Maternal Mental Health and Child Development

Transcription

1 Maternal Mental Health and Child Development Sonya Krutikova Jere Behrman Ian Bennett Whitney Schott Presentation at the BMGF/HICHD/GCC research symposium 30th June 2015

2 What We Do Examine the relationship between maternal mental health and two dimensions of child development - health and cognitive in a developing country context. Examine multiple stages in childhood (not just post-natal) Utilise wealth and longitudinal design of YL data to examine the sensitivity of the estimates to key identifying assumptions Draw conclusions on the assumptions that we need to make in order to claim causal identication with these observational data

3 What We Do Examine the relationship between maternal mental health and two dimensions of child development - health and cognitive in a developing country context. Examine multiple stages in childhood (not just post-natal) Utilise wealth and longitudinal design of YL data to examine the sensitivity of the estimates to key identifying assumptions Draw conclusions on the assumptions that we need to make in order to claim causal identication with these observational data

4 What We Do Examine the relationship between maternal mental health and two dimensions of child development - health and cognitive in a developing country context. Examine multiple stages in childhood (not just post-natal) Utilise wealth and longitudinal design of YL data to examine the sensitivity of the estimates to key identifying assumptions Draw conclusions on the assumptions that we need to make in order to claim causal identication with these observational data

5 What We Do Examine the relationship between maternal mental health and two dimensions of child development - health and cognitive in a developing country context. Examine multiple stages in childhood (not just post-natal) Utilise wealth and longitudinal design of YL data to examine the sensitivity of the estimates to key identifying assumptions Draw conclusions on the assumptions that we need to make in order to claim causal identication with these observational data

6 Outline Motivation & existing evidence Measuring maternal mental health Analytical Framework Results Extensions

7 Outline Motivation & existing evidence Measuring maternal mental health Analytical Framework Results Extensions

8 Outline Motivation & existing evidence Measuring maternal mental health Analytical Framework Results Extensions

9 Outline Motivation & existing evidence Measuring maternal mental health Analytical Framework Results Extensions

10 Outline Motivation & existing evidence Measuring maternal mental health Analytical Framework Results Extensions

11 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

12 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

13 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

14 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

15 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

16 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

17 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

18 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

19 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

20 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

21 Motivation Evidence that maternal mental health can have signicant eects on children's development (cognitive, nutritional, health, socio-emotional). HIC's: psychosocial, behavioural and emotional domains LMIC's: associations with physical growth and illness also found But: Existing evidence based on developed country samples (/highly selected samples in developing countries) Most evidence based on cross-sectional studies of correlations Focus on PND Big identication challenges Hard to think of an RCT (other than interventions treating maternal depression) Hard to think of instruments How well can we do with one of the best observational data-sets available to answer this question in developing countries?

22 Measuring Maternal Mental Health Use the Self-Reporting Questionnaire (SRQ-20) = list of 20 statements to which women answer yes/no Developed by Harding et al (1980) for WHO as screening tool for developing countries based originally on large samples in Colombia, India, Senegal, and Sudan (70's-80's) Items selected based on existing screening instruments Validated in 21 languages incl. Spanish and contexts incl. Latin America Standard way of using these data = score is number of yes's. Score above 8 = clinically signicant levels of symptoms

23 Measuring Maternal Mental Health Use the Self-Reporting Questionnaire (SRQ-20) = list of 20 statements to which women answer yes/no Developed by Harding et al (1980) for WHO as screening tool for developing countries based originally on large samples in Colombia, India, Senegal, and Sudan (70's-80's) Items selected based on existing screening instruments Validated in 21 languages incl. Spanish and contexts incl. Latin America Standard way of using these data = score is number of yes's. Score above 8 = clinically signicant levels of symptoms

24 Measuring Maternal Mental Health Use the Self-Reporting Questionnaire (SRQ-20) = list of 20 statements to which women answer yes/no Developed by Harding et al (1980) for WHO as screening tool for developing countries based originally on large samples in Colombia, India, Senegal, and Sudan (70's-80's) Items selected based on existing screening instruments Validated in 21 languages incl. Spanish and contexts incl. Latin America Standard way of using these data = score is number of yes's. Score above 8 = clinically signicant levels of symptoms

