Youth transitions to productive activities in Ethiopia - Editorial Express

Youth transitions to productive activities in Ethiopia: gendered path dependencies and reciprocal links in adolescents’ time allocation over time

Zlata Bruckauf ([email protected]) and Yekaterina Chzhen ([email protected])* UNICEF Office of Research – Innocenti, Piazza SS Annunziata 12, Florence 50122, Italy *Corresponding author

Abstract

Using time-use data for Ethiopia from the first four rounds of the Young Lives longitudinal survey, this study analyses the dynamic nature of children’s gendered time allocation between the ages of 12 and 19. After documenting the diverging paths of boys’ and girls’ time use as they grow older, the study analyses the degree of path dependency and the relative strengths of reciprocal relationships between typical daily hours spent in four types of activities: study (at and outside school), household production (care and domestic chores), paid work, and unpaid help on a family farm or business. Using a three-wave auto-regressive cross-lagged structural equation model, the analysis shows that study in and out of school and unpaid help on a family farm or business are the most self-reinforcing of the four types of activities, controlling for a host of individual and household-level characteristics. Paid work shows significant rank order stability effects for boys only. While boys who engage in more hours of paid work spend fewer hours in education at a later age, hours of study in and out of school are not associated with future hours of paid work for either boys or girls. These findings underline a cumulative nature of skills formation but point to potential failures of the school system and the labour market in Ethiopia to translate learning into paid work at age 19. The results suggest that for Ethiopia to capitalise on its demographic dividend, it is not only necessary to recognise the life-course pattern of adolescents’ time allocation to paid and unpaid work but also evaluate the productive capacity of accumulated skill sets.

Key words: time-use dynamics, skills, children, adolescents, Young Lives, Ethiopia.

1

Introduction Youth is a time of opportunity and increasing agency, but for millions of adolescents in Africa it is also a time of greater pressure to balance personal aspirations for work and education with immediate family needs. Supporting young people’s productive activities and/or training is one of the most imperative policy tasks formulated by the African Union in the ‘Youth Decade Plan of Action 2009-2018’ (2011). While recognizing the multidimensionality of the challenges that adolescents face as well as the gender dynamics of their development, the current policy debate tends to focus on ways to turn the ‘youth bulge’ into an economic powerhouse by matching the young labour force with appropriate labour market conditions and opportunities (O'Higgins, 2017). To achieve this policy makers need to understand better the complex relationship of children’s time allocation to education and labour across different stages of childhood. Time is an important individual resource, but children and young people might not exert full control over it, as their time use responds to complex intra-household dynamics and family needs (Gershuny & Sullivan, 2014; Hofferth & Sanderberg, 2001; Camfield, 2014; Boyden et al., 2016, Gagen et al., 1999), parental education (Bianchi & Robinson, 1997), and low income (Amin & Chandrasekhar, 2012). Children’s time use also depends directly and indirectly on a myriad of changing circumstances that adversely (e.g. shocks) or positively (e.g. pubic transfers) affect household welfare (Woldehanna & Hagos, 2015; Borga, 2015; Hoddinott, Gilligan, & Taffesse, 2009; Woldehanna, 2009; Woldehanna, Second Author, & Third Author, 2008; Pereznieto & Escobal, 2006). Research evidence almost universally confirms that children’s time use is highly gendered, especially with regards to household chores and care responsibilities, reflecting the prevailing gender divisions and norms in the family (Bruckauf & Reese, 2017; Hu, 2015; Alvarez & Miles-Touya, 2012). By the time boys and girls reach late adolescence and face the choice between different types of work and further education or training, they may already be on different and diverging tracks. By crowding out schooling, children’s engagement in ‘hidden’ forms of child labour (e.g. care and domestic chores for girls and help on a family farm/business for boys) can have detrimental effects on their adult outcomes (Webbink et al 2012). Yet, the livelihood for many children and young people in low-income countries such as Ethiopia makes daily work an economic reality that not only reflects reciprocal connections and a sense of responsibility to the family but also a coping strategy within the household (Morrow et al, 2014). 2

Using longitudinal cohort data for Ethiopia from the Young Lives study of childhood poverty this paper examines the dynamic nature of adolescents’ time allocation between ages 12 and 19. The study focuses on self-reinforcing and reciprocal longitudinal associations between boys’ and girls’ time use in a low-income country context, while accounting for a range of individual and household characteristics. The study addresses the following two main questions: 1. What is the extent of path dependency in boys’ and girls’ time allocation to study, care and domestic work, unpaid work, and paid work over the course of adolescence? 2. What is the direction and strength of reciprocal associations between boys’ and girls’ time spent in different types of activity (i.e. study, care and domestic work, unpaid work, and paid work) as children grow older? The study adds to the emerging literature on transitions to adulthood in lower-income country settings, where the markers of a successful transition to adulthood are less clear than in industrialised countries (AfDB, OECD, UNDP, UNECA, 2012). To our knowledge, it is the first study that models differences in several categories of children’s time allocation jointly and over time. It advances our understanding of the degree of self-reinforcement in children’s timeuse trajectories, complementarities and trade-offs between different forms of labour and schooling as children grow older. It does so through a gender lens by allowing all associations to vary between boys and girls. The analysis contributes thematically to two SDG Targets. By focusing on the individual pathways to paid work and education the study relates to the Target 8.6, which calls on all countries to substantially reduce the proportion of youth not in employment, education and training. The analysis of gender differences in care and domestic chores links directly to Target 5.4 on the promotion of shared responsibility of unpaid domestic and care work within the household.

Children’s time use in Ethiopia Ethiopia is the second most populous, one of the poorest and fastest-growing economies in the Sub-Saharan Africa (World Bank, 2017). In the recent decade, its economy has shown resilience to persistent weather shocks by sustaining high and consistent economic growth. For example, in 2015/2016 it grew 8% in real terms. The country pursued Agricultural Development-Led Industrialisation and an active public investment program with the objective 3

of reducing poverty (World Bank, 2015). Moreover, the Productive Safety Net Program (PSNP) introduced in 2005 has targeted rural households facing chronic food insecurity enabling them to resist shocks, create assets and become food self-sufficient. It consists of public works for individuals able to work, and cash transfers, direct food packages or combination of both for those who are unable to supply labour (World Food Program, 2012). Evaluations of PSNP programmes showed their positive impact on reducing household vulnerability, food insecurity, and distress sale of assets among others (Hoddinott et al., 2009, 2012). Overall, effective safety nets, growth in the agricultural sector and public investments

