OCC) for each trajectory group.Supplementary Tables S2-5 provide further details and results from the model selection process.

Multinomial Regression Analyses
In the second stage of analysis, multinomial logistic regression modelling was used to characterize the clusters of women following each of the different UL trajectories, according to baseline (wave 2) demographic characteristics.

Logistic Regression
In our final analysis stage, we used the identified trajectory groups as predictors of various employment outcomes in wave 19 (the wave following the end point of the trajectory).To do this, five separate logistic regression models (adjusted for the confounders detailed as covariates above) were utilised, assessing the associations between UL trajectory group membership and our five nominated indicators of PE and LFD in wave 19.

Baseline (Wave 2)
From our baseline analytic sample, all 1,381 women aged between 25 and 35 years of age in 2002 (wave 2) had sufficient data to be included in the UL GBTM groups.GBTM handles missing data by fitting the model using maximum likelihood estimation and assigns an individual's probability of group membership based on their data.Therefore, whilst respondents must have sufficient data (determined by the model) to be included in the GBTM trajectories, some missing UL data over the course of the trajectory (waves 3-18) does not preclude their inclusion.
We also controlled for several baseline covariates at wave 2. This was for the five logistic regression models (analysis stage 3).There was minimal missingness for covariates, with only two variables having any missing data (country of birth and education, both only 0.07% of sample respectively).  1, except for the LFD variable, not all 862 women were captured in each of the PE variables being analysed, with numbers varying depending on the outcome being measured.This was mainly due to employment specific questions only being asked to those employed/in labour force at wave 19, but also due to some minimal missing outcome data for some of the PE variables.The respective n for each of the PE variables being examined are recorded in Tables 1 and 2.

Descriptive Statistics
Table 1 presents the sample characteristics of our cohort.The mean age of women at baseline (wave 2) was 30 years, with over half (58%) reporting being in either a couple with children under the age of 15 (47%), or as a lone parent with children under 15 years (11%).In contrast, 31% reported no children (either couple with no children or lone person).Educational level was evenly distributed across the four categories, with a slightly higher number of women (29%) having a bachelor's degree or above.Of the disposable income quintiles, just over half of the cohort (55%) were in the two highest quintiles.Most women (89%) had no underlying long term health condition, disability, or impairment.Two thirds lived in a major city and 81% were born in Australia.Sample characteristics of wave 19 indicators of PE are also presented in Table 1.At wave 19 (mean age 47), 83% were employed, 41% of whom were working part-time and 16% of whom were casual workers.Approximately a third of the women who were employed reported low job control and low job security.

Trajectory Analysis
A 4-group trajectory model was identified as the best-fitting model, with the optimal shape of each trajectory (polynomial components) found to be quadratic, quartic, quartic, and cubic (2,4,4,3) respectively.Figure 1 shows the four different trajectory groups of UL load (hours/week) as a function of age over time, from a mean age of 30 (2002) to a mean age of 47 (2018).As per Fig. 1, trajectory groups were categorised as persistent low (33% of cohort), decreasing (37%), increasing umbrella (19%), and persistent high (11%).For the persistent low group, UL time remained reasonably low and stable (average at around 20 h/week) for the course of the trajectory.In contrast, the decreasing group started high  (averaging at around 60 h/week) for the first five waves, before gradually decreasing to a mean of just over 30 h/week by wave 18.Whereas the increasing umbrella group started off relatively low (an average of about 30 h/week) before steeply increasing to a mean of almost 80 h/week midway through the trajectory, then starting to fall again (following a falling trajectory for the last 5 or so waves).Notably, this group was still undertaking relatively high amounts of UL (a mean of over 50 h/week) at wave 18. Lastly, the persistent high group are named as such, as whilst the trajectory shows a decreasing trend overall, starting at over 90 h/week and doing over 60 h/week at wave 18, compared to the other groups their time burden remains persistently high over the whole 17 years.Table S6 shows participant characteristics according to trajectory group assignment.Details on model fit are contained in supplementary tables S1-4, however for each group, the mean posterior probability and odds of correct classification respectively were: 0.86 and 11.1 (persistent low), 0.86 and 9.8 (decreasing), 0.88 and 44.0 (increasing umbrella (19%), and 0.90 and 75.0 (persistent high) (Table S5).

