BACKGROUND

Missing data is a major concern in ALSPAC and other longitudinal studies. Many analyses assume that data are missing at random (MAR), but this assumption is hard to verify from data and is often not very plausible.

ALSPAC has sources of data that offer the opportunity to learn more about the missing data mechanism, and in particular about whether MAR is true:

1. Repeated attempts to contact participants at each wave.

2. External sources of data from schools and SATs.

Repeated attempts have previously been used to learn about missing data mechanisms , , but not in longitudinal data sets.

OBJECTIVES

1. To explore the missing data mechanism for key variables on drugs, alcohol and conduct disorder.

2. To develop appropriate statistical methods, in particular for handling repeated attempts data at multiple waves.

DATA

Main outcomes: drug use, alcohol use and conduct disorder derived from clinic visits at ages 11, 13, 15 and 17, and questionnaires at intermediate ages.

Related outcomes: schools data at years 6 and 11; SATs data at ages 11 and 14.

Auxiliary variables: the number of letters sent before each clinic visit happened (invitation, reminder, last chance) and before each questionnaire was returned (initial, reminder). Ideally we would like to use further information on whether "defaulters" were phoned and dates of letters, but we understand this is not currently available.

Covariates including sex, socio-economic status, ethnicity.

PLAN OF ANALYSIS

1. Descriptive statistics, including associations of the main outcomes with the related outcomes and the auxiliary variables.

2. Using related outcomes: assuming the main outcomes are MAR when the related outcomes are included in the model, explore the missing data mechanism when the related outcomes are not in the model. This could be done using multiple imputation.

3. Using repeated attempts: develop and fit models for repeated attempts at multiple waves. Hence explore developmental trajectories (perhaps using latent class models).

Paul Clarke and Michael Spratt are doing a related project focussing on repeated attempts at a single time point, and we will work with them to ensure that the two projects complement each other. We will work with Matt Hickman (drugs and alcohol strand) and Glyn Lewis ("risky behaviours" strand) to ensure that this project fits well with their strands.

PROPOSED OUTPUTS

1. A statistical paper describing the models for repeated attempts at multiple waves.

2. An epidemiological paper describing what has been learned about missing data in ALSPAC

This project examines the role that hyperbolic discounting plays as a determinant of unhealthy

birth outcomes. If pregnant women discount hyperbolically, a reduction in personal delivery costs (via public insurance) will exacerbate the disparity between each woman's short- and long-run rates of time preference, leading to greater engagement in unhealthy behaviors with long-run impacts. This disparity in rates of time preference will also make it more costly for women to maintain traditional commitment mechanisms designed to keep them on their optimal path to achieving a healthy birth outcome.

THEORETICAL FRAMEWORK

Assume there exists a time t* (less than T) during pregnancy at which a woman is indifferent between two

choices with different rewards: (a) engaging frequently in risky behavior(s) and experiencing a premature and/or low birth weight (LBW) baby at time T (denoted by C) and (b) engaging minimally in risky behavior(s) and not experiencing a premature and/or LBW baby at time T0 (denoted by C0), where T less than T0. We assume that C0 has the greater payoff and that p(t) (p0(t)) is the probability density of a birth before T (T0). Furthermore, we assume that C0 has both a non-zero probability of a premature and a late birth, and that T and T0 are the anticipated birth dates.

When there is greater time to birth, I hypothesize the woman exhibits patience to obtain the reward of a healthy birth outcome (C0). However, as the time to birth grows shorter and shorter, she becomes increasingly impatient and seeks instant gratification via any number of risky behaviors. My research question is the following: To what extent, if at all, does hyperbolic discounting occur in pregnancy?

I assume that p(v; t) is the probability density that C is realized at time v greater than t such that p(t; t) greater than p0(t; t) for t less than T with the assumption that if the density of premature birth at time v drops off monotonically as we move away from the anticipated realization time (T or T0), then the inequality seems reasonable as the density should have dropped off less at t for C than for C0. Since progression of pregnancy up to 39 weeks reduces the expected payoff from C (due to the reduced probability of having a premature birth), it follows that C p(t; t) less than C0 p0(t; t) for t less than T. Given these previous assumptions, we can then show that up to a certain point t* (say, the beginning of the third trimester of pregnancy), a woman might have engaged in protective behaviors such as quitting smoking and drinking alcohol upon learning she was pregnant only to return to those behaviors later in the pregnancy (sometime after t* ) given their addictive nature

(see Dasgupta and Maskin [1] for the proof of this).