in basic services were the drivers of Ethiopia’s poverty reduction from 44 % in 2000 to 30 % in 2011 according to the national absolute poverty headcount (World Bank, 2015). Despite this impressive progress in reducing total poverty rates, children and adolescents remain the most vulnerable groups in the population, with 32.4% (13 million) children under 18 living in poverty compared with 29.6% of the total population (CSA, UNICEF and OPM, 2015). Monetary poverty and the demand for child labour in poor households are some of the major factors preventing children’s (and particularly girls’) participation in schooling beyond primary level. Although opportunities for productive activities are increasing, due to the shift in the labour market structure towards more non-agricultural sectors, youth unemployment rates and duration are high in Ethiopia (Seid, Taffesse, & Ali, 2016). For instance, 21% of males age 15-24 in Addis Ababa were unemployed in 2012, 36% among those who have just graduated from high school (World Bank, 2015). Young people’s labour market experiences in Ethiopia are similar to other low-income countries in the region with prevailing participation in informal and/or precarious employment, including self-employment, unpaid help on a family farm or business, and temporary jobs for private companies (O'Higgins, 2017). Children living in households headed by adults working in the informal sector have the largest poverty gap (CSA, UNICEF and OPM, 2015). This implies the transmission of vulnerability and precarious conditions through the generations. A particular concern is unemployment among young skilled labour (college graduates), although this interacts with the special dynamics of migration from rural to urban areas. According to a recent review by the Brookings Institute (Seid, Taffesse, & Ali, 2016), there is a mismatch between the increasing supply of youth entering the labour market and the availability of secure employment opportunities. The trajectories of young girls and boys and their opportunities for productive activities in late adolescence are strikingly different. A recent study based on data from the Demographic and 4

Health Surveys shows that only 20% of girls who are employed as domestic maids by nonrelative households are in school. Girls are employed by wealthier households in urban areas for babysitting, cooking and other chores (World Bank, 2015). It is argued that households’ decisions to invest in girls’ schooling are based on the rational assumption that as a female she will typically spend five times more time in household work than a man (Gable, 2013). This translates into a high opportunity cost of taking up market employment. Regardless of the data source, methods and thematic focus, evidence from Ethiopia unequivocally points out to strong gender division of roles in the household pronounced in the types of activities boys and girls engage in. Both qualitative and quantitative evidence shows that younger boys and girls do similar work – herding smaller animals, fetching water and wood, looking after siblings (Orkin, 2012), but as they grow older, differences in the amount of time allocated become more significant (Woodhead, Dornan & Murray, 2013). Using the Young Lives Round 2 data for 11-12-year-olds, Heissler and Porter (2010) found no gender differences in the total hours of work but very clear gendered distribution of tasks: girls spend more time in domestic work than boys did in farming. This is consistent with earlier findings of the Ethiopian Rural Household Survey (Admassie, 2003) and other studies using YL Ethiopia dataset and age groups (Woldehanna, 2009). A mixed methods study (Camfield, 2014) using data for children aged 14-15 from the Young Lives Round 3 suggests that girls replace mothers and resume reproductive and productive responsibilities from an early age. A number of studies examined the correlates of children’s time use in Ethiopia. Persistent economic insecurity and limited access to credit affect the intra-household allocation of time and money (Boyden et al., 2016). Rural and urban children in Ethiopia engage in different types of work (Poluha, 2007). Household wealth and possession of assets tend to reduce the likelihood of children’s participation in household chores and help on a family farm (Orkin, 2012; Webbink et al 2012). Children’s participation in schooling is sensitive to shocks (e.g. weather shocks, death or illness of a household member), with varying effects by gender. Woldehanna and Hagos (2015) found that a drought, crop failure and pest infestation had a negative effect on children’s school completion rates, with girls affected to a greater degree. The negative effect of the livestock death was associated with the reduction in the demand for child labour in the household, thus having a positive effect on school completion in the short run, especially for boys. Meanwhile, the death or illness of a household member had a negative effect on primary school completion in both rural and urban areas. Heissler and Porter (2010) found that social norms mediate the differentiated behavioural response to adverse events: boys 5

and girls substitute adults in housework according to gender. For example, girls work more when their mothers are ill, and boys work more on domestic tasks when their fathers fall ill. Qualitative studies found a strong sense of reciprocal connection within the household, a sense of collective responsibility for livelihoods reflected in the distribution of roles and responsibilities by gender and age (Grivello and Van de Gaag, 2016). Children’s work within and for the household and close kin is a way of social integration into their community and preparation for adulthood (Boyden et al., 2016, Morrow, 2015). It is a continuous process that can be affected by a range of external events and changing circumstances. Overall, the stock of evidence from Ethiopia and other low-income settings points to entrenched gender ‘specialisation’ in the distribution of children’s time across various activities, the consistent effects of family wealth, credit constrains, family composition, community and shocks on children’s time allocations. Yet it remains unclear whether behaviours early in life self-reinforce and carry on into young adulthood. There is also a gap in evidence on the reciprocal influence of different activities on each other over the course of childhood and adolescence. Using Young Lives data from Ethiopia, India and Vietnam, Borga (2015) found that children’s hours of paid work, unpaid work, domestic chores, and care activities were all associated with lower cognitive test scores both directly and indirectly, through crowding out schooling and studying. To account for potential trade-offs between different activities the present analysis models relationships between an exhaustive and mutually exclusive array of children’s time allocations.

Conceptual framework and hypotheses There is strong empirical and theoretical evidence on the cumulative and complementary nature of skills formation over the critical stages of children’s life. The notion of ‘skills beget skills’ formulated within the Technology of Skills Formation framework (Carneiro, Cuhna & Heckman, 2003; Cuhna and Heckman, 2008) explains evidence on skill formation through selfproductivity and complementarity of human capital investment. The theory presents skill formation as a life-course, multiplier process (Cuhna et al., 2006). Self-productivity implies that skills produced at one stage of childhood augment the skills attained at a later stage. Moreover, skills acquired in one period will persist at later periods. In other words, skills are self-reinforcing. 6

