This study investigates the relationship between multidimensional household poverty and cognitive and behavioural development during the formative years of childhood (from 9 months to 9 years), using nationally representative longitudinal data from Ireland for the cohort of children born in 2007-2008. The results indicate substantial inequalities in Irish children's cognitive and behavioural outcomes at age 9 by multidimensional poverty duration. Children with at least one spell in poverty (out of four interviews) have worse cognitive and behavioural outcomes. Dynamic structural equation models provide evidence in support of a hybrid family investment/family stress model. Although family investment processes account for some of the cumulative effects of childhood poverty on cognitive outcomes, family stress processes help explain the links between poverty and both cognitive and behaviour outcomes in early childhood. Overall, poverty is strongly related to child outcomes over time via the direct effects of current poverty on child outcomes and path dependency in both poverty and child outcomes over time. There are also indirect effects via the two child outcomes reinforcing each other as children grow older (with the effects of behaviour problems dominating those of cognitive ability), even as the parental investment and maternal stress pathways become less pronounced.

A large quantitative research literature has documented the association between poverty and children's outcomes within a range of well-being indicators, including cognitive ability, education, and social and behavioural development (Van Lancker and Vinck 2019). Children from low income households do worse, on average, in cognitive tests (Gregg et al.2008) and in emotional and behavioural development assessments (Blanden et al.2007; Davies and Woitach 2008). Meanwhile, qualitative research has emphasised the negative consequences of poverty for children's well-being and relationships (Ridge 2002; Treanor 2020). However, it is difficult to disentangle any effect that household income may have in and out of itself from that of its correlates, such as parental education, occupational class, or neighbourhood quality.

Moreover, to the extent that income impacts children's outcomes, the specific pathways and mechanisms remain poorly understood. In a systematic review of experimental, quasi-experimental and observational longitudinal studies on the relationship between household income and child outcomes, Cooper and Stewart (2021) noted both direct causal effects of household income on children's cognitive and behavioural development and indirect effects via parental well-being and the home environment. However, many of these studies were from the United States, and the evidence base was richer for cognitive than non-cognitive outcomes. The positive effects of income were more pronounced among those whose incomes were lower to start with.

At the same time, low household income is not necessarily a sufficient indicator of poverty as a predictor of child outcomes. Poverty is a multidimensional concept that – at least in the European Union – is usually measured using indicators of low disposable household income, enforced lack of socially perceived necessities (i.e. material deprivation), and difficulty making ends meet (i.e. financial strain) (Guio et al.2018). A study using birth cohort data for the United Kingdom found that a combination of all three of these indicators of poverty was associated with the highest levels of child behaviour problems (Schenck-Fontaine and Panico 2019). It highlighted the importance of considering the multidimensional nature of poverty in analyses of child development. A recent study using Irish birth cohort data showed that it is financial strain rather than household income that is associated with behaviour problems in middle childhood and adolescence (Gibbons et al.2023). An earlier study using US data found that material hardship (a latent factor based on deprivation and financial strain indicators) mediated the effect of household income on family processes (Gershoff et al.2007).

The present study adds to this literature by investigating the association between multidimensional poverty and children's cognitive and behavioural development between the ages of 9 months and 9 years in Ireland. Using four waves of longitudinal data for the 2008 birth cohort from the nationally representative ‘Growing Up in Ireland’ (GUI) study in a dynamic structural equation modelling framework, we aim to disentangle the direct contemporaneous effects of poverty on child outcomes from the indirect effects accruing over time via earlier poverty, outcomes, parental investments, and parental stress. By observing children (within their households) at the ages of 9 months, 3, 5, and 9 years, we elucidate the roles of parental investments and parental stress as mediators of poverty during the formative years of childhood. The period up to 9 years of age is critical for cognitive, social, emotional, and physical development (see Britto et al.2013).

The two main theoretical perspectives to analyse the effects of poverty (and socio-economic inequality more broadly) on children's outcomes are the Family Investment Model (FIM) and the Family Stress Model (FSM). The FIM posits that parents use their financial, social, and human capital to foster their children's skills and well-being (Bradley and Corwyn 2002; Heckman 2006; Lareau 2011). Meanwhile, the FSM postulates the links between economic hardship, economic pressure, parental psychological distress, inter-parental and parent–child relationships, and, finally, child outcomes (see Masarik and Conger 2017).

Both models enjoy substantial multi – and inter-disciplinary empirical support, but they are usually employed separately. FIM-based studies tend to analyse differences in cognitive ability and educational outcomes (Duncan et al.2017). FSM is primarily used to predict child mental health, subjective well-being, and behavioural adjustment (Chzhen et al.2021; Ponnet 2014), but several FSM-based studies focussed on education-related outcomes (Justice et al.2019; Schoon et al.2010b).

The two models can also be incorporated into a hybrid one to analyse cognitive and behavioural development. A growing evidence base provides support for doing so. For example, Yeung et al. (2008) examined both family investment and family stress processes as mediators of household income effects on cognitive and behavioural outcomes at age 3-5, using data from the 1997 Child Development Supplement of the US Panel Study of Income Dynamics. They found that the home learning environment played a significant role in mediating the effect of income on children's cognitive outcomes, while maternal emotional stress and parenting practices did so for behaviour outcomes, but the model including all the mediating pathways performed better than the ones focusing on separate pathways. Meanwhile, Gershoff et al. (2007) estimated ‘hybrid’ models with parental investment, parenting stress and parenting behaviour as mediators of family income effects on children's cognitive and behaviour outcomes at the age of 6, using US data from the Early Childhood Longitudinal Study. They too found that parental investment mediated the effects of household income on children's cognitive outcomes, but parenting stress and behaviours did so for both behavioural outcomes and cognitive outcomes. More recently, Layte (2017) used longitudinal data from the UK Millennium Cohort Study to analyse the role of the home learning environment and parental mental health as mediators in the relationship between social class, child cognitive ability and behavioural problems, and teacher-assessed educational performance using. Parental mental health and parental investments (e.g. reading to the child at age 3) had direct effects both on child cognitive ability and behavioural adjustment, supporting a hybrid FIM-FSM model.

A growing body of evidence from longitudinal studies has found the link between household poverty and children's outcomes. Using British birth cohort data, Dickerson and Popli (2016) identified large cumulative negative effects of persistent poverty on children's cognitive development between the ages of 3 and 7. Hansen and Joshi (2007) showed that by the age of three, some disadvantaged children were lagging a full year behind their better-off peers in terms of cognitive development, social skills, and school readiness. The results are supported by Schoon et al. (2010a), who further suggested that the significant negative relationship between persistent financial hardship and children's cognitive development could be mitigated by positive parenting in the UK.

Birth cohort studies show that it is not only differences between poor and non-poor households that are related to inequalities in children's outcomes, but there is an income gradient more generally. Using data for Ireland (i.e. GUI), McMullin et al. (2020) documented a substantial household income gradient in children's cognitive ability at the ages of 3 and 5. Washbrook et al. (2014) showed that higher income in early childhood was associated with higher cognitive ability scores and fewer behaviour problems in middle chilhood, based on English birth cohort data from the Avon Longitudional Study of Parents and Chilren. In a cross-country comparative study of children's early cognitive development, Bradbury et al. (2019) found larger gaps in children's test scores between high-income and middle-income households in the US than in Australia, Canada or the UK, while the gaps between middle-income and low-income households were largely similar. They explain this finding as evidence of a greater concentration of non-monetary resources among high-income households in the US.

Longitudinal studies have also been invaluable for establishing the associations between family resources more generally and children's cognitive development. Bradbury et al. (2015) found that educational achievement was strongly tied to family resources and investment in children during the pre-school period, such as centre-based childcare, books available at home, and home-learning activities in Australia, Canada, the US, and the UK. Caro et al. (2009) tested the trajectory of Canadian students’ academic achievement from 7 to 15 years old and found that the socio-economic status gradient in academic achievement was stable in early childhood but increased during school years. Meanwhile, Skopek and Passaretta (2021) found that achievement gaps by parental education in Germany were already large by the time children started school and remained stable throughout the school years. Similar findings were documented for the Netherlands and the UK (Passaretta et al.2022). A recent analysis of GUI data by McGinnity et al. (2022) showed that occupational social class differentials in reading achievement at age 9 persisted even after accounting for vocabulary scores at age 3, early childhood education and care attendance, and length of school exposure.

Poverty also contributes to differences in children's mental health and social and behavioural development from an early age. By the age of 14, British children with at least one spell in income poverty had a higher risk of stress, emotional and behavioural difficulties, and low subjective wellbeing than those who had not been in income poverty (Rees 2019). Meanwhile, British children living in families experiencing both material deprivation and financial strain were at the highest risk of behaviour problems at age 7 (Schenck-Fontaine and Panico 2019). Financial strain was associated with greater externalised behavioural difficulties in Irish adolescents (Gibbons et al.2023).

We build on this literature by integrating the FIM and FSM frameworks to analyse the consequences of multidimensional poverty for both cognitive and behavioural development in early to middle childhood. First, we document differences in children's outcomes at age 9 by the number of waves they had been in multidimensional poverty. Second, we model the mechanisms through which household poverty affects cognitive and behavioural development between the ages of 9 months and 9 years.

We operationalise the main FIM channel as parental activities stimulating children's learning (e.g. reading to the child at age 3) and the key FSM pathway as parenting stress reported by the mother. FIM predicts that poverty will affect children's cognitive and behavioural development primarily via parental activities rather than directly (Hypothesis 1), while FSM emphasises parental psychological distress and inter-family relationships as the key mediators (Hypothesis 2). A hybrid model suggests that both pathways underlie the association between poverty and child outcomes over time (Hypothesis 3). We also explore the possibility that cognitive and behavioural outcomes reinforce each other over time (Hypothesis 4).

The Irish context is particularly relevant to the study of longer-term consequences of poverty for child development because of the economic crisis of 2008-2011 that followed more than ten years of the ‘Celtic Tiger’ boom. Between the onset of the economic crisis in 2008 and the first signs of economic recovery in 2013, the levels of material deprivation, income poverty and financial strain have risen sharply, particularly among families with children (Nolan and Maître 2017). Thus, using longitudinal data for 2008-2017, we focus on a period of substantial macro-economic turbulence and fluctuation in household incomes when many families with children would experience economic hardship at least once. Indeed, other studies using data from GUI documented reductions in household incomes (Layte and McCrory 2018), subjective perceptions of the recession (Reinhard et al.2018), widespread parental job loss (Mari and Keizer 2021), and increased difficulty making ends meet (Watson et al.2016) at GUI Wave 2 in 2011.

Substantial fluctuation in poverty indicators in the GUI sample over the course of the Great Recession helps analyse the short and longer-term consequences of multidimensional poverty for children. Indeed, other studies have utilised the exogenous shock of the Great Recession and GUI's data collection spanning the crisis and its aftermath to better understand the role of changing household economic circumstances for children. For example, Sprong et al. (2022) described the socio-demographic characteristics of Irish households with children that experienced different trajectories of economic hardship during the recession and recovery. Mari and Keizer (2021) estimated the effects of parental job loss on child cognitive ability and behaviour problems. Layte (2022) analysed the effects of financial strain on children's performance in state examinations. Gibbons et al. (2023) studied the effects of changes in household income, financial strain, and material deprivation on internalised and externalised behavioural difficulties in early to middle childhood and in adolescence.

Growing up in Ireland and the study sample

We use data for the ‘08 Cohort from the GUI study (https://www.growingup.gov.ie/). It collects a rich set of household demographic and socio-economic indicators as well as child cognitive and behavioural outcomes. The ‘08 cohort is a nationally representative sample of around 11,000 children born in 2008, randomly sampled from the Child Benefit register that includes all children resident in Ireland. GUI families were visited for the first time in 2008/2009 when the study child was 9 months old (Wave 1) and followed up again when the child was 3 years old (Wave 2, 2011), 5 years old (Wave 3, 2013) and 9 years old (Wave 4, 2017). See Thornton et al. (2013) for further details of the GUI sampling methodology.

