Essays on Early Childhood Development, Essays (university) of Childhood Development

Download Essays on Early Childhood Development and more Essays (university) Childhood Development in PDF only on Docsity! Essays on Early Childhood Development Pamela Jervis Ortiz A thesis submitted for the Degree of Doctor of Philosophy (P.hD) to the University College London September 2016 Declaration I, Pamela Jervis Ortiz confirm that the work presented in this thesis is my own. Where information has been derived from other sources, I confirm that this has been indicated in the thesis. Funding acknowledgments This thesis has been made possible through the scholarship “Programa de Formación de Capital Humano Avanzado Becas Chile” from The National Commission for Science and Technology CONICYT, Government of Chile, the Institute Fiscal Studies Scholarship from ESRC, the WM Gorman Schol- arship from the Department of Economics at UCL and the “Early Child Development Programs: E↵ective Interventions for Human Development” from the Eunice Kennedy Shriver National Institute of Child Health and Human Development. Contents 1 Introduction 10 2 Disentangling the Determinants of Early Childhood Cognitive Development in Chile 18 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.2 A Literature Review . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3 A Simple Theoretical Framework of Childhood Development . . 29 2.4 Data and Measures . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.4.1 Survey Description . . . . . . . . . . . . . . . . . . . . . 33 2.4.2 Factor Analysis . . . . . . . . . . . . . . . . . . . . . . . 41 2.4.2.1 Wealth score . . . . . . . . . . . . . . . . . . . 44 2.4.2.2 Socioeconomic Status (SES) score . . . . . . . . 46 2.4.2.3 Home Assessment scores . . . . . . . . . . . . . 46 2.5 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.5.1 Socioeconomic Gradients and Childhood Development . 53 2.5.2 Estimation strategy . . . . . . . . . . . . . . . . . . . . . 55 2.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.6 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . 61 3 The myths behind the Technology of Human Capital Forma- tion in Childhood 63 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.2 Estimation of the Technology of Human Capital Formation . . 67 3.2.1 Measurement error . . . . . . . . . . . . . . . . . . . . . 70 3.2.2 Endogeneity of Inputs . . . . . . . . . . . . . . . . . . . 73 3.2.3 Unscented Kalman Filter . . . . . . . . . . . . . . . . . 74 3 3.3 Data and Empirical Specifications . . . . . . . . . . . . . . . . 76 3.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.3.2 Empirical Specifications . . . . . . . . . . . . . . . . . . 81 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 3.4.1 Measurement error: Noise and Signal . . . . . . . . . . . 84 3.4.2 Parental Investment Equations and Identification . . . . 85 3.4.3 Estimation of the Technology of Human Capital Forma- tion: Specifications . . . . . . . . . . . . . . . . . . . . . 87 3.5 Conclusions and further work . . . . . . . . . . . . . . . . . . . 92 4 Parental Beliefs and Investments in Human Capital 94 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 4.2 Parental beliefs about child development . . . . . . . . . . . . . 98 4.2.1 Latent variables and measurement . . . . . . . . . . . . . 100 4.2.2 Construction of Scenarios . . . . . . . . . . . . . . . . . 100 4.2.3 Survey Questions . . . . . . . . . . . . . . . . . . . . . . 102 4.3 Using the beliefs questions . . . . . . . . . . . . . . . . . . . . . 103 4.3.1 Scaling: The Relationship Between Hi,0, Hi,1 and Age . . 104 4.3.2 Using information from a latent factor model . . . . . . 105 4.3.3 Using beliefs data: developmental delays and rates of return to investment . . . . . . . . . . . . . . . . . . . . 106 4.3.4 From answers to beliefs to beliefs about a production function . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 4.4 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.5 Identification of the Model . . . . . . . . . . . . . . . . . . . . 111 4.5.1 Measurement of Child Development . . . . . . . . . . . . 111 4.5.1.1 Scales of Child Development . . . . . . . . . . . 111 4.5.1.2 Item Response Theory . . . . . . . . . . . . . . 112 4.5.2 Estimation of the Technology of Skill Formation and Utility Function . . . . . . . . . . . . . . . . . . . . . . . 115 4.6 Data and Empirical Results . . . . . . . . . . . . . . . . . . . . 116 4.6.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.6.2 Empirical Results . . . . . . . . . . . . . . . . . . . . . . 118 4.6.2.1 Beliefs about the Returns to Investments . . . . 118 4 A.0.1Home Assessments: Proportion answering yes . . . . . . . . . . 175 A.0.2Kernel densities of latent traits: One investment input, 7-23mths, Specification b . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 A.0.3Kernel densities of latent traits: One investment input, 48- 58mths, Specification b . . . . . . . . . . . . . . . . . . . . . . . 203 A.0.4Maternal Investment: time and didactic materials . . . . . . . . 204 A.0.5Expected Value Functions for periods T to t = 1 . . . . . . . . . 205 A.0.6Solution for the optimal family decisions, x axis is the child’s cognitive skills . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 7 List of Tables 2.1 IRT analysis Act 2012 . . . . . . . . . . . . . . . . . . . . . . . 49 3.1 Parental Investment Equations, Age 24-47 mths . . . . . . . . . 86 3.2 Production Function of Cognitive and Non-cognitive skills: One investment input, Age 24-47 mths, Specification 1 . . . . . . . . 87 3.3 Production Function of Cognitive and Non-cognitive skills: One investment input, di↵erent age stages, Specification 2 . . . . . . 89 3.4 Production Function of Cognitive and Non-cognitive skills: Mul- tiple investments, Age 24-47 mths, Specification 2, 3 and 4 . . . 91 4.1 Answers about Maternal Investment . . . . . . . . . . . . . . . 103 4.2 Returns on Investment and SE characteristics . . . . . . . . . . 120 4.3 Investment and Returns on Investment . . . . . . . . . . . . . . 122 4.4 Maternal Investment . . . . . . . . . . . . . . . . . . . . . . . . 122 4.5 Production Function Estimates: Perceived Median and ”True” . 123 4.6 Production Function Estimates and SE characteristics . . . . . . 124 5.1 Child Outcomes and Parental Investment . . . . . . . . . . . . . 141 5.2 Calibration Values . . . . . . . . . . . . . . . . . . . . . . . . . 152 5.3 Structural Parameters . . . . . . . . . . . . . . . . . . . . . . . 154 A.1 Kaiser-Meyer-Olkin measure . . . . . . . . . . . . . . . . . . . . 168 A.2 IRT analysis Lear Mat 2010 . . . . . . . . . . . . . . . . . . . . 176 A.3 IRT analysis Act 2010 . . . . . . . . . . . . . . . . . . . . . . . 177 A.4 IRT analysis Resp 2010 . . . . . . . . . . . . . . . . . . . . . . . 178 A.5 IRT analysis Inv 2010 . . . . . . . . . . . . . . . . . . . . . . . 179 A.6 IRT analysis Lear Mat 2012 . . . . . . . . . . . . . . . . . . . . 180 A.7 IRT analysis Resp 2012 . . . . . . . . . . . . . . . . . . . . . . . 181 8 A.8 Children’s cognitive eedp test z-score: Linear regression model . 185 A.9 Children’s cognitive bdi test z-score: Linear regression model . . 186 A.10 Children’s cognitive tepsi test z-score: Linear regression model . 187 A.11 Children’s cognitive tvip test z-score: Linear regression model . 188 A.12 Children’s cognitive cbcl1 test z-score: Linear regression model . 189 A.13 Children’s cognitive tadi test z-score: Linear regression model . 190 A.14 Children’s cognitive bdi test z-score: Linear regression model . . 191 A.15 Children’s cognitive tvip test z-score: Linear regression model . 192 A.16 Children’s cognitive cbcl1 test z-score: Linear regression model . 193 A.17 Children’s cognitive tvip+2012 test z-score: Value added esti- mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 A.18 Children’s cognitive cbcl1+2012 test z-score: Value added esti- mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 A.19 Descriptive statistics age 7-23 mths . . . . . . . . . . . . . . . . 196 A.20 Descriptive statistics age 24-47 mths . . . . . . . . . . . . . . . 197 A.21 Descriptive statistics Multiple Investments age 24-47 mths . . . 198 A.22 Descriptive statistics age 48-58 mths . . . . . . . . . . . . . . . 199 A.23 Total Variance in Measures: Signal and Noise . . . . . . . . . . 200 9 Recently, the most advanced research in the field has turned to the credible estimation of structural models that can inform social policies in more compre- hensive ways about relevance, heterogeneous impacts, and the optimal timing of programs. Human capital investments are a↵ected by both service provision (e.g., health and educational institutions) and by individual and familial decisions, with interactions between them that may mean that the most e↵ective policies may involve improving both service provision and familial inputs. A life-cycle perspective is important because there are likely to be important complemen- tarities over the life cycle, with investments earlier in the life-cycle increasing returns to investments later in the life cycle. International comparisons are informative about how successfully Chile is approaching global frontiers in human capital related to health and skills. Such comparisons suggest, for ex- ample, that in many respects such as PISA and PPVT (receptive vocabulary) tests Chile on average performs well in comparison with other LAC countries, but continues to lag considerably behind the European and East Asian coun- tries on the global frontier. Within Chile a high degree of inequality persists in most dimensions of human resources, as well as in income and wealth, with considerable human resource gaps across income groups that start very early in the life-cycle (e.g., 0.8 SD in PPVT scores for 3-5 year olds between top and bottom quartiles) and tend to persist or even increase over the life cycle. This thesis dissertation aims to provide enough evidence to increase the en- tire distribution of Chilean human capital towards the global frontier, but at the same time to focus mainly on improving the human capital of the disadvan- tages Chileans to reduce social exclusion, inequality and poverty. International experiences with various programs over the life cycle are informative about what appears to be “best practice” and therefore merits consideration for the Chile. But best practices from other contexts cannot just be eciently trans- ferred without adjustments and modifications to Chile because the Chilean context di↵ers from other contexts. Using models as the ones that I develop in this thesis dissertation it is pos- sible to understand the mechanisms behind decision-making and to simulate policies ex-ante that are crucial to address a set of research questions as: What are the processes (biological, neurological, psychological) that govern the com- 12 ponents of human capital? How do acquired skills generate new skills and how does this process di↵er by the developmental stage of the child? What are the determinants of parental investments in children and what are the constraints they face? Is the lack of knowledge (or awareness) of the potencial returns, the lack for time or monetary resources, or their beliefs the reason why par- ents behave in one or another way? And what is the relative importance of these constraints? Do public investments in human capital (for example, pre- schooling or child care services) compensate for or replace parental investments (money and time)? To address all these questions, it is indispensable to develop more com- plex economic analysis to simulate the e↵ects of alternative policies and to understand the mechanisms behind decision-making. To do so in the Chilean context, this thesis dissertation provides a better insight of the state-of-the- art of numerical methods and computer technology to develop new techniques posed by modern economic models. Some of these ex-ante policies to be simulate will be based on current social policies in Chile and associated with early childhood development as extensions of maternity leave, extending parental leave to fathers, parenting programs of- fered by Chile Grows with You as well as a cash transfer (Asignación Familiar) to the poorest families. I divide this thesis dissertation into five chapters, each related to my re- search questions. In chapter 2, I present a theoretical framework and empirical analysis to contribute to the debate about the determinants of early childhood development in a developing country from Latin America: Chile. The Eco- logical Environment theoretical model for childhood was proposed to define the determinants of early childhood. This chapter aims to disentangle the de- terminants behind early childhood development based on multiple empirical strategies through the use of the first and second wave of a recent longitu- dinal survey, which was designed to characterise the child development. The data contains information about demographics, family’s background, cogni- tive, socioemotional and physical measures for mothers and children under five years old and home assessment environment. The determinants of early childhood development, particularly, cognitive and non-cognitive skills, are studied through the estimation of contemporaneous and value-added cognitive 13 and non-cognitive production functions, as well as the use of factor analysis such as item response theory for reducing the number of inputs. Three main results arise: (1) there are significant socioeconomic gradients in all cognitive tests between poorest and richest quintiles, which lead to a liability among disadvantaged children. Once controlling by observables, the gradient starts to decrease and in some cases to lose significance; (2) there is a significant e↵ect of mother’s characteristics and family background at later stage devel- opment (above 24/30 months old) measured principally by mother’s education, age and cognitive skills, if the family is a two-parent family, the presence of younger/older children as well as home environment measures by parent-child activities, learning materials, parental involvement and verbal and emotional responsibility scores. The later stage development also adds a significant ef- fect on attending a preschool. The previous determinants drive the fall in the socioeconomic gradient in both stages; and (3) regarding the non-cognitive skills, for both waves, the results are similar, there are socioeconomic gradi- ents that are still significant after controlling for all the variables. If the child is male, have a negative and significant e↵ect as well if they attend to preschool. Mother’s education and age have positive and significant impact meanwhile having younger children in the household have an adverse and significant ef- fect. Having both parents have a positive impact as well as child’s weight at birth and the mother’s cognition level. For the first time, all the sub scales of the mother’s socioemotional test are (positively) correlated with the child’s so- cioemotional skills. The home environment continues presenting positive and significant e↵ect on child’s development. Chapter 3 is based in a co-authored paper with Italo Lopéz (RAND). We characterise the process of human capital accumulation in early years. Genet- ics, environment and parental investment at di↵erent stages of early years of childhood a↵ect the formation of human capital skills. Only when these chan- nels are adequately incorporated in the study of the human capital formation will be possible to tackle early gaps in childhood and formulating ecient public policies. Despite these recent advances, there is still very little known about the return to cognitive and non-cognitive skills in developing countries. Recent studies have demonstrated how multiple factors relate in a complex way (Cuhna et. al. (2007, 2010)) through the use of technologies of skill 14 Chapter 4 is based on a co-authored paper with Orazio Attanasio (UCL and IFS) and Flávio Cunha (Rice University). We shed light on the importance of maternal subjective beliefs in explaining the heterogeneity in maternal choices of investments in the development of their children. Subjective beliefs about the production function of skills in early childhood development is crucial since parents may have biased beliefs about the returns to investments, which is cru- cial to pin down in designing policies aimed at remediating poor investments. To determinate the importance of maternal subjective beliefs, we first show how to convert the answers to a specific set of questions into estimates of ex- pected rates of returns on specific investment and then relate these estimates to actual maternal behaviour, then we formulate and estimate a model in which mothers have subjective beliefs about the technology governing the formula- tion of skills in early childhood development, drawing on detailed and unique data for the identification of the model from an early childhood intervention ran in Colombia, in which, home visitors paid weekly visits to randomly chosen households with the aim of promoting child cognitive and non-cognitive devel- opment and improving mother-child interactions. The intervention targeted poor households with children aged 12 to 24 months at baseline and lasted 18 months. We find that parents think that the productivity of investment is much higher for low initial conditions than higher initial conditions. Some findings are worth being discussed. We have elicited maternal beliefs about the production function. We have shown how to relate answers about devel- opmental age under di↵erent scenarios to beliefs about returns to investment and parameters of the production function. We find that parents think that the productivity of investment is much higher for low initial conditions than higher initial conditions. We want to extend this approach and estimate si- multaneously the production function, the perceived production function and the investment strategy. In the last chapter, I develop a dynamic structural model estimated with rich longitudinal data from Chile, in which I integrate a children’s human capital model with multiple stages of childhood into a dynamic framework to explain parental investment decisions, modeling quality parental investment time and children’s technology skill formation accounting for unobserved het- erogeneity (income shocks). Parents maximise a constrained model, choosing 16 consumption and quality time with their child and monetary investments in a sequential decision problem using a unitary model.This way, I explore potential mechanisms: First, the e↵ect of parental preferences when they make decisions in each period of a child’s life in terms of his/her developmental outcome mea- sure as cognitive and non-cognitive skills; Second, I analyse the constraints parents face when they are taking their decisions in terms of monetary and quality time with their child; and third, the importance of addressing expecta- tions driving investment choices. An important contribution to the literature of child development is a two-step procedure used to eliminate the presence of measurement error in the data for the inputs in the production function as well as integrating a life cycle model into the analysis and hence accounting for the endogeneity (correlation with the unobserved shocks) of investments. In the first stage, I estimate a measurement model based on a linear dynamic factor model and exploit cross-equation restrictions (covariance restrictions) proving that I can identify all of them. In the second step, I estimate together with the dynamic and stochastic structural model that incorporate parental choices based on the overall description of the mechanisms through which parental in- vestment is modified and a↵ects the human capital formation of their children, adding restrictions that involve weaker assumption than those derived from the literature, as well as allowing for simulations of the most e↵ective targeting policies for Early Child Development compensating the most disadvantaged children. 17 Chapter 2 Disentangling the Determinants of Early Childhood Cognitive Development in Chile 2.1 Introduction Early childhood development is essential for improving the wellbeing and wel- fare of society as a whole. An increasing body of studies in neuroscience, psychology and economics, shows that the first three years of life are critical for the future development of children. In particular, Heckman et al. (2006) find that cognitive and non-cognitive abilities are essential to the future per- formance on a series of social and labour market outcomes: enrolment rates, wages, work experience, crime rates, early pregnancies, drug use in the labour market, among others. Hence, early stimulation of these abilities plays a cru- cial role in the child’s future and the social development of future generations. Heckman and Masterov (2007) show in a cost-benefit analysis that a right in- vestment in early life has a higher economic return than the same investment for adults. This e↵ect is even stronger for most disadvantaged children. Disentangling the determinants of early childhood development is crucial for understanding how early investment improve abilities in early childhood. Recent studies have shown that focusing on socioeconomic gradient only is far 18 which makes early stimulation particularly beneficial. At this stage, environ- mental stimulation in children results in the generation of new neural connec- tions that alter brain organisation. Consequently, non-appropriate or lack of stimulation at all in early childhood, not only prevents the growth of neural connections but also makes their number diminish progressively. I revise this evidence in what follows. One of the most cited theoretical references in cognitive development be- longs to the psychologist and philosopher Jean Piaget. His Theory of Cognitive Development exposed in the 30’s postulates that children go through specific stages as their relationship between intellect and ability to see matures (Chap- man, 1988). He mentions four stages: Sensorimotor, Preoperational, Concrete Operations, and Formal Operations. The first stage occurs between birth and two years of age, when children begin to understand the information they re- ceive through their senses and also to develop their ability to interact with the world.1 The second stage occurs between two and seven years old. Chil- dren learn how to interact with their environment in a more complex manner through the use of words and images.2 Even so, Piaget’s approach is a good tool to characterise the cognitive development at childhood, a new generation of neuroscientist argue that this theory neglected brain functions and that it vastly underestimated infant cognitive development (Catherwood, 1999 and Mark Johnson, 1999). Nowadays, developmental psychologist e↵orts have been aided by develop- mental neuroscientists whose conclusions about brain growth complement the findings of behavioural scientists and reflect the importance of the brain in the cognitive development in children. Thompson (2001), states that by the sixth prenatal month, nearly all of the billions of neurones that populate the mature brain have been created, with new neurones generated at an average rate of more than 250,000 per minute. These neurones produce far more synapses with other neurones forming great potential for the developing brain and finished with stimulating experiences activating specific neural synapses, and there- fore dropping those that are not enabled progressively over time. Through 1Stage marked by Object Permanency, defined as the ability to understand that these objects do in fact continue to exist. 2Stage marked by Egocentrism and Conservation, defined as the ability to understand that quantity does not change if the shape changes. 21 this principle of use it or lose it the architecture of the developing brain be- comes adapted to environmental stimulation. A similar principle is used to explain the early development of memory ability. Nelson (1995), Diamond et al. (1994) and Dawson (1994) document the growth of early categorisation and thinking skills and early emotional development. It is also possible to establish that early years of children are critical because the brain has its most signifi- cant level of brain plasticity, which makes the nervous system to show greater resilience and organic and functional reorganisation. This allows children to adapt to the environment and to generate new neural connections by altering brain organisation through influence received from environmental stimulation (OECD, 2007). Van der Gaag (2005) also finds that newborns have significantly more neu- rones than a three-year-old child doubling the number they will have as adults. Also, he claims that the brain consists of neural pathways keen to develop par- ticular skills, which if not properly stimulated, do not reach their full potential and are lost. Moreover, a considerable number of recent studies in neuro- science find that the interactive influences of genes and experience shape the architecture of the developing brain, and the active ingredient is the serve and return nature of children’s engagement in relationships with their parents. This process directly impacts the infant’s future productivity, and hence, the social development of future generations (The National Scientific Council on the Developing Child, 2007 and Centre on the Developing Child, 2010). This evidence is consistent with the existence of critical periods in which the brain is particularly e↵ective against certain types of learning. In the case of language, this time ranges from birth to 3 years old. In the case of logical-mathematical, sensitive periods go from 1 to 4 year old (UNICEF, 2004). Not only neuroscience and psychology studies emphasise the importance of providing appropriate stimulation during early childhood, but also economic research in recent years has studied this phenomenon as a public policy tool that helps improve a number of social welfare indicators. Heckman et al. (2006) analysed in the United States how the development of cognitive and non-cognitive abilities are critical to the future performance of a series of social and labour market themselves: enrolment rates, wages, work experience, crime rates, early pregnancies, drug use in the labour market, among others. 22 Therefore, it is crucial to analyse two main points in the economic liter- ature. First, if cognitive and non-cognitive abilities genuinely benefit from interventions during early childhood; and second, if this is the case, what are the relevant determinants of early childhood development. In the first point, the economic literature has focused to show through a cost-benefit analysis, that a right investment in early life has a higher eco- nomic return than the same investment for adults (Carneiro and Heckman, 2003). This e↵ect is even stronger for the most disadvantaged children that live in a poor environment (Heckman and Masterov, 2007). In this sense, it is essential to focus on the benefits of short and medium-term interventions generated during early childhood. Currie (2001) states that the benefits can easily compensate between 40% and 60% of the costs of programs implemented on a large scale so that even low long-term benefits are enough to pay the in- vestment. Furthermore, if we do not intervene early on the most vulnerable children, the cost of investing in them when they are adults is so high that it becomes prohibitive (Heckman, 2006; Cunha and Heckman, 2007 and Behrman et al., 2006). There are also studies from organisations that found that when two children born in di↵erent families and socioeconomic environments, like those children living in families in the first and fifth quintile, necessarily face di↵erent opportunities that lead to di↵erent educational and socioeconomic outcomes (OECD, 2009). Once we establish that cognitive and non-cognitive development begins early in childhood and that the most a↵ected, without early interventions, are the most vulnerable children, the next step is to focus on the e↵ect of socioeconomic status, parental-child interactions patterns, child-care centres, and home-learning environments, among others, as the main sources that af- fect children’s cognitive and non-cognitive development. The economics lit- erature has focused on understanding these variables principally using Re- gression Analysis. Now, I review the literature that applies this technique to the study the determinants of child’s development. The majority esti- mate the e↵ects of these variables by linear regressions, without taking into account the potential endogeneity issues that arise depending on the vari- ables used. Few studies have tried to deal with this issue using variables to proxy for unmeasured endowments, sibling di↵erences, fixed e↵ects models, 23 nutrition chronic reached 2.9%, with a higher percentage in the low category of socioeconomic status. The environmental stimulation variable proved to be the most influential in cognitive performance with a di↵erence of 48 points in the scales of the HOME between the low and high socioeconomic status (CLACYD, 2000 and CLACYD, 2002). In another study in the same country, Piacente et al. (1990) use a survey of 1,521 children less than 5 years old and 920 mothers with characteristics of families in marginal areas to determine the psychological development and nutritional status of children. The results showed, when they compare the performance using a control sample consist- ing of non-poor children, that they had higher scores in the di↵erent areas of psychological development, particularly in language. In Paraguay, Peairson et al. (2008) study the e↵ect of the Pastoral del Nio program, which served a non-random sample of children aged 24 months old or younger from poor rural Paraguayan areas. They show that program children scored