EMPIRICAL FRAMEWORK

The dependent variables will include number of times per day smoked, number of glasses of wine / beer

drank per day, illicit drug use, folic acid use, coffee consumption, amount of exercising, desire to lose weight, and mental status (i.e., depression, anxiety, etc.). Independent variables will include Knowledge from medical professional of the negatives and positives of certain health behaviors, External Environmental variables (previous abortion, spouse smokes, work exposure to smoke, dieting/desire to lose weight, etc.), Previous Health issues (mother was low-birth weight herself and/or mother smoked during her pregnancy, family illnesses), and Demographics (race, age, income, education, etc.). I use the trimester time (i.e., given by 8 weeks, 18 weeks, and 32 weeks) and mother's pre-pregnancy BMI as stratification variables. Thus one empirical specification would be the following:

smoke_ij = B_1 + B_2 * K + B_3 * EE + B_4 * PH + B_5 * D + e_i,

where i = 1, 2, or 3 is the trimester indicator and j = 1, 2, 3, or 4 is the BMI category. The hypothesis of interest is that knowledge of the effects of smoking will have less of an impact on a woman's actual behavior as she grows impatient by the third trimester, her external environment will inuence her to smoke more, and previous health issues will not be as significant in avoiding smoking behavior by the third trimester. This specification would extend analogously to the other behaviors mentioned.

Acne vulgaris is a common inflammatory disease of the pilosebaceous glands characterized by inflamed (papules, pustules, nodules and cysts) and noninflamed lesions (comedomes) on the face, neck, chest and back. The prevalence of acne in adolescents has been reported as being between 35% and 90%, depending on the method of classification, with peak incidence occurring at between 14 and 17 years in females, and 16 and 19 years in males (Stathakis et al, 1997; Kilkenny et al. 1998). As well as the physical pain and discomfort that acne produces, it is well documented that acne sufferers experience higher levels of psychological distress, are more self-conscious and have a poorer self-image than age-matched controls. A possible role for genetic factors in the development of acne is based on the observation that relatives of affected individuals are at increased risk of developing acne compared with unrelated individuals. However, related individuals share similar environments as well as genes. By using the classical twin design which compares the similarity of monozygotic (MZ) twin pairs with that of dizygotic (DZ) twin pairs I was able to show previously that acne is a highly heritable trait (Evans et al. 2005).

No group has performed a genome-wide association study of acne to date. This study will combine the power of the ALSPAC, QIMR and TwinsUK cohorts to detect common variants underlying risk of acne/acne severity in a meta-analysis of genome-wide association studies. Variants which are shown to be associated with acne will be followed up using single SNP genotyping in the remainder of the ALSPAC cohort for whom genome-wide data is not available.

Objectives

1. To develop and apply latent class and other statistical models that combine symptoms and physiological measurements to define and refine asthma phenotypes in children in the Avon Longitudinal Study of Parents & Children (ALSPAC) study.

2. To develop and apply methods to validate and replicate phenotypes in other large, population-based cohort studies and to compare phenotype definitions and their associations with asthma risk factors.

3. To examine associations between genetic variants identified through genomewide association studies of asthma with asthma phenotypes, in order to understand the biological pathways leading to different phenotypes.

Summary of proposal

Several studies have demonstrated the utility of the latent class approach in identifying phenotypes that have clear associations with clinically relevant outcomes [1-4]. Different approaches to phenotype definition appear to elucidate different aspects of the relationship between symptoms and asthma-related markers. There remains a need to develop these approaches to further investigate the validity and cross-cohort replicability of proposed phenotypes, and their suitability as proxies for underlying pathophysiological constructs of asthma.

The ALSPAC cohort (13,000 children born 1991-92) includes longitudinal reporting of respiratory symptoms (cough, wheeze, breathlessness, trigger factors) collected prospectively from birth to 17 years and direct measurements of lung function, bronchial responsiveness and allergy at different ages. In my previous work on the ALSPAC study I used reported wheezing at 7 time points to define wheezing phenotypes, in a "single dimensional" approach [1]. To refine and improve our understanding of asthma phenotypes, I will investigate "multi-dimensional" approaches in which different types of phenotypic information (for example, frequency and duration of respiratory symptoms, atopy and bronchial responsiveness) are included in latent class models. A multidimensional approach that uses a full range of symptoms, measures of allergy and measures of lung function to define phenotypes may better cover the distinctive pathophysiological features that represent the complexity and heterogeneity of asthma. Latent class models will be fitted using Mplus software (with which I am already familiar). Pre-selection of the most relevant variables using Principal Components Analysis may be required, to limit the number of parameters in the model.