The amount of time children spend in different activities can be seen as a proxy for the acquisition of a certain range of cognitive and non-cognitive skills (with exception of sleep which fulfils a biological function). For example, by engaging in play and leisure activities children learn self-control and social skills among others. When children engage in caring for their siblings and relatives they develop skills in self-organization, focus, responsiveness to the needs of others etc. These are arguably different sets of skills compared to those that children acquire through relatively specialized tasks of herding the household cattle or farming. So, it can be argued that children in low-income countries for whom work on a family farm or help in household production acquire a range life-skills that are as important as cognitive and noncognitive skills gained through schooling or other human investment. Those skills can be highly specialized which reinforces their persistence and self-productivity over the stages of the childhood. Therefore, we expect a path dependency in adolescents’ time allocation throughout adolescence, especially to activities which are more specialized in nature (such as helping on a family farm). H1: There is a self-reinforcement pattern in children’s time allocations to the same activity between ages 12-19. On the other hand, young people’s time use choices respond to the costs and benefits faced by the household economy, including the opportunity cost of forgone benefits of alternative activities (Becker, 1965; Chez and Becker, 1975). While Borga (2015) documented the tradeoffs between children’s labour and schooling in low income country contexts, there is less evidence on trade-offs between different forms of labour. The life-course perspective adds a temporal dimension to the basic time allocation model integrating immediate costs and benefits with their future pay-offs (Gurven & Kaplan, 2006). It allows to account for the differential skills associated with developmental stages of child development or functional age. Thus, within the household production children will be allocated tasks that are appropriate for their level of skills, strength and capacity at a certain stage of development (Bock, 2002). We can assume that there is a trade-off between children’ time allocations that produce higher future or present returns. For example, schooling and extra study are expected to provide increased rates of return in the future but they entail an opportunity cost in terms of forgone earnings or household production as well as have direct costs (e.g. fees, uniforms, or school supplies). 7

Therefore, we expect to observe a set of associations between the hours spent on different types of labour and schooling over time:

H2:

There are complementarities and trade-offs between in children’s time allocations to

different activities between ages 12-19. •

H2.1: Hours of schooling and extra study at ages 12 and 15 will positively contribute to young people’s hours of paid work at age 19.

•

H2.2: More time spent in paid and unpaid work at earlier ages will negatively affect children’s hours of schooling and extra study at older ages.

•

H2.3: More time spent in care giving and household chores at earlier ages will negatively affect children’s hours of schooling and extra study, paid and unpaid work at older ages.

Data and methods We use data for Ethiopia from the first four waves of the Young Lives (YL) study 20022013/14. YL is an international study of childhood poverty, following the lives of 12,000 children over 15 years in four countries: Ethiopia, India (Andhra Pradesh), Peru, and Vietnam. In each study site YL follows two cohort samples: 1,000 children born in 1994 (Older Cohort) and 2,000 children born between 2000 and 2011 (Younger Cohort). At the start of the study in 2002, these children were 8 years old and between 6 months and 18 months old, respectively. The first four rounds of YL data are currently available to researchers, with the data from the fifth round in preparation for public release. While not designed to be nationally representative, YL surveys provide rich data about the diverse characteristics of the cohort children as well as their households and communities over time. As such, YL is valuable for examining the dynamics of children’s well-being over time (see www.younglives.org.uk for more information about the study). Panel attrition in the four-wave YL dataset for Ethiopia is low by international standards for longitudinal studies: 2.2% for the Younger Cohort and 8.4% for the Older Cohort (Pankhurst and Woldehanna 2014). However, only 90% of Older Cohort girls are present in all four waves, compared with 97% of boys. While nearly all Older Cohort girls (99%) are present in the first three waves of the survey, i.e. between the ages of 8 and 15, a substantial 10% of these girls drop out by the time of the fourth round at age 19.

8

Time-use information was collected in Rounds 2 to 4 for both cohorts. For the Younger Cohort, information was collected from the caregiver in Rounds 2 to 4. For the Older Cohort, data were collected from the caregiver in Round 2 and from the children directly in Rounds 3 and 4. Time-use is measured using stylized questions. Cohort members (or their caregivers) allocate hours to each of the following eight activities within 24 hours on a typical day (i.e. not a weekend or holiday): 1 

Care for others (younger children, ill household members)



Household chores (fetching water, firewood, cleaning, cooking, washing, shopping)



Tasks on a family farm, cattle herding, other family business, sheepherding, piecework or handicrafts done at home



Activities for pay or for money outside of household or for someone not in the household



At school/college/University (including travel time)



Studying at home or extra tuition outside the home



Leisure: playing, seeing friends, using the internet, etc.



Sleep at night

The hours of providing unpaid care tend to be positively correlated with domestic chores, while the time spent at school goes together with out-of-school study (see Table 6 in the appendix). We combine these pairs into variables denoted as “care/chores” and “school/study”. We omit leisure and sleep from the multivariate modelling because they are collinear with the rest of the activities. As expected, the remaining four types of activity tend to be negatively correlated with each other – chores/care, school/study, paid work, and help with a family farm or family business (“unpaid work” from here on as a short-hand). We use a three-wave auto-regressive cross-lagged structural equation model with four endogenous variables denoting the number of hours spent in each of four activities on a typical day: school/study, paid work, unpaid work, and chores/care (see equations 1-4). We focus on children in the Older Cohort, aged 12, 15 and 19. We re-estimate the model for the Younger Cohort as a robustness check.

1

See http://www.younglives.org.uk/sites/www.younglives.org.uk/files/et-r4-yc-childquestionnaire.pdf (Section 2). 9

School or studyit = α1t + β1 School or studyi,t−1 + γ1 Paid work i,t−1 + δ1 Unpaid work i,t−1 + θ1 Chores or carei,t−1 + ∑kj=1 σjt X1jit + εit (1) Paid work it = α2t + β2 Paid work i,t−1 + γ2 School or studyi,t−1 + δ2 Unpaid work i,t−1 + θ2 Chores or carei,t−1 + ∑kj=1 σj X2jit + ϵit (2) Unpaid work = α3t + β3 Unpaid work i,t−1 + γ3 Paid work i,t−1 + δ3 School or studyi,t−1 + θ3 Chores or carei,t−1 + ∑kj=1 σj X3jit + μit (3) Chores or careit = α4t + β4 Chores or carei,t−1 + γ4 Paid work i,t−1 + δ4 Unpaid work + θ4 School or studyi,t−1 + ∑𝑘𝑗=1 𝜎𝑗 𝑋4𝑗𝑖𝑡 + 𝜋𝑖𝑡 (4) Where, i=1, …. N; t=3, 4 – the last two rounds of the YL study. We thus estimate eight equations at a time separately for children in each cohort. All equations are estimated simultaneously using maximum likelihood, allowing to assess the global likelihood and fit of the model. The standard errors are adjusted for geographic clustering to account for the complex survey design. α1t – α4t are intercepts in the four equations; β1 – β4 are the autoregressive (i.e. rank order stability) coefficients; and γ1- γ4, δ1- δ4, θ1- θ4 are the cross-lagged (i.e. reciprocal) parameters to be estimated. The vectors X1-X4 include k controls that denote the characteristics of the household that may influence children’s time allocation to studying (in and out of school) and/or labour (paid work, help on a family farm or business, care, and domestic chores). We control for economic factors that may affect a household’s decision to engage children in labour rather than schooling in the absence of perfect capital markets by including a composite wealth index. Normalised to range from 0 to 1, the wealth index is based on housing quality, possession of consumer durables, and access to services. It is the only time invariant covariate in the model because we use the values at the first round of the study in order to capture the influence of the material living conditions prevailing at the time the children were youngest. We also control for the