significantly higher than non-program children on the mental development index portion of the Bailey Scales of Infant Development II test at 0-4 months old and also at 20-24 months old. The most important variables that explain this result include health, nutrition, and education variables. In Brazil, Halpern et al. (1996) use a sample of 20% (1,400 children) of all children born in 1993 in hospitals from the Rio Grande do Sul, Brazil, these children were followed through home visits during the first year having nutritional status and Denver II Test. A 34% of the children assessed at 12 months failed this test, where failure was associated with birth weight and family income variables that were strongly related to the potential risk of developmental delays at the age of 12 months old. As reviewed in this section, the major problem in achieving an excellent cognitive and non-cognitive development is the existence of socioeconomic gra- dients that a↵ect all the others variables that determine the socioeconomic status. They included parental-child interactions patterns, physical status, child-care centres, and home-learning environment among others. Today, Chile is one of the major industrialised countries in Latin America. In fact, in 2007, it achieved the status as the region’s richest country regarding GDP per capita. It also has low unemployment rates, a stable monetary policy, political transparency and stability. However, according to the OECD (2011) 26 Chile is the country with greater inequalities among its population regarding income relative to other OECD countries.3 In this context, public policies aimed at early childhood are intended to tackle the problem of inequality. Since 2006, the country experienced two policies that are designed to improve protection of children in Chile that is essential to the future of vulnerable children. The international community, through UNESCO, highlights this fact in its “Education for All Global Moni- toring Report” published in 2010 where it is claimed that Chile has begun to implement a strategy for the development of children focused on health and education. Its purpose is to provide early childhood care and education to all children under five years, focusing primarily on those belonging to the two poorest quintiles. The first element is the increase in the supply of childcare centres nation- wide, while the second is the implementation of Chile Grows with You program (Vega, 2011) that provides free access to childcare centres and parenting advice for children among the poorest 60% and between3-months-old to 4-years-old. A study by the United Nations Development Program (UNDP) examines the implementation of the program and its e↵ect on the female labour market. It can be concluded that the e↵ect of increased access to child-care centres on female labour supply is essential and results in considerable improvements in household income and levels of poverty (UNDP, 2008). Another line of work that looks at improving the knowledge of childhood development is to increase the availability of high quality and detailed data on children. A partnership formed in 2007 between the Junta Nacional de Jardines Infantiles (JUNJI), an organisation of the Ministry of Education and the Centro de Estudios de Desarrollo y Estimulación Psicosocial (CEDEP), set out to implement a longitudinal study to evaluate the e↵ects of participation in the child care program on child’s development outcomes as measured by the Spanish version of the Battelle Developmental Inventory. Characteristics of a child’s family and private child-care establishments were gathered at the beginning of the study. Another study from JUNJI, UNICEF and UNESCO 3The OECD noted that the Gini coecient that measures inequality in Chile of 0.50 when the average organisation is 0.31. Also, 18.9% of Chileans are poor; a number surpassed only by Mexico (third) and Israel, far from the 10% across the OECD. 27 is the Encuesta Nacional Primera Infancia survey which goal was to collect information about early childhood (from birth to 5 years 11 months old) for de- scribing the development of 6,500 children to respond relevant public policies. The most important source of data on children is the Encuesta Longitudinal de Primera Infancia (Early Childhood Longitudinal Survey (ELPI)) survey, which is a longitudinal study and allows researchers to assess the impact of early childhood policies and provide valuable information for the evaluation and design of social policies in this field. It is based on a representative sam- ple of 15,000 children under 5 years old and their families for the first wave, during the second wave, the sample is 18,000 as included a refreshing sample from 0 to 3 years old. This survey provides demographic information, mea- sures of cognitive and non-cognitive abilities for children and their mothers, anthropometrics and the quality of care for children provided at home among others. Noboa and Urzúa (2010) use quasi-experimental methodologies and took information from the Longitudinal Survey conducted by JUNJI in 2007. They determined through a series of indexes of cognitive and non-cognitive abilities, the average e↵ect that has on children who attend to public child-care centres. They use a random sample of 41 public child-care centres, and the treatment group was randomly selected from children. The final sample consisted of 331 children between 5 to 14 months old, attending a public child-care centre in April 2007 and the control group was randomly selected from healthy children who did not attend any child-care centre but attended nearby health clinic finished with a sample of 151 children. The purpose was to match socioeco- nomic conditions across groups (roughly the same age). They found that while attending the initial e↵ect was negative, over time the e↵ect on all areas of development measures was positive. Another study with considerable less quality in data is one written by Riquelme del Solar (2003). The author created a test on basic abilities for calculus introduction (TIC) for children between 5 to 6 years old. The study obtained that 32 items according to five cognitive abilities constitute this in- strument proposed by J. Piaget. The author uses a sample of 60 children attending public and private preschool in equal proportion. Even with the ob- served data, the study was able to display immediately that a child belonging 28 circuitry to the capacity of children’s empathy is a↵ected by the environment and the accumulated experiences, that start early in the prenatal period and that extend through the years of early childhood (Nelson, 1995; Diamond et al., 1994 and Centre on the Developing Child, 2010 among others). The second system is called Mesosystems, and it includes the relationships between two or more settings in which the developing person participates. It contains the interrelations of two or more environments where the developing person actively participates. Therefore, it is a system of microsystems. These systems form when the person enters into a new environment. For childhood development, this system could be called Family System. Parents and other regular caregivers are essential assets of the influence of the environment dur- ing early childhood. Children grow and thrive in the context of relationships that provide love, security, responsive interaction, and stimulation for explo- ration. Without these relationships, the development is disrupted, and the consequences can be severe and permanent. Hence the early development of children depends on the welfare of their parents. With regard to the influence of parenting, it is essential to consider that several legacies are given to children and daily routines for such ordinary activities as sleeping, eating, playing with the values that prevail in their conceptions of discipline, role gender, religious or spiritual values among others, and the contexts that frame the cognitive, linguistic, and socioemotional and therefore influence the acquisition of specific abilities or behaviours (Shonko↵ and Phillips, 2000). The third system is called Exosystems and refers to one or more environ- ments in which the person in development is not included directly, but to those environments in which events that a↵ect the person. For childhood de- velopment, this system could be called Environment/Neighbourhood System. It refers to the physical environment, culture, values, social capital, social net- works, and geography, among others that are involved in child development. They include health system, child-care centres, preschool and primary estab- lishment, basic services, the workplace of the parents, the circle of friends of parents, among others. The last system is called Macrosystem and refers to cultural and ideo- logical frameworks that a↵ect or transversely a↵ect the other three systems. It gives certain uniformity in form and content and also some di↵erence from 31 Figure 2.1: Ecological Environment for Childhood Development other environments influenced by other dissimilar cultural or ideological frame- works. As Bronfenbrenner notice, in society, the structure and substance of the three other systems tend to be similar, but among di↵erent social groups, the constituent systems may have notable di↵erences. Therefore, analysing and comparing the child, family and environment/neighbourhood systems that characterise di↵erent social classes, ethnic and religious groups, is possible to systematically describe and distinguish the ecological properties of these social contexts. In the case of Childhood Development, this system could be called Society System. Latin America and particularly Chile still have two signifi- cant issues that still generate di↵erences among children: income inequality and sex discrimination. Relative to the first, Chile is the country with higher inequalities among its population regarding income with respect other coun- tries OECD members. In the second issue, Chile has cultural or ideological frameworks that have e↵ect in the child on the gender archetype that is devel- oped through the biological roles and identity, cognitive and social influence by sex stereotypes. In this context, it is possible to provide tools that detect cases of violence and promote social activities for children under conflict reso- lution based on dialogue without discrimination. UNESCO (2010) finds that all this gender-based violence is the result of what children experience daily in their homes. They notice that women, mothers and girls do more tasks inside and outside the home than men and children. This is the foundation of the problems detected by scholars. Figure 2.1 shows the four systems that are the basis for cognitive abili- 32 ties, language acquisition and empathy with other humans. In other words, they are the pillars of childhood development. Once the child, familiar, envi- ronment/neighbourhood and society systems are understood, it is possible to disentangle the determinants of early childhood development and therefore an ecient and correct intervention. 2.4 Data and Measures One of the major limitations for disentangling the determinants of childhood development in Latin America and particularly in Chile is the lack of high quality and detailed data on children under 5 years old. In this section, I present new data that allows researchers to assess the impact of early childhood policies and to provide valuable information for the evaluation and design of social policies in this field since this data contains variables determining children’s cognitive and non-cognitive abilities. Throughout this section, I develop an empirical analysis that allows me to generate relevant measures of early childhood using Principal Components Analysis, Factor Analysis and Item Response Theory. 2.4.1 Survey Description The Early Childhood Longitudinal Survey is a longitudinal survey that collects information about children and their families in the first few years of the child’s life. The first wave was conducted in 2010 and the second one in 2012. It is a representative survey of children from urban and rural areas who were born on January 1, 2006, and August 31, 2009, in the first wave, and that is a representative sample. The sample includes di↵erent cohorts of children, distinguished by year of birth. The second wave follows the first wave but also add a refreshing sample of around 3,000 children born on September 1, 2009, and December 31, 20115. The sample design corresponds to a stratified sample, where strata are constructed by clustering of districts that had similar socioeconomic status. 5The sampling frame for the first wave corresponds to 1.297.822 birth records from March 1, 2004, and August 31, 2009, however, the final sampling frame used information from January 1, 2006, and August 31, 2009 (total 44 months) due to the test’s timing. 33 mental domains: personal-social, adaptive, motor, communication, and cog- nition. These five domains are further divided into twenty-two separate sub- domains. The personal-social domain is composed of: adult interaction, ex- pressions/feelings/a↵ect, self-concept, peer interaction, coping, and social role. The adaptive domain includes attention, eating, dressing, personal responsi- bilities, and toileting. The motor domain is composed of muscle control, body coordination, locomotion, fine muscle, and perceptual motor. The communi- cation domain includes: receptive and expressive. Finally, the cognitive do- main is composed of: perceptual discrimination, memory, reasoning/academic skills, and conceptual development. The test results can be analysed using z and T scores and the ratios of standard deviations as the basis for conclu- sions concerning the strengths and weaknesses of development (De la Cruz and González, 1998). Was applied only in the first wave. Battelle Developmental Inventory, Screening Test, 2nd ed. - (BDI 2012) Psychomotor skills (for children 6-84 months old). The BDI 2012 includes the same areas as the BDI 2010. Has 96 items (two per each age level) extracted from the full version of the BDI and it is a screening test that evaluates the child development from 0-8 years old. The objective is to assess the fundamental skills development in five areas (personal and social, adaptive, motor, communication and cognitive). Was applied only in the second wave. Test de Vocabulario en Imágenes Peabody, Hispanic America adap- tation (TVIP) Language skills (for children 30-84 months), is the Span- ish version of the Peabody Picture Vocabulary Test (PPVT). This test mea- sures the child’s comprehension and understanding of vocabulary using relating words to an illustration. The scale should be viewed primarily as an achieve- ment test since it shows the extent of Spanish vocabulary acquisition of the subject. Also, it may be considered to be a screening test of scholastic aptitude (verbal ability or verbal intelligence), or as one element in a comprehensive test battery of cognitive processes when Spanish is the language of the home and community into which the subject was born, has grown up, and resides; and when Spanish is, and has been, the primary language of instructions at school (Dunn et al., 1986). Was applied only in both waves. 