The next step will be validation and replication of the phenotypes identified in analyses of ALSPAC data. Three population-based childhood cohorts have agreed to contribute data to these analyses: The Stockholm Children Allergy and Environmental Prospective Birth Cohort Study (BAMSE) [5] (4,089 children, born 1994); The prevention and incidence of asthma and mite allergy cohort (PIAMA) [6] based in Groningen (3291 children, born 1996-97) and The Western Australian Pregnancy Cohort (Raine) Study [7] (2868 children, born 1989-91). All four cohorts, including ALSPAC, have respiratory symptoms from repeat questionnaires and lung function measurements, bronchial challenge and allergy tests at different intervals through childhood.

In order to validate and replicate the derived asthma phenotypes in other cohorts, I will have to take into account differences in the nature, number and timing of measures among the cohorts. I will use cubic splines (nonlinear functions defined by smoothly joined piecewise polynomials) [8] to derive parametric curves representing the ALSPAC-derived asthma phenotypes. These curves will be used to interpolate model parameters appropriate to the number and timing of measurements in each dataset.

The accuracy and generalisability of the models will be validated across the independent cohorts. To investigate phenotypic variation among cohorts I will measure the loss of fit by using deviance (-2 x log-likelihood) differences. Such an approach has been used previously to derive generalisable prognostic models for the prognosis of HIV-infected patients starting antiretroviral therapy [9]. The performance of candidate latent class models in single cohorts will be measured by comparing the fit using the parameters estimated from the full data set with the fit using parameters re-estimated from single cohort data alone. The best-generalising model is defined as that for which the loss of fit in single cohort data is smallest.

ALSPAC, BAMSE, PIAMA and the cross-sectional study MAGICS/ISAAC (Hannover, 1,400 children) will have comparable genomewide data available through participation in the GABRIEL Study [10]. GABRIEL is a genomewide association study involving 14 European countries. Common genetic variants that are associated with childhood asthma have been identified, and more are likely to be reported over the next 12 to 24 months. Replication among these cohorts will enable me to investigate associations of candidate-gene variants in known functional pathways with different asthma phenotypes, and hence to determine if there is genetic evidence for biological differences between derived phenotypes. Between-phenotype variation in associations with genetic variants will also provide objective evidence for the validity of derived phenotypes.

AIMS

Overall aim

To investigate the association of suicide and suicide attempt in parents with suicidal behaviour, mental health, social and academic functioning in offspring.

Specific objectives

1. To investigate whether:

a) Exposure to parental self harm (fatal or non-fatal) confers an independent risk for self harm, suicidal thoughts, suicidal plans, suicidal attempts, psychiatric symptoms, diminished social and academic functioning beyond that observed in offspring exposed to parental mental or physically disabling illness and offspring of parents who have no history of self harm or mental and/or physically disabling illness.

b) Exposure to parental death by suicide confers an independent risk for self harm, suicidal thoughts, suicidal plans, suicidal attempts, psychiatric symptoms, diminished social and academic functioning beyond that of exposure to parental death by a cause other than suicide.

2. To examine whether outcomes in offspring i.e. self harm, suicidal thoughts, suicidal plans, suicidal attempts, psychiatric symptoms, social and academic indices differ depending on the gender of the self harming parent and/or gender of offspring.

3. To examine whether age of exposure to self harm of a parent modifies the relationship between exposure to self harm and self harm, suicidal thoughts, suicidal plans, suicidal attempts, psychiatric symptoms, social and academic functioning in offspring.

4. To assess whether all these exposure factors (gender of proband and offspring, age of exposure, type of self harm) are related to the age of first onset of self harm, suicidal thoughts, suicidal plans, suicidal attempts, behaviour or mental health problem in the offspring.

5. To examine whether exposure to different forms of self-harm in parents (i.e. suicide versus attempted suicide) is associated with specific self-harming behaviours (i.e. suicidal thoughts, suicidal plans, suicidal attempts, self harm with no intention to die), psychiatric symptoms, social and academic parameters in offspring.