10

household’s self-reported ability to raise an amount of money2 in one week as a measure of current financial constraints and the availability of support networks. We include an indicator for livestock ownership as a measure of demand for labour tending animals and for the presence of children under six as a measure of demand for care. Gender of the household head is included to account for potential gender role replication in the family. As female headed households are less likely to contain any male adults aged 18-60, we expect that girls in female-headed households are more likely to engage in more hours of paid work. We control for the type of area (urban vs rural) and for region to account for structural differences in access to paid work and schooling as well as in demand for children’s unpaid labour. Finally, we control for an indicator of self-reported experience of a drought during the past year. Because very few droughts were reported in Round 4, we only include it to predict children’s time allocation in Round 3. We have tested for a range of other factors found to be important to children’s time allocation in Ethiopia but excluded them from the final model because their inclusion does not improve the model fit. These included children’s health, parental literacy, land ownership, composition of the household by age and sex (in addition to the presence of young children), and an array of other environmental, income and demographic shocks. 𝜀𝑖𝑡 , 𝜖𝑖𝑡 , 𝜇𝑖𝑡 , and 𝜋𝑖𝑡 are the idiosyncratic error terms from equations (1) to (4), respectively. We assume that the error terms (𝜀𝑖𝑡 , 𝜖𝑖𝑡 , 𝜇𝑖𝑡 , 𝜋𝑖𝑡 ) have the means of zero, conditional on the covariates; are uncorrelated with the exogenous covariates in X1-X4; and are not autocorrelated. However, they may be correlated with each other within each time period: covariances are estimated for each of the six pairs of error terms per period. As a full graphical representation of the system of four equation would be too unwieldy, Figure 1 illustrates the path diagram with auto-regressive and cross-lagged effects of hours of schooling and paid activity from Equations (1) and (2). The coefficient β12 is the net effect of the hours of study at t3 on the hours of study at t4, controlling for the hours in paid work at t3. In other words, for any two children with a one hour difference in the hours of study in the previous round of the survey and who spent the same number of hours in paid activity, β12 is the predicted difference in their hours of study in the current period. The autoregressive coefficients reflect rank order (or positional inter-subject) stability: a large and significant

2

500 Birr in Round 4 and 260 Birr in Round 3. 11

coefficient would suggest that the same children who allocated more hours to a given activity in the previous period remain the ones who spend more hours on this activity in the current period. Meanwhile, the cross-lagged reciprocal parameter γ21 is the effect of the hours of paid work at t3 on the hours of study at t4, controlling for the hours of study at t3. In other others, we would predict the hours of study to differ by the value of γ21 in the current period for any two children whose hours of paid work in the previous period differed by one hour and who reported the same number of study hours in that period. Cross-lagged coefficients indicate the direction and relative strengths of reciprocal relationships over time. It has to be noted that the model treats as predetermined the initial measures of the endogenous variables reported in Round 2 of the YL, when time-use questions had been asked for the first time. All potential prior influences are absorbed in the means, variances and co-variances of the Round 2 measures of time-use (hence the double-headed arrow between schooling and paid work in Figure 1). Extending this framework to the full four-equation model with exogenous covariates, the partial effects of the hours spent in any of the four activities in the previous period are interpreted as holding constant the hours spent in the other three activities and controlling for the effects of key individual and household characteristics. Autoregressive cross-lagged panel models are used extensively in developmental research (see Selig and Little 2012). By using the longitudinal structure of the dataset and comparing the significance and the strength of reciprocal coefficients over time, the model helps establish the temporal causal ordering of different activities and the relative strength of their effects on each other. This is done purely in the tradition of “Granger causality”: z Granger causes y if past z is useful, in addition to past y, for predicting yt (see Wooldrige 2006). This allows testing the hypothesis that, for example, the number of hours in paid labour at age 15 predicts the number of hours of schooling at age 19 over and above the influence of past hours of study. In addition, we can test if the effect of paid labour on schooling is stronger than the reciprocal effect of the hours of schooling on labour between the ages 15 and 19, everything else being equal. The model does not allow disentangling any potential contemporaneous relationships between the hours spent on different activities. We cannot test if devoting an extra hour a day to paid labour at age 19 causes a reduction in the number of hours of study at the same time. Instead we assume that the hours allocated to each of the four activities are determined jointly at any 12

given time, influenced by the current needs of the household and constrained by the 24 hours of the day. We allow the residuals in every equation to be correlated with the other three in every period to reflect that the same unobserved characteristics could influence all these activities. We perform a range of robustness checks on our models. To determine if the same model could be estimated parsimoniously for a combined sample of girls and boys, we test for the equality of parameters across groups. Given the increasingly gendered divisions of labour as children age, documented in previous research on Ethiopia, we expect at least some parameters to be statistically significantly different for girls and boys. To check if there may be an unobserved time-invariant influence on the propensity to engage in a particular activity over time, we test a model in which we allow the residual variances of the endogenous variables denoting the same type of activity to be auto-correlated (e.g. ε2 and ε3). If the model fit does not improve, we revert to the baseline specification where the residuals are not auto-correlated and the auto-regressive parameters alone measure any path dependency in children’s time use as they grow older.