36 Test de Desarrollo Psicomotor (TEPSI) Psychomotor skills (for chil- dren 24-60 months old), not in English, could be translated as Psychomotor Development Test. TEPSI is a screening test, i.e., an assessment list showing the level of performance in the psychomotor development of children between two and five years old, in terms of a statistical norm established by age group and whether this performance is standard, or is under expectations through the observation of the child’s behaviour in situations proposed by the exam- iner. The TEPSI yields results at global as well as sub-scale levels, which are coordination, language and motor functions (Haeussier and Marchant, 2003 and Wechsler, 1974). Was applied only in the first wave. Test of Learning and Child Development (TADI) Psychomotor skills (for children 6-80 months old), not in English. TADI is a Chilean instrument that allows measuring what children know, and what they do, according to four dimensions of development: language, cognition, motor and socio emotionality, each of which constitutes a separate scale. Therefore, the TADI allows to evaluate learning and development globally, covering the four dimensions, or by each dimension separately (CEDEP, CIAE, 2012). Was applied only in the second wave. Wechsler Adults Intelligence Scale (WAIS) Intelligence/cultural level and memory skills/processing speed/short-term auditory memory (for primary caregivers) through the Vocabulary and Digit Span sub-scales respectively. The WAIS measures human intelligence reflected in both verbal (which mea- sures the subject’s knowledge of word meaning) and digital distance (ability to recall digits from memory, performance based on the maximum length of a list of digits the subject can remember) abilities. It is based on the belief that intelligence is a global construct, which reflects a variety of measurable skills and that can be considered in the context of the overall personality. It has been demonstrated that the test provides highly accurate measurements and has a high predictive capacity regarding the future behaviour of an individual. The WAIS is also administered as part of a battery test to make inferences about personality and pathology; both through the content of specific answers and patterns of subtest scores (Apfelbeck and Hermosilla, 2000). Was applied 37 only in the first wave. Social-Emotional Measures Child Behaviour Checklist (CBCL1) (for children 18-60 months old) Obtain standardised ratings, and descriptive details of children’s function- ing, as seen by parents/caregivers providing results for three general scales: Total Problems, Internalising and Outsourcing There are 7 syndrome scales designates as Emotionally Reactive, Anxious/Depressed, Somatic Complaints, Withdrawn, Sleep Problems, Attention Problems, and Aggressive Behaviour (Achenbach and Rescorla, 2000). Was applied only in both waves. The Big Five Inventory (BFI) (for all primary caregivers) The BFI is a self-report inventory designed to measure the Big Five dimensions. The five factors are Openness, Conscientiousness, Extroversion, Agreeableness, and Neuroticism. It contains 44 items and consists of short phrases with relatively accessible vocabulary (Casullo, 2000). Was applied only in the first wave. All Cognitive and Social-Emotional measures have a raw score as well as T scores. As the analysis is within the sample I have internally standardised each measure, taking advantage of the sample size per month, using non-parametric estimation for age and hence I remove the age e↵ect. For this, the first step is to use kernel-weighted local polynomial smoothing methods to regress the raw score on child’s age (in months) and recover the mean. The second step is to use again the same estimation method to regress the square of the residuals in the first regression on child’s age and recover the variance. Finally, for calculating the z-score, I subtract the mean and divide by the standard deviation for each raw score. Figures 2.3 and 2.4 shows the z-scores for the cognitive and socioemotional tests. 38 of the Environment (HOME). It has items that can be related to i) parent- child interactions as reading books, singing songs, taking the child outside the home, playing with the child, spending time with the child, etc. and ii) learning materials. 2.4.2 Factor Analysis In this subsection, I develop an empirical analysis to generate relevant measures for analysing early childhood using Principal Components Analysis (Pearson, 1901 and Hair et al., 1987), Factor Analysis (Child, 1973) and Item Response Theory. Principal Component Analysis (PCA) is defined as a linear combination of variables used in the analysis. In this method, the new variables that are generated are independent of each other and hence are uncorrelated indexes or components. With this technique, it is possible to reduce the number of variables that we have originally. The newly created components can explain much of the total variability of the data. The weights for each principal compo- nent are given by the eigenvectors of the correlation matrix, while the variance for each principal component is given by the eigenvalue of the corresponding eigenvector (Hair et al., 1987).7 Alternatively, Factor Analysis (FA) is defined as a latent variable model, not as PCA but that aspires to reduce the number of variables. In this case, the factors (instead of the components) derived from the analysis are assumed to represent the original processes that result in the correlations between the variables. The factors are continuous and follow a multivariate normal distri- bution. In FA, it is necessary to choose an option for estimation among the principal factor (default option), the iterated principal factor (same as the prin- cipal factor but that is computed iteratively), the principal component factor (same as PCA), and maximum likelihood. For the two methodologies (PCA and FA), the number of extracted components/factors is defined by a couple of common methods used to select components/factors. The first method is a screen plot looking for the elbow in the chart, while the component/factor 7Notice that categorical data are not suitable for PCA, as the categories are converted into a quantitative scale which does not have any meaning and also it is important to notice that if the data has been standardised, then PCA should use the covariance matrix. 41 before the elbow is the number of components/factors to keep. The second one for deciding is to just keep every component/factor with the associated eigenvalue greater than one. The logic here is that each variable accounts for a variance of 1 so if a component/factor accounts for a variance of more than 1, then it accounts for more variance than any one of the original observed variables could (Hair et al., 1987). The classical analysis of items does not provide information about how the level of ability interacts with the performance required by the item to be answered correctly. This situation is resolved by the implementation of Item Response Theory (IRT). The Item Response Theory assumes that responses to the items of a test or inventory can be explained from a latent trait, or that the underlying latent trait can be explained from the answers to the items of a test or inventory. One of the key concepts of Item Response Theory is the Item Characteristic Curve (ICC), which is represented by a graph that shows the probability of responding correctly to an item as a function of the performance of the underlying latent trait in an item from the test or inventory. The ICC is a curve which increases a person’s ability increases since it has been more likely to answer the item correctly. The importance of the ICC regarding the classic indicators of diculty and discrimination is that it allows visualising how the probability of responding correctly to an item depends on the level of the latent trait that the evaluated has. The application of Item Response Theory is valid under the fulfilment of the following assumptions: • Local Independence: The probability of responding correctly to the item interacts only with the skill level, and not the result of another factor (tracks that give other items within the test or inventory). • Unidimensionality: It is defined regarding statistical dependence between the items of a test or inventory. Specifically, the requirement that a test/inventory is unidimensional is that the statistical dependence be- tween items can be explained by a dominant latent trait. This means that a test is unidimensional if their items are statistically dependent for entire testing population, and for a dominant latent trait. • The latent trait is continuous and normally distributed. 42 • Conditional on the latent trait, the responses to any two items are inde- pendent of each other. The most popular models of the Item Response Theory are logistic models, which di↵er by the number of parameters to be estimated and can only be applied to binary responses items. I use the standard two-parameter logistic (2PL) IRT model for recovering the underlying latent trait which assumes that the interviewee did not use random to answer an item correctly. Therefore, the probability of correctly answering an item depends only on the skill level of this person and the characteristics of the item, which are: diculty and discrimination, since it is assumed that the random tends to zero. In particular, let the probability of the person i having a value of 1, which indicates a positive or ”correct” response, for item j be: Pr(Yi,j = 1 | ↵j, j, ✓i) = e↵j✓i+j 1 + e↵j✓i+j , (2.1) where ✓i denote the underlying latent trait by interviewee i. The parameter ↵j represents the discrimination of item j which tells how fast the probability of success on that item responds to small changes in the latent trait, close to the diculty of that item. The higher the discrimination parameter, the bet- ter that item discriminates, around its diculty value, between interviewees of similar levels of the latent trait. Using the IRT parameterisation; the trans- formation is aj = ↵j and bj = j/↵j, where the parameter bj is the diculty. The item is more dicult the higher it’s level of bj and therefore the lower the level of j. Based on a factor analysis, three measures will be studied below in the as- sumption that these are associated with early childhood development as shown in Section 2.3: Wealth factor, SES factor, and Home Assessments factors. For the first two, the idea is to construct two factors that measure the same (wealth and socioeconomic status factor). The chosen factor has the best performance according to principally to the Kaiser-Meyer-Olkin measure of sampling ade- quacy. 43 2.4.2.2 Socioeconomic Status (SES) score As there is no consensus in which multivariate analysis is the optimal to con- struct an SES score, I use the PCA method9 because it does not require multi- variate normal distribution assumption. McLoyd (1998) proposes to construct an SES score using home information of education, occupation and income. The education variable is defined as the parental education measure in years of schooling starting in 0 (no education), while the occupational variable is defined as the parental occupational position starting from unpaid workers to employers. Finally, the income variable is defined as the per capita income, in other words, the total familiar income divided by the number of people in the home. For SES, I also construct a score for each sample, where the first principal component explains on average 44% of the variation in these vari- ables (using multiple imputation techniques before generating the factor when there are missing values). The SES factor is placed into quintiles where the first quintile is the poorest. Nevertheless, once the score is constructed (not only by PCA but also by FA) the Kaiser-Meyer-Olkin measure is between 0.50 to 0.59 leading a miserable sampling adequacy. This does not result in some components to be used by the scree plot criterion, which leads to choosing the use of only the wealth score as the measure of socioeconomic status in the empirical analysis. Appendix 2 shows the relevant statistics and figures for screen plot criterion that supports the decision. 2.4.2.3 Home Assessment scores A crucial determinant of early childhood development is the information that is possible to collect from the child’s environment. As shown in Section 2.3, one of the key systems for childhood development is the Family System in which parents and other regular caregivers are important assets of the influence of the environment during early childhood as well as household equipment. The most worldwide known instruments are the HOME inventory as well as the FCI. Table A3.1 and Table A3.2 in Appendix 3 present the statistics of the items from the original instruments that are included in the survey and used 9Independently of the factor strategy all methods create similar scores. The pairwise correlation between PCA and the two FA is about 0.92-0.99 depending on the sample. Appendix 2 shows these results using the first wave of the survey. 46 in this analysis for both waves. Regarding the HOME, they are grouped in 2 subscales for the first wave: emotional and verbal responsivity and paternal involvement with the child and in 2 subscales for the second wave: emotional and verbal responsivity, and learning materials. Regarding the FCI, two other measures related to parent-child activities and learning materials in the home were constructed with information included in the first wave and a parent-child activities factor using the second wave10. Figure A.0.1 in Appendix 3 show the proportion answering yes for each item. For creating the home assessment scores, in particular for the HOME in- ventory, I use a two-stage procedure. Firstly, I use a Factor Analysis initially as exploratory regarding the number of factors that can be extracted from only the HOME inventory as it contains multiples subscales11. Two scores are extracted according to the screen plot criterion. Each score is related with each subscale. Table A3.3 in Appendix 3 presents the factor analysis for both waves. For the second step, I use a standard two-parameter IRT model to analyse the extent to which the set of di↵erent items for both the HOME and FCI can be used to estimate one or more underlying latent traits represent- ing the quality of the home environment (HE). Note that each of the items j of these inventories are a discrete binary variable. Thus, let define the score HEk,⇤i,j in the following way: HEk,⇤i,j = a HE,k j (✓i b HE,k j ) + u HE,k i,j , (2.2) where uHE,ki,j is the measurement error that has logistic distribution and let ✓i be normally distributed with mean zero and variance 2. The parameters aHE,kj and b HE,k j capture the diculty and the informativeness of a given item in each scale k. Let HEki,j 2 {0, 1} denote the observed score for person i in item j from the HE scale k. It follows that: HEki,j = ( 0, if HEk,⇤i,j  0, 1, if HEk,⇤i,j > 0. 