Our study sample includes those who took part in each of these four waves (N = 7,507). To account for non-random attrition across the waves, we re-weight our data using the weighting factors provided in the GUI dataset specifically for this balanced sample. This is important because children from less socio-economically advantaged backgrounds were under-represented at Wave 4 and the longitudinal weights adjust for that (Quail et al.2019). However, these weights do not correct for non-random non-response to survey questions. Income variables are particularly vulnerable to item non-response (see Table 1). Therefore, we estimate our multivariate models (i.e. dynamic structural equation models) using the maximum likelihood with missing values estimator. It draws on all the available information instead of listwise deleting any observation with a missing value on any of the variables in the analysis. Therefore, our analytic sample includes all 7,507 households who took part in all four waves, even if they did not give valid responses to every question. For comparison, a balanced sample based on listwise deletion of missing values would include only 5,000 households.

Table 1. 
Measures of outcomes, predictors, and mediators.
Age 9 monthsAge 3 yearsAge 5 yearsAge 9 years
Cognitive ability ASQ-2 BAS BAS DPRT-R; SAT 
Behaviour problems ASQ-2 SDQ SDQ SDQ 
Poverty Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. 
Home learning environment Frequency of talking to the baby (1 item). Frequency of age-appropriate cognitively stimulating activities. Frequency of age-appropriate cognitively stimulating activities.  
Parental stress and parent-child relationship Parenting Stressors sub-scale. Parenting Stressors sub-scale. Parenting Stressors sub-scale.  
Age 9 monthsAge 3 yearsAge 5 yearsAge 9 years
Cognitive ability ASQ-2 BAS BAS DPRT-R; SAT 
Behaviour problems ASQ-2 SDQ SDQ SDQ 
Poverty Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. Disposable household income below 60% of the median wave-specific median; material deprivation; financial strain. 
Home learning environment Frequency of talking to the baby (1 item). Frequency of age-appropriate cognitively stimulating activities. Frequency of age-appropriate cognitively stimulating activities.  
Parental stress and parent-child relationship Parenting Stressors sub-scale. Parenting Stressors sub-scale. Parenting Stressors sub-scale.  

Wave 1 data collection coincided with the onset of the financial and economic crisis in Ireland. Yet it was in fact at Wave 2 (2011) that households with children experienced some of its worst effects, which were compounded by the austerity measures the government introduced after the ‘bail-out’ by the European Union and the International Monetary Fund in 2010. According to monthly seasonally adjusted unemployment statistics for the 25-74 age group (CSO 2021), unemployment peaked at 13% between Waves 2 and 3 and fell to its pre-crisis levels of around 5% by the time of the age 9 data collection in 2017.

Measures: outcomes, predictors, and mediators

When the study child was 9 months old, the primary caregiver answered a battery of questions from the Ages and Stages Questionnaire, 2nd edition (ASQ-2). The primary caregiver is almost always the mother of the study child (99.9% at Wave 1, 98.4% at Wave 2, 97.9% at Wave 3 and 97.4% at Wave 4). We refer to the primary caregiver as the mother from here onwards for simplicity. ASQ-2 is a screening tool to check an infant's progress in the areas of communication, gross motor skills, fine motor skills, problem solving, and personal-social development (Squires et al.1999). Following Nixon et al. (2013), we rely on the 10-month interval ASQ-2 questionnaire as the one most appropriate for 9-month-olds. The continuous scores for each sub-scale (0-60) load on one latent factor, which accounts for 43% of the total variance in these items in the unweighted Wave 1 sample. We use this as a measure of both cognitive development and behavioural development at 9 months.

Trained interviewers administered age-appropriate cognitive ability tests in the child's home at ages 3, 5 and 9. Two British Ability Scales (BAS) tests were used at ages 3 and 5: Naming Vocabulary and Picture Similarity (Elliot et al.1996). We use the corresponding standardised test scores available in the GUI user database that adjust for the differential difficulty of the items. Following Jones and Schoon (2008) and Chzhen and Bruckauf (2019), who used BAS scores from the UK Millennium Cohort Study, we derive latent cognitive ability scores using principal component analysis in each wave. The two sub-scales load on one factor, which explains 70% of the total variance in these sub-scales at age 3 and 64% at age 5.

A different batch of cognitive assessments were administered at age 9: the Drumcondra Primary Reading Test Revised (DPRT-R) and the Selective Attention Test (SAT). The DPRT-R is based on the Irish national primary school curriculum (Educational Research Centre 2007). We use the logit scores provided in the GUI user database, which adjust for the difficulty of the items and the child's age at interview. Meanwhile, the SAT assesses children's ability to locate 80 small symbols on a map within a minute, while ignoring distracting information. It measures attentional capacity rather than reading ability or curriculum knowledge (Manly et al.1999). We use the number of correctly identified symbols out of 80 alongside the DPRT-R logit score to derive a latent cognitive ability factor. The resulting factor explains 56% of the total variance. We normalise it to have the mean of 100 and the standard deviation of 15 for use in descriptive analyses (Figure 3).

We use the Strengths and Difficulties Questionnaire (SDQ) (Goodman 1997) to measure children's behavioural development at ages 3, 5 and 9. We employ the total score on each of the four (mother-reported) sub-scales: emotional symptoms, conduct problems, hyperactivity/inattention, and peer relationship problems. Responses to each item include ‘Certainly true’, ‘Somewhat true’, and ‘Not true’, resulting in the scores from 0 to 10 on each sub-scale and from 0 to 40 in total. Higher scores correspond to more difficulties. The four sub-scales load on one factor in each wave, explaining 45% of the total variance at age 3, 48% at age 5 and 52% at age 9.

Each wave of the GUI study collects data on annual disposable household income (from all sources) after taxes and benefits, adjusted for household size and composition using the Irish equivalence scale. This scale is very similar to the modified OECD scale used by Eurostat. The national relative income poverty measure in Ireland is the share of the population living in households with annual disposable equivalised income below 60% of the national median. The measure is rooted in Townsend's (1979) theory of relative deprivation, as it benchmarks one's own position in the national income distribution to the level of income that would approximate ‘the minimum acceptable way of life in the member state in which they live’ (European Council 1984, p. 1) According to this measure, the share of children under 18 in Ireland living in income-poor households fluctuated around 20% between 2008 and 2017 (Central Statistics Office 2019).

Yet we do not have a reliable measure of the national median income to replicate the official Irish relative income poverty indicator in the GUI sample. Official poverty statistics use data from the nationally representative cross-sectional annual Statistics on Income and Living Conditions survey, which differs substantially from the GUI study in its sampling methods and the dates of data collection. Therefore, we use 60% of the wave-specific median as the income poverty line in each wave. This produces the rates of 20% in Wave 1, 16% in Wave 2, 16% in Wave 3 and 17% in Wave 4. The falling or stable poverty rate during a recession is an artefact of linking the poverty line to the national (or sample) median income, which fell during the recession. Yet this income poverty measure still captures the spirit of Townsend's relative poverty approach because it allows us to compare the outcomes of the poorest children to their better-off peers even at a time of changing macro-economic circumstances.

However, the issue of the poverty line aside, low household income does not automatically mean that people are living in poverty (Mack and Lansley 1985; Townsend 1979). For example, some households can draw on savings and other assets in times of hardship (Lister 2004), while even higher income households may have significant needs that their incomes do not meet. Therefore, we also include measures of material deprivation and financial strain in the analysis. We define financial strain using the mother's response to the question about the household's ability to make ends meet. This ranges from 1 ‘with great difficulty’ to 6 ‘very easily’. We recoded this into a binary variable where 1 indicates ‘with difficulty or great difficulty’ and 0 ‘with some or no difficulty’.

We measure material deprivation using nine items referring to the household's ability to afford goods or activities that are customary in the Irish society. These are: eating meals with meat, chicken, or fish (or a vegetarian equivalent) every second day; having a roast joint at least once a week; buying new rather than second-hand clothes; each household member having a warm waterproof coat; two pairs of strong shoes; replacing worn-out furniture; keeping the home adequately warm; having family for friends for a drink or meal once a month; buying presents for family or friends at least once a year. Households that report enforced lack of two or more of these nine items are counted as materially deprived in our descriptive analysis (see Figures 3 and 4). We use the ordinal (0-9) material deprivation measure in our structural equation models.

To create a measure of multidimensional poverty history for our descriptive analyses (Figures 3 and 4), we count the number of waves in which the study child lived in a multidimensionally poor household. We construct a measure that draws on low income, material deprivation and financial strain: those who are subject to at least two of these ‘dimensions’ are counted as multidimensionally poor. We then combine those who were in multidimensional poverty in all four waves (from 9 months until 9 years) as well as those who experienced poverty in any three waves into the group that experienced persistent poverty. As a result, 78.5% of the age 9 (Wave 4) sample who took part in all four waves were never in multidimensional poverty, 13.4% experienced it once, 5.6% twice (‘intermittent’) and the remaining 2.5% were in persistent multidimensional poverty.

Parental investment in the child's early learning was assessed at ages 3 (Wave 2) and 5 (Wave 3) using questions about the frequency of age-appropriate cognitively stimulating activities. However, very little such information is available at Wave 1. We use the question about how often the child's mother talked to the baby while doing other things, scored on a 5-point-scale from 1 ‘Never’ to 5 ‘Always’.

At Wave 2 (age 3), the mother reported how many days in an average week anyone at home: reads to the child; helps the child learn ABC or alphabet; helps the child learn numbers or counting; helps the child learn songs, poems, or nursery rhymes; plays board games with the child; paints, draws or colours with the child. These items are scored on a 7-pont scale from 0 to 7 days a week. They form a reliable scale (α = 0.70) and load on one latent factor (accounting for 40% of the total variance).

At Wave 3 (age 5), the mother reported how often she: played with the child using toys, games, or puzzles; visited the library; read to the child; used a computer with the child in educational ways; and went on educational trips together (e.g. to museums, farms). Each item was scored on a 5-point scale from 1 ‘Never’ to 5 ‘Every day’. These five items form a scale with modest internal consistency (α = 0.45) but removing any of these items does not increase the reliability of the scale, so we kept them all. All these items load on one factor (accounting for 33% of the total variance).

We use self-reported information on parenting stress. It is measured using the six-item parenting stressors sub-scale of the Parental Stress Scale (Berry and Jones 1995). It potentially taps into both the parental psychological distress and disrupted parenting aspects of the FSM. Mothers are asked if they agree or disagree with statements regarding their relationship with the child: Caring for my child sometimes takes more time and energy than I have to give; The major source of stress in my life is my child; Having a child leaves me little time and flexibility in my life; Having a child has been a financial burden; It is difficult to balance different responsibilities because of my child; Having a child has meat having too few choices and too little control over my life. The questions are scored on a 5-point scale from 1 ‘Strongly Agree’ to 5 ‘Strongly Disagree’. We reverse code these items so that higher values refer to greater stress. These parenting stress items load on one factor, accounting for 43% of the total variance at Wave 1, 47% at Wave 2 and 49% at Wave 3. The Cronbach's alpha for each wave is acceptable: 0.73 for Wave 1, 0.77 for both Wave 2 and Wave 3.

Maternal depression – measured using an 8-item version of the Centre for Epidemiological Studies Depression Scale – could be used instead of parenting stress to operationalise a key FSM pathway. However, maternal depression exhibits less variation over time than parenting stress and may have a shared unobserved genetic component with children's behavioural difficulties. Having re-estimated our models using maternal depression instead of parenting stress, our findings were qualitatively the same (see Table S9 in Supplementary Materials). Table 1 summarises the definition of outcomes, predictors and mediators used in each wave.

Table 2 shows summary statistics for each wave. The rates of material deprivation and financial strain peaked at Wave 3 (2013). Most children in the sample live with two parents rather than one, have a mother with upper secondary education or lower rather than tertiary education, and have no or just one other child under 18 in the household (who is not necessarily a full sibling) rather than two or more.