Results Raw gender differences in time-use Gender differences in the rate of children’s participation in caregiving and help on a family farm of business emerge as early as age 5 and get larger as children get older. For the Older Cohort, it steadily widens with the largest disparity observed at age 19. While both girls and boys substantially reduce their participation in care at age 19, twice as many girls (34%) remain engaged in caring than boys (15%) (Table 3-5 in the Appendix). In line with previous research on Ethiopia we observe differences in the participation rates between urban and rural locations. A greater share of girls in rural areas engage in caring, household chores and help on a family farm or business compared to urban areas (Figure 2). Conversely, girls in urban locations spend more time studying in and out of school. There are no significant differences between locations in girls’ participation in paid work. Not surprisingly, the largest difference in participation rate is observed between rural and urban for boys in help on a farm or family business (44 percentage points). Using a pooled sample of two YL cohorts, our descriptive analysis shows a widening gender gap in children’s time allocation on domestic chores between ages 5 and 19. While 5-year-old 13

children contribute a non-negligible amount of their time to household chores, both boys and girls increase their average hours of household chores substantially from age 8. By age 19, girls in the Older Cohort spent 2.6 time more time in domestic chores than boys. Consistent with the evidence on traditional social and gender roles attached to specific domestic tasks in African societies, we find the reverse gender dynamics in unpaid work (Figure 3). From age 8 boys contribute more time to the household’s economic livelihood. The gender gap in time allocated to this activity is highly significant across all age groups but the trajectory is non-linear. The relative gender gap in the amount of time helping on a family farm or business widens between ages 12 and 15, but narrows somewhat between ages 15 and 19. This is due to an increase in girls’ participation in unpaid work over this period (from 20% at age 15 to 24% at age 19) and the longer hours spent by those who participate compared to previous periods. It is possible that girls have to substitute for boys (brothers) who moved to paid work activities at this age. Time allocated to schooling follows an expected trajectory relevant to children’s developmental stages between ages 5 and 19. Girls tend to spend significantly longer hours in school than boys at ages 12 (Younger Cohort) and age 15 (Older Cohort) (Figure 3). Dynamic relationships between different time-use categories Table 1 shows the estimated coefficients from Equations 1-4 for the Older Cohort children. They are aged 12 in Round 2, 15 in Round 3 and 19 in Round 4. All parameters are allowed to vary between girls and boys. The model fits the data very well (SRMR=0.014) and explains a sizeable 74% of the total variation across all estimated equations. Allowing for autocorrelations in the disturbance terms does not improve the fit of the model, so the results are presented with the disturbances correlated within time but not across time. Boys who engage in more hours of schooling or studying, paid work, or unpaid work tend to devote more hours to these activities at later ages compared to their peers, everything else being equal. The stability effects are not trivial in size, ranging from around 20 minutes to 30 minutes per hour spent in the previous wave. This suggests that adolescent boys acquire task-specific skills and continue specialising in these activities as they grow older. However, the autoregressive coefficients between the ages of 12 and 15 are not statistically different from those between 15 and 19, suggesting that the extent to which each of these three types of activity self-reinforce remains stable during adolescence.

14

The results for girls show a more muted picture. The rank order stability effects of unpaid work are significantly lower than for boys both between the ages of 12 and 15 and between 15 and 19. This suggests a lower rate of self-reinforcing skill acquisition in unpaid work for girls, who are less likely to engage in help on a family farm or business in the first place (see Table 3 in the Appendix). Contrary to our Hypothesis 2.1, we do not observe any reciprocal lagged effects of schooling on paid work for boys or girls. This suggests that the skills acquired at school and in extra studying do not translate effectively into paid labour in Ethiopia, even between the ages of 15 and 19. In line with our Hypothesis 2.2, boys who engage in more hours of paid work at an earlier age devote fewer hours to schooling at both 15 and 19. However, hours of unpaid work do not have statistically significant effects on schooling. This suggests that paid activity presents a greater threat to boys’ education than unpaid help on a family farm or business. Neither paid nor unpaid work has a reciprocal effect on schooling for girls. The data do not lend support to our “burden of care and domestic work” Hypothesis 2.3, as there is no reciprocal lagged effect of domestic chores and care on any of the other three activities for boys or girls. Moreover, there is no evidence of persistence in the hours of household chores and care over time. These activities are more influenced by current demands of the household, such as the presence of children under the age of five. Meanwhile, both unpaid work and schooling have negative effects on the hours of domestic chores and care for boys between the ages of 12 and 15. Although the effect sizes are modest, they indicate that the boys who had specialised less in schooling and unpaid work at younger ages would be more likely to take up the female-dominated household production tasks later on. The partial effects of the covariates are of the direction and size consistent with the findings in the YL literature on Ethiopia. Household wealth tends to be negatively associated with the hours of paid work for boys and domestic and care chores for girls at age 15. Livestock ownership is positively related to the hours of unpaid work for boys at age 15; the present of young children with domestic chores and care, especially for girls. Nineteen-year-old girls in female headed households engage in more hours of paid work and fewer hours of domestic chores and care. There are substantial regional and rural/urban differences. Table 2 shows the estimates for the Younger Cohort, based on the measures taken at the ages of 5, 8 and 12. Paid work is excluded from the estimation because very few children engage in 15

paid activity at the ages of 5 and 8. The model fits the data very well and explains a substantial 80% of variability in the hours of schooling and study, unpaid work, and care/household chores. Children who invest an additional hour in a particular activity at an earlier age tend to allocate significantly more time to this activity at a lager age relative to their peers, all else being equal, but these effects are modest in size compared to the results for the Older Cohort. There are no significant gender differences in the rank order stability coefficients, suggesting that gender specialisation that we observed among the Older Cohort boys in unpaid work tend to emerge in later adolescence. There are no reciprocal lagged effects between the hours devoted to three different activity types. The only exception is the negative effect of the hours of schooling on the hours of care and domestic work for girls between the ages of 8 and 12, and the negative effect of care and domestic chores on schooling for girls between the ages of 8 and 12, everything else being equal. Although this suggests that schooling protects from extra hours of care and domestic work at primary school ages, while care and domestic work exerts a burden on schooling at secondary school ages, these effects are substantively small in size.