There has been speculation that an atopic predisposition might increase susceptibility to cancer, although convincing evidence has been lacking. An exception is prostate cancer, for which a positive association with atopy1, and especially dust mite sensitisation2, has been reported, but not explained. A strong predictor of prostate cancer risk is raised levels of IGF-13. The observation that Th2 cytokines promote the expression of IGF-14 is in keeping with the hypothesis that atopy may increase risk of prostate cancer. However, it is also possible that IGF-1 may drive atopy. Laboratory studies have suggested that IGF-1 promotes Th2 responses and raised IgE5;6. Furthermore, IGF-1 is a primary mediator of the effects of growth hormone, and cord blood IGF-1 levels are positively associated with size at birth7. This may have relevance to the, albeit inconsistent, evidence linking enhanced fetal growth with atopy8;9. Interestingly, IGF-1 levels fall with increasing parity10, raising the possibility that IGF-1 might partly explain the relation between parity (birth order) and atopy in the offspring. However, epidemiological studies on the relation between IGF-1 and atopy are lacking.

Childhood Obesity and Severe Antenatal Maternal Stress

Obesity is a complex public health problem that is a growing threat to children's health, as well as a current and future drain on National Health Service (NHS) resources. Obesity already costs the NHS directly around £1 billion a year and the UK economy a further £2.3 to £2.6 billion in indirect costs. If this present trend continues, by 2010 the annual cost to the economy could be £3.6 billion a year.

By the IOTF's international standards nearly 27.5% of girls and 22% of boys in UK aged 2-15 were found to be overweight, including 5.5% of boys and 7.2% of girls who were obese in 2002. The IOTF analysis indicates a marked acceleration in the trend from the mid-1980s onwards. Using the national Body Mass Index percentiles approach (adopted by the Department of Health assuming 15% overweight including 5% obesity in 1990) 30.7% of girls and 30.3 % of boys were overweight, including 16% who were obese, by 2002.(See DH weblink)Furthermore, childhood obesity is one of the key predictors of adult obesity. Evidence of the long term effectiveness of obesity interventions is mixed (see Mend website; ) once obesity is established (see Oude Luttikhuis et al 2009). Identifying early risk factors for child obesity may have important preventive implications for future public health strategy.

In humans, adverse early life experiences such as childhood trauma, neglect and abuse are associated with the development of psychosocial disorders (Heim & Nemeroff, 1999; Kaufman et al., 2000; Heim et al., 2002; MacDonald et al., 2008). In animal models, an adverse early life experience causes abnormalities in neurodevelopment (Pickering et al., 2006), increased stress response as measured by plasma corticosterone (Meaney et al., 1993; Pihoker et al., 1993), and behavioural deficits (Pryce and Feldon, 2003; Pryce et al., 2005; Slotten et al., 2006) in adulthood.

Numerous studies report relationships between palatable food (rich in sugar and saturated fat), and emotional states and stress. In humans and in animal models, emotions influence food intake and choice of food and palatable foods high in fat and carbohydrate improve mood and modify stress responsivity. For example, subjects report a preference for palatable food rich in saturated fat and sugar during negative emotions (Macht, 2008) and adult individuals subjected to psychological stress in a laboratory setting are reported to eat greater amounts of palatable food compared to control groups (Oliver et al., 2000). In adults, depression and anxiety can also be linked to compulsive behaviours and craving for palatable food which induce feelings of pleasure (Kelley et al., 2005; Paterson and Markou, 2007). A study in UK school children reported that higher self-reported stress scores were associated with increased intake of fatty food compared to fruits and vegetables (Cartwright et al., 2003). These data suggest that the direction of the association is that humans subjected to stressful circumstances (chronically and acutely) are more likely to crave palatable, high saturated fat and high sugar containing foods or 'comfort foods'.