25 Measuring Maternal Mental Health Use the Self-Reporting Questionnaire (SRQ-20) = list of 20 statements to which women answer yes/no Developed by Harding et al (1980) for WHO as screening tool for developing countries based originally on large samples in Colombia, India, Senegal, and Sudan (70's-80's) Items selected based on existing screening instruments Validated in 21 languages incl. Spanish and contexts incl. Latin America Standard way of using these data = score is number of yes's. Score above 8 = clinically signicant levels of symptoms

26 Strengths & Weaknesses of the Measure Harpham et al 2003 Measuring Mental Health; Health Policy and Planning: Pluses: correlates with other measures, good specicity (healthy also healthy according to diagnostic instruments); 10 min and can be administered by lay interviewers; available in many languages and cut-os determined for many contexts; respondents like Minuses: Measures symptoms and respondent's inclination to report symptoms; 30 day recall can lead to more bias with some respondents; dierential misclassication (false positives found to be more likely among women and less educated)

27 Strengths & Weaknesses of the Measure Harpham et al 2003 Measuring Mental Health; Health Policy and Planning: Pluses: correlates with other measures, good specicity (healthy also healthy according to diagnostic instruments); 10 min and can be administered by lay interviewers; available in many languages and cut-os determined for many contexts; respondents like Minuses: Measures symptoms and respondent's inclination to report symptoms; 30 day recall can lead to more bias with some respondents; dierential misclassication (false positives found to be more likely among women and less educated)

28 Strengths & Weaknesses of the Measure Harpham et al 2003 Measuring Mental Health; Health Policy and Planning: Pluses: correlates with other measures, good specicity (healthy also healthy according to diagnostic instruments); 10 min and can be administered by lay interviewers; available in many languages and cut-os determined for many contexts; respondents like Minuses: Measures symptoms and respondent's inclination to report symptoms; 30 day recall can lead to more bias with some respondents; dierential misclassication (false positives found to be more likely among women and less educated)

29 YL Data - SRQ 20 Summary Stats Measured in Round 1 in all countries and all rounds in Peru Mean SRQ20 Score N Mean sd Mean SRQ20 Score Round ,011 Round ,958 Round ,901 Elevated score (SRQ20>=8) Round ,011 Round ,958 Round ,901 Elevated score (SRQ20>=10) Round ,011 Round ,958 Round ,901

30 YL Data - SRQ-20 Items Round 1 Round 2 Round 3 Highest in all 3 rounds Did you often have headaches? Do you feel nervous, tense or worried? Do you feel unhappy? Are you easily frightened? Lowest in all 3 rounds Has the thought of ending your life been on your mind? Have you lost interest in things? Are you unable to play a useful part in life? Do you nd it dicult to enjoy your daily activities?

31 YL Data - Factor Analysis Round 1 Round 2 Round3 Did you often have headaches Was your appetite poor Did you sleep badly Were you easily frightened Did your hands shake Did you feel nervous, tense and worried Was your digestion poor Did you have trouble thinking clearly Did you feel unhappy Did you cry more than usual Did you nd it dicult to enjoy your daily activities Did you nd it dicult to make decisions Did your daily work suer Were you unable to play a useful part in life Did you lose interest in things Did you feel you were a worthless person Were things so bad that you felt that you just couldn't go on Did you feel tired all the time Were you easily tired Eigenvalues

32 Conceptualisation Use continuous measure for core analysis Think of this more as a measure of stress? (more on that in the next presentation)

33 Conceptualisation Use continuous measure for core analysis Think of this more as a measure of stress? (more on that in the next presentation)

34 Analytical Framework Adopt the production function framework within which: the accumulated stock of human capital (health, cognitive skills) is a function of child, parental and pre-school/school inputs (esp. cognitive production function). Allow production functions for health and cognitive skills to dier: Y ija = Y a (P ij (a), I ij (a), S ij (a), µ ij0, µ j0, ε ija ) (1) Y ija = Outcome of child i from household j at age a P ij (a) = Parental input history as of age a I ij (a) = Individual input history as of age a S ij (a) =School input history as of age a µ ij0 = child specic endowments at birth µ j0 = mother specic genetic endowment at birth ε ija = µ ija + ν ja household and individual level shocks/measurement error