10Table A3.1 and Table A3.2 show the items of each measure. 11Independently of the factor strategy all methods create similar indexes. The pairwise correlation between PCA and FA is about 0.80-0.83 depending on the sample. All variables are standardised. Responses are incorporated to the maximum likelihood extraction method with 50 iterations and with varimax rotation instead of promax due to the correlation. 47 The estimation is by maximum likelihood and uses GaussHermite quadra- ture to approximate the log likelihood using 7 points. Letting for each k: pi,j = Pr(Yi,j = 1 | bj, aj, ✓i) and qi,j = 1 pi,j, the conditional density for person i is: f(yi | B, ✓i) = JY j=1 p Yi,j i,j q 1Yi,j i,j , (2.3) where yi = (Y1,i, ..., YJ,i), B = (b1, ..., bJ , a1, ...., aJ). The log likelihood is the sum of the I individual log likelihoods: Li(B) = Z 1 1 f(yi | B, ✓i)(✓i)d✓i, (2.4) where (·) is the standard normal density function. The IRT estimation of each scale k are present in Appendix 3 in Tables A.2 to A.7. The scores generated are for Learning materials, Parent-Child Inter- action and Emotional and Verbal Responsivity for both waves, and Paternal Involvement only for the first wave. Table 2.1 shows the results for Parent-Child Interaction in the second wave. Using the IRT parameterisation described above, the parameter ↵j = aj and bj = j/↵j, where the parameter bj is the diculty. The item is more dicult the higher it’s level of bj and therefore the lower the level of j. 48 In this case, all the items are informative as they have significant and a high ↵0s, which corresponds to a steeper slope and should result in a better item fit, and some are more dicult than others but all significant. This is not the same as all the scores where most of them have only loose items but all significant. An informative score would have items that have discriminatory power as well as variability in diculty, under this type of score one can identify di↵erent types of quality environment. Figure 2.7 shows the Item Characteristic Curves (ICCs) for each scale k, most of them show the variation of the ICCs in the negative side of the graph which means that even though they are informative, they are only informative to explain low levels of the quality environment. In this sense, all the results presented in Section 2.5 regarding home assessments are the lower bound. 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_eedp_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_tepsi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_bdi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_tvip_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_cbcl1_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_eedp_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_tepsi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_bdi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_tvip_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_cbcl1_2010 Learing Materials & Activities 2010 Expected score 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_tadi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_bdi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_tvip_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Lear_Mat_cbcl1_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_tadi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_bdi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_tvip_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Activities_cbcl1_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_tadi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_bdi_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_tvip_2012 0 2 4 6 8 10 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_cbcl1_2012 Learing Materials Activities and Responsivity 2012 Expected score 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_eedp_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_tepsi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_bdi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_tvip_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta Responsivity_cbcl1_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta PC’s Involvement_eedp_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta PC’s Involvement_tepsi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta PC’s Involvement_bdi_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta PC’s Involvement_tvip_2010 0 2 4 6 8 E xp e ct e d S co re −4 −2 0 2 4 Theta PC’s Involvement_cbcl1_2010 Responsivity & Maternal Involvement 2010 Expected score Figure 2.8: Home Assessments: Test Characteristic Curves (TCCs) Continuing with the IRT models, Figure 2.8 plot the test characteristic curves (TCCs) which are the expected scores for the latent traits for each 51 scale k and Figure 2.9 show the test information functions (TIFs) which are the sum up of all the item information functions that tells us how well the estimation of the latent trait can estimate household locations. The scores provide maximum information for households approximately located at ✓ =- 0.5. As we move away from that point in either direction, the standard error of the TIF increases and the instrument provides less precise information about ✓. .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_eedp_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_tepsi_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_bdi_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_tvip_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_cbcl1_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Activities_eedp_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Activities_tepsi_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Activities_bdi_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Activities_tvip_2010 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 In fo rm a tio n −4 −2 0 2 4 Theta Activities_cbcl1_2010 Learing Materials & Activities 2010 Test information Standard error .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_tadi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_bdi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_tvip_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Lear_Mat_cbcl1_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Activities_tadi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Activities_bdi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Activities_tvip_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Activities_cbcl1_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_tadi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_bdi_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_tvip_2012 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 1 0 2 0 3 0 4 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_cbcl1_2012 Learing Materials Activities and Responsivity 2012 Test information Standard error .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_eedp_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_tepsi_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_bdi_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_tvip_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta Responsivity_cbcl1_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta PC’s Involvement_eedp_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta PC’s Involvement_tepsi_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta PC’s Involvement_bdi_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta PC’s Involvement_tvip_2010 .2 .4 .6 .8 1 S ta n d a rd E rr o r 0 2 4 6 8 1 0 In fo rm a tio n −4 −2 0 2 4 Theta PC’s Involvement_cbcl1_2010 Responsivity & Maternal Involvement 2010 Test information Standard error Figure 2.9: Home Assessments: Test Information Functions (TIFs) 2.5 Methodology Throughout this section, I work out a methodology to disentangle the deter- minants of cognitive and socioemotional childhood development and hence to assess the impact of public policies that intervene optimally and successfully early childhood development. To achieve this purpose, I first analyse children’s cognitive and socioemotional measures for each sample defined by age groups 52 of children. Then, I study the socioeconomic gradients that may exist. Finally, I perform regression analysis for each child’s cognitive test. 2.5.1 Socioeconomic Gradients and Childhood Devel- opment The main goal of the first wave of the survey is to assess the cognitive and socioemotional abilities as the physical status of children below 5 years old and their mothers or primary caregivers. The first wave of the ELPI has information about 15,175 households, who have on average 1.1 children under 5 years old, 4.7 people living in the household, an average monthly income per capita equal to 149.92 pounds, 46.6% are male, 40.2% live in the Metropolitan Region (which contains Chile’s capital), 35.5% are employed and only 4.2% is composed of a household head aged 45 years old or older. This is because the sampling households for the first wave are those who had a child born on January 1, 2006, and August 31, 2009, so they are mostly households with young household heads.12 The first wave of the ELPI has an achievement of 91.6% of measurement assessment, which means, that a total of 13,895 households agreed to be as- sessed, and as explained in Section 2.4 during the first wave four cognitive and one socioemotional test was applied resulting in five samples that can be used for the empirical analysis (see Table A4 2 in Appendix 4). For the sec- ond wave, I use only the panel data and no the information collected in the refreshment sample for children below 3 years old. The number of households from the panel data that were also interviewed in the second wave was 12,898, so the attrition level is approximately 15.0%. During the second wave, three cognitive and one socioemotional test was applied resulting in four samples that can be used for the empirical analysis (see Table A4 3 in Appendix 4). There are one cognitive and one socioemotional test that was applied in both waves, the TVIP and the CBCL1 respectively. Most of the primary caregivers are the mother of the child, in fact, a 98.0% so from now on the mother or primary caregiver will be treated indistinctly as mothers. As we have several tests for di↵erent ages, I divide the sample into two 12For details see Table A4.1 and Figure A4.1 in Appendix 4. 53 where hk,t,i denote the cognitive or socioemotional measure k = c, s for child i at age t, the hk,t=0,i vector represent the Child System which is de- fined by the initial conditions given by genetic, in this case, I use the child’s physical information at birth (the z-score for weight and length). For the Family System, we use three vector, the P ck,t,i which includes maternal educa- tion as well as the mother’s cognitive test (WAIS), vector P sk,t,i which includes parental mental health through mother’s socioemotional test (BFI) and the vector HEk,t,i which includes all the variables that fall into maternal char- acteristics and family background categories, in particular, mother’s age and employment status as well as indicators for younger or older children in the household and if the family is a two-parent family. Other variables included are the home assessment using the scores described in Section 2.4.2.3 which includes parental style and a measure of household equipment: emotional and verbal responsivity, paternal involvement, learning materials and parent-child activities. The Environment/Neighbourhood is measured using the indicator if the child attends an educational centre that somehow takes into account the child-care or preschool centres access and equipment which is addressed by the vector CCk,t,i. Finally, area’s type (urban/rural) and the socioeconomic gradient are used as measures of socioeconomics’ equity in the Society System through the vector Xk,t,i 13. In the estimation procedure, there are some diculties that need to be addressed, firstly, we do not observe hk,t, Pc, P c, P s and HEt directly. Indeed, most of the measures used to create factors for cognitive, socioemotional and home assessments (investment) inputs are measured with error. One way to address the measurement error is the use of IRT analysis for the estimation of unbiased investments parameters. Another issue is that problems of multi- collinearity and endogeneity may arise for this I assume that all the omitted factors are orthogonal to the included input measures. A second way to ad- dress these issues is to use a value-added specification (Todd and Wolpin, 2003, 2006) as the following: 13Table A4 2 and A4 3 show descriptive analysis of the sample of children depending on the applied test for the first and second wave. 56 hk,t,i = 0 + khk,t1,i + h0hk,t=0,i + pcP c k,t,i + psP s k,t,i + heHEk,t,i + heHEk,t1,i + ccCCk,t,i + xXk,t,i + µk,t,i, (2.6) Nine models are estimated using specification (2.5) and two using specifi- cation (2.6) due to the availability of lagged information for one cognitive and one socioemotional test, the TVIP and CVCL1 respectively as well as lagged information for HE. 2.5.3 Results Tables A.8 to A.16 in Appendix 4 present the contemporaneous linear regres- sion models using ordinary least squares and where the dependent variable is the children’s cognitive or socioemotional test z-score in each wave. Tables A.8 to A.12 in Appendix 4 are for the first wave meanwhile tables A.13 to A.16 are for the second wave. Note that the contemporaneous linear regression models are also for di↵erent age stage. The early (later) stage is presented in Ta- bles A.8 (EEDP) and A.9 (BDI 2010) (Tables A.10 (TEPSI) to A.16 (CBCL1 2012)) in Appendix 4 14. In each estimation, I obtain the socioeconomic gra- dient in children’s test scores. This allows me to confirm the results found it in the Section 2.5.1 about the existence of socioeconomic gradients (specifica- tion 1) and successively add to the model estimation demographics, mother’s characteristics and other family background (specification 2), child’s physical (specification 3), mother’s cognitive and socioemotional tests (specification 4) and child’s home environment (specification 5). In the early stage specifications, Table A.8 show that children in the fifth quintile have higher average EEDP z-scores (di↵erence of 0.276 standard devia- tions) than children in the bottom quintile statistically at 1%. Hence, a socioe- conomic gradient in cognitive, psychomotor development is present. However, this gradient begins to disappear and to lose significance since the specification is controlling by other variables like hk,t=0,i, Xk,t,i, and CCk,t,i. In general, the 14Regression check and deal with multicollinearity, nonlinearities and heteroskedasticity. Also included dummies variable created for each explanatory variable with missing values (coded as a one if the related explanatory variable is missing and zero otherwise). 