Dallman (2005) described the phenomenon of 'comfort food' or 'comfort eating' as self-medication with food.Adam and Epel, 2007 propose that stress-stimulated consumption of palatable food is a form of reward-based eating because palatable high-fat diet or 'comfort food' modulates hypothalamic-pituitary-adrenal (HPA) axis activity and indirectly blunts the stress response. In other words, the craving for such foods derives from their effect on the HPA axis which the presence of the stress has activated acutely or chronically. Animal data support the notion that consuming such a diet can have a beneficial effect by modulating the stress response: rats consuming molar sucrose solution and sweetened vegetable shortening (lard), in addition to regular chow, showed attenuated ACTH responses to restraint stress and lower plasma corticosteronecompared to those consuming chow alone (Pecoraro et al., 2004;la Fleur 2005). Similarly, rats exposed to lard for 2 h per day for 7 weeks also showed a decreased stress response compared to rats consuming chow only (Kinzig et al., 2008). Alterations in hippocampal GR mRNA and hypothalamic CRH mRNA, key mediators of HPA axis activity, are thought to be the mechanisms by which these diets alter the stress response (e.g. Morris 2010). Thus, data suggest that palatable high fat diets or comfort food have a role in ameliorating stress and possibly stress-related behaviour whether given chronically or as an option.

These studies focus on postnatal events in humans and animal models. It is well recognised that prenatal maternal stress alters HPA axis sensitivity in rodent models (e.g. Kapoor et al 2008; Liu et al 1997). Our work and others has provided evidence that in utero exposure to severe maternal stress can increase the risk of neuropsychiatric disorder i.e. schizophrenia and affective disorder (Khashan et al 2008; Khashan et al submitted) both of which are considered in part to result from altered HPA axis sensitivity.

Research Question

Does pregnancy-related and early postnatal exposure to domestic violence predict increased BMI in offspring of exposed mothers in the ALSPAC cohort?

Hypotheses

Primary hypothesis: a child exposed to severe early life stressors is more likely to become overweight or obese than children not exposed in this way manifest as a rightwards shift in the BMI (and or DXA derived fat mass) distribution by age 15 of exposed versus unexposed children.

Secondary hypotheses: i) as a result of exposure to maternal stress in utero, HPA axis sensitivity in exposed children is greater than in unexposed children (tested using salivary cortisol measures) ii) overweight or obesity occurs through increased intake of palatable or comfort foods in exposed versus unexposed children iii) intake of a high saturated fat and high carbohydrate diet reduces or modulate HPA axis sensitivity iv) overweight or childhood obesity is associated with reduced HPA axis sensitivity (mean salivary cortisol levels) in exposed children compared to exposed non-obese or normal weight children

Future hypotheses i) We will examine whether we may use birth weight (BW) of children exposed and unexposed as a proxy for early life stress. We would then examine BW and its association with later life BMI through the use of genotypes associated with it via the presence of key genetic loci associated with birth weight(Freathy et al 2010).

Method

We wish to use a model of early life stress indexed by the maternal experience of adverse life events and of domestic violence before conception, during pregnancy or in the first year of life, in order to assess effects on childhood growth and rates of overweight and obesity. We shall examine the distribution of the BMI and body composition (DXA) of children exposed in utero and postnatally to maternal domestic abuse and compare it to that of unexposed children and assess HPA axis sensitivity using salivary cortisol levels in obese/overweight and normal weight exposed children. Variables assigning abuse status to individual participants have already been derived (see below) and we seek permission to examine these in light of anthropometric, serological and genetic data.

All analysis will be undertaken collaboratively by Nic Timpson at the MRC CAiTE centre and as such we have not requested data transfer.

Exposure variables: Antenatal domestic violence has been assessed in the ALSPAC cohort at 18 weeks gestation. Participants were asked two questions on whether their partner had been emotionally cruel, and whether their partner had physically hurt them since the start of the pregnancy and to state how much it affected them on a scale from 'no effect' to 'affected a lot'. A woman will be considered to have experienced antenatal domestic violence if she responded positively (any level of effect) to either physical or emotional cruelty since the start of the pregnancy. Questions asking if their partner had been emotional or physically cruel have also been used to define postnatal domestic violence and this was reported at 2, 8, 21 and 33 months after the child was born, in each case referring to the period since the last questionnaire. Violence from the partner to the mother, partner to the child and mother to the child was reported by the mother in the same way at these times. The repeated responses will be summarised into a variable indicating the number of time points that violence was experienced since the birth, a separate variable for each of: violence occurring to the mother, to the child from partner or child from the mother. Other stressful life events will be assessed using the life events scale covering 13 life events including serious illness/death in the family, marital difficulties, financial problems etc. These were recorded by the mother at 8, 21 and 33 months after the birth.