Multinomial Regression Analysis: UL Trajectory Characteristics
Table 2 presents the results of the adjusted multinomial models of the probability (relative risk) for women's membership in each of the trajectory groups according to their baseline (wave 2) characteristics.The relative risk ratios (RRRs) indicate the relative risk of membership in each of the trajectory groups compared to the persistent low (reference) group for each unit change in the predictor (baseline wave 2) variable.Women who had dependent children (< 15yrs) at baseline were at greater risk of being categorised into the decreasing group (lone parents RRR = 3.3, part of a couple RRR = 6.6) and persistent high group (in couple RRR = 7.3), compared to the persistent low group.Correspondingly, those who were childless or had older children (> 15yrs) were at less risk of being in the higher UL groups.The only other key group characteristic that was associated with greater risk of being categorised into a higher UL group was educational level, specifically having a bachelor's degree or higher, the RRR being 1.5 and 2.4 for decreasing and increasing umbrella groups respectively.Lastly, a one-year increase in age was associated with a 1.08 increase in the relative risk of membership in increasing umbrella group versus membership in persistent low group.
In contrast, key characteristics associated with lower risk of being categorized into other groups relative to persistent low were: being in the middle-income quintiles (RRR = 0.36 Fig. 1 Estimated trajectories for total unpaid labour time (optimal 4-group model) for women from waves 2-18 (n = 1381).The plotted curves show the estimated mean unpaid labour load (in hours/week) as a function of age at each annual wave.The estimated population proportion in each trajectory group is also displayed for 3rd quintile and 0.44 for 4th quintile) and having a long-term health condition (RRR = 0.51) for increasing umbrella; and for being in the 3 highest household income quintiles (RRR = 0.29 for 3rd quintile and 0.21 for 4th quintile and 0.29 for 5th quintile) and being from a non-English speaking country (RRR = 0.32) for persistent high.

Logistic Regression Analysis: Precarious Employment and Labour Force Detachment in Wave 19
Table 3 presents the results of the five separate/individual adjusted logistic regression analysis models assessing the association between UL trajectory group membership and our five indicators of PE and LFD.Compared to the persistent low group, those in the persistent high group had 2.68 times greater odds (95% CI 1.40-5.12) of being not employed (unemployed or "hidden" unemployed).Moreover, compared to the persistent low group, those in the decreasing, increasing umbrella and the persistent high group all had significantly higher odds of being employed part-time compared to full-time (OR 1.57 (95% CI 1.02-2.41);OR 3.85 (95% CI 2.38-6.25)and OR 4.72 (95% CI 2.55-8.72)respectively).
Likewise, compared to the persistent low group, those in the increasing umbrella group and the persistent high group had significantly greater odds of being a casual worker (compared to permanent, OR 2.57, 95% CI 1.27-5.19,and OR 2.87, 95% CI 1.21-6.80respectively).There was some weak evidence that those in the persistent high group had greater odds of having low job control (OR 1.77, 95% CI 0.94-3.34)compared to the persistent low group.Lastly, although it appeared that compared to the persistent low UL trajectory group, membership in all other groups had higher odds of job insecurity, the confidence intervals contained the null.