Table 2. 
Summary statistics.
Age 9 monthsAge 3 yearsAge 5 yearsAge 9 years
M (SD)NM (SD)NM (SD)NM (SD)N
Low household income (ref. 60% of the median or above) 0.20 7,507 0.16 7,507 0.16 7,507 0.17 7,507 
Material deprivation (ref. not materially deprived) 0.06 7,489 0.07 7,500 0.12 7,500 0.07 7,483 
Financial strain (ref. not reporting financial strain) 0.13 7,501 0.22 7,502 0.26 7,503 0.13 7,487 
Material deprivation 0.25 (0.8) 7,489 0.33 (0.8) 7,500 0.47 (1.1) 7,500 0.3 (0.8) 7,483 
Multidimensionally poor (1 or more dimensions) 0.27 7,485 0.33 7,497 0.37 7,498 0.26 7,477 
Multidimensionally poor (2 or more dimensions) 0.09 7,485 0.11 7,497 0.14 7,498 0.09 7,477 
ASQ_problem solving score (0-60) 46.1 (13.3) 7,094       
ASQ_gross motor score (0-60) 32.1 (16.7) 7,481       
ASQ_fine motor score (0-60) 51.5 (11.1) 7,277       
ASQ_communication score (0-60) 44.8 (11.4) 7,463       
ASQ_personal social score (0-60) 43.9 (11.8) 7,408       
COG_Picture Similarity (20-80)   52.6 (10.8) 7,356 58.3 (10.68) 7,452   
COG Naming Vocabulary (20-80)   50.7 (12.7) 7,153 55.1 (12.1) 7,437   
COG Drumcondra reading logit (45-135)       98.5 (15.2) 7,250 
COG Selective attention test (0-80)       31.9 (9.30) 7,252 
Parent's report SDQ_emotional problems_parent report (0-10)   1.41 (1.42) 7,504 1.61 (1.73) 7,505 2.11 (2.08) 7,498 
SDQ_conduct problems (0-10)   2.20 (1.84) 7,503 1.54 (1.52) 7,505 1.17 (1.41) 7,499 
SDQ_hyperactivity problems (0-10)   3.26 (2.20) 7,502 3.43 (2.52) 7,504 3.25 (2.63) 7,499 
SDQ_peer problems (0-10)   1.22 (1.40) 7,504 1.05 (1.36) 7,504 1.14 (1.55) 7,499 
SDQ_total score (0-40)   8.08 (4.65) 7,502 7.63 (5.01) 7,504 7.67 (5.67) 7,498 
Teacher's report SDQ_emotional problems_teacher report (0-10)     1.35 (1.87) 7,071 1.58 (2.07) 6,856 
SDQ_conduct problems (0-10)     0.77 (1.37) 7,073 0.69 (1.40) 6,857 
SDQ_hyperactivity problems (0-10)     3.16 (2.88) 7,072 2.73 (2.75) 6,857 
SDQ_peer problems (0-10)     1.03 (1.50) 7,070 0.96 (1.57) 6,857 
SDQ_total score (0-40)     6.31 (5.45) 7,070 5.97 (5.79) 6,855 
Parental investment         
Talking to baby (1-5) 4.5 (0.9) 7,506       
ABC or alphabet (0-7)   3.9 (22.4) 7,504     
Numbers or counting (0-7)   5.2 (2.0) 7,504     
Songs, poems or nursery rhymes (0-7)   5.2 (2.0) 7,503     
Paint, draw, colour or play-doh (0-7)   5.0 (2.0) 7,504     
Read to child (0-7)   5.5 (2.0) 7,504 4.5 (0.9) 7,505   
Play games (1-5)   4.3 (2.3) 7,501 4.1 (0.9) 7,506   
Visit library (1-5)     2.2 (1.1) 7,504   
Use computer in educational ways (1-5)     2.5 (1.3) 7,506   
Educational visits (1-5)     2.9 (0.7) 7,506   
Maternal stress         
Time & Energy (1-5) 3.1 (1.3) 7,501 2.3 (1.2) 7,425 2.3 (1.2) 7,408   
Main source (1-5) 1.9 (1.0) 7,501       
Time & flexibility (1-5) 3.0 (1.2) 7,496 1.5 (0.8) 7,426 1.4 (0.8) 7,408   
Financial burden (1-5) 2.2 (1.0) 7,492 2.1 (1.1) 7,424 1.9 (1.0) 7,406   
Responsibilities (1-5) 2.5 (1.1) 7,485 1.7 (0.9) 7,424 1.6 (0.9) 7,409   
Choice & control (1-5) 2.0 (0.9) 7,500 1.9 (1.0) 7,423 1.7 (0.9) 7,409   
Doing enough (1-5)   2.9 (1.2) 7,423 3.0 (1.3) 7,409   
Female child (ref. male) 0.49 7,507       
Mother has tertiary education (ref. upper secondary or less) 0.27 7,503 0.30 7,495 0.29 7,505   
Three or more children in the family (ref: one or two) 0.25 7,507 0.35 7,507 0.44 7,507   
Single parent family (ref: couple family) 0.16 7,507 0.17 7,507 0.16 7,507   
Age 9 monthsAge 3 yearsAge 5 yearsAge 9 years
M (SD)NM (SD)NM (SD)NM (SD)N
Low household income (ref. 60% of the median or above) 0.20 7,507 0.16 7,507 0.16 7,507 0.17 7,507 
Material deprivation (ref. not materially deprived) 0.06 7,489 0.07 7,500 0.12 7,500 0.07 7,483 
Financial strain (ref. not reporting financial strain) 0.13 7,501 0.22 7,502 0.26 7,503 0.13 7,487 
Material deprivation 0.25 (0.8) 7,489 0.33 (0.8) 7,500 0.47 (1.1) 7,500 0.3 (0.8) 7,483 
Multidimensionally poor (1 or more dimensions) 0.27 7,485 0.33 7,497 0.37 7,498 0.26 7,477 
Multidimensionally poor (2 or more dimensions) 0.09 7,485 0.11 7,497 0.14 7,498 0.09 7,477 
ASQ_problem solving score (0-60) 46.1 (13.3) 7,094       
ASQ_gross motor score (0-60) 32.1 (16.7) 7,481       
ASQ_fine motor score (0-60) 51.5 (11.1) 7,277       
ASQ_communication score (0-60) 44.8 (11.4) 7,463       
ASQ_personal social score (0-60) 43.9 (11.8) 7,408       
COG_Picture Similarity (20-80)   52.6 (10.8) 7,356 58.3 (10.68) 7,452   
COG Naming Vocabulary (20-80)   50.7 (12.7) 7,153 55.1 (12.1) 7,437   
COG Drumcondra reading logit (45-135)       98.5 (15.2) 7,250 
COG Selective attention test (0-80)       31.9 (9.30) 7,252 
Parent's report SDQ_emotional problems_parent report (0-10)   1.41 (1.42) 7,504 1.61 (1.73) 7,505 2.11 (2.08) 7,498 
SDQ_conduct problems (0-10)   2.20 (1.84) 7,503 1.54 (1.52) 7,505 1.17 (1.41) 7,499 
SDQ_hyperactivity problems (0-10)   3.26 (2.20) 7,502 3.43 (2.52) 7,504 3.25 (2.63) 7,499 
SDQ_peer problems (0-10)   1.22 (1.40) 7,504 1.05 (1.36) 7,504 1.14 (1.55) 7,499 
SDQ_total score (0-40)   8.08 (4.65) 7,502 7.63 (5.01) 7,504 7.67 (5.67) 7,498 
Teacher's report SDQ_emotional problems_teacher report (0-10)     1.35 (1.87) 7,071 1.58 (2.07) 6,856 
SDQ_conduct problems (0-10)     0.77 (1.37) 7,073 0.69 (1.40) 6,857 
SDQ_hyperactivity problems (0-10)     3.16 (2.88) 7,072 2.73 (2.75) 6,857 
SDQ_peer problems (0-10)     1.03 (1.50) 7,070 0.96 (1.57) 6,857 
SDQ_total score (0-40)     6.31 (5.45) 7,070 5.97 (5.79) 6,855 
Parental investment         
Talking to baby (1-5) 4.5 (0.9) 7,506       
ABC or alphabet (0-7)   3.9 (22.4) 7,504     
Numbers or counting (0-7)   5.2 (2.0) 7,504     
Songs, poems or nursery rhymes (0-7)   5.2 (2.0) 7,503     
Paint, draw, colour or play-doh (0-7)   5.0 (2.0) 7,504     
Read to child (0-7)   5.5 (2.0) 7,504 4.5 (0.9) 7,505   
Play games (1-5)   4.3 (2.3) 7,501 4.1 (0.9) 7,506   
Visit library (1-5)     2.2 (1.1) 7,504   
Use computer in educational ways (1-5)     2.5 (1.3) 7,506   
Educational visits (1-5)     2.9 (0.7) 7,506   
Maternal stress         
Time & Energy (1-5) 3.1 (1.3) 7,501 2.3 (1.2) 7,425 2.3 (1.2) 7,408   
Main source (1-5) 1.9 (1.0) 7,501       
Time & flexibility (1-5) 3.0 (1.2) 7,496 1.5 (0.8) 7,426 1.4 (0.8) 7,408   
Financial burden (1-5) 2.2 (1.0) 7,492 2.1 (1.1) 7,424 1.9 (1.0) 7,406   
Responsibilities (1-5) 2.5 (1.1) 7,485 1.7 (0.9) 7,424 1.6 (0.9) 7,409   
Choice & control (1-5) 2.0 (0.9) 7,500 1.9 (1.0) 7,423 1.7 (0.9) 7,409   
Doing enough (1-5)   2.9 (1.2) 7,423 3.0 (1.3) 7,409   
Female child (ref. male) 0.49 7,507       
Mother has tertiary education (ref. upper secondary or less) 0.27 7,503 0.30 7,495 0.29 7,505   
Three or more children in the family (ref: one or two) 0.25 7,507 0.35 7,507 0.44 7,507   
Single parent family (ref: couple family) 0.16 7,507 0.17 7,507 0.16 7,507   

Source: Growing Up in Ireland ‘08 Cohort. Longitudinal weights used.

Estimation

To account for the dynamic nature of child development, we draw on the value-added plus lagged inputs model of skill formation for both cognitive ability and behavioural functioning (Todd and Wolpin 2007). The child's cognitive (see Figure 1) and behavioural outcomes (see Figure 2), measured in each wave, are a function of previous outcomes and previous ‘inputs’ (i.e. parental investment in cognitively stimulating activities and maternal stress). Following Dickerson and Popli (2016), who modelled children's cognitive development between the ages of 3 and 7 in a dynamic structural equations framework using data from the UK Millennium Cohort Study, we use current multidimensional poverty (from here onwards: poverty, for simplicity) as a predictor of child outcomes in each wave. This allows us to disentangle the direct contemporaneous effects of poverty on child outcomes from the indirect effects accruing over time via the effects of poverty on parental investment and parental stress. Thus, cognitive ability (behaviour difficulties) at the age of 9 months is our baseline outcome, predicted only by household poverty at that time. From age 3 to age 9, however, child outcomes are predicted by prior outcomes, prior parental ‘inputs,’ and current poverty. However, to allow for persistence of poverty, we also include auto-regressive pathways from poverty in one period to poverty in the subsequent period.
Figure 1. 

Main hypothesised relations between poverty, parental investment and child cognitive ability (see Equations 1 and 3).

Figure 1. 

Main hypothesised relations between poverty, parental investment and child cognitive ability (see Equations 1 and 3).

Close modal
Figure 2. 

Main hypothesised relations between poverty, maternal stressors, and child behaviour problems (see Equations 2 and 4).

Figure 2. 

Main hypothesised relations between poverty, maternal stressors, and child behaviour problems (see Equations 2 and 4).

Close modal

Using the counterfactual approach to mediation analysis (see VanderWeele 2015), poverty is the exposure or treatment, parental investment and parental stress are the mediators, and child cognitive ability and behaviour problems are the outcomes. In this causal framework, the direct effect of poverty is its ‘impact’ at a fixed level of the mediators. It is the expected difference in the outcome associated with a change in poverty level if parental investment/stress are the same at all levels of poverty. Meanwhile, the indirect effect is the effect of changes in poverty (or material well-being) that act solely through parental investment or stress.