35 Estimation Start with linear approximation Y ija = F ija β 1 + F ija 1 β F ij1 β a + αµ ij0 + γµ j0 + ε ija (2) Challenges: Missing data on inputs (P ij (a), I ij (a), S ij (a)) Endowments µ ij0, µ j0 are hard to observe but are not orthgonal to inputs Shocks to child development ε ija are not orthogonal to inputs Measurement error

36 Estimation Start with linear approximation Y ija = F ija β 1 + F ija 1 β F ij1 β a + αµ ij0 + γµ j0 + ε ija (2) Challenges: Missing data on inputs (P ij (a), I ij (a), S ij (a)) Endowments µ ij0, µ j0 are hard to observe but are not orthgonal to inputs Shocks to child development ε ija are not orthogonal to inputs Measurement error

37 Estimation Start with linear approximation Y ija = F ija β 1 + F ija 1 β F ij1 β a + αµ ij0 + γµ j0 + ε ija (2) Challenges: Missing data on inputs (P ij (a), I ij (a), S ij (a)) Endowments µ ij0, µ j0 are hard to observe but are not orthgonal to inputs Shocks to child development ε ija are not orthogonal to inputs Measurement error

38 Estimation Start with linear approximation Y ija = F ija β 1 + F ija 1 β F ij1 β a + αµ ij0 + γµ j0 + ε ija (2) Challenges: Missing data on inputs (P ij (a), I ij (a), S ij (a)) Endowments µ ij0, µ j0 are hard to observe but are not orthgonal to inputs Shocks to child development ε ija are not orthogonal to inputs Measurement error

39 Our Approach I believe serious attention to two words would sweeten the atmosphere of econometric discourse. These are whimsy and fragility. In order to draw inferences from data as described by econometric tests, it is necessary to make whimsical assumptions. The professional audience consequently and properly withholds belief until an inference is shown to be adequately insensitive to the choice of assumptions...if we are to make eective use of our scarce data resources, it is therefore important that we study fragility in a much more systematic way. Edward Leamer, 1983, American Economic Review: Take the Con Out of Econometrics.

40 Taking a Leaf Out of Leamer's Book Utilise the wealth of Young Lives data & longitudinal dimension to estimate a number of dierent models Keep track of the assumptions for each of these Determine which assumptions estimates are sensitive to... And which assumptions none of these models allow to test Take careful stock of just how whimsy and fragile our estimates are and the assumptions under which we could claim causality with these data

41 Taking a Leaf Out of Leamer's Book Utilise the wealth of Young Lives data & longitudinal dimension to estimate a number of dierent models Keep track of the assumptions for each of these Determine which assumptions estimates are sensitive to... And which assumptions none of these models allow to test Take careful stock of just how whimsy and fragile our estimates are and the assumptions under which we could claim causality with these data

42 Taking a Leaf Out of Leamer's Book Utilise the wealth of Young Lives data & longitudinal dimension to estimate a number of dierent models Keep track of the assumptions for each of these Determine which assumptions estimates are sensitive to... And which assumptions none of these models allow to test Take careful stock of just how whimsy and fragile our estimates are and the assumptions under which we could claim causality with these data

43 Taking a Leaf Out of Leamer's Book Utilise the wealth of Young Lives data & longitudinal dimension to estimate a number of dierent models Keep track of the assumptions for each of these Determine which assumptions estimates are sensitive to... And which assumptions none of these models allow to test Take careful stock of just how whimsy and fragile our estimates are and the assumptions under which we could claim causality with these data

44 Taking a Leaf Out of Leamer's Book Utilise the wealth of Young Lives data & longitudinal dimension to estimate a number of dierent models Keep track of the assumptions for each of these Determine which assumptions estimates are sensitive to... And which assumptions none of these models allow to test Take careful stock of just how whimsy and fragile our estimates are and the assumptions under which we could claim causality with these data

45 The Models We Estimate Contemporaneous: Only current inputs matter / inputs are unchanging over time; observed inputs are orthogonal to endowments/shocks. Y ija = κ 0 + F ija κ 1 + ɛ ija (3) Cumulative: Observables fully capture the past history of inputs; observed inputs are not correlated with endwoments/shocks: Y ija = θ 0 + F ija θ 1 + F ija 1 θ F ij1 θ a + ɛ ija (4) Cumulative + VA: Eect of unobserved inputs and endowments declines geometrically with age at constant rate ψ 2 & contemporaneous unobserved inputs are uncorrelated with observed ones or lagged outcome. Y ija = ϑ 0 +F ija ϑ 1 +F ija 1 ϑ F ij1 ϑ a +ψ 2 Y ija 1 +ɛ ija (5)