16

Discussion and conclusion This study aimed to provide evidence on the dynamic nature of children and adolescents’ time use in Ethiopia. This is realised by examining the interplay and self-reinforcement effect of children’s time allocation decisions across paid and unpaid work, schooling, care and domestic chores through life-course and gender perspectives. Consistent with earlier studies we see that from an early age boys and girls in Ethiopia develop gender specific task specialisations. The gender gap in average time allocated is most prominent in caregiving activities and unpaid work (help on a family farm, business etc.). Furthermore, rural or urban location plays a significant role reflecting structural constrains and the types of work boys and girls undertake. Drawing on The Technology of Skills Formation framework (Cuhna and Heckman, 2008) we hypothesised that by spending time on different activities at earlier stages children develop a set of skills which are cumulative and self-reinforcing over time. This leads to path dependencies in their time allocation decisions during adolescent years. We find strong support to this hypothesis for both boys and girls with respect to schooling and unpaid work between the ages of 12 and 19. We also observe this in the younger cohort of children between ages 5 and 12, but the effects are smaller in magnitude. The positive and self-reinforcing effect of education gives a very unambiguous policy message on the importance of expanding access to formal education at the pre-primary level, particularly to rural and most vulnerable groups of children. These groups are likely to face limited formal childcare options and monetary constrains in accessing private pre-school provisions in Ethiopia. While the expansion of preschool provision will have a positive ‘ripple effect’ on the hours of schooling at the later stages of childhood, our results suggest that it is never too late to invest in more learning. Even at ages 12 and 15 one extra hour of schooling will result in more time studying at age 19. It is an important finding particularly in the context of a higher risk of school drop-out among Ethiopian children and particularly rural boys (UNESCO, 2015, Woldehanna et al., 2011). The findings on path dependency in time spent on unpaid and paid work (the latter applicable to boys only) during adolescent years raises a number of policy questions. Confirming our hypothesis of early specialisation and self-reproduction of skills, it indicates that the timing and the type of support young people (particularly boys) need to receive is crucial. Policy needs to recognise that young people’s time allocation around ages 19 is entrenched in the past choices and economic behaviour at age 12 and earlier. It is also supported by a strong sense of social responsibility to the family and awareness of the risks of not working (Morrow et al., 17

2014). The labour market should consider transferability of accumulated skills but differentiate between different work activities that children and young people engage in and their future productive potential. We do not find self-reinforcement effect of time spent on caregiving and domestic chores suggesting that children’s time contribution to these activities is not rigid over time. Given highly gendered nature of these tasks, this can be explained by a greater pressure on girls to meet the current or immediate family care needs and demand for labour often serving as available time resource at the time of persistent crises (Camfield, 2014). This is supported by the lack of substantial reciprocal effects between girls’ time allocation to household production, education, paid and unpaid work over time. Overall, girls’ time allocation in Ethiopia seems to be more contemporaneous in nature, but with a corresponding gender attribute (Heissler & Porter, 2010). Policy interventions at the household level (e.g. public safety net programmes) were shown to be effective in reducing girls’ contribution to care and chores (Woldehanna 2009). But community level measures (e.g. access to pre-school provision) and infrastructure which meet household demand for caregiving and domestic chores can further release girls’ as well as boys’ time for other activities. It could also be that our analysis does not capture the type of jobs that girls perform for money thus missing the path dependency in the specific skill set that girls develop. Moreover, 10% of girls in the Older Cohort dropped out of the panel between the ages of 15 and 19. Although they did not differ in their average hours of care and domestic tasks at age 15, they had allocated significantly more time to household production at age 12. Contrary to our expectation, we find that more time spent in schooling at ages 12 or 15 does not guarantee more time in paid work at age 19. One of the plausible explanations of this result is the mismatch between the skills acquired during the compulsory school years and the requirements of the Ethiopian labour market leading to high levels of unemployment among educated youth (Seid, et al 2016). Another explanation is the low quality of educational programs or other failures of the school system in Ethiopia which do not support cognitive and non-cognitive skills formations expected by the labour market (Joshi & Verspoor, 2013). This is supported by perceptions at the family level (Boyden et al., 2016). It is also possible, that our analysis does not capture the full effect of schooling due to the age group boundaries. Indeed, Young Lives data shows that 59% of young people still report participation in schooling or training at age 19.

18

Finally, we find a detrimental effect of time spent by boys on paid work at ages 12 and 15 on their time in schooling or further training at age 19. This important result has multiple policy implications. Acknowledging that performing activities for pay might be the only way for poor children in Ethiopia to support the immediate family needs and continue studying (Boyden et al., 2016), the findings call for a robust evaluation of the type of skills available to children and young people through paid work from the early age. This has to be supported by policies which ensure that at age 12 and 15 all boys and girls have a fair access a high-quality learning that builds a vital platform for their future economic activity. It is one of the ways to prevent a potential trap in low skills and low paid jobs that do not lead to improvements in productivity of the youth labour force that the country needs. The transition to productive activities of rural boys is of particular concern as the burden of early specialisation in farming activities and interrupted schooling creates a ‘ceiling’ on their labour market prospects. The findings of this analysis feed into a broader methodological discussion on the value of children’s time use as an important social indicator capturing choices and behaviours within the household. The evidence presented in this study confirmed the dynamic nature of children’s time allocation and underlined the cumulative and persistent nature of skill formation. It provides a valuable contribution to an active debate on how to support the youth in developing countries in their transitions to productive work. References AdfDB, OECD, UNDP, UNECA (2012). Chapter 6: Promoting Youth Employment in Africa Economic Outlook, available at: www.africaneconomicoutlook.org/en/indepth/Youth_Employment Admassie, A. (2003) 'Child Labour and Schooling in the Context of a Subsistence Rural Economy: Can They Be Compatible?', International Journal of Educational Development, Vol. 23: 167–185. African Union (2011). Youth Decade Plan of Action (2008-2018). Accelerating Youth Empowerment for Sustainable Development. Available at: http://www.un.org/en/africa/osaa/pdf/au/african_youth_decade_2009-2018.pdf Alvarez, B. & Miles-Touya, D. (2012). Exploring the relationship between parents’ and children’s housework time in Spain, Review of Economics of the Household, Vol. 10: 299-318. Amin, S., & Chandrasekhar, S. (2012). Looking beyond universal primary education, Asian Population Studies, Vol. 8(1): 23-38, http://dx.doi.org/10.1080/17441730.2012.646820 Becker, G.S. (1965). A Theory of the Allocation of Time, The Economic Journal, 75 (299): 493-517. Bianchi, S.M. & Robinson, J. (1997). What Did You Do Today? Children's Use of Time, Family Composition, and the Acquisition of Social Capital, Journal of Marriage and Family, Vol.59 (2):332-344. 19