However, in our dynamic ‘lagged inputs’ model, it is prior parental investment/stress that mediate prior poverty, while contemporaneous poverty has direct effects on children's outcomes. Thus, while the mediators of theoretical interest are parental investment and parenting stress, we have additional mediators in the model (prior outcomes and current poverty) to account for the cumulative nature of children's cognitive and behavioural development and for persistence in poverty over time. For example, outcomes at the age of 3 are a function of 1) the outcomes at age 9 months, which mediate the effect of poverty at that time, 2) current poverty, which mediates the effect of poverty at 9 months and 3) parental investment/stress at 9 months, which also mediate the effect of poverty in that period. If the direct contemporaneous effect of poverty is not statistically significantly different from zero, we would conclude that the effects of poverty are primarily via the mediating mechanisms, in line with Hypotheses 1 and 2. In other words, for children with similar levels of ability or behavioural problems, parental ‘inputs’ and poverty - all measured at an earlier age – there would be no differences in current ability/behaviour by current household poverty.

To interpret the estimates of this model in a causal framework, several assumptions need to be met. First, there should be no exposure-mediator interactions. This would mean that the direct effect of poverty on child outcomes is the same across all levels of parental investment/parenting stress. However, the model does not include a direct pathway from prior poverty to current outcomes, only indirect ones, while current poverty is directly associated with current outcomes, without mediators. Second, there should be no unmeasured confounders in the link between poverty and the mediators or between mediators and the outcomes. This assumption is investigated in the robustness checks.

We model child outcomes, household poverty, parental investment, and maternal stress as latent (unobserved) variables that are represented by a set of manifested (observable) variables. Children's cognitive development is measured using ages and stages questionnaires (at 9 months, Wave 1), Naming Vocabulary and Picture Similarity from the British Ability Scales (age 3 and 5, Waves 2-3), and the Drumcondra Primary Reading Test-Revised and the Selective Attention Test (age 9, Wave 4). Child behaviour problems are measured using ages and stages questionnaires in Wave 1 and the four SDQ sub-scales in Waves 2-4. Note that higher ASQ scores refer to age-appropriate development, while higher SDQ scores refer to greater behaviour problems. For the items that are similar across any two subsequent waves, we allow the error terms to be correlated (i.e. Naming Vocabulary scores at ages 3 and 5, Picture Similarity scores at ages 3 and 5; SDQ sub-scales at ages 3 and 5, and 5 and 9).

Poverty is measured using three binary variables in each wave: disposable household income (adjusted for household size and composition using the Irish equivalence scale) below 60% of the wave-specific median, material deprivation, and financial strain. Parental investment is measured using the frequency of parental engagement in age-appropriate cognitively stimulating activities with the child in Waves 1-3. Maternal stress is measured using 6 items of the parenting stressors sub-scale of the Parental Stress Scale (Berry and Jones 1995) in Waves 1-3.

We estimate a series of structural equation models (SEM) separately for latent cognitive ability (1) and behaviour problems (2) in Stata 17 using the maximum likelihood with missing values estimator. Instead of using modification indices to obtain a model with the best possible fit, we specify the hypothesised pathways and accept the results if the standardised root mean squared residual (SRMR) is below 0.08 (Hu and Bentler 1999). The SEM approach allows us to estimate multiple structural and measurement equations simultaneously and calculate both direct and indirect structural effects.

(1)
(2)
where,
  • - COGt(BPt): Latent cognitive ability (or behaviour problems) of the child at time t;

  • - PIt1 (or STRESSt1): Parental investment (or maternal stress) in the preceding wave;

  • - POVtC,B: multidimensional poverty;

  • - λt,: coefficients of time-varying predictors;

  • - εt: Error terms, assumed to be independent across children/families and over time.

Similarly, parental investment (maternal stress) is a latent variable associated with some covariates. These include time-varying household poverty and two family characteristics: whether there are three or more children in the household (including the study child) and whether the mother has a co-resident partner. Both large family status and lone parent status are associated with child poverty in Ireland (Maître et al.2020) and they can also affect parental investment and maternal stress. The estimates of the effects of poverty are thus net of family structure, a potential confounder.
(3)
(4)
Figures 1 and 2 illustrate the main hypothesised relations between predictors and outcomes embedded in the above equations. Cognitive ability (behaviour problems) are a function of current poverty, prior cognitive ability (behaviour problems), and prior parental investment (maternal stress). The latter are a function of current poverty. From Wave 2 onwards, poverty is a function of prior poverty status. This means that experiencing poverty in early childhood can affect later child outcomes both directly (i.e. concurrently) and indirectly via the effects on prior outcomes (and, thus, prior poverty) and prior home learning environment (maternal stress).
Finally, to address all four hypotheses simultaneously, we test a hybrid FIM+FSM model for cognitive ability and behaviour problems which includes prior maternal stress and home learning environment (Figure 3). This is a more complex model than the first two because it includes both child outcomes simultaneously and allows for the reciprocal cross-lagged pathways between them. Cognitive ability at time t is now a function of not only cognitive ability at time t-1 (as well as poverty at time t, parental investment at t-1 and maternal stress at t-1), but also behavioural problems at t-1, and vice versa (see Equations 5 and 6). Since both cognitive ability and behavioural problems at Wave 1 are measured by developmental progress at 9 months (ASQ-2), it predicts these outcomes at age 3, and the cross-lagged portion of the model involves the period from age 3 to age 5 to age 9. Comparing the magnitude and direction of the reciprocal cross-lagged coefficients helps establish the relative strength of their effects on each other (Selig and Little 2012). Figure 3 represents this structural equation model graphically.
(5)
(6)
We also carry out a series of robustness checks. First, we check for dynamic complementarity of child outcomes and parental inputs/stress by allowing child outcomes at an earlier age to influence parental inputs/stress at a later age. Second, we re-estimate our main models separately by maternal educational status measured when the child was 9 months old (those without a tertiary degree vs those with a tertiary degree) to investigate if the pathways of influence are the same in both groups or if maternal education is an important moderating variable. Third, we check if the behaviour problems findings would differ if internalised and externalised symptoms were analysed separately. Fourth, we re-estimate the main models with controls for maternal education to assess the sensitivity of the coefficients to another potential confounder (in addition to family structure). We do not include maternal education in our main results to avoid over-controlling.
Figure 3. 

Main hypothesised relations between poverty, parental investment, maternal stressors, child cognitive ability and child behaviour problems (see Equations 5 and 6).

Figure 3. 

Main hypothesised relations between poverty, parental investment, maternal stressors, child cognitive ability and child behaviour problems (see Equations 5 and 6).

Close modal

Data are available to apply via: https://www.growingup.ie/, and our replication package is available online (http://doi.org/10.5281/zenodo.8268737)

Poverty duration and child outcomes at age 9

Figure 4 shows differences by income poverty duration in mean cognitive ability scores that are derived from two separate age-appropriate tests and are standardised to have a mean of 100 (SD = 15) at age 9. Children in households that never experienced two or more dimensions of poverty (i.e. low income, material deprivation or financial strain) have statistically significantly higher cognitive ability scores than those who experienced poverty in just one wave (‘one-off’), two waves (‘intermittent’) or three or four waves (‘persistent’) from the age of 9 months onwards (p < 0.001). These are substantively large differences: children who were never in poverty score one-half of a standard deviation higher on the cognitive ability scale than those in persistent poverty. However, there are no statistically significant differences in cognitive ability scores between those in one-off, persistent, or intermittent poverty.
Figure 4. 

Differences in child cognitive ability scores at age 9 by income poverty history.

Note: Error bars represent 95% Confidence Intervals.

Figure 4. 

Differences in child cognitive ability scores at age 9 by income poverty history.

Note: Error bars represent 95% Confidence Intervals.

Close modal
Figure 5 shows differences in mean SDQ scores by poverty duration. The SDQ score is a sum of the emotional, conduct, hyperactivity, and peer problems sub-scales, where higher scores indicate greater behaviour problems. Children who have never been in poverty have lower SDQ scores than their counterparts who experienced poverty at least once (p < 0.001). However, there are no statistically significant differences in SDQ scores among those who experienced poverty once or more than once.
Figure 5. 

Differences in child Strengths and Difficulties Questionnaire scores at age 9 by income poverty history.

Note: Error bars represent 95% Confidence Intervals.

Figure 5. 

Differences in child Strengths and Difficulties Questionnaire scores at age 9 by income poverty history.

Note: Error bars represent 95% Confidence Intervals.

Close modal

Poverty, home learning environment, maternal stress and child outcomes

Table 3 reports the standardised coefficients from a dynamic structural equation model of cognitive ability (see Equations 1 and 3, and Figure 1). It focuses on the family investment pathways between poverty and child cognitive development (see Hypothesis 1).

Table 3. 
Standardised coefficients of latent predictors from a structural equation model of cognitive ability and parental investment (n = 7,507).
Current wavePrevious wave
Cognitive abilityParental investmentPovertyCognitive ability
WaveBSEBSEBSEBSE
Age 9 months Poverty −0.007 0.029 −0.027 0.027 0.874*** 0.022   
Large family   −0.041** 0.016     
Lone parent   0.018 0.024     
Cognitive ability (ASQ)       0.229*** 0.023 
Parental investment       0.032 0.020 
Age 3 Poverty −0.295*** 0.026 −0.072 0.028 0.918*** 0.020   
Large family   −0.112*** 0.019     
Lone parent   0.057* 0.023     
Cognitive ability       0.721*** 0.021 
Parental investment       0.049* 0.020 
Age 5 Poverty −0.070** 0.024 −0.280*** 0.033 0.874*** 0.027   
Large family   −0.217*** 0.021     
Lone parent   0.036 0.030     
Cognitive ability       0.842*** 0.049 
Parental investment       0.085* 0.032 
Age 9 Poverty −0.102*** 0.030       
Current wavePrevious wave
Cognitive abilityParental investmentPovertyCognitive ability
WaveBSEBSEBSEBSE
Age 9 months Poverty −0.007 0.029 −0.027 0.027 0.874*** 0.022   
Large family   −0.041** 0.016     
Lone parent   0.018 0.024     
Cognitive ability (ASQ)       0.229*** 0.023 
Parental investment       0.032 0.020 
Age 3 Poverty −0.295*** 0.026 −0.072 0.028 0.918*** 0.020   
Large family   −0.112*** 0.019     
Lone parent   0.057* 0.023     
Cognitive ability       0.721*** 0.021 
Parental investment       0.049* 0.020 
Age 5 Poverty −0.070** 0.024 −0.280*** 0.033 0.874*** 0.027   
Large family   −0.217*** 0.021     
Lone parent   0.036 0.030     
Cognitive ability       0.842*** 0.049 
Parental investment       0.085* 0.032 
Age 9 Poverty −0.102*** 0.030       

Coefficient of determination (CD) = .839. Standardised Root Mean Square Residual (SRMR) = .055. Estimates of control variables coefficients, factor variances, error variances, error covariances, intercept estimates and disturbances are omitted for clarity. Total effects of poverty on cognitive ability at 9 years: −0.299*** at 9 months; −0.340*** at 3 years; and −0.172*** at age 9.

***p < .001. **p < .01. *p < .05.

Everything else being equal, household poverty is statistically significantly negatively associated with the child's cognitive ability at the ages of 3, 5 and 9 years, but not at the age of 9 months. Children's cognitive ability is associated with their previous test scores: a 1 SD difference in cognitive ability scores at the age of 9 months is associated with 0.23 SD (p < 0.001) higher cognitive scores at age 3, on average; a 1 SD difference in cognitive ability scores at age 3 is associated with a 0.72 SD (p < 0.001) higher predicted score at age 5, and a 1 SD difference at age 5 is associated with a 0.84 SD (p < 0.001) difference at age 9.