57 children’s age (in months) and being male have positive and negative signifi- cant e↵ects on EEDP respectively, mother’s age and if the child attends to a child-care are significant and negative suggesting that younger children are not translating what they receive in the child-care to improvement in the EEDP. The main result is the drop in the socioeconomic gradient due to the inclusion of child’s physical information at birth (weight), mother’s socioemotional tests (BFI Openness), and the parent-child, involvement and responsibility scores which have significantly big and positive e↵ects on EEDP z-scores. In partic- ular, both involvement and responsibility one standard deviation of the scores increase the EEDP z-scores by 0.246 standard deviation. On the other hand, Table A.9 show that the BDI has a z-score that is 0.39 standard deviation higher for children in the bottom quintile, which finished with a z score equal to 0.057 and losing significance when controlling for all variables (specification 5). The latter implies that a bigger socioeconomic gradient is present respect to EEDP due to the results’ stability in this test. Child’s age and being male have positive and negative significant e↵ects on EEDP respectively. The fall in socioeconomic gradient is caused by the inclusion of physical information at birth (weight), urban household, mother’s age and cognition (only verbal WAIS), socioemotional tests (BFI Openness), and by the learning materials, parent-child, involvement and responsibility scores, which have positive e↵ects on child’s general development measured by the BDI. For both, the EEDP and BDI, maternal employment status does not a↵ect. In the later stage specification, Table A.10 show the TEPSI has a z score that is 0.649 standard deviations higher when compared to children in the bottom quintile, who have a 0.213 z-score without losing any significance after controlling for all variables by decreasing their e↵ectiveness. In general, the children’s age (in months) is not significant for most of the specifications and being male have adverse and significant e↵ects on TEPSI, mother’s age and if the child attends to preschool are significant and positive suggesting a comple- mentary e↵ect between preschools and household environment, as the learning materials, parent-child, involvement and responsibility scores, have positive ef- fects on child’s general development measured by the TEPSI. TableA.11 show that children in the fifth quintile have higher average TVIP z-scores (di↵erence of 0.912 standard deviations) than children in the bottom quintile. This stage 58 2.6 Concluding remarks In this chapter, I contributed to the discussion about the determinants of early childhood cognitive development through theoretical and empirical evidence. This chapter is one of the few studies that deal with this challenging concern in Chile and developing countries, mainly due to the inadequate information about the topic. I take advantage of a longitudinal survey designed to char- acterise children development, using its first two waves and that contains a set of cognitive, socioemotional and anthropometrics measures for both the children and their mothers that help to fill the emptiness of such evidence due to the typical unobserved parental abilities issue, measurement error and endogeneity, which allows a proper monitoring of early childhood. The determinants of early childhood development, particularly, cognitive and non-cognitive skills, are studied through the estimation of contempora- neous and value-added cognitive and non-cognitive production functions, as well as the use of factor analysis such as item response theory for reducing the number of inputs using a theoretical framework. I show that when the socioeconomic wealth score is included as a whole, the overall association with cognitive and socioemotional child’s development can be obtained for early and later stages. This is also supported by em- pirical analysis indicating significant socioeconomic gradients in all cognitive and socioemotional tests between the poorest and richest quintiles. The latter implies that disadvantaged children are liable. Once controlling by other ob- servables, the gradient starts to decrease and in some cases to lose significance; there is a significant e↵ect of mother’s characteristics and family background at later stage development (above 24/30 months old) measured principally by mother’s education, age and cognitive skills, if the family is a two-parent family, the presence of younger/older children as well as home environment measures by parent-child activities, learning materials, parental involvement and verbal and emotional responsibility scores. The later stage development also adds a significant e↵ect on attending a preschool. The previous determinants drive the fall in the socioeconomic gradient in both stages. Regarding the non-cognitive skills, for both waves, the results are similar, there are socioeconomic gradients that are still significant after controlling for 61 all the variables. If the child is male, have an adverse and significant e↵ect as well if they attend to preschool. Mother’s education and age have positive and significant e↵ect meanwhile having younger children in the household have a negative and significant e↵ect. Having both parents have a positive e↵ect as well as child’s weight at birth and the mother’s cognition level. For the first time, all the subscales of the mother’s socioemotional test are (positively) cor- related with the child’s socioemotional skills. The home environment continues presenting positive and significant e↵ect on child’s development. A common limitation of previous studies and also this one is that they have failed to control for potential endogeneity problem fully: parents that work may be systematically di↵erent from parents who do not work, and the child’s skills itself may influence parental decisions of whether to work or not, moreover, parents are heterogeneous in their skill endowments, the constraints they face, and their tastes, therefore is crucial to allow parental decisions to depend on these unobserved heterogeneous characteristics of both parents, as well as, measurement error problem. For this reason, in the next chapters, I try to deal with these issues including a measurement error system in the estimation of the production function for human capital and estimating a model of parental investment choices jointly with a child production function for cognitive and socioemotional abilities. These results raise important questions about the relevance about quantity v/s quality evaluation, about the ideal form of early interventionism, about the implications on the optimal design, and about the implementation of an optimal and successful public policy designed to improve child development, for example, between the optimal time for mother to enter or to go back to the labour market as well as when to send the children to child-care centres or preschool institutions. 62 Chapter 3 The myths behind the Technology of Human Capital Formation in Childhood 3.1 Introduction Recent research spotlights the e↵ect of early influences and investments by parents during early childhood on brain blooming, learning skills and non- cognitive abilities suggesting that the first years of life are a crucial period for children’s development. Non-cognitive abilities are associated with patience, perseverance, temperament, motivation, self-control and self-esteem among others. Heckman et al. (2006) find that cognitive and non-cognitive abilities are critical to the future performance on a series of social and labour market outcomes: enrolment rates, wages, work experience, crime rates, early preg- nancies, drug use in the labour market, among others providing evidence of the e↵ect of cognitive and non-cognitive abilities in socioeconomic success. Nonetheless, gaps in both cognitive and non-cognitive skills emerge at early ages before preschool and remain constant in the life-cycle; consequently, a clear comprehension of the di↵erent channels for the human capital formation process in childhood is essential for improving childhood development. Only when these channels, mostly related to genetics, investments and environment, 63 a second contribution is to analyse if complementarities change with age in a context of a developing country. A third contribution is concerning the optimal number of dimensions, are two dimensions enough or should we expand the system to three or four (cognition, socio-emotional, language, health, etc.)? The estimation of these technologies of human capital formation in child- hood allows answering appropriately what investment practices or strategies are more ecient in the creation of cognitive and non-cognitive skills (and other dimensions). The previous chapter uses the methodology proposed by Todd and Wolpin, 2003, 2006 and finds that mother’s skills, the home environment as well as at- tending to preschool are the most important determinants of skill formation in childhood. This chapter attempts to contribute to the growing early childhood literature through addressing some of the myths behind the estimation of the technology of human capital and the real returns on parental investment for di↵erent stages during childhood. This chapter also contributes from previous research as include a rich Chilean data to apply the state-of-the-art methodology in the estimation of the production function. I use data from a recently longitudinal survey designed in Chile to characterise children development, using its first two waves, which were collected in 2010 and 2012, and that contains a set of cognitive, socioemo- tional and anthropometrics measures for both the children and the mothers or primary caregivers. There is also information about home assessment using the Home Observation for Measurement of the Environment (HOME) inventory and the Family Care Indicators (FCI) that help to measure intermediate out- comes and mediators for early childhood development. This information deals with the typical unobserved parental abilities issue, which allows a proper mon- itoring of early childhood. Chile is one of the major industrialised countries in Latin America, but at the same time, is the country with higher inequali- ties among its population regarding income relative to other OECD countries (OECD, 2011). This fact leads perhaps too strong socioeconomic gradients by cognitive and non-cognitive development. This chapter proceeds as follows: Section 2 presents the framework of parental investments and the technology of human capital in childhood and how is possible to deal with measurement error and endogeneity of inputs. 66 Section 3 presents the survey data and the multiple empirical specifications for the estimation of the technology of human capital in childhood. Section 4 presents the main results. Section 5 concludes. 3.2 Estimation of the Technology of Human Capital Formation There is growing recognition that multiple skills are essential predictors and likely determinants of success in many aspects of life. Although a variety of methods are used to measure these skills, there is no agreement on the best ways to do so. Parental environments and investments at di↵erent ages and stages of childhood development determine skills. Recent studies have demon- strated how multiple factors relate in a complicated way (Cuhna et al. (2007, 2010)). Understanding the factors a↵ecting the evolution of multiple skills is crucial for the design of e↵ective public policies in developing countries. As pointed out by Cuhna et al. (2010), it is necessary to estimate a multistage technology to capture di↵erent development phases in the life cycle of a child. For this, they identified a more general nonlinear technology by extending linear state space and factor analysis to a nonlinear setting. This allows elim- inating the assumption that early and late investments are perfect substitutes over the feasible set of inputs. To estimate the technology of human capital, I use the two-step procedure proposed by Cuhna et al. (2010). In this technology, inputs at each t produce outputs at t+1 and children’s development is driven by a certain number of la- tent factors which are reflected in measurements. Similarly, investment is also driven home environmental measurements. In the first step, as some issues need to be addressed, such as the presence of endogeneity (correlation with the unobserved shock) and measurement error in the data for the inputs in the technology, I estimate the measurement model of the latent factors based on a nonlinear dynamic factor model and exploit cross-equation restrictions (covariance restrictions) proving that I can identify all of them. In the second step, I estimate together with the dynamic factor model and the non-linear technologies. I consider initially two technologies, one for the production of 67 cognitive skills and one for the production of non-cognitive skills. One innova- tion in the estimation is to add a new multi-dimensional parental investment, specifically, measured as material resources (monetary investment) and quality time investment as well as cognitive stimulation and socioemotional support. A second possible innovation in the estimation is to increase the set of tech- nologies including also a third and even fourth skill, for example, health or language1. I estimate di↵erent specifications of the technology of skill formation for cognitive and non-cognitive skills in childhood, for all of them the production function is assumed to be the Constant Elasticity of Substitution (CES) to allow for complementarities of inputs. Specification 1 assumes one input for measuring investment for only one stage of child’s development, Specification 2 estimates previous specification but for multiple stages of child’s develop- ment, there are 3 child age stages at t = 0 and hence 3 more at t = 1, each of them accounting for a di↵erent childhood stage, the gap between t = 0 and t = 1 is approximately 20 months. The first age stage is for children aged 7-23 months, the second one for children aged 24-47 months and the last one for children aged 48-58 months. The decision behind the stages is mostly due to the di↵erent skills that are measured at each stage. Specification 3 assumes multi-dimensional parental investment, specifically, measured as material re- sources (monetary investment) and quality time investment and Specification 4 assumes multi-dimensional parental investment, specifically, measured as cognitive stimulation and socioemotional support. The explanation about the procedure behind the estimation of the technol- ogy of skill formation is based in Specification 1, the results hold for the other specifications as well. Specification 1 incorporates only one input for measur- ing investment, parental investment it in child skills at age t, that accounts for several measures, in particular, parent-child activities, learning materials, emotional and verbal responsivity, paternal involvement, acceptance, child’s discipline and activities related with the program Chile Grows with You. The program (Vega, 2011) provides free access to childcare centres and parent- ing advice for children among the poorest 60% and between3-months-old to 1I’m currently in the estimation process of a system of three skills which includes cogni- tion, health and non-cognitive skills. 