Bock, J. (2002). Evolutionary Demography and Intrahousehold Time Allocation: Schooling and Children’s Labor among Okavango Delta People of Botswana. American Journal of Human Biology, Vol. 14: 206-221. Borga, L.C. (2015). Children's Own Time Use and its Effect on Skill Formation, Working Paper 534, Center for Economic Research and Graduate Education, Charles University, Prague. Boyden, J., Porter, K., Zharkevich, I., Heissler, K. (2016). Balancing School and Work with New Opportunities: Changes in Children’s Gendered Time Use in Ethiopia (2006-2013). Young Lives Working Paper 161. Bruckauf, Z. & Rees, G. (2017). Children’s involvement in housework: Is there a case of gender stereotyping? Evidence from the International survey of Children’s Well-being, Innocenti Research Brief, 2017-17. Available at: https://www.unicef-irc.org/publications/898/ Camfield, L. (2014). Growing Up in Ethiopia and Andhra Pradesh: The Impact of Social Protection Schemes on Girls’ Roles and Responsibilities, European Journal of Development Research, Vol. 26, 107–123. doi:10.1057/ejdr.2013.36 Chiappori, P.A.& Lewbel, A. (2015). Gary Becker’s , “A Theory of the Allocation of Time”, The Economic Journal, Vol. 125 (583): 410-442., DOI: 10.1111/ecoj.12157 CSA, UNICEF and OPM (2015). Child Well-Being in Ethiopia. Analysis of Child Poverty Using the HCE/WMS 2011 Database. Available at: https://www.unicef.org/ethiopia/Child_Poverty_Report_January_2016.pdf Cuhna, F., & Heckman, J.J. (2008). Formulating, Identifying and Estimating the Technology of Cognitive and Noncognitive Skill Formation, The Journal of Human Resources, Vol. 43 (4): 738-782, doi: 10.3368/jhr.43.4.738 Elder, G.H. (1998). The Life Course as Developmental Theory, Child Development, Vol. 69: 1-12. Fiorini, M. & Keane, M. P. (2012). How the Allocation of Children’s Time Affects Cognitive and Non-Cognitive Development, Journal of labour Economics, Vol. 32, N.4: 787-836 Gager, C., Cooney, T.M., Call, K.T. (1999). The effects of family characteristics and time use on teenagers’ household labor, Journal of Marriage and Family, Vol. 61(4): 982-994. Gable, S. (2013). Girls and Income Growth in Ethiopia. Girls’ Hub, Ethiopia, available at: http://www.girleffect.org/media?id=3030 Grivello, G. & van der Gaag, N. (2016). Between Hope and a Hard Place: Boys and Young Men Negotiating Gender, Poverty and Social Worth in Ethiopia, Young Lives Working Paper 160. Gurven, M., & Kaplan, H. (2006). Determinants of Time Allocation across the Lifespan, Human Nature, Vol.17 (1): 1-49. O’Higgings (2017). Rising to the Youth Employment Challenge, International Labour Organisation (ILO), Available at: http://www.ilo.org/wcmsp5/groups/public/---dgreports/--dcomm/---publ/documents/publication/wcms_556949.pdf Gershuny, J. & Sullivan, O. (2014). Household structure and housework: assessing the contributions of all household members, with a focus on children and youths, Review of Economics of the Household, Vol. 12:7-27.

20

Joshi, R.D., & Verspoor, A. (2013). Secondary Education in Ethiopia. Supporting Growth and Transformation, A World Bank Study, The World Bank, Washington, D.C. Available at: https://elibrary.worldbank.org/doi/pdf/10.1596/978-0-8213-9727-5 Heissler, K & Porter, C. (2010). Know Your Place: Ethiopian Children’s Contributions to the Household Economy, Young Lives Working Paper 61. Hoddinott, J., Gilligan, D.O., Taffesse, A.S. (2009). The Impact of Ethiopia’s Productive Safety Net Program on Schooling and Child Labor, Available at SSRN: https://ssrn.com/abstract=1412291 or http://dx.doi.org/10.2139/ssrn.1412291 Hoddinott, John, Guush Berhane, Daniel O. Gilligan, Neha Kumar, and Alemayehu Seyoum Taffesse (2012). The Impact of Ethiopia’s Productive Safety Net Programme and Related Transfers on Agricultural Productivity. Journal of African Economies 21(5): 761–786. Hofferrth, S.L. & Sandberg, J.F. (2001). How American Children Spend Their Time, Journal of Marriage and Family, Vol. 63, No.2: 295-308. Hu, Y. (2015). Gender and Children’s Housework Time in China: Examining Behavior Modeling in Context, Journal of Marriage and Family, Vol. 77 (5): 1126-11:43, DOI: 10.1111/jomf.12225. Morrow, V. (2015) ‘Intersections of School, Work, and Learning: Children in Ethiopia, India, Peru, and Vietnam, in T. Skelton (ed.) Handbook of Geographies of Children and Young People, Singapore: Springer. Morrow, V., Tafere, Y., Vennam, U. (2014). Changes in Rural Children’s Use of Time: Evidence from Ethiopia and Andhra Pradesh, in Bourdillon, M., & Boyden, J. (Eds.) Growing Up in Poverty. Findings from Young Lives. Palgrave Macmillian UK. Orkin, K. (2012). Are Work and Schooling Complementary or Competitive for Children in Rural Ethiopia? A Mixed-methods study, Young Lives Paper 77. Pankhurst, A. and Woldehanna, T. (2014) Young and development: preliminary findings from Round 4 in Ethiopia. Young Lives Factsheet. https://www.younglives.org.uk/content/youthand-development-preliminary-findings-round-4-ethiopia. Perenznieto, P. and J. Escobal (2006). Children, Time and Poverty: The Impact on Children of Easing Poor Families’ Credit Constraints, paper presented at the XVI ISA World Congress of Sociology, Durban, 23-29 July. Poluha, E. ed. (2007). The World of Boys and Girls in Rural Ethiopia, Addis Ababa: Forum of Social Studies Selig, J. and Little, T. (2012) “Autoregressive and cross-lagged panel analysis for longitudinal data.” In Handbook of Developmental Research Methods. Edited by Laursen, B., Little, T. and Card, N. The Guildford Press. Seid, Y., Taffesse, A.S., Ali, S.N. (2016). Ethiopia – an agrarian economy in transition. Understanding the African Lions- Growth Traps and Opportunities in Six Dominant African Economies, Brookings Institute, United Nations University. Available at: https://www.brookings.edu/wpcontent/uploads/2016/08/global_20160816_ethiopia_economy.pdf UNESCO (2015). Education for All 2015 National Review, Ethiopia. Available at: http://unesdoc.unesco.org/images/0023/002317/231724e.pdf