Meanwhile, parental investment at an earlier age is positively associated with cognitive test scores at the ages of 5 and 9, but these effects are substantively small. Everything else being equal, a 1 SD difference in the parental investment score at age 3 is associated with a 0.05 SD (p < 0.05) higher predicted cognitive ability score at age 5, and a 1 SD difference in the parental investment score at age 5 is associated with a 0.09 SD (p < 0.05) higher predicted cognitive ability score at age 9.

Thus, poverty can influence current cognitive ability not only directly but also indirectly via the prior parental investment and cognitive ability pathways as well as due to poverty persistence over time. Controlling for large family and lone parent status, poverty is associated with lower parental investment scores at age 5 (B = −0.28; p < 0.001), but not at the earlier waves. The total (direct and indirect) effect of poverty on cognitive ability at age 9 is a substantial −0.30 SD (p<0.001) at age 9 months, −0.34 SD (p < 0.001) at age 3, and −0.17 SD (p < 0.001) at age 5. This is due to the direct effects of poverty on cognitive ability at ages 3, 5 and 9, indirect effects of poverty via the effect of parental investment at age 5 on cognitive ability at age 9 and strong path dependency in both poverty and cognitive ability over time.

Table 4 shows the standardised coefficients from a dynamic structural equation model of behaviour problems (see Equations 2 and 4, and Figure 2), which focuses on the family stress pathways between poverty and children's behavioural development (Hypothesis 2). Poverty is associated with statistically significantly higher behaviour problems at ages 3, 5, and 9, everything else being equal. Poverty also predicts maternal stress when the child is 9 months, 3 years, and 5 years old, with standardised coefficients ranging from 0.21 SD to 0.28 SD (p < 0.001). In turn, maternal stress measured in the previous wave is associated with statistically significantly higher problem behaviour scores at ages 3, 5 and 9, with the largest association at age 3 (B = 0.25, p < 0.001). There is also substantial stability in behavioural problems over time. For example, a 1 SD difference in behaviour problem scores at age 3 is associated with a 0.74 SD higher predicted score at age 5 (p < 0.001), while a 1 SD difference at age 5 is associated with 0.72 SD higher scores at age 9 (p < 0.001), everything else being equal. The cumulative effects of poverty on behaviour problems at 9 years via prior problems and maternal stress, as well as path dependency in poverty and behaviour problems over time, is 0.26 SD (p < 0.001) for poverty experienced at 9 months, 0.26 SD (p < 0.001) at 3 years and 0.14 SD (p < 0.01) at 5 years.

Table 4. 
Standardised coefficients of latent predictors from a structural equation model of behaviour problems and maternal stress (N = 7,507).
Current waveNext wave
Behaviour problemsMaternal stressPovertyBehaviour problems
WaveBSEBSEBSEBSE
Age 9 months Poverty −0.031 0.029 0.276*** 0.030 0.859*** 0.022   
Large family   0.041 0.018     
Lone parent   0.028 0.027     
Behaviour problems (ASQ)       −0.153*** 0.025 
Maternal stress       0.248*** 0.024 
Age 3 Poverty 0.224*** 0.027 0.232*** 0.029 0.920*** 0.02   
Large family   −0.182*** 0.015     
Lone parent   0.075** 0.024     
Behaviour problems       0.742*** 0.021 
Maternal stress       0.099*** 0.02 
Age 5 Poverty 0.082** 0.025 0.207*** 0.031 0.859*** 0.026   
Large family   −0.167*** 0.015     
Lone parent   0.099*** 0.024     
Behaviour problems       0.718*** 0.018 
Maternal stress       0.057** 0.020 
Age 9 Poverty 0.074*** 0.023       
Current waveNext wave
Behaviour problemsMaternal stressPovertyBehaviour problems
WaveBSEBSEBSEBSE
Age 9 months Poverty −0.031 0.029 0.276*** 0.030 0.859*** 0.022   
Large family   0.041 0.018     
Lone parent   0.028 0.027     
Behaviour problems (ASQ)       −0.153*** 0.025 
Maternal stress       0.248*** 0.024 
Age 3 Poverty 0.224*** 0.027 0.232*** 0.029 0.920*** 0.02   
Large family   −0.182*** 0.015     
Lone parent   0.075** 0.024     
Behaviour problems       0.742*** 0.021 
Maternal stress       0.099*** 0.02 
Age 5 Poverty 0.082** 0.025 0.207*** 0.031 0.859*** 0.026   
Large family   −0.167*** 0.015     
Lone parent   0.099*** 0.024     
Behaviour problems       0.718*** 0.018 
Maternal stress       0.057** 0.020 
Age 9 Poverty 0.074*** 0.023       

Coefficient of determination (CD)  = 0.831. Standardised Root Mean Square Residual (SRMR) = .079. Estimates of control variables coefficients, factor variances, error variances, error covariances, intercept estimates and disturbances are omitted for clarity. Total effects of poverty on behaviour problems at 9 years: 0.262*** at 9 months; 0.260*** at 3 years; and 0.135*** at age 9.

***p < .001. **p < .01. *p < .05.

Table 5 reports the standardised coefficients from a hybrid model of cognitive ability and behaviour problems (see Hypothesis 3) that incorporates both the family investment and family stress pathways. It also allows the two child outcomes to predict each other over time (see Hypothesis 4). Poverty is directly associated with both cognitive ability and behaviour problems at ages 3, 5 and 9, with the coefficients comparable in size to those from the separate models of cognitive ability (Table 3) and behaviour problems (Table 4). Similarly, the auto-regressive paths for poverty, cognitive ability and behaviour problems are similar to those in the separate models. It is the added pathways that make the hybrid model different. Thus, everything else being equal, higher maternal stress when the child is 9 months old is associated with greater behaviour problems (but not cognitive ability) at age 3, while prior parental investment (i.e. the frequency of talking to the infant) is not a statistically significant predictor of either outcome. Since poverty predicts greater maternal stress when the child is 9 months old, maternal stress mediates the effects of poverty on behaviour problems at age 3. Similarly, poverty is associated with lower parental investment and greater maternal stress at age 3, but only maternal stress is then associated with greater behaviour problems at age 5. Parental investment at age 3 is not significantly associated with either child outcome at age 5. This is likely because both prior child outcomes are now controlled for simultaneously. Notably, behaviour problems at age 3 predict statistically significantly lower cognitive development at age 5 (B = 0.09, p<0.01). Thus, one of the pathways between poverty at age 3 and child outcomes at age 5 is via behaviour problems at age 3.

Table 5. 
Standardised coefficients of latent predictors from a structural equation model of cognitive ability, behaviour problems, maternal stress and parental investment (N = 7,505).
Current waveNext wave
Cognitive abilityBehaviour problemsParental investmentMaternal stressPovertyCognitive abilityBehaviour problems
BSEBSEBSEBSEBSEBSEBSE
Age 9 months Poverty −0.039 0.029 −0.039 0.029 −0.041 0.027 0.28*** 0.018 0.871*** 0.021     
Large family     −0.038 0.016 0.00041 0.018       
Lone parent     0.025 0.025 0.023 0.027       
Cognitive and behavioural development (ASQ)           0.225*** 0.023 −0.157*** 0.024 
Parental investment           0.027 0.020 0.008 0.018 
Maternal stress           −0.016 −0.016 0.240*** 0.025 
Age 3 Poverty −0.293*** −0.027 0.244*** 0.027 −0.097** 0.029 0.24*** 0.03 0.923*** 0.020     
Large family     −0.108*** 0.018 −0.185*** 0.015       
Lone parent     0.066** 0.024 0.070** 0.024       
Cognitive ability           0.707*** 0.021 −0.035 0.023 
Behaviour problems           −0.085** 0.027 0.733*** 0.021 
Parental investment           0.039 0.021 −0.001 0.02 
Maternal stress           0.037 0.021 0.099*** 0.02 
Age 5 Poverty −0.059* 0.025 0.077** 0.026 −0.30*** 0.033 0.21*** 0.031 0.863*** 0.026     
Large family     −0.218*** 0.021 −0.169*** 0.015       
Lone parent     0.05 0.030 0.094*** 0.024       
Cognitive ability           0.842*** 0.048 −0.080*** 0.022 
Behaviour problems           −0.223*** 0.035 0.715*** 0.02 
Parental investment           0.034 0.033 0.076** 0.027 
Maternal stress           0.125*** 0.026 0.048* 0.02 
Age 9 Poverty −0.076* 0.032 0.074** 0.026           
Current waveNext wave
Cognitive abilityBehaviour problemsParental investmentMaternal stressPovertyCognitive abilityBehaviour problems
BSEBSEBSEBSEBSEBSEBSE
Age 9 months Poverty −0.039 0.029 −0.039 0.029 −0.041 0.027 0.28*** 0.018 0.871*** 0.021     
Large family     −0.038 0.016 0.00041 0.018       
Lone parent     0.025 0.025 0.023 0.027       
Cognitive and behavioural development (ASQ)           0.225*** 0.023 −0.157*** 0.024 
Parental investment           0.027 0.020 0.008 0.018 
Maternal stress           −0.016 −0.016 0.240*** 0.025 
Age 3 Poverty −0.293*** −0.027 0.244*** 0.027 −0.097** 0.029 0.24*** 0.03 0.923*** 0.020     
Large family     −0.108*** 0.018 −0.185*** 0.015       
Lone parent     0.066** 0.024 0.070** 0.024       
Cognitive ability           0.707*** 0.021 −0.035 0.023 
Behaviour problems           −0.085** 0.027 0.733*** 0.021 
Parental investment           0.039 0.021 −0.001 0.02 
Maternal stress           0.037 0.021 0.099*** 0.02 
Age 5 Poverty −0.059* 0.025 0.077** 0.026 −0.30*** 0.033 0.21*** 0.031 0.863*** 0.026     
Large family     −0.218*** 0.021 −0.169*** 0.015       
Lone parent     0.05 0.030 0.094*** 0.024       
Cognitive ability           0.842*** 0.048 −0.080*** 0.022 
Behaviour problems           −0.223*** 0.035 0.715*** 0.02 
Parental investment           0.034 0.033 0.076** 0.027 
Maternal stress           0.125*** 0.026 0.048* 0.02 
Age 9 Poverty −0.076* 0.032 0.074** 0.026           

Coefficient of determination (CD)  = 0.852. Standardised Root Mean Square Residual (SRMR) = .066. Estimates of control variables coefficients, factor variances, error variances, error covariances, intercept estimates and disturbances are omitted for clarity. Total effects of poverty on cognitive ability at 9 years: −0.324*** at 9 months; −0.342*** at 3 years; and −0.116*** at age 9. Total effects of poverty on behaviour problems at 9 years: 0.278*** at 9 months; 0.273*** at 3 years; and 0.111*** at age 9.

***p < .001. **p < .01. *p < .05.

Further, behaviour problems at age 5 are associated with lower cognitive development at age 9 (B = −0.22, p < 0.001), controlling for prior cognitive development, parental investment, and maternal stress, as well as current poverty. Similarly, cognitive ability at age 5 is associated with fewer behaviour problems at age 9 (B = −0.08, p < 0.001), controlling for prior behaviour problems, parental investment, and maternal stress, as well as current poverty. The cross-lagged coefficient of behaviour problems is larger in absolute terms than the reciprocal coefficient of cognitive ability, and a Wald test shows that the two coefficients are statistically significantly different from each other (p < 0.01). This suggests that behaviour problems dominate the mutually reinforcing relationship between the two aspects of child development: behaviour problems are a better predictor of cognitive ability than vice versa.

Meanwhile, controlling for prior development outcomes and current poverty, parental investment at age 5 is associated with greater behaviour problems at age 9 (B = 0.08, p < 0.01) and not with cognitive ability, while maternal stress is a significant predictor of both higher cognitive ability (B = 0.13, p<0.001) and greater behaviour problems (B = 0.05, p < 0.05). These are substantively small associations, and some of them have counterintuitive signs. This may be due to the inclusion of the cross-lagged pathways between the two child outcomes, which already contain the effects of parental investment and maternal stress at the earlier ages.