Discussion
Women are over-represented in PE and experience weaker labour force attachment across the globe, including in Australia (International Labour Office (ILO), 2016; International Trade Union Confederation (ITUC), 2011; Pennington, 2020).Despite gender inequality in unpaid work being considered the key underpinning mechanism that influences gender gaps in labour outcomes (Ferrant et al., 2014), examining how UL (particularly as an exposure over time) is related to indicators of employment has been largely unexplored.This study sought to address this gap.
To our knowledge, our study is the first of its kind to identify trajectories of unpaid labour over women's prime working years and examine how these trajectories predict indicators of PE and LFD later in working life.
Drawing on 17 waves of Australian data (HILDA), this study identified four distinct trajectories of UL for women, spanning much of their prime working years from aged 25-35yrs at baseline to 42-52yrs at wave 18.Four distinct trajectories were identified: persistent low, decreasing, increasing umbrella, and persistent high.We found that women exposed to high amounts of UL over their prime working-age years had increased probability of PE and LFD later in prime working life.
There was some distinguishing characteristics of the different trajectory groups.Household structure was one, whereby relative to the persistent low group, being in the higher UL groups was associated with having dependent children at baseline.This is not surprising given we expect women with children at baseline would already be undertaking relatively higher amounts of UL (due to increased care requirements), which then decreased for some (decreasing group) over the ensuing 17 years (presumably as children became more independent).Meanwhile, a smaller subset sustained persistently high levels of UL time (persistent high group).The sustained high UL in this latter group may be due to multiple factors, including those women having larger families and/or providing ongoing/additional elder/other care, as well as childcare.Educational level was also relevant, and it follows that women who have undertaken higher education are more likely to enter motherhood later than those less educated (given their time in education is longer), accounting for their increased likelihood of being in the increasing umbrella group.Further to this logic, it is slightly less clear why the decreasing group (those considered more likely to already have children at baseline) may be over-represented in higher education status.Nevertheless, a decreasing trajectory shape does fit (less UL into the future), given that educated women are known to spend less time on domestic work and tend to share domestic work more equitably (Coltrane, 2000;Hertog et al., 2021).It also makes sense that those in wealthier households are less likely to be in the persistent high group given they have the financial resources to outsource elements of UL.Lastly, it follows that women who have a long-term health condition are more likely to be in the persistent low given they are likely less able to do higher UL (healthy carer effect).
In line with other Australian research (Dinh et al., 2017), our results indicate that a key driver for women's employment precarity is the responsibility they shoulder for UL.Indicators of PE and LFD at wave 19 (aged 43-53 yrs) for this cohort of women were associated with these distinct trajectory groups and, in line with our hypothesis, women with greater exposure to UL over time experienced greater odds of PE and LFD.Relative to the persistent low UL group, we found all other groups (with relatively higher UL demands) were more likely to be precariously employed at wave 19.All had higher odds of being part-time, and those women in the persistent high and increasing umbrella groups also had higher odds of being casual workers, compared to the persistent low group.In addition, those in the persistent high group had greater odds of not being employed at all, and of having low job control (approaching significance) in wave 19.It is unsurprising that this persistent high group faced greater LFD at wave 19, given their average UL time remained very high at wave 18.With UL in excess of a mean of 60 h/week (remaining largely unchanged across the 17 years), the most likely reason for this group to not be actively searching for work/being employed at wave 19 is the ongoing pressures and requirements of care provision (Baum & Mitchell, 2010).
In contrast to the results for the persistent high UL group, the findings for the decreasing and increasing umbrella groups are more perplexing.These two groups were found to have greater odds of being in PE despite being beyond the peak of their UL demands (as evidenced by their trajectories), acknowledging that UL time remains relatively high for the increasing umbrella group at wave 18.This suggests that the even when UL demands subside (substantially or even somewhat post the intensive parenting years), a scarring effect overlays women's ability to recover gainful and less precarious employment.Multifactorial explanations likely drive this disadvantage including structural barriers and biases (such as employer perception and gendered ageism), as well as lowered confidence and fractured work histories (Jones, 2019;Rochon et al., 2021).Nonetheless, it leaves women on a trajectory of precarity in their later working years, hurtling towards retirement with a stymied ability to make up for lost ground (Riach, 2019).
To further explore this possibility and contextualise our findings, we investigated the reasons given by our cohort for working part-time at wave 19 (additional variable in HILDA dataset).Of those in the decreasing group, it is noteworthy that 24% cited their reason for part-time work being they could not find full-time work, or that part-time hours were a requirement of their jobs.Caregiving was another common reason given.Of those in the persistent high and increasing umbrella groups, the predominant reason given for part-time work was care/family responsibilities (65% and 55% respectively).This aligns with European, UK and Chinese research reporting that, even after the intensive child-rearing years are over, women continue having to juggle and trade-off paid work with unpaid care (Birchall & Holt, 2022;Chai et al., 2021;Opree & Kalmijn, 2012).This continuation of gendered paid work trade-offs can be clearly explained by relative resource theories.By late middle life, most women will possess considerably less bargaining "power" around prioritising their paid work compared to their partner, having traded off comparable employment and career gains during intensive child-rearing years.Qualitative research from the UK reveals interesting nuance in women's attitudes to more flexible (precarious) work in their fifties and beyond (Loretto & Vickerstaff, 2015).Themes of "fitting in" paid work around care (as well as "fitting in" to partner's retirement plans and /or employer expectations later in life), and "helping out" remain dominant amongst these older women, with care responsibilities shifting from care of own children to care of children's children for example.Whereas, for others, after a lifetime of flexible working to accommodate UL demands, escaping low quality and unsatisfying jobs was a predominant theme, rejecting flexible working in later life in a bid to renew or continue building their careers (Loretto & Vickerstaff, 2015).
Overall, our findings align with the limited other empirical evidence in this space.Whether studies have examined total UL division (Samtleben & Müller, 2021), provision of informal unpaid care (Lilly et al., 2007;Smith et al., 2020), or levels of housework responsibility (Peutere et al., 2017), higher UL is consistently associated with more PE (reduced hours/part time) and weaker labour force attachment for women.Crucially, as a recent Australian study demonstrated, women's unequal time in domestic work and care is likely the single most important determinant of the gender workhour gap (Doan et al., 2021).The same study also reported that women's paid work hours would also increase (all else being equal), if women worked in the same industries and had the same job security as men do (Doan et al., 2021).This points to reducing occupational segregation as an important part of the solution.However, a more challenging issue to address is job security, given it is inherently linked to the construct of the ideal (full-time permanent) worker, which has long held significant barriers for women.Ultimately, high UL drives time scarcity (in line with time availability theory), and this time poverty in combination with strong socio-cultural norms ("doing gender") is imposing a cohort acceptance of greater employment precarity [Dinh et al., 2017;International Labour Office (ILO), 2016].Crucially, this exposure to PE not only has lasting economic consequences for women, but also poses significant risks for health and wellbeing (Benach et al., 2014).Importantly, we acknowledge our findings may not be generalisable to other countries.As noted in our introduction, Australia has relatively weak family policies and correspondingly high part-time employment rates amongst women.Moreover, deeply entrenched gender stereotypes pertaining to the division of unpaid labour (especially care) are more persistent in the Australian context than in other more egalitarian countries, such as the Scandinavian nations (Craig & Mullan, 2011).It may be that the introduction of generous non-transferable parental leave entitlements, in combination with a robust universal childcare policy in Australia (King et al., 2023;Remeikis, 2021) would see a significant shift towards more equitable UL demands on women, and resultantly less PE and LFD into the future.
Prior discourse on how best to assess PE has discussed the merits of both objective and subjective indicators (both employed in this study) in examining multi-dimensional PE (Vosko et al., 2009).Objective job characteristics such as casual employment and hours worked/week are definitive and easily collected.Nonetheless, interpreting the precariousness of part-time hours/work is commonly discussed.We argue that "choice" is a word that seems over-utilised when it comes to gender and reduced paid work hours.Layers of multiple and competing constraints shape (as well as dictate) "choice", such as availability of affordable childcare, in addition to the deep socio-cultural expectations around mothering and care provision.It could be argued that in many parts of the world, including Australia, that there is a cohort experience of having no other "choice" but to choose precarity (PT work hours or more casualised work) to accommodate UL responsibilities.Distinct from employment status or hours worked, other indicators of PE (such as job security and job control in our study) are more subjective, informed by questions based on participants perceptions of their jobs.In our study, apart from low job control (approaching significance) for those in the persistent high UL group, our results for these two more subjective measures of PE were largely insignificant.We speculate that despite the more objective measures telling us otherwise, older women's social conditioning to almost expect greater employment precarity has potentially contributed to the null findings on these more subjective measures.Given many of these women have likely been exposed to greater employment precarity over their whole lives, we theorise their baseline acceptance of low job security or job control affects their subjective reporting of these indicators in later life, with considerably lower expectations than their male counterparts.We also propose that some of the indicators (and questions that inform) PE measures around job security and job control are somewhat "masculinized", drawn predominantly from historical standard employment relationship constructs surrounding the ideal worker that underpins a "malebreadwinner" model of household labour (Vosko et al., 2009).Thus, whilst beyond the scope of this paper, it is possible that how we interrogate PE as a gendered phenomenon needs to be re-imagined, providing fertile ground for future research.