21

Webbink, E., Smits, J., de Jong, E. (2012) Hidden child labor: determinants of housework and family business work of children in 16 developing countries, World Development 40(3): 631642. Woodhead, M., Dornan, P., Murray, H. (2013). What inequality means for children, Young Lives Policy paper, Available at: http://www.younglives.org.uk/content/what-inequalitymeans-children-1 Woldehanna, T., & Hagos, A. (2015). Economic shocks and children’s dropout from primary school: Implications for Education Policy in Ethiopia, Africa Education Review, Vol 12(1): 2847, DOI: 10.1080/18146627.2015.1036548 Woldehanna, T., Gudisa, R., Tafere, Y., Pankhurst, A. (2011). Understanding Changes in the Lives of Poor Children: Initial Findings from Ethiopia, Young Lives, Round 3 Survey Report, Ethiopia. Available at: https://assets.publishing.service.gov.uk/media/57a08aebed915d3cfd000a02/round-3-surveyreport_ethiopia.pdf Woldehanna, T. (2009). Productive safety net programme and children’s time use between work and schooling in Ethiopia. Young Lives working paper N.40. Woldehanna, T., Johns, N., Tefera, B. (2008). The invisibility of Children’s paid and unpaid work. Implications for Ethiopia’s national poverty reduction policy, Childhood, Vol. 15 (2):177-201 Wooldridge, J. (2006) Introductory econometrics: a modern approach. Thomson SouthWestern. World Bank (2017). The World bank in Ethiopia. Country overview, Available at: http://www.worldbank.org/en/country/ethiopia/overview World Bank Group (2015). Ethiopia Poverty Assessment 2014. Poverty Global Practice, Africa Region, Report No. AUS6744, Available at: https://openknowledge.worldbank.org/handle/10986/21323 World Food Program (2012). Ethiopia. Productive Safety Net Programme (PSNP). Fact Sheet. Available at: https://www.wfp.org/sites/default/files/PSNP%20Factsheet.pdf

22

Figures Figure 1 - Path diagram of auto-regressive and cross-lagged relationships between hours of schooling and paid activity over three waves

ε2 School (t2)

β11 γ21

School (t3)

β12 γ22

School (t4)

γ21

γ11 Paid activity (t2)

ε3

β21

β22

Paid activity (t3) є2

Paid activity (t4) є3

Figure 2 - Participation rate (%) of girls and boys at age 19 in various activities by the type of settlement. Girls

100.0

Urban girls

Rural girls

Boys

100.0

80.0

80.0

60.0

60.0

40.0

40.0

20.0

20.0

0.0

0.0

Urban boys

Rural boys

Source: Young Lives Round 4

23

Figure 3 – Average hours spent on various activities across ages (two cohorts) and across the whole sample of children (including zeros)

Source: Young Lives Round 2-4

24

Tables Table 1 - Structural equation model of children’s time allocation between ages 12, 15 and 19

Stability effects (t-1) Boys Girls Reciprocal effects (t-1) Paid work (boys) Paid work (girls) School/study (boys) School/study (girls) Unpaid work (boys) Unpaid work (girls) Care/chores (boys) Care/chores (girls) Controls (t) HH wealth index (Age 8) (boys) HH wealth index (Age 8) (girls) HH cannot raise money (boys) HH cannot raise money (girls) HH has livestock (boys) HH has livestock (girls) Children aged 0-5 in HH (boys) Children aged 0-5 in HH (girls) Female HH head (boys) Female HH head (girls) HH reports drought last year (boys) HH reports drought last year (girls) Urban (boys) Urban (girls) Region (ref: Addis Ababa) Amhara (boys) Amhara (girls) Oromia (boys) Oromia (girls) SNNP (boys) SNNP (girls) Tigray (boys) Tigray (girls) Intercept (boys) Intercept (girls) N SRMR Total R-squared

Age 15 School/ study

Paid work

Unpaid work

Care/ chores

Age 19 School/ study

Paid work

Unpaid work

Care/ chores

0.30*** 0.39***

0.46* 0.12

0.33*** 0.16***

0.04 0.10*

0.28* 0.32**

0.33** 0.16

0.47*** 0.21*

0.08* -0.04

0.07 0.04 0.01 -0.02

-0.10 -0.10 -0.12*** -0.03 -0.09** 0.06

-0.37* -0.13

0.08 0.04 0.02 0.11*

-0.05 0.18 -0.05 -0.19* -0.08 -0.20

-0.42* 0.05

-0.05 0.09 0.06 0.06

-0.04 -0.00 -0.00 -0.04 0.07 0.00

1.71 1.09 -0.32 -0.21 0.86* -0.08 -0.31 0.04 -0.15 0.23 -0.45 -0.12 0.39 0.87

-2.49** -0.10 0.52 0.64 -0.87* -0.15 -0.01 -0.09 -0.05 0.02 -0.02 0.27 0.67*** -0.48*

0.00 0.14 -0.41* -0.07 0.59** 0.04 0.19 -0.14 -0.00 -0.01 0.64* 0.20 -1.07*** 0.03

-1.93*** -0.89 -2.09*** -0.01 -1.57*** 0.20 -1.02* -0.27 5.84*** 3.96*** 873 0.014 0.737

0.13 -0.04 0.23 0.36* 0.34 0.04 0.40 0.15 1.12 0.31

0.13 0.33* 0.95** 0.07 -0.41** -0.00 0.45* 0.31 1.25* 0.21

-0.20 0.09 -0.10 -0.05

-0.01 -0.11 -0.01 -0.09 0.15 0.10

0.57 -2.17*** 0.18 -0.28 -0.28 0.36 0.22*** 0.64*** -0.13 -0.18 -0.26 -0.32 -0.34 0.17

4.27 5.33* 0.06 -0.97** 1.14** 1.17 0.17 -0.34 0.69* 0.28

-5.18*** -1.03 0.37 0.72* -1.28 -0.74*** 0.27 -0.15 -0.84** 0.72*

-0.21 -0.96 -0.28 0.05 0.50 -0.32 -0.71** 0.09 -0.31 -0.10

0.28 -2.05 0.06 -0.09 -0.15 -0.06 0.12 0.90* 0.12 -0.76*

-0.35 0.91

1.93* -0.21

-1.02* -0.71**

0.05 0.22

0.38 1.52*** 0.47 0.82*** 1.30*** 1.48*** 0.12 1.39*** 2.77*** 3.11***

0.37 0.56 -1.49* 0.25 -0.47 1.27 -2.50*** -0.59 3.00* 0.91

-0.69 -1.13* 0.81 -1.44** -0.83 -1.71*** 1.12 -0.77 2.67** 3.02**

0.09 0.24 -0.38 0.17 0.90* 0.22 0.77*** 0.41* 1.65 -0.08

0.97** 1.12* 1.17*** 2.00*** 1.76*** 1.78* 0.11 1.01* 0.89 4.95***

-0.08 0.00

-0.05 0.13

SRMR: standardized root mean squared residual. *** p