Finally, the cumulative effects of poverty at the age of 9 months, 3 years, and 5 years on the age 9 outcomes are comparable to those in the separately estimated models, even though the pathways of influence are now more complicated. Poverty is still strongly related to child outcomes over time via the direct effects of current poverty on child outcomes and path dependency in both poverty and child outcomes over time. There are also indirect effects via the two child outcomes reinforcing each other as children grow older, even as the parental investment and maternal stress pathways are less pronounced.

Robustness checks

We checked for dynamic complementarity of child outcomes and parental inputs/stress by allowing child outcomes at an earlier age to influence parental inputs/stress at a later age. Table S1 in Supplemental Information reports the results of a structural equation model of cognitive ability identical to the one in Equation 1 but with an additional pathway from cognitive ability at age 3 to parental investment age at 5. This coefficient is positive and statistically significant: a 1SD difference in cognitive ability at age 3 is associated with 0.2 SD greater parental investment score at age 5 (p < 0.001), everything else being equal. In the next period, parental investment at age 5 is no longer statistically significantly associated with cognitive ability at age 9. However, the total effects of earlier poverty on cognitive ability at age 9 are nearly identical to those in the main model.

Similarly, Table S2 shows that behavioural problems at age 3 are significantly associated with greater maternal stress at age 5, but the total effects of poverty at earlier ages on behavioural problems at age 9 are similar to the estimates in the main model. This alternative specification suggests that poverty at age 9 months is associated with greater maternal stress, which in turn is associated with greater child behavioural problems at age 3. These, in turn, are related to greater maternal stress at age 5. Meanwhile, poverty has a direct effect on behavioural problems at ages 3, 5 and 9. Overall, these alternative models indicate that the relationship between parental investment (stress) and cognitive (behavioural) development may be mutually reinforcing and at least in part triggered by poverty at an early age.

In a second series of checks, we re-estimated our main models allowing the coefficients to vary by maternal educational status (those without a tertiary degree vs those with a tertiary degree at Wave 1). Table S3 shows the results for cognitive ability. Although the findings are similar to those in the main model in Table 3, there are several differences to be noted. The effect of parental investment at age 9 months (i.e. frequency of talking to the infant) on cognitive ability at age 3, the effect of poverty at age 3 on parental investment at age 3, and the effects of poverty on cognitive ability at ages 5 and 9, are all greater for children whose mothers are university educated. The total effects of poverty on cognitive ability at age 9 are then larger in higher educated families. Meanwhile, the effect of poverty at age 3 on behaviour problems are larger in higher educated families, even as the effect of poverty on maternal stress at ages 3 and 5 is larger for lower educated families (Table S4). The total effects of poverty on behaviour differences at age 9 are comparable for the two sub-groups.

To check if the associations between poverty and child outcomes may be confounded by other dimensions of family socio-economic status, we re-estimated our main models (Equations 1-4) with additional controls for maternal educational attainment (i.e. those without a tertiary degree vs those with a tertiary degree at Wave 1). Table S4 shows children whose mothers are tertiary educated have substantially higher cognitive ability scores, on average, at ages 3, 5, and 9 years (p < 0.001). Maternal education in associated with statistically significantly higher parental investment at ages 3 (p < 0.05) and 5 (p < 0.001). However, the direct contemporaneous effects of poverty are only slightly attenuated compared to the corresponding estimates in Table 3 and the total effects of poverty are only somewhat smaller. Meanwhile, Table S7 shows that children whose mothers are tertiary educated have lower behaviour problems scores at age 3 (p < 0.001), but there are no statistically significant associations at ages 5 or 9. Tertiary educated mothers have higher levels of parenting stress (p < 0.001) in all the waves. The direct and total effects of poverty on behaviour problems are somewhat attenuated compared to the corresponding estimates in Table 4, but remain of the same sigh and statistical significance.

We also checked if the behaviour problems findings would differ if internalised and externalised symptoms were analysed separately. Tables S7 and S8 in Supplemental Information report the results of a specification equivalent to that in Table 4 separately for internalised and externalised symptoms, respectively. Poverty is associated with higher internalised symptoms scores at ages 3 and 9, while prior maternal stress is associated with higher internalised symptoms at ages 3, 5 and 9. The cumulative effects of poverty for internalised symptoms at age 9 via maternal stress and prior symptoms are 0.25 SD (p < 0.001) for poverty experienced at 9 months, 0.26 SD (p < 0.001) at 3 years, and 0.13 SD at age 5.

Poverty is associated with higher externalised symptoms scores at ages 3 and 5, while prior maternal stress is associated with higher externalised symptoms at ages 3, 5 and 9. The total effects of poverty on externalised symptoms at age 9 via maternal stress and prior symptoms are 0.22 SD (p < 0.001) for poverty experienced at 9 months, 0.22 SD (p < 0.001) at 3 years, and 0.10 SD (p < 0.001) at 5 years. These results suggest that dividing behaviour problems into internalised and externalised symptoms does not alter the main findings, while producing less precisely estimated coefficients in dynamic SEM models.

This study investigated the relationship between child poverty and child development, using a unique Irish longitudinal dataset for the period 2008-2017, when many households experienced at least one instance of economic hardship. Our key contributions are two-fold. First, we establish the differences in children's cognitive and behavioural outcomes at age 9 by the number of instances of multidimensional poverty since the age of 9 months. We demonstrate that even one instance of poverty is harmful to child development. Second, unlike other studies that documented similar patterns using longitudinal data from birth cohort studies in Ireland and the UK (Maître et al.2020; Rees 2019), we investigate the theoretical mechanisms through which poverty affects child development over time. We focus on the family investment and family stress processes as well as the potentially mutually reinforcing relationships between children's cognitive and behavioural development outcomes. We are contributing to the longitudinal studies literature that draws on both family investment and stress processes in explaining socio-economic differences in children's cognitive and behavioural outcomes in early to middle childhood (Layte 2022, 2017) and to the related body of evidence that operationalises poverty as not solely based on household income (Gershoff et al.2007; Gibbons et al.2023; Schenck-Fontaine and Panico 2019).

First we document substantial inequalities in Irish children's cognitive and behavioural outcomes by multidimensional poverty duration. Children with at least one spell in poverty (out of four interviews) since the age of 9 months have substantially worse cognitive and behavioural outcomes at age 9. Poverty is measured as household experience of two or more dimensions of economic hardship simultaneously (i.e. low income, financial strain, or material deprivation). We use this conservative definition of multidimensional poverty for descriptive statistics. In multivariate analyses, poverty is a latent factor based on the three indicators of economic hardship.

Overall, these results suggest that persistent multidimensional poverty is particularly harmful to children's outcomes, but even a one-off spell in poverty can matter. This is in line with findings in Maître et al. (2020), who documented similar patterns using data from Growing Up in Ireland, and Rees (2019), who found lower wellbeing in terms of cognitive ability, physical health, and emotional and behavioural difficulties among children with more extensive histories of poverty in the UK.

Second, we test hypotheses derived from the family investment and family stress literatures using dynamic structural equation models to understand these channels of influence. Although the links between poverty and adverse children's outcomes are well documented descriptively (Van Lancker and Vinck 2019), our study is among the few to analyse the specific channels through which poverty affects children's cognitive and behavioural development. In a series of models estimated separately for: 1) cognitive ability, poverty and parental investment and 2) behaviour problems, poverty and maternal stress, we find that family investment processes – based on information about the cognitively stimulating activities that parents engage in with their young children – account for some but not all of the cumulative effects of childhood poverty on cognitive ability test scores. Similarly, family stress processes – manifested in maternal stress – play a key role in explaining the links between poverty and behaviour problems in early to middle childhood, but do not fully explain the relationship between poverty and child outcomes over time. These results only partially support our initial hypotheses 1 and 2 because our family investment and stress measures do not fully account for the association between poverty and child development outcomes. This could be because we included these measures in separate models.

In contrast, the more complex model, which includes both child development outcomes and both mediating channels simultaneously, sheds light on all four of our hypotheses. Here we find some evidence in support of a hybrid family investment and family stress model. Poverty is associated with lower parental investment and greater parental stress at ages 3 and 5. Controlling for prior cognitive ability and behaviour problems, prior maternal stress is associated with greater behavioural problems at age 5 but not with cognitive ability, while prior parental investment is no longer associated with either of the two child outcomes at that age.

The picture gets more complicated in the period between age 5 and age 9, when the two child outcomes already contain the effects of poverty, parental investment, and maternal stress at the earlier ages. Prior maternal stress is associated with greater behaviour problems at age 9 as well as with greater cognitive ability. This can be interpreted as follows: among the children with similar levels of poverty at age 9, cognitive and behavioural outcomes at age 5 as well as parental investment at age 5, those whose mothers experienced greater levels of parenting stress (when the children were 5) had higher cognitive ability scores at age 9. Although the temporal ordering of the measures helps preclude reverse causality, this may still indicate that greater cognitive ability leads to greater parental anxiety. Meanwhile, although parental investment at age 5 is positively associated with cognitive ability at age 9, we also observe a small but statistically significant association between higher intensity of home learning activities at age 5 and greater behaviour problems at age 9. Since most children are already at school at age 5 (McGinnity et al.2022), this may suggest a compensatory relationship: parents invest more in children who are more at risk of behaviour problems. Although further work is needed to focus specifically on the potentially compensatory effects of parental investment and stress processes, this is in line with Washbrook et al.'s (2014) finding that some of the income-mediating pathways can have opposing associations different child outcomes. Importantly, we find that behavioural problems are stronger predictors of cognitive ability than vice versa, suggesting that greater behaviour problems may impede children's ability to learn, everything else being equal, while greater cognitive ability is not necessarily a protective factor for the development of behaviour problems. Further, both parental investment and maternal stress are related to behavioural problems. This indicates that behavioural problems may mediate the effects of poverty and family processes on child cognitive development.

Overall, our hybrid model results contrast somewhat with Layte (2017) who found more conclusive evidence in support of a hybrid model using data from the first four sweeps (up to age 7) from the UK Millennium Cohort Study. Both the home learning environment and maternal psychological distress directly affected both child cognitive ability and behaviour problems, controlling for parental social class and household income. In contrast, we find that maternal stress plays a more important role as a mediator of poverty overall. Some of the differences in our findings may be due to different research questions, data, measures, and methods used. Importantly, Layte (2017) focused on social class rather than poverty or material well-being. More research is needed to contrast the roles of family investment and family processes in children's cognitive and behavioural development over time. Moreover, it would be useful to have data on children's own perceptions of hardship and economic pressure, complementing parental reports. A recent adaptation of the standard FSM that relied on data from both parents and children showed that economic hardship can affect children more directly and immediately than via parental distress (Chzhen et al.2021).

Our analysis has several limitations. First, although we used longitudinal weights designed to account for differential panel attrition and a maximum likelihood with missing values estimator to adjust for item non-response, our findings are based on nationally representative data for a specific cohort (i.e. those born in 2008 and resident in Ireland at the age of 9 months). These findings need to be replicated using longitudinal data for a different country, cohort, or period.

Second, we could only rely on the measures available in the GUI dataset. Thus, we do not have measures of child cognitive ability or behaviour problems at the ages of 9 months. We used ages and stages questionnaires as baseline measures for both outcomes, which is a valid measure of early developmental delays (Gollenberg et al.2010). We showed that ASQ at 9 months is strongly predictive of both higher cognitive ability scores and lower behaviour problem scores at age 3. We also rely on parent-reported measures of children's behaviour problems, which could have been biased downward (i.e. towards fewer problems). Teacher-reported measures are available when the children are aged 5 and 9, but they have more missing values than parental reports. Similarly to other studies (Goodman et al.2010), parent and teacher reports are moderately but statistically significantly (p<0.001) correlated for each sub-scale age ages 5 and 9 (see Tables S10-S11 in the Annex). It does not appear that parents tend to under-report their children's behaviour problems: teacher reports of SDQ sub-scales are lower, on average, than parental reports (see Table 2).