46 The Models We Estimate Contemporaneous: Only current inputs matter / inputs are unchanging over time; observed inputs are orthogonal to endowments/shocks. Y ija = κ 0 + F ija κ 1 + ɛ ija (3) Cumulative: Observables fully capture the past history of inputs; observed inputs are not correlated with endwoments/shocks: Y ija = θ 0 + F ija θ 1 + F ija 1 θ F ij1 θ a + ɛ ija (4) Cumulative + VA: Eect of unobserved inputs and endowments declines geometrically with age at constant rate ψ 2 & contemporaneous unobserved inputs are uncorrelated with observed ones or lagged outcome. Y ija = ϑ 0 +F ija ϑ 1 +F ija 1 ϑ F ij1 ϑ a +ψ 2 Y ija 1 +ɛ ija (5)

47 The Models We Estimate Contemporaneous: Only current inputs matter / inputs are unchanging over time; observed inputs are orthogonal to endowments/shocks. Y ija = κ 0 + F ija κ 1 + ɛ ija (3) Cumulative: Observables fully capture the past history of inputs; observed inputs are not correlated with endwoments/shocks: Y ija = θ 0 + F ija θ 1 + F ija 1 θ F ij1 θ a + ɛ ija (4) Cumulative + VA: Eect of unobserved inputs and endowments declines geometrically with age at constant rate ψ 2 & contemporaneous unobserved inputs are uncorrelated with observed ones or lagged outcome. Y ija = ϑ 0 +F ija ϑ 1 +F ija 1 ϑ F ij1 ϑ a +ψ 2 Y ija 1 +ɛ ija (5)

48 A Quick VAM Interlude Growing evidence to suggest that this specication of VAM models (Dynamic OLS) yields reliable estimates: Guarino et al (2014): simulations Studies nding similar results using experimental design vs VAM (DOLS): Andrabi et al (2011): eectiveness of private schools in Pakistan Muralidharan & Sundararaman (2013): eectiveness of contract teachers Deming et al (2014): school eects comparing VAm to school choice lottery eects

49 A Quick VAM Interlude Growing evidence to suggest that this specication of VAM models (Dynamic OLS) yields reliable estimates: Guarino et al (2014): simulations Studies nding similar results using experimental design vs VAM (DOLS): Andrabi et al (2011): eectiveness of private schools in Pakistan Muralidharan & Sundararaman (2013): eectiveness of contract teachers Deming et al (2014): school eects comparing VAm to school choice lottery eects

50 A Quick VAM Interlude Growing evidence to suggest that this specication of VAM models (Dynamic OLS) yields reliable estimates: Guarino et al (2014): simulations Studies nding similar results using experimental design vs VAM (DOLS): Andrabi et al (2011): eectiveness of private schools in Pakistan Muralidharan & Sundararaman (2013): eectiveness of contract teachers Deming et al (2014): school eects comparing VAm to school choice lottery eects

51 A Quick VAM Interlude Growing evidence to suggest that this specication of VAM models (Dynamic OLS) yields reliable estimates: Guarino et al (2014): simulations Studies nding similar results using experimental design vs VAM (DOLS): Andrabi et al (2011): eectiveness of private schools in Pakistan Muralidharan & Sundararaman (2013): eectiveness of contract teachers Deming et al (2014): school eects comparing VAm to school choice lottery eects

52 A Quick VAM Interlude Growing evidence to suggest that this specication of VAM models (Dynamic OLS) yields reliable estimates: Guarino et al (2014): simulations Studies nding similar results using experimental design vs VAM (DOLS): Andrabi et al (2011): eectiveness of private schools in Pakistan Muralidharan & Sundararaman (2013): eectiveness of contract teachers Deming et al (2014): school eects comparing VAm to school choice lottery eects

53 And Some More Models Within Child Within Family Help as long as we can assume that endowment and shock eects does not vary with age or time elapsed Within Child assumptions about persistence of lagged outcome eects especially problematic for skills