Strengths and Limitations
Several important limitations are acknowledged with respect to this study.Firstly, reality is infinitely more complex than the trajectories interrogated in this study.Given GBTM characterises trajectories as continuous distributions (signifying distinct groups), it does not account for variation or any heterogeneity within the women in these groups.Secondly, adjustment for confounding when trajectory groups are studied as the exposure in relation to future outcomes (as we have done) poses unique challenges.In controlling for all confounders at baseline, we acknowledge our inability to account for the time-varying confounders over the course of the trajectory, and thus cannot be sure of what extent or direction of bias may result.Thirdly, our variables were self-reported.Self-reporting of time spent in UL is particularly prone to biases, including recall bias and over-estimating bias.Moreover, given many UL tasks are done simultaneously (such as cooking while caring for children), the possibility of "double counting" is also a limitation.However, given all these potential issues are considered non-differential, any impact on our results is minimised.Fourthly, sample attrition may be a limitation as 62% of the original 2002 analytic sample responded in 2019.This may have attenuated results given those with a poorer labour market position may have been less likely to respond.In addition, we acknowledge that a bidirectional relationship likely exists between UL and poorer employment outcomes, conceding that our methodology does not address this reciprocal relationship, but that it remains a key future avenue of research.Lastly, the relationship between our UL trajectories and PE may be modified by important factors, such as social class.Whilst we controlled for measures of social class in our models (i.e., education and income), disentangling how social class in Australia (and other factors) may influence the association between UL and PE remains another important avenue for future research.
Strengths include our analytic approach, our study being novel in employing GBTM of UL on future employment outcomes.Rather than use one-time point exposure measures, we allowed for women's UL demands to change over time, shedding new light on the patterns of UL experienced by women over their prime working years.We also utlised a large ongoing longitudinal dataset that is representative of the Australian population.Lastly, this study was novel in not only applying a gender lens to PE, and interrogating the relationship between UL and PE, but was also strengthened by utility of multiple indictors of PE, not oft done, in line with PE being a complex and multi-dimensional construct.