Third, our results may not necessarily be interpreted as causal, despite engaging with causal theories of child development and causal mediation frameworks. Our dynamic structural equation models estimate outcomes between the ages of 9 months and 9 years, using key predictors from prior waves, but reverse causation in later years may still be a problem if parents adjust their home learning actions based on their perceptions of their child's cognitive ability, or if maternal mental health is affected by the child's behaviour difficulties. While Dickerson and Popli (2016) found no evidence of reverse causality in parental inputs and cognitive ability in a similar study using UK data, our robustness checks show that higher cognitive ability at age 3 is associated with greater parental investment at age 5. Similarly, greater behaviour problems at age 3 are associated with greater maternal stress at age 5. However, the cumulative effects of childhood poverty remain robust to alternative specifications of the intermediate pathways. Poverty at the earlier ages triggers a series of reactions that culminate in substantial cognitive and behavioural inequalities by age 9 via both family investment and family stress processes, with some unexplained direct effects of current poverty remaining.

Furthermore, we cannot discount the possibility of unmeasured confounders. We included family size and lone parent status in the baseline (age 9 months) equations because there are differences by family structure in poverty, parental investment/stress and child outcomes. When we further controlled for maternal education in sensitivity analyses, the results were qualitatively the same. However, there are other potential confounders that we cannot account for using our data, such as genetic information.

We also assume that there are no interactions with potential confounders. However, we have found some evidence for our dynamic structural equation models performing somewhat differently in for households with lower and higher educated mothers. When we re-estimated our main models (Figures 1 and 2) allowing all the coefficients to vary by maternal education (i.e. with or without a university degree), we observed that the total effects of poverty on cognitive ability are somewhat larger in higher educated families, with no sub-group differences behaviour problems. Meanwhile, Mari and Keizer (2021) found larger effects of parental job loss on children's behaviour problems at age 5 among those with tertiary-educated mothers, with no sub-group differences for cognitive ability, using the same survey. This suggests that differences in material hardship may translate into greater inequalities in children's outcomes among higher educated families, who are at a lower risk of poverty overall. If education is a proxy for the non-income dimensions of socio-economic status, this implies a moderating relationship. Since socio-economic status is a complex concept that taps into different aspects of social position (Bradley and Corwyn 2002), further work is needed to disentangle the relationship between different aspects of household socio-economic status (i.e. material well-being, education and occupational social class), family inputs and child outcomes.

There are other avenues for further research. Given the complex dynamic nature of socio-economic inequalities in family processes and child development, sociologists could rely more on the causal mediation methods developed in epidemiology. For example, there are approaches for handling multiple mediators (VanderWeele and Vansteelandt 2014) and mediator-outcome confounding (Wodtke and Zhou 2020). A recent application of the latter in sociology is Mari and Keizer's (2021) analysis of parental job loss and children's outcomes. Furthermore, advances in genetics (e.g. polygenic scores) and sociogenomics (Mills and Tropf 2020) offer novel directions for the study of socio-economic inequalities in child development.

Concluding remarks

Despite these limitations, a key strength of our study lies in analysing both cognitive and behavioural outcomes during an important developmental window from infancy (9 months) to middle childhood (9 years). We use data from a unique Irish study for the period 2008-2017, when many households experienced hardship due to the Great Recession and austerity policies. We show that experiencing poverty from age 3 is particularly harmful to both cognitive and behavioural development. This is due to indirect cumulative effects over time via family processes (i.e. the parental investment in cognitively stimulating activities; parenting stress) and strong persistence in cognitive and behavioural development. This is consistent with an interdisciplinary body of work highlighting the early years as a key developmental window and calling for a special focus on the early years in anti-poverty policies (see Dickerson and Popli (2016)). However, we also find substantial direct effects of poverty on child outcomes at later ages, even after controlling for prior outcomes and prior parental ‘inputs’. This means that public policies need to continue supporting families past the earlier ages. In fact, we do not observe any differences by household poverty in children's developmental milestones at age 9 months, but only from age 3 onwards.

Our findings have other potential policy implications. Since parental investment and parenting stress mediate the effects of poverty, interventions can be designed that improve their levels among the poor. Thus, poorer households need to be enabled to engage in cognitively stimulating activities with their children while experiencing less parenting stress. However, we also observed direct associations between poverty and children's outcomes even after adjusting for these mediators. This suggests that a more effective approach may be tackling poverty itself. Thus, rather than prioritising parent-focussed behavioural interventions, tax-benefit systems can redistribute incomes to reduce relative income poverty while accessible and affordable services can reduce financial strain and material deprivation. A better understanding of the family processes through which poverty affects children's outcomes is valuable without implying that efforts to improve disadvantaged children's outcomes must focus solely on these processes, while ignoring poverty and wider socio-economic inequality.

Thus our study contributes to the emerging literature that combines the family investment and stress processes in explaining socio-economic differences in children's outcomes (Layte 2022, 2017). It also speaks to the broader literature on child poverty and its consequences, highlighting the value of longitudinal birth cohort data. Although cross-sectional studies highlight the factors associated with higher risks of poverty at a point in time, they cannot distinguish between groups who spend only a brief time in poverty and those who remain poor. Longitudinal studies demonstrate that while most income poverty is short-term, substantial minorities suffer from persistent poverty (Goodin et al.1999; Headey 2008; Rees 2019). Persistent lack of income not only leads to debt (Lea et al.1995) but also exacerbates long-term anxiety and stress (Rowlingson and McKay 2005). As children depend on their caregivers and are not responsible for their household circumstances, especially at younger ages, child poverty is both unfair and inefficient (Brooks-Gunn and Duncan 1997). It undermines meritocracy and equality of opportunity, standing in the way of all children reaching their full potential.

The authors wish to thank Richard Layte, Jan Skopek, and PhD students at the Department of Sociology at Trinity College Dublin for their feedback on earlier versions of this study, as well as participants in the 2021 European Consortium for Sociological Research Annual Conference, the 2021 Growing Up in Ireland conference, and the 2023 Spring Meeting of the Research Committee 28 on Social Stratification and Mobility of the International Sociological Association. Growing Up in Ireland (GUI) is funded by the Department of Children and Youth Affairs (DCYA). It is managed by DCYA in association with the Central Statistics Office (CSO). Results in this report are based on analyses of data from Research Microdata Files provided by the Central Statistics Office (CSO). Neither the CSO nor DCYA take any responsibility for the views expressed or the outputs generated from these analyses.

No potential conflict of interest was reported by the author(s).