54 And Some More Models Within Child Within Family Help as long as we can assume that endowment and shock eects does not vary with age or time elapsed Within Child assumptions about persistence of lagged outcome eects especially problematic for skills

55 And Some More Models Within Child Within Family Help as long as we can assume that endowment and shock eects does not vary with age or time elapsed Within Child assumptions about persistence of lagged outcome eects especially problematic for skills

56 And Some More Models Within Child Within Family Help as long as we can assume that endowment and shock eects does not vary with age or time elapsed Within Child assumptions about persistence of lagged outcome eects especially problematic for skills

57 Assumptions We Can Test Only current inputs matter = A1 Observed history of inputs captures full history = A2 Inputs are uncorrelated with individual endowments and shocks=a3a Inputs are uncorrelated with parental endowments and shocks = A3b Test A1 A2 A3a A3b κ aols,ha = θ aols,ha? θ aols,ha = ϑ aols,ha? θ aols,ha = θ aife,ha? θ aols,ha = θ asfe,ha?

58 Assumptions We Can Test Only current inputs matter = A1 Observed history of inputs captures full history = A2 Inputs are uncorrelated with individual endowments and shocks=a3a Inputs are uncorrelated with parental endowments and shocks = A3b Test A1 A2 A3a A3b κ aols,ha = θ aols,ha? θ aols,ha = ϑ aols,ha? θ aols,ha = θ aife,ha? θ aols,ha = θ asfe,ha?

59 Assumptions We Can Test Only current inputs matter = A1 Observed history of inputs captures full history = A2 Inputs are uncorrelated with individual endowments and shocks=a3a Inputs are uncorrelated with parental endowments and shocks = A3b Test A1 A2 A3a A3b κ aols,ha = θ aols,ha? θ aols,ha = ϑ aols,ha? θ aols,ha = θ aife,ha? θ aols,ha = θ asfe,ha?

60 Assumptions We Can Test Only current inputs matter = A1 Observed history of inputs captures full history = A2 Inputs are uncorrelated with individual endowments and shocks=a3a Inputs are uncorrelated with parental endowments and shocks = A3b Test A1 A2 A3a A3b κ aols,ha = θ aols,ha? θ aols,ha = ϑ aols,ha? θ aols,ha = θ aife,ha? θ aols,ha = θ asfe,ha?

61 Health and Cognitive Skill Production Functions Level Child endowments at birth Child characteristics Shocks Parental inputs - time Parental inputs -quality of time Parental inputs - materials Child time-use Cognitive skill production function birth size, pre-natal care, conditions at birth, birth-spacing (age gap between index child and next oldest child), mother's age Age, sex, ethnicity, birthorder, health, nutrition At least one negative shock in the last x years Parents live in the household, child-care in infancy parental education, parenting style (age 1), maternal self-esteem and self-ecacy, household size wealth index, vaccinations, educational materials at home time-allocation (ages 5 & 8), pre-school/school Health production function birth size, pre-natal care, conditions at birth, birth-spacing (age gap between index child and next oldest child), mother's age Age, sex, ethnicity, birthorder, health At least one negative shock in the last x years Parents live in the household, child-care in infancy parental education, household size wealth index, vaccinations, dietary diversity

62 Descriptive Stats: Endowments Mean Std Dev Endowments at birth Diculties during pregnancy Dicult labour Number of ante-natal visits< Four weeks plus premature Two or more tetanus injections (antenatal) Gave birth at home Child hospitalised at birth Birth size (maternal report: 1-5) Child received measles vaccine

63 Maternal Mental Health Estimates:Health maternal mental health score (age 1) maternal mental health score (age 5) maternal mental health score (age 8) ZHFA (age 1) ZHFA (age 5) ZHFA (age 5) ZHFA (age 5) ZHFA (age 8) ZHFA (age 8) ZHFA (age 8) * (0.021) (0.022) (0.023) (0.026) (0.024) * (0.022) (0.024) (0.017) (0.027) (0.014) (0.025) (0.024) (0.016) contemporaneous controls x* x x cumulative controls x x cumulative controls + lagged x x outcome sentinel site xed eects x x x x x x x R-squared

64 Other Inputs (cumulative specication) Health (haz): Less than 4 antenatal visits; born at home, birth size, mother age >25 (ages 5-8), parental education, relative health, wealth index Cognitive (PPVT): born at home, received measles vaccine, parental education, haz, wealth, chores instead of play time - (age 8), pre-school/school +