Outcome variables:

Child Biometric Data: Information collected on birthweight and gestational age will be used along with the height, weight, ethnicity of the mother and parity of the child, to determine the birthweight centile of the child using the Gestation Network calculator; children in the lowest 10% will be counted as small for gestational age (SGA).BMI recorded atages21 months, 4, 7, 9, 11 and13 years of age usingheight and weight measurements in GP reports; as well asDXA fat mass at the same ages.

HPA axis sensitivity: we shall use salivary cortisol measures collected at ages xxx and compare mean and SD between exposed and unexposed children. (need to check availability).

Dietary records at age 10 yrs will be used to account for dietary/calorie intake.

Potential confounders: The following maternal characteristics will be measured as potential confounders of any association between stressors and child biometric measures: maternal age at the birth of the child, maternal socioeconomic status as assessed by the following variables: maternal education, (recorded at 32 weeks gestation simplified from a 5 item response into 3 categories: low (CSE, vocational), medium (O-level), high (A-level, degree)) and homeownership status, grouped as: own/mortgage versus rented (Cleland et al 2010; Crawford et al 2010). Smoking and alcohol use during pregnancy were determined at 18 weeks gestation. Women are considered to have smoked during pregnancy if they had smoked any form of tobacco in the 2 weeks prior to completion of the 18 week questionnaire. Women were also questioned on their alcohol use during pregnancy at 18 weeks gestation, responses were categorised into less than 1 glass (unit) per day and 1 or more glasses (units) per day.

Analysis

Simple analyses will be used to assess the association between severe early life stress exposure, measured at 18 weeks gestation and up to 33 weeks postnatally, and the biometric outcome variables.Multivariable analysis will be conducted using logistic regression to adjust for the following possible confounding factors: SES (maternal education, homeownership status), maternal marital status, maternal age, parity, ethnicity and paternal depression. Biometric variables will be handled as continuous variables initially to look for shifts in the distribution of weight in the exposed and unexposed groups by age. Bivariable associations between overweight, antenatal domestic violence and baseline characteristics of the mother and child will be assessed using chi-squared or t-tests. The association between antenatal domestic violence and child overweight will be further investigated using logistic regression. Generally, the data will be analysed in three stages: (i) calculation of unadjusted odds ratio of severe early life stressor (domestic violence and life events score) in relation to overweight; (ii) adjustment for possible confounding factors; (iii) addition of potential mediators to the model. To investigate the possible mediating effect of maternal depression on the association, mother with recorded depression (EPDS score greater than 13) will be excluded in a sensitivity analysis. A variable for domestic violence exposure i.e. no violence, antenatal violence only, antenatal and postnatal violence, postnatal violence only, will also be used to give a direct comparison of the different levels of exposure. If numbers are sufficient, gender interactions will also be explored.

A second set of analyses using a Chi square tests will assess the secondary hypotheses: the relationship between exposure and mean salivary cortisol levels. The divide the cases into exposed overweight/obese and exposed normal weight and

Multivariable analysis will be conducted using logistic regression to adjust for the following possible confounding factors: SES (maternal education, homeownership status), maternal marital status, maternal age, parity, smoking, ethnicity and paternal depression. Bivariable associations between salivary cortisol levels, antenatal domestic violence and baseline characteristics of the mother and child will be assessed using chi-squared or t-tests. The association between antenatal domestic violence and childhood salivary cortisol levels will be further investigated using logistic regression. Finally, dietary intake will be assessed as high or low in palatable food and/or calorie and a chi square test used to examine the association between maternal stress/domestic violence and dietary intake. Multivariable analysis will be conducted using logistic regression to adjust for the following possible confounding factors: SES (maternal education, homeownership status), maternal marital status, maternal age, smoking, parity, ethnicity and paternal depression.

Missing Data: We shall use the (MICE) with the ice command in Stata to impute missing values using chained equations. This will be carried out on the full dataset including all variables used in this analysis as well as additional variables known to be related to non-response.

The goal of this project is to take a behavioral genetics approach to skill in emotion recognition accuracy. The ALSPAC team provided its participants with Dr. Stephen Nowicki's emotion recognition test called the Diagnostic Analysis of Nonverbal Accuracy (DANVA), in which respondents judged the emotions contained in facial expressions and vocal tones. Children completed this diagnostic during the Focus@8 session, and Main Carers completed this diagnostic in the Teen Focus 1 or Teen Focus 1 FastTk session. I propose to conduct two analyses of these existing data:

1. Convergence among twins

This analysis would compare the correlation of scores among Monozygotic twins (identical twins who share 100% of the same DNA) versus Dizygotic twins (fraternal twins who share 50% of their DNA). Greater convergence for MZs than DZs would suggest that there is a biological component to the development of emotional skill.