68 yk,t,j = µk,t,j + ↵k,t,jln(hk,t) + "k,t,j, (3.3) where µk,t,j is the mean and the vector "k,t,j captures measurement error. For each measurement system, the goal is to recover the latent factor hk,t which is error-free. Several assumptions are needed for the identification of the measurement system where the covariance restrictions in measurement error are crucial: 1. Define a factor scale ↵k,t,1 = 1 in each system, which means we need to standardise the factor loading of one measurement per latent factor equal to 1; 2. Define the factor location by setting E(ln(hk,t)) = 0; 3. "k,t,j are uncorrelated across all measurements; 4. E("k,t,j) = 0; and 5. There are at least 3 measures for each factor. The identification draws from the Kotlarsky theorem that states that with two independent mea- surements per factor, the distribution of the underlaying trait and the measurement error can be identified non-parametrically up to a change of location. Using these assumptions it is straightforward to proof the identification for the children’s skills, maternal skills and parental investment latent factors. Let’s use hc,t as an example. I proceed in several steps, first let’s consider the following fourth measurements for the cognitive latent factor at t: yc,t,1 = µc,t,1 + ln(hc,t) + "c,t,1 yc,t,2 = µc,t,2 + ↵c,t,2ln(hc,t) + "c,t,2 yc,t,3 = µc,t,3 + ↵c,t,3ln(hc,t) + "c,t,3 yc,t,4 = µc,t,4 + ↵c,t,4ln(hc,t) + "c,t,4, 71 Initially, we can identify the factor mean for the first measurement: µc,t,1 = E[yc,t,1] as E(ln(hk,t)) = 0 and E("k,t,j) = 0. The second step is to identify the factor loadings for the measurement system, for this, first I calculate de covariance using assumptions 3 and 4 for the (3.3), (3.4) and (3.6), relationship (3.5) combines results from (3.3) and (3.4) and relationship (3.7) combines results from (3.4) and (3.6): cov(yc,t,1, yc,t,2) = ↵c,t,2var(ln(hc,t)) (3.4) cov(yc,t,1, yc,t,3) = ↵c,t,3var(ln(hc,t)) (3.5) ↵c,t,2 = ↵c,t,3[cov(yc,t,1, yc,t,2)/cov(yc,t,1, yc,t,3)] (3.6) cov(yc,t,2, yc,t,3) = ↵c,t,2↵c,t,3var(ln(hc,t)) (3.7) = ↵c,t,3cov(yc,t,1, yc,t,2), (3.8) Finally, rearranging I can (3.7) and solving for ↵c,t,3 I can then identify all the factor loadings of the measurement system: ↵c,t,1 = 1 ↵c,t,3 = cov(yc,t,2, yc,t,3)/cov(yc,t,1, yc,t,2) ↵c,t,2 = cov(yc,t,2, yc,t,3)/cov(yc,t,1, yc,t,3) ↵c,t,4 = cov(yc,t,2, yc,t,4)/cov(yc,t,1, yc,t,2), Then, the variance of the underlaying trait is derived with the previous results and defined as: var(hc,t) = [cov(yc,t,1, yc,t,2)/cov(yc,t,2, yc,t,3]cov(yc,t,1, yc,t,3) var("c,t,1) = var(yc,t,1) dvar(ln(hc,t)) var("c,t,2) = var(yc,t,2) ↵2c,t,2 dvar(ln(hc,t)) var("c,t,3) = var(yc,t,3) ↵2c,t,3 dvar(ln(hc,t)) var("c,t,4) = var(yc,t,4) ↵2c,t,4 dvar(ln(hc,t)) 72 I can follow the same logic in (3.3) and assumptions for identification of the other children’s skills, maternal skills and parental investment. 3.2.2 Endogeneity of Inputs The use of Instruments Variables or within-child/family fixed e↵ect estimators have been implemented for addressing the problem. For the later, it is needed to have multiple observations for the child at di↵erent ages or assume that children in the same family have a common heritable component. As explained before, this problem arises because parents adjust their investment decisions to unobserved shocks or inputs that are not observed by the econometrician. Cuhna et al. (2010) propose a solution involving exclusion restrictions based on economic theory in the spirit of control function procedure. Hence, for identification of the model I need some exclusion restrictions, at least as many as endogenous variables are in the model, that determines parental investment decisions but not to be included in the technology of skills. To solve this I use therefore exclusion restrictions, say zt, that is contained in the state variables at period t but the reverse is not true, meaning, zt it is not contained in the children’s skill, maternal skills and an invariant-time heterogeneity component, ⇡ observed by parents before making investment decisions but they exogenously a↵ect the household budget constraint. zt con- tains average comuna-level female (FW ) and male (MW ) wages as they change the household budget constraint at t and hence parental investment decisions, but they are not included in the production function estimation. At least one exclusion restriction is needed to identify the model and thus the parental in- vestment decision but for the case of multi-dimensional investment at least 2 exclusion restrictions will be needed, for this case, I also add to the set the average comuna-level unemployment (U) and the average region-level of invest- ment prices (IP )3. The variable ⇡ is assumed to be distributed independently among children and observed by the parent before making investment decisions and hence observed by the econometrician. The log-linear investment policy function, for the case I use two exclusion restrictions, is defined as: 3All the variables in zt are assumed to be measured without error. 73 3.3 Data and Empirical Specifications 3.3.1 Data I use new data from Chile that allows researchers to assess the impact of early childhood policies and to provide valuable information for the evaluation and design of social policies in this field since this data contains variables determin- ing children’s cognitive and non-cognitive abilities. The Encuesta Longitudinal de Primera Infancia (Early Childhood Longitudinal Survey (ELPI)) survey, which is a longitudinal study based on a representative sample of 15,000 chil- dren under 5 years old and their families for the first wave, during the second wave the sample is 18,000 as included a refreshing sample from 0 to 3 years old. The first wave was conducted in 2010 and the second one in 2012. It is a representative survey of children from urban and rural areas who were born on January 1, 2006, and August 31, 2009, in the first wave, and that is a representative sample. The sample includes di↵erent cohorts of children, distinguished by year of birth. The second wave follows the first wave but also add a refreshing sample of around 3,000 children born on September 1, 2009, and December 31, 2011. This survey provides demographic information, measures of a set of cog- nitive, non-cognitive and anthropometrics measures for both the children and the mothers or primary caregivers. There is also information about home as- sessment using the Home Observation for Measurement of the Environment (HOME) inventory and the Family Care Indicators (FCI) that help to mea- sure intermediate outcomes and mediators for early childhood development. Tests applied in both waves measured the development of children in di↵erent areas, such as motor, cognitive, language, emotional and social areas. The following describes the tests used in this chapter as measures of the children’s cognitive and non-cognitive skills, maternal cognitive and non-cognitive skills and parental investment for the di↵erent technologies specifications. The de- scriptive statistics are presented in Appendix 5 in Tables A.19 to A.22 for each age stage. In general, the measurements used for t are the ones applied in the ELPI 2010 and for t+ 1 the ones applied in the ELPI 2012. 76 Children’s Cognitive Measures Independently of the child’s age the ELPI has information about cognitive skills. For children aged 0-23 months old there is the test Escala de Evaluación de Desarrollo Psicomotor (EEDP) which is a Psychomotor Development Rating Scale. The test measures the performance and the reaction of the child to certain situations to be resolved for which a certain level of psychomotor development is required (Rodŕıguez et al., 2008). The measure used is called fc eedp in Table A.19. Another test used for the same age range is the Battelle Developmental Inventory (BDI 2010). The BDI provides three developmental domains related with cognition: cognition (fc bdi cog), communication (fc bdi com) and motor (fc bdi m). These three domains are further divided into twenty-two separate sub-domains. The mo- tor domain is composed of muscle control, body coordination, locomotion, fine muscle, and perceptual motor. The communication domain includes: receptive and expressive. Finally, the cognitive domain is composed of: perceptual dis- crimination, memory, reasoning/academic skills, and conceptual development (De la Cruz and González, 1998). For children aged 24-60 months old, the test Test de Desarrollo Psicomotor (TEPSI) measures the psychomotor Develop- ment Test. TEPSI is a screening test which yields results at global as well as sub-scale levels regarding coordination (fc tepsi coo), language (fc tepsi l) and motor (fc tepsi m) functions (Haeussier and Marchant, 2003 and Wech- sler, 1974). The subtest coordination evaluates the ability of the child to take or manipulate objects and draw, through behaviours such as build towers with cubes, threading a needle, recognise and copy geometric figures, draw a human figure among others. The subtest language evaluates aspects of understanding and expression of this, through behaviours such as naming objects, defining words, verbalise or describe actions scenes depicted in films and the subtest motricity evaluates the child’s ability to manage their own bodies through be- haviours like picking up a ball, hopping, walking on tiptoe or stand on one foot for a while. The previous three tests were applied only in the first wave. The Test de Vocabulario en Imágenes Peabody, Hispanic America adaptation (TVIP) is a language skill which is the Spanish version of the Peabody Picture Vocabulary Test (PPVT). This test measures the child’s comprehension and understanding of vocabulary using relating words to an illustration since age 30 months old. The scale (fc tvip l) should be viewed primarily as an achieve- 77 ment test since it shows the extent of Spanish vocabulary acquisition of the subject. Also, it may be viewed as a screening test of scholastic aptitude (ver- bal ability or verbal intelligence), or as one element in a comprehensive test battery of cognitive processes when Spanish is the language of the home and community into which the subject was born, has grown up, and resides; and when Spanish is, and has been, the primary language of instructions at school (Dunn et al., 1986). Was applied in both waves. During the second wave, two more tests were applied, the Battelle Developmental Inventory, Screening Test, 2nd ed. (BDI 2012) which includes the same areas as the BDI 2010. Has 96 items (two per each age level) extracted from the full version of the BDI and it is a screening test that evaluates the child development from 0-8 years old. The objective is to evaluate the fundamental skills development in three areas in cognition: cognition (fc bdi cog), communication (fc bdi com) and motor (fc bdi m). The second one is the Test of Learning and Child De- velopment (TADI) which is a Chilean instrument that allows measuring what children know, and what they do, according to three dimensions of cognition development: language (fc tadi l), cognition (fc tadi c) and motor (fc tadi m), each of which constitutes a separate scale (CEDEP, CIAE, 2012). Children’s Non-cognitive Measures For children aged 0-23 months old there are 2 subscales provided by the Battelle Developmental Inventory (BDI 2010) which are personal-social (fn bdi ps) and adaptive (fn bdi ad). The personal-social domain is composed of: adult interaction, expressions, feelings, a↵ect, self-concept, peer interaction, coping, and social role. The adaptive domain includes attention, eating, dressing, personal responsibilities, and toi- leting. There is also a scale provided by the Ages and Stages Questionnaires for socioemotional areas (ASQ: SE) applied for children 6, 12 and 18 months old. It is a parental report and helps to identify possible problems in the social and a↵ective development of the child. The tests (fn asq) addresses seven behavioural areas: self-regulation, compliance, communication, adap- tive functioning, autonomy, a↵ect, and interaction with people (Squires et al., 2006). The previous socioemotional tests were applied only in the first wave. The Child Behaviour Checklist (CBCL1) is applied for children older than 18 months old and obtain details of children’s functioning, as seen by 78 instruments, toys, books, pencils, dolls, educational games, etc. 3.3.2 Empirical Specifications Specification 1 incorporates only one input for measuring investment, parental investment it in child skills at age t, that accounts for several measures, in particular, parent-child activities, learning materials, emotional and verbal re- sponsivity, paternal involvement, acceptance, child’s discipline and activities related with the program Chile Grows with You. Hence, this specification includes a single-dimensional input for parental investment. hc,t denotes cog- nitive skills of the child at age t, hn,t denotes non-cognitive skills of the child at age t. Pc and Pn represent maternal cognitive and non-cognitive skills respec- tively, A is the total factor productivity which is a function of the covariates mentioned in (3.2) and ⌘t are shocks and/or unmeasurable inputs like the use of childcare or preschool. The technology for skill k = c, nc at period t (where t = 24 47 months) is: hk,t+1 = Ake ⌘t [k,ch c,k c,t + k,nch nc,k nc,t + k,pcP pc,k c + k,pncP pnc,k nc + k,ii i,k t ] 1 k , (3.16) where k,m 2 [0, 1], P m k,m = 1 for m 2 [c, nc, pc, pn, i]. ⌘t is assumed to be normally distributed and serially independent over all t. This formulation assumes that the formation of skills depend on initial conditions, Ak, the stock of skills in period t, parental investment at t, it, maternal skills, Pk, and shocks in period t, ⌘t. This specification is estimated with and without using the control function approach described in Section 3.2.2. Specification 2 is the same as Specification 1, but now it is estimated for di↵erent child stages: • 7-23 months in ELPI 2010; 33-51 months in ELPI 2012 • 24-47 months in ELPI 2010; 44-79 months in ELPI 2012 81 • 48-58 months in ELPI 2010; 69-83 months in ELPI 2012 In particular, the technology for skill hk,t+1,s for k = c, nc at period t and stage s is: hk,t+1,s = Ak,se ⌘t,s [k,c,sh c,k,s c,t,s + k,nc,sh nc,k,s nc,t,s + k,pc,sP pc,k,s c + k,pnc,sP