Berry
,
J. O.
and
Jones
,
W. H.
(
1995
) ‘
The parental stress scale: initial psychometric Evidence
’,
Journal of Social and Personal Relationships
12
:
463
72
. doi: .
Blanden
,
J.
,
Gregg
,
P.
and
Macmillan
,
L.
(
2007
) ‘
Accounting for intergenerational income persistence: noncognitive skills, ability and education
’,
The Economic Journal
117
:
C43
C60
.
Bradbury
,
B.
,
Corak
,
M.
,
Waldfogel
,
J.
and
Washbrook
,
E.
(
2015
)
Too Many Children Left Behind: The U.S. Achievement gap in Comparative Perspective
,
New York
:
Russell Sage Foundation
.
Bradbury
,
B.
,
Waldfogel
,
J.
and
Washbrook
,
E.
(
2019
) ‘
Income-Related gaps in early child cognitive development: Why Are they larger in the United States than in the United Kingdom, Australia, and Canada?
’,
Demography
56
:
367
90
. doi: .
Bradley
,
R. H.
and
Corwyn
,
R. F.
(
2002
) ‘
Socioeconomic status and child Development
’,
Annual Review of Psychology
53
:
371
99
. doi: .
Britto
,
P. R.
,
Engle
,
P. L.
and
Super
,
C. M.
(
2013
)
Handbook of Early Childhood Development Research and its Impact on Global Policy
,
New York
:
Oxford University Press
.
Brooks-Gunn
,
J.
and
Duncan
,
G. J.
(
1997
) ‘
The effects of poverty on children
’,
The Future of Children
7
:
55
. doi: .
Caro
,
D. H.
,
McDonald
,
J. T.
and
Willms
,
J. D.
(
2009
) ‘
Socio-economic status and academic achievement trajectories from childhood to adolescence
’,
Canadian Journal of Education
32
:
558
90
.
Central Statistics Office
(
2019
)
Ireland's UN SDGs 2019 - Report on indicators for goal 1 no poverty [WWW Document].
URL https://www.cso.ie/en/releasesandpublications/ep/p-sdg1/irelandsunsdgs2019-reportonindicatorsforgoal1nopoverty/poverty/ (accessed 3.25.22).
Central Statistics Office
(
2021
)
Seasonally adjusted monthly unemployment [WWW Document].
URL https://ws.cso.ie/public/api.restful/PxStat.Data.Cube_API.ReadDataset/MUM01/XLSX/2007/en (accessed 2.21.22).
Chzhen
,
Y.
and
Bruckauf
,
Z.
(
2019
) ‘
Household income and sticky floors in children's cognitive development: evidence from the United Kingdom millennium cohort Study
’,
Longitudinal and Life Course Studies
10
:
307
26
. doi: .
Chzhen
,
Y.
,
Howarth
,
C.
and
Main
,
G.
(
2021
) ‘
Deprivation and intra-family conflict: children as agents in the family stress Model
’,
Journal of Marriage and Family
84
:
121
40
. doi: .
Cooper
,
K.
and
Stewart
,
K.
(
2021
) ‘
Does household income affect children's outcomes? A systematic review of the Evidence
’,
Child Indicators Research
14
:
981
1005
. doi: .
Davies
,
P. T.
and
Woitach
,
M. J.
(
2008
) ‘
Children's emotional security in the interparental relationship
’,
Current Directions in Psychological Science
17
:
269
74
.
Dickerson
,
A.
and
Popli
,
G. K.
(
2016
) ‘
Persistent poverty and children's cognitive development: evidence from the UK millennium cohort Study
’,
Journal of the Royal Statistical Society. Series A (Statistics in Society)
,
535
58
.
Duncan
,
G. J.
,
Magnuson
,
K.
and
Votruba-Drzal
,
E.
(
2017
) ‘
Moving beyond correlations in assessing the consequences of Poverty
’,
Annual Review of Psychology
68
:
413
34
. doi: .
Educational Research Centre
(
2007
) ‘
Drumcondra primary reading test–Revised.
’ Levels 3–6 administration manual and technical manual [WWW Document].
Elliot
,
C. D.
,
Smith
,
P.
and
McCulloch
,
K.
(
1996
) ‘
British ability scales second edition (BAS II) [WWW Document].
’ NFER-Nelson, Windsor. URL (accessed 6.4.21).
European Council
(
1984
) ‘
Council decision of 19 December 1984 on specific community action to combat poverty, CELEX1
’,
Official Journal of the European Communities.
Gershoff
,
E. T.
,
Aber
,
J. L.
,
Raver
,
C. C.
and
Lennon
,
M. C.
(
2007
) ‘
Income Is Not enough: incorporating material hardship into models of income associations With parenting and child Development
’,
Child Development
78
:
70
95
. doi: .
Gibbons
,
R. A.
,
Sprong
,
S.
and
Chzhen
,
Y.
(
2023
) ‘
Growing Up in the great recession: The effects of three dimensions of economic well-being on child behavioral difficulties from ages 3 to 17
’,
Journal of Youth and Adolescence
52
:
1024
38
. doi: .
Gollenberg
,
A. L.
,
Lynch
,
C. D.
,
Jackson
,
L. W.
,
McGuinness
,
B. M.
and
Msall
,
M. E.
(
2010
) ‘
Concurrent validity of the parent-completed ages and stages questionnaires, 2nd Ed. with the bayley scales of infant development II in a low-risk sample
’,
Child: Care, Health and Development
36
:
485
90
. doi: .
Goodin
,
R. E.
,
Goodin
,
R. E.
,
Headey
,
B.
,
Muffels
,
R.
and
Dirven
,
H.-J.
(
1999
)
The Real Worlds of Welfare Capitalism
,
Cambridge
:
Cambridge University Press
.
Goodman
,
R.
(
1997
) ‘
The strengths and difficulties questionnaire: a research note
’,
Journal of Child Psychology and Psychiatry
38
:
581
6
.
Goodman
,
A.
,
Lamping
,
D. L.
and
Ploubidis
,
G. B.
(
2010
) ‘
When to Use broader internalising and externalising subscales instead of the hypothesised five subscales on the strengths and difficulties questionnaire (SDQ): data from British parents, teachers and Children
’,
Journal of Abnormal Child Psychology
38
:
1179
91
. doi: .
Gregg
,
P.
,
Janke
,
K.
and
Propper
,
C.
(
2008
)
Handedness and Child Development
,
Bristol
:
Centre for Market and Public Organisation 43
.
Guio
,
A.-C.
,
Gordon
,
D.
,
Marlier
,
E.
,
Najera
,
H.
and
Pomati
,
M.
(
2018
) ‘
Towards an EU measure of child deprivation
’,
Child Indicators Research
11
:
835
60
. doi: .
Hansen
,
K.
and
Joshi
,
H.
(
2007
) ‘
Millennium cohort study second survey A user's guide to initial findings
’,
Centre for Longitudinal Studies
198
.
Headey
,
B.
(
2008
) ‘
Poverty is low consumption and low wealth, not just low income
’,
Social Indicators Research
89
:
23
39
. doi: .
Heckman
,
J. J.
(
2006
) ‘
Skill formation and the economics of investing in disadvantaged children
’,
Science
312
:
1900
2
.
Hu
,
L.
and
Bentler
,
P. M.
(
1999
) ‘
Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives
’,
Structural Equation Modeling: A Multidisciplinary Journal
6
:
1
55
. doi: .
Jones
,
E. M.
and
Schoon
,
I.
(
2008
) ‘
Child cognition and behaviour.’ Millennium Cohort Study Third Survey: A user
's guide to initial findings
118
144
.
Justice
,
L. M.
,
Jiang
,
H.
,
Purtell
,
K. M.
,
Schmeer
,
K.
,
Boone
,
K.
,
Bates
,
R.
and
Salsberry
,
P. J.
(
2019
) ‘
Conditions of poverty, parent–child interactions, and toddlers’ early language skills in Low-income Families
’,
Maternal and Child Health Journal
23
:
971
8
. doi: .
Lareau
,
A.
(
2011
)
Unequal Childhoods: Class, Race, and Family Life
,
Cambridge
:
Univ of California Press
.
Layte
,
R.
(
2017
) ‘
Why do working-class kids do worse in school? An empirical test of two theories of educational disadvantage
’,
European Sociological Review
33
:
489
503
. doi: .
Layte
,
R.
(
2022
) ‘
Does family economic strain reduce child educational achievement? A longitudinal assessment using the great recession in Ireland
’,
European Sociological Review
38
:
408
24
. doi: .
Layte
,
R.
and
McCrory
,
C.
(
2018
) ‘
Fiscal crises and personal troubles: the great recession in Ireland and family processes
’,
Social Psychiatry and Psychiatric Epidemiology
53
:
987
1001
. doi: .
Lea
,
S. E.
,
Webley
,
P.
and
Walker
,
C. M.
(
1995
) ‘
Psychological factors in consumer debt: money management, economic socialization, and credit use
’,
Journal of Economic Psychology
16
:
681
701
.
Lister
,
R.
(
2004
) ‘A politics of recognition and respect: involving people with experience of poverty in decision-making that affects their lives’, in
The Politics of Inclusion and Empowerment
,
Springer
, pp.
116
38
.
Mack
,
J.
and
Lansley
,
S.
(
1985
) ‘
Poor Britain / Joanna Mack, and Stewart Lansley ; foreword by A.H. Halsey.
’ G. Allen & Unwin London.
Maître
,
B.
,
Privalko
,
I.
and
Watson
,
D.
(
2020
) ‘
Social transfers and deprivation in Ireland: A study of cash and non-cash payments tied to.
’ ESRI and the Department of Social Protection, Dublin.
Manly
,
T.
,
Robertson
,
I. H.
,
Anderson
,
V.
and
Nimmo-Smith
,
I.
(
1999
)
The Test of Everyday Attention (TEA-CH)
,
Bury St. Edmunds, England
:
Thames Valley Test Company
.
Mari
,
G.
and
Keizer
,
R.
(
2021
) ‘
Parental job loss and early child development in the great Recession
’,
Child Development
92
:
1698
716
. doi: .
Masarik
,
A. S.
and
Conger
,
R. D.
(
2017
) ‘
Stress and child development: a review of the family stress Model
’,
Current Opinion in Psychology
13
:
85
90
. doi: .
McGinnity
,
F.
,
McMullin
,
P.
,
Murray
,
A.
,
Russell
,
H.
and
Smyth
,
E.
(
2022
) ‘
Understanding differences in children's Reading ability by social origin and gender: The role of parental Reading and pre- and primary school exposure in Ireland
’,
Research in Social Stratification and Mobility
81
:
100729
. doi: .
McMullin
,
P.
,
McGinnity
,
F.
,
Murray
,
A.
and
Russell
,
H.
(
2020
) ‘
What you do versus who you are: home-learning activities, social origin and cognitive skills among young children in Ireland
’,
European Sociological Review
36
:
610
25
. doi: .
Mills
,
M. C.
and
Tropf
,
F. C.
(
2020
) ‘
Sociology, genetics, and the coming of Age of Sociogenomics
’,
Annual Review of Sociology
46
:
553
81
. doi: .
Nixon
,
E.
,
Swords
,
L.
and
Murray
,
A.
(
2013
) ‘
Parenting and infant development: infant cohort.
’ Stationery Office / Department of Children and Youth Affairs, Dublin.
Nolan
,
B.
and
Maître
,
B.
(
2017
) ‘
Children of the Celtic tiger during the Economic Crisis: Ireland. Children of austerity.
’ Impact of the great recession on child poverty in rich countries
146
169
.
Passaretta
,
G.
,
Skopek
,
J.
and
van Huizen
,
T.
(
2022
) ‘
Is social inequality in school-age achievement generated before or during schooling? A European perspective
’,
European Sociological Review
jcac005
), doi: .
Ponnet
,
K.
(
2014
) ‘
Financial stress, parent functioning and adolescent problem behavior: An actor–partner interdependence approach to family stress processes in Low-, middle-, and high-income Families
’,
Journal of Youth and Adolescence
43
:
1752
69
. doi: .
Quail
,
A.
,
O’Reilly
,
C.
,
Watson
,
D.
,
McNamara
,
E.
,
O’Mahony
,
D.
and
Murray
,
A.
(
2019
)
Growing Up in Ireland and a summary guide to cohort ‘08 at 9 years (wave 5 of the infant cohort at 9 year) 35.
Rees
,
G.
(
2019
) ‘
Poverty and children's wellbeing at 14 years old [WWW Document]
.’ Child Poverty Action Group. URL https://cpag.org.uk/welfare-rights/resources/article/poverty-and-children%E2%80%99s-wellbeing-14-years-old (accessed 5.11.21).
Reinhard
,
E.
,
Layte
,
R.
,
McCrory
,
C.
,
Panico
,
L.
and
Avendano
,
M.
(
2018
) ‘
The great recession and the health of young children: A fixed-effects analysis in Ireland
’,
American Journal of Epidemiology
187
:
1438
48
. doi: .
Ridge
,
T.
(
2002
)
Childhood Poverty and Social Exclusion: From a Child's Perspective, Studies in Poverty, Inequality, and Social Exclusion
,
Bristol
:
Policy Press
.
Rowlingson
,
K.
and
McKay
,
S.
(
2005
) ‘
Lone motherhood and socio-economic disadvantage: insights from quantitative and qualitative evidence
’,
The Sociological Review
53
:
30
49
.
Schenck-Fontaine
,
A.
and
Panico
,
L.
(
2019
) ‘
Many kinds of poverty: three dimensions of economic hardship, their combinations, and children's behavior problems
’,
Demography
56
:
2279
305
. doi: .
Schoon
,
I.
,
Cheng
,
H.
and
Jones
,
E.
(
2010a
) ‘
Resilience in children's development. children of the twenty-first century
’,
The First Five Years
2
:
235
48
.
Schoon
,
I.
,
Hope
,
S.
,
Ross
,
A.
and
Duckworth
,
K.
(
2010b
) ‘
Family hardship and children's development: the early years
’,
Longitudinal and Life Course Studies
1
:
209
22
. doi: .
Selig
,
J. P.
and
Little
,
T. D.
(
2012
) ‘Autoregressive and cross-lagged panel analysis for longitudinal data. In B. Laursen, T. D. Little, & N. A. Card (Eds.),
Handbook of Developmental Research Methods
(pp.
265
78
),
New York, NY, US
:
The Guilford Press
.
Skopek
,
J.
and
Passaretta
,
G.
(
2021
) ‘
Socioeconomic inequality in children's achievement from infancy to adolescence: The case of Germany
’,
Social Forces
100
:
86
112
. doi: .
Sprong
,
S.
,
Gibbons
,
R. A.
and
Chzhen
,
Y.
(
2022
) ‘
Divergent trajectories: three dimensions of child poverty during the great recession in Ireland
’,
Longitudinal and Life Course Studies
1
:
1
10
. doi: .
Squires
,
J.
,
Bricker
,
D. D.
and
Potter
,
L.
(
1999
)
The ASQ user's guide / by Jane Squires, LaWanda Potter, and Diane Bricker., 2nd ed. ed.
Paul HBrookes, Baltimore, MD.
Thornton
,
M.
,
Williams
,
J.
,
McCrory
,
C.
,
Murray
,
A.
and
Quail
,
A.
(
2013
) ‘
Design, instrumentation and procedures for the infant cohort at wave one (9 months)
’,
Department of Children and Youth Affairs
107
.
Todd
,
P. E.
and
Wolpin
,
K. I.
(
2007
) ‘
The production of cognitive achievement in children: home, school, and racial test score gaps
’,
Journal of Human Capital
,
91
136
.
Townsend
,
P.
(
1979
)
Poverty in the United Kingdom: A Survey of Household Resources and Standards of Living
,
Oakland
:
Univ of California Press
.
Treanor
,
M. C.
(
2020
)
Child Poverty: Aspiring to Survive
,
Bristol
:
Policy Press
.
VanderWeele
,
T.
(
2015
)
Explanation in Causal Inference: Methods for Mediation and Interaction
,
New York
:
Oxford University Press
.
VanderWeele
,
T.
and
Vansteelandt
,
S.
(
2014
) ‘
Mediation analysis with multiple Mediators
’,
Epidemiologic Methods
2
:
95
115
. doi: .
Van Lancker
,
W.
and
Vinck
,
J.
(
2019
)
The Consequences of Growing up Poor
,
Routledge
:
Routledge International Handbook of Poverty
.
Washbrook
,
E.
,
Gregg
,
P.
and
Propper
,
C.
(
2014
) ‘
A decomposition analysis of the relationship between parental income and multiple child outcomes
’,
Journal of the Royal Statistical Society. Series A (Statistics in Society)
177
:
757
82
.
Watson
,
D.
,
Whelan
,
C. T.
,
Maitre
,
B.
and
Williams
,
J.
(
2016
) ‘
Socio-economic variation in the impact of the Irish recession on the experience of economic stress among Families
’,
The Economic and Social Review
47
:
477
98
.
Wodtke
,
G. T.
and
Zhou
,
X.
(
2020
) ‘
Effect decomposition in the presence of treatment-induced confounding: A regression-with-residuals Approach
’,
Epidemiology
31
:
369
. doi: .
Yeung
,
W. J.
,
Linver
,
M. R.
and
Brooks-Gunn
,
J.
(
2008
) ‘
How money matters for young children's development: parental investment and family Processes
’,
Child Development
73
:
1861
79
. doi: .

Author notes

Edited by Katya Ivanova

Supplemental data for this article can be accessed online at https://doi.org/10.1080/14616696.2023.2275592.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the use is non-commercial and the original work is properly cited. For a full description of the license, please visit https://creativecommons.org/licenses/by-nc/4.0/legalcode.

Supplementary data