For this analysis, we would need to be able to identify who are the 195 pairs of twins (this number of twins was reported by the ALSPAC team, e.g., http://www.bristol.ac.uk/alspac/sci-com/resource/recruit/). For each of these 195 pairs of twins, we would also need to be able to identify whether they are identical/MZ or fraternal/DZ. (This determination of MZ vs. DZ could be made using genetics data if it has not already been done for other purposes.)

2. Convergence between main carers and children

This analysis would establish the correlation of scores between children and their main carers. It would be valuable to identify which carers were biologically related vs. unrelated to the children, and to compare the magnitude of correlations for each type. If there is a sufficient number of biologically unrelated carers to make the comparison, greater convergence among related vs. unrelated carers would suggest that there is a biological component to the development of emotional skill. Even if there is not a sufficient number of biologically unrelated carers, the correlation between children and their main carers would be a valuable descriptive statistic that could be compared with the correlation for fraternal/DZ twins, given that both types of relatives share 50% of their DNA.

Control variables: For the above analyses, it would be valuable to include control variables of the sex of the child, their ethnicity, and their income / socio-economic status.

The effect of breastfeeding on health and developmental outcomes: an instrumental variable approach

by

Emla Fitzsimons, Senior Research Economist, Institute for Fiscal Studies

Marcos Vera-Hernandez, Economics Lecturer, University College London

The causal effects of breastfeeding on children's outcomes such as health and cognitive development are little understood. The problem is that breastfeeding is a choice made by parents, and so any associations between breastfeeding and outcomes may be confounded by broader socio-economic factors. Even if we control for these as best we can, there is still the concern that there are unobserved factors that simultaneously affect the decision to breastfeed and children's outcomes.

If well conducted, randomized trials are known to provide the most robust evidence on causal effects. However, a breastfeeding randomized trial would probably be very costly and unethical. In the absence of a randomised controlled trial, economists have long turned to the method of instrumental variables in order to identify causal effects. It is a technique that recognises that breastfed and non-breastfed children would have different average outcomes even regardless of whether or not they were breastfed, most likely because they differ in background characteristics. This technique works by assumming that there is one variable, called an instrumental variable, that (1) predicts breastfeeding participation, (2) only affects outcomes through affecting breastfeeding, and (3) is not correlated with the unobserved background characteristics that affect outcomes. The credibility of the resulting estimate thus hinges on the instrument being suitable. We believe that day of birth satisfies all of these requirements. Our hunch is that it satisfies (1) because day of birth impacts on the amount of breastfeeding support/advice received whilst in hospital. For instance, the intensity of support available at weekends is generally less than that available during the week, which may subsequently affect breastfeeding rates. We have no reason to believe it has a direct effect on outcomes, thus satisfying (2) (a possible concern is with planned Caesarean sections, which we would exclude from the analysis). And because time of delivery is a fairly random event, we believe it satisfies requirement (3).

It is for this reason that we are seeking access to the data already collected by the ALSPAC study. It is an ideal data set to use to address our research questions: on the one hand it has extremely detailed information on pregnancy, labour, delivery and infant-feeding practices; and on the other, very detailed information on child measurement and assessment.

Apart from the information contained in the questionnaires and the child measurements, the methodology also requires access to the variable day of birth.

We anticipate that we could use many of the variables collected under the ALSPAC study. The information collected before birth will be used to check the suitability of the instrumental variable (day of the week should be uncorrelated with background variables). The longer the list of variables that we will present, the more credible our instrumental variable will be. We will also use information on child questionnaires and measurements to build our outcome variables (child's health and development). We will also use parents' information collected after birth to consider the pathways through which breastfeeding could affect the outcomes (for instance, whether parental investments in the child follow compensating or reinforcement behaviours)

At this preliminary stage of the research it is difficult to pin down exactly which variables we would use. Moreover, the results of the estimations will inform us on what other variables to use. For these reasons, ideally we would like to have access to :(1) day of birth, (2) all variables in the child and parents questionnaires collected at any moment in time, (3) assessments (4) linkage with school data, (5) variables that have been derived from biological samples at any moment in time