pnc,k,s nc + k,i,si i,k,s t,s ] 1 k,s , (3.17) In this way, it is possible to make a distinction between periods of parental investments an the children’s stages of development. Specification 3 incorporates multiple inputs for measuring investment based on material resources and quality time in child skills at age t. The items used for measuring parental investment in material resources are from the FCI, in particular, items that ask about if the child has a space in which to keep their toys and belongings (SpecialPlace), has at least one toy with wheels that can be raised (WheelsToys), there are age-appropriate learning equipment (LearEquip) and has 3 or more books of his/her property (Books3). For measuring quality time with the child these are also using the FCI, in particular, items ask about reading stories or looking at picture books with child (LookBooks), telling stories to child (Stories), taking the child to parks, zoo or museums (GoOut) and spending time with child talking and/or drawing (Drawing). hc,t denotes cognitive skills of the child at age t, hn,t denotes non- cognitive skills of the child at age t. Pc and Pn represent maternal cognitive and non-cognitive skills respectively, A is the total factor productivity and ⌘t are shocks and/or unmeasurable inputs. The technology for skill k = c, nc at period t (where t = 24 47 months) is: hk,t+1 = Ake ⌘t [k,ch c,k c,t + k,nch nc,k nc,t + k,pcP pc,k c + k,pncP pnc,k nc + k,qtqt qt,k t + k,imr mr,k t ] 1 k , (3.18) 82 where qtt is the quality time the mother and/or father spends with the child at age t and mrt is material resources in child skills at age t. Table A.21 presents the descriptive statistics for all the investments items available in the survey. Specification 4 incorporates multiple inputs for measuring investment based on cognitive stimulation and emotional support in child skills at age t. The items used for measuring cognitive stimulation are from the FCI, in particular, items that ask about if the child has toys to push and pull toys (PushPull), has toys for role-playing (RolePlay), there are age-appropriate learning equipment (LearEquip), and there are literary and musical material (Musical). The items for measuring emotional support are from the HOME - Emotional and verbal responsivity scale, in particular, items that ask about if the mother sponta- neously vocalises to child at least twice (Vocalises) during the interview, keeps within his visual range and look often (Visual), spontaneously praises child at least twice (Praises) and caresses or kisses child at least once (Caresses). hc,t denotes cognitive skills of the child at age t, hn,t denotes non-cognitive skills of the child at age t. Pc and Pn represent maternal cognitive and non-cognitive skills respectively, A is the total factor productivity and ⌘t are shocks and/or unmeasurable inputs. The technology for skill k = c, nc at period t (where t = 24 47 months) is: hk,t+1 = Ake ⌘t [k,ch c,k c,t + k,nch nc,k nc,t + k,pcP pc,k c + k,pncP pnc,k nc + k,iccs cs,k t + k,iees es,k t ] 1 k , (3.19) where cst is the cognitive stimulation provided by parents for the child at age t and est is emotional support provided by parents in child skills at age t. Table A.21 presents the descriptive statistics for all the investments items available in the survey. 83 and maternal’s skills are significant and explain important part of the reason why parents invest in their children for all the specifications except for the Emotional Support. These findings explain the existence of skill gaps even at early ages and conditional on initial levels of skills in children as mothers with a higher level of cognitive and non-cognitive skills invest more in their children. I also find that investments decrease in any dimension if the level of unemployment increases. I find that female wages have a positive impact on investments, particularly, on material and emotional support. The e↵ect of prices is not significant and it has a small impact on investments. I perform sensitivity analysis using di↵erent combinations of the possible exclusion restrictions and I find that all of them work as instruments. Nev- ertheless, there is one parental investment equation, the one for emotional support, which do not present instruments that are too strong as the ones for the rest of investment equations. Table 3.1: Parental Investment Equations, Age 24-47 mths Log Log Material Log Quality Log Cognitive Log Emotional Investment Resources Time Stimulation Support Cognitive Skill t,c 0.246 0.142 0.102 0.114 0.019 (0.017) (0.019) (0.018) (0.018) (0.019) Non-cognitive Skill t,nc 0.047 0.059 0.026 0.043 0.110 (0.015) (0.017) (0.016) (0.016) (0.016) Mother’s Cognitive Skill pc 0.159 0.140 0.049 0.149 0.019 (0.014) (0.016) (0.015) (0.015) (0.016) Mother’s Non-cognitive Skill pnc 0.124 0.179 0.150 0.158 0.105 (0.016) (0.018) (0.017) (0.017) (0.018) Log unemployment Ut -0.102 -0.109 -0.044 -0.106 -0.002 (0.015) (0.017) (0.016) (0.017) (0.017) Log average female wages FW t 0.070 0.112 -0.027 0.086 0.125 (0.024) (0.027) (0.018) (0.026) (0.027) Log average male wages MW t 0.016 0.009 0.065 0.047 -0.049 (0.024) (0.027) (0.026) (0.026) (0.027) Log investment prices Pt 0.001 0.032 -0.011 0.033 -0.012 (0.015) (0.017) (0.016) (0.016) (0.017) Unobserved Heterogeneity ⇡ -1.870 0.896 1.290 -2.389 -2.521 (1.092) (0.124) (1.067) (0.730) (0.826) F-test (p-values) FW t and Ut 27.97 30.52 4.06 27.54 11.23 0.0000 0.0000 0.0172 0.0000 F-test (p-values) FW t, MW t, Ut and Pt 24.30 30.28 5.27 33.96 9.55 0.0000 0.0000 0.0004 0.0000 0.0000 Note: Standard errors in parentheses are obtained through bootstrapping. 86 3.4.3 Estimation of the Technology of Human Capital Formation: Specifications For the estimation of the parameters of the technology of human capital for di↵erent stages in childhood, I use the estimation strategy outlined in Section 3.2.3. In all estimations, standard errors are obtained through bootstrap- ping. Table 3.2 show the estimates of the cognitive and non-cognitive skill CES technologies for children at t aged between 24-47 months and between 44-79 months at t + 1 based in Specification 1, the first two columns present the result for cognitive first without using the control function approach for dealing with the endogeneity of investment meanwhile the second use this method. The same follows for columns 3 and 4. The comparison of the es- timations with and without control function approach shows that the e↵ect of self-productivity decreases for the formation of child’s non-cognitive and instead the cross-productivity increases (column 3 against 4). Table 3.2: Production Function of Cognitive and Non-cognitive skills: One investment input, Age 24-47 mths, Specification 1 Cognitive t+1 Non-Cognitive t+1 without control fn with control fn without control fn with control fn Cognitive Skill t,c 0.605 0.692 0.123 0.339 (0.027) (0.059) (0.070) (0.062) Non-cognitive Skill t,nc 0.038 0.087 0.405 0.182 (0.012) (0.040) (0.023) (0.055) Mother’s Cognitive Skill pc 0.034 0.016 0.019 0.003 (0.043) (0.008) (0.011) (0.005) Mother’s Non-cognitive Skill pnc 0.007 0.005 0.165 0.244 (0.004) (0.030) (0.067) (0.010) Investment i 0.200 0.317 0.231 0.288 (0.028) (0.075) (0.040) (0.034) Complementary parameter 0.106 0.543 0.012 0.788 (0.121) (0.113) (0.054) (0.148) Elasticity of substitution 1/(1 ) 1.119 2.187 1.012 4.715 Note: Standard errors in parentheses are obtained through bootstrapping. For the formation child’s cognitive skills self and cross-productivity increase (column 1 and 2), mother’s cognitive skills decrease in the formation of both child’s skills meanwhile mother’s non-cognitive skills increases in the formation of child’s non-cognitive skills. There is substantial evidence of the e↵ect of 87 parental investment in early childhood development and also support the fact that parental investment is endogenous. 0 .0 5 .1 .1 5 k− d e n si ty −20 −10 0 10 20 Child’s cognitive skills t 0 .0 2 .0 4 .0 6 .0 8 .1 k− d e n si ty −10 −5 0 5 10 Child’s non−cognitive skills t 0 .0 5 .1 .1 5 .2 k− d e n si ty −10 −5 0 5 10 Mother’s cognitive skills 0 .0 2 .0 4 .0 6 .0 8 k− d e n si ty −30 −20 −10 0 10 20 Mother’s non−cognitive skills 0 .0 5 .1 .1 5 .2 k− d e n si ty −10 −5 0 5 Parental Investment 0 .0 5 .1 .1 5 k− d e n si ty −15 −10 −5 0 5 10 Child’s cognitive skills t+1 0 .0 2 .0 4 .0 6 .0 8 .1 k− d e n si ty −15 −10 −5 0 5 10 Child’s non−cognitive skills t+1 Figure 3.1: Kernel densities of latent traits: One investment input, Age 24-47 mths, Specification 1 with control function Comparing now the formation of cognitive and non-cognitive skills in chil- dren under the control function approach, the results provide evidence about the importance of the stock of child’s skills as well as early investment in childhood development. Regarding self-productivity, there is evidence for the formation of both child’s cognitive and non-cognitive skills, but the e↵ect is stronger for cognitive skills, which means that an increase in the stock of skills at t have more prominent e↵ects on the formation of the same skills at t + 1. Mother’s skills have relative low persistence in future cognitive skills, instead of, for future child’s non-cognitive skills the mother’s non-cognitive skills have a more significant e↵ect. The complementary parameter is bigger for future 88 Table 3.4: Production Function of Cognitive and Non-cognitive skills: Multiple investments, Age 24-47 mths, Specification 2, 3 and 4 Cognitive t+1 Non-Cognitive t+1 Inv Money/Time Cog/Non-cog Inv Money/Time Cog/Non-cog Cognitive Skill t,c 0.692 0.664 0.739 0.339 0.269 0.214 (0.059) (0.052) (0.042) (0.062) (0.058) (0.044) Non-cognitive Skill t,nc 0.087 0.075 0.074 0.182 0.108 0.091 (0.040) (0.027) (0.033) (0.055) (0.051) (0.043) Mother’s Cognitive Skill pc 0.016 0.021 0.012 0.003 0.003 0.002 (0.038) (0.010) (0.007) (0.005) (0.005) (0.002) Mother’s Non-cognitive Skill pnc 0.005 0.004 0.003 0.244 0.037 0.000 (0.030) (0.012) (0.015) (0.010) (0.012) (0.018) Investment t,m 0.317 0.288 (0.075) (0.034) Material resources Investment mr 0.123 0.135 (0.032) (0.046) Quality time Investment qt 0.112 0.223 (0.028) (0.031) Cognitive Stimulation Investment cs 0.091 0.097 (0.022) (0.031) Emotional Support Investment es 0.081 0.235 (0.042) (0.098) Complementary parameter 0.543 0.448 0.596 0.788 0.778 0.787 (0.113) (0.135) (0.124) (0.148) (0.236) (0.224) Elasticity of substitution 1/(1 ) 2.187 1.811 2.477 4.715 4.499 4.706 Note: Standard errors in parentheses are obtained through bootstrapping. Specification 3 and Specification 4 incorporate multiple input for measuring investment. Table 3.4 show the estimates of the cognitive and non-cognitive skill CES technologies for children at t with 7-23 months for Specifications 1, 3 and 4 in column 1, 2 and 3 respectively for both future child’s skills. There is still evidence of self-productivity for future child’s skills, though the e↵ect of self-productivity is bigger for non-cognitive skills. There is also evidence of cross-productivity for the future non-cognitive skills. As before, mother’s cognitive skill has low persistence in both child’s future skills and the mother’s non-cognitive skills a↵ect only for fostering the non-cognitive skills, but this e↵ect disappears once that the emotional support investment is incorporated in the estimation. In terms of the e↵ects of separating the investment in material resources and quality time in child skills at age t the results show that material resources is important for determining future child’s cognitive skills and quality time for determining future child’s non-cognitive skills which are the similar result found by Attanasio et al. (2015). 91 Figure 3.2 shows the densities of the latent traits derived from the mea- surement error system and used for the estimation of the cognitive and non- cognitive skill CES technologies for children at t for Specifications 3 and 4. Regarding the e↵ects of separating the investment in cognitive stimulation and emotional support in child skills at age t, the results show that there is not much return regarding cognitive stimulation meanwhile the return of emotional support is higher on future child’s non-cognitive skills. 3.5 Conclusions and further work During the last years, an increasing body of research has focused on the de- velopment of early human capital mostly because gaps in early skills translate into long-term gaps in social and economic inequality. Unfortunately, the hu- man capital formation is a complicated process which is a multi-dimensional and dynamic process, the dimensions of human capital interact both within and across periods, and several unobserved inputs are crucial for estimating correctly the technology of human capital. Some of the key questions are how the skills develop in children’s human capital? What is ”true” technology of skill formation and how does it change? Are the inputs of the technology mal- leable? Do public policies influence child’s outcomes through the inputs of the technology of skill formation? This chapter provides some evidence for answering the previous answers. Exploiting the rich panel structure of the Encuesta Longitudinal de Primera Infancia (Early Childhood Longitudinal Survey (ELPI)) survey I find evidence about the importance of the stock of child’s skills as well as early investment in childhood development. Comparing the formation of cognitive and non- cognitive skills in children dealing or not with endogeneity, there is substantial evidence of the e↵ect of parental investment in early childhood development and also support the fact that parental investment is endogenous. Based on the estimation of the same production function but for di↵erent age stages, the primary result is how parental investment foster cognitive skills between 24-47 months concerning the first and the latest stage instead for future non- cognitive skills the parental investment have the same e↵ect for all the age stages. There is evidence of cross-productivity for both skills which raises 92 for the latest stage. Regarding the e↵ects of separating the investment in material resources and quality time in child skills at age t, the results show that material resources are essential for determining future child’s cognitive skills and quality time for determining future child’s non-cognitive skills. Finally, splitting the investment in cognitive stimulation and emotional support in child skills at age t, the results show that there is not much return regarding cognitive stimulation meanwhile the return of emotional support is higher on future child’s non-cognitive skills. Nevertheless, there are still some limitations in the analysis and further work can be done, for example, using the structural parameters of the tech- nology for simulating policies as well as attempt to characterise the process of human capital accumulation in early years for multiple dimensions (adding health or language skills) to analyse if the return of investment is biased or not. 93