65 Other Inputs (cumulative specication) Health (haz): Less than 4 antenatal visits; born at home, birth size, mother age >25 (ages 5-8), parental education, relative health, wealth index Cognitive (PPVT): born at home, received measles vaccine, parental education, haz, wealth, chores instead of play time - (age 8), pre-school/school +

66 Maternal Mental Health Estimates: Cognitive Skills - Verbal maternal mental health score (age 1) maternal mental health score (age 5) maternal mental health score (age 8) PPVT (age 5) PPVT (age 5) PPVT (age 8) PPVT (age 8) PPVT (age 8) * (0.017) (0.025) (0.023) * ** (0.021) (0.022) (0.020) (0.021) *** ** *** (0.019) (0.024) (0.023) contemporaneous controls (with endowments) x x cumulative controls (with endowments) x x cumulative controls + lagged x outcome sentinel site xed eects x x x x x R-squared

67 Estimates: Cognitive Skills - Quantitative maternal mental health score (age 1) maternal mental health score (age 5) maternal mental health score (age 8) CDA (age 5) CDA (age 5) Maths (age 8) Maths (age 8) Maths (age 8) (0.27) (0.025) (0.024) ** (0.028) (0.030) (0.022) (0.022) (0.018) (0.021) (0.020) contemporaneous controls (with endowments) x x cumulative controls (with endowments) x x cumulative controls + lagged x outcome (cda) sentinel site xed eects x x x x x

68 Other Inputs (cumulative specication) Health (haz): Less than 4 antenatal visits; born at home, birth size, mother age >25 (ages 5-8), parental education, relative health, wealth index Cognitive (PPVT): born at home, received measles vaccine, parental education, haz, wealth, chores instead of play time - (age 8), pre-school/school +

69 Other Inputs (cumulative specication) Health (haz): Less than 4 antenatal visits; born at home, birth size, mother age >25 (ages 5-8), parental education, relative health, wealth index Cognitive (PPVT): born at home, received measles vaccine, parental education, haz, wealth, chores instead of play time - (age 8), pre-school/school +

70 Extensions Psychosocial dimension Within child and family xed eects Linearity of eects - elevated levels measure Linearity of production function - implies that inputs are perfect substitutes for eachother. Strongly rejected in a number of contexts incl YL India (Attanasio et al, 2015) and Colombia (Attanasio et al, 2015b). Accounting for complementarities between inputs is highly important CES production function. Measurement - is there a more ecient way to control for dierent groups of inputs?

71 Extensions Psychosocial dimension Within child and family xed eects Linearity of eects - elevated levels measure Linearity of production function - implies that inputs are perfect substitutes for eachother. Strongly rejected in a number of contexts incl YL India (Attanasio et al, 2015) and Colombia (Attanasio et al, 2015b). Accounting for complementarities between inputs is highly important CES production function. Measurement - is there a more ecient way to control for dierent groups of inputs?

72 Extensions Psychosocial dimension Within child and family xed eects Linearity of eects - elevated levels measure Linearity of production function - implies that inputs are perfect substitutes for eachother. Strongly rejected in a number of contexts incl YL India (Attanasio et al, 2015) and Colombia (Attanasio et al, 2015b). Accounting for complementarities between inputs is highly important CES production function. Measurement - is there a more ecient way to control for dierent groups of inputs?

73 Extensions Psychosocial dimension Within child and family xed eects Linearity of eects - elevated levels measure Linearity of production function - implies that inputs are perfect substitutes for eachother. Strongly rejected in a number of contexts incl YL India (Attanasio et al, 2015) and Colombia (Attanasio et al, 2015b). Accounting for complementarities between inputs is highly important CES production function. Measurement - is there a more ecient way to control for dierent groups of inputs?

74 Extensions Psychosocial dimension Within child and family xed eects Linearity of eects - elevated levels measure Linearity of production function - implies that inputs are perfect substitutes for eachother. Strongly rejected in a number of contexts incl YL India (Attanasio et al, 2015) and Colombia (Attanasio et al, 2015b). Accounting for complementarities between inputs is highly important CES production function. Measurement - is there a more ecient way to control for dierent groups of inputs?