Conclusion
Across the life course, women are disproportionately found in precarious and insecure employment arrangements.Our results indicate that a key driver for this is the disparate responsibility they shoulder for UL.Moreover, we argue that UL is a pertinent (but to date largely overlooked) factor in predicting PE and LFD into the future.Our GBTM study has shown that when women are exposed to high amounts of UL over their prime working-age years, this increases their probability of PE and LFD later in prime working life.This study contributes to the emerging evidence base that ongoing inequity in the division of UL has considerable long-term implications for gender inequality in the paid labour force, and points to the urgent need for collective political will and policy attention on and around these issues.

Table 1
Sample characteristics Participant baseline characteristics at wave 2 (n = 1381) Of the 1,381 participants at baseline in 2002, 862 (62%) participated in the 19th wave of the HILDA survey in 2019.As reported in Table

Table 1
(continued)For each of the 5 employment outcomes measured at wave 19 varied, due to missingness/nonresponse * n

Table 2
Results of multinomial logistic regression model: relative risk ratios (RRR) for trajectory group membership according to wave 2 demographic variables (n = 1381) *Non-dependent children or dependant students(> 15)^Contains 3 collapsed categories: other related (no children < 15), group household unrelated and multi-family

Table 3
Results of logistic regression models* assessing the relationship between women's unpaid labour trajectory groups and our five different indicators of employment precarity at wave 19 *All models adjusted for country of birth, education, income, place of residence, household structure, long term health condition and age (wave 2)