Chapter (2) ("Does family income affect children's health?") investigates whether there is a socioeconomic gradient in child health or, in other words, whether there are gaps in child health that are related to differences in family income. When one considers, for example, a broad set of confounding variables, unobserved heterogeneity and the dynamics of health production. Many studies provide evidence for a positive relationship between child health and family income, which is referred to as the child health/ family income gradient, and suggest that this relationship becomes stronger as children grow older (see for example Case et al., 2002; Currie and Stabile, 2003; Condliffe and Link, 2008). Despite this strong empirical support for the existence of a protective effect of income on child health, little is known on the mechanisms underlying the gradient. While earlier studies find that the gradient is robust with respect to consideration of confounding variables like, for example, parental education, parental physical health, child's initial health stock or simple genetic mechanisms (e.g. Case et al., 2002; Currie et al., 2007; Propper et al., 2007), more recent analyses show that the gradient is less robust if confounding variables as well as the dynamic of health capital accumulation are taken into account (e.g. Murasko, 2008 and Khanam et al., 2009). Results obtained by Murasko (2008) and Khanam et al. (2009) suggest that parental, and in particular maternal, health are important determinants of child health. Although the child health/ family income gradient has attracted considerable attention since the influential study of Case et al. (2002), several unsolved issues remain. More recent studies refer to the health capital model provided by Grossman (1972, 2000) and adopt a dynamic framework (Murasko, 2008 and Khanam et al., 2009). However, these studies do not apply appropriate dynamic estimators and do not account for unobserved heterogeneity. The analysis at hand contributes to the literature by applying a dynamic panel estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) to comprehensive British birth cohort data from the Millennium Cohort Study. In addition to this, particular attention is paid to differences between household types. The System GMM estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) in conjunction with comprehensive data, allows us to consider unobserved heterogeneity and endogenous right-hand side variables as well as to disentangle the mechanisms underlying the gradient. Initial cross-section results are in line with the preceding literature as there is strong evidence for a protective effect that grows stronger up to the age of seven. The results also stress the role of maternal health, although this could not fully explain the gradient. The longitudinal results corroborate the theoretical predictions and show a significant path dependency of child health. The protective effect of income is significantly less compared to cross-sectional results. Again, maternal health proves important and leads to a disappearance of the gradient. These findings confirm evidence presented by Khanam et al. (2009). Furthermore, the analysis shows marked differences with respect to family type. In summary, the results suggest paying particular attention to parental health rather than family income. Parental ill health could influence child health in different ways, like providing less or lower quality care. Strong path dependency of child health recommends policies that intervene in early childhood. Chapter (3) (“Does illness have an impact on child development?”) deals with the question of whether child health affects the development of cognitive and noncognitive skills during early childhood. Early empirical analyses of this matter focus on physical health and cognitive skills as well as educational outcomes and find strong evidence for ill health having negative consequences on child development and educational outcomes (for example Edwards and Grossman, 1979, Shakotko et al., 1980, Rosenzweig and Wolpin, 1994 and Korenman et al., 1994). Nonetheless, these findings have been subject to severe criticism due to inappropriate or missing consideration of methodological issues like unobserved heterogeneity or reversed causation (e.g. Kaestner and Corman, 1995). Kaestner and Corman (1995) and Behrman and Lavy (1997) address these problems and find few or no significant effects of child health on developmental outcomes. Currie and Stabile (2006) stress the importance of mental health, in terms of ADHD, and incorporate other developmental outcomes. They provide strong evidence for mental health having a negative on various outcomes. Their results additionally suggest that mental health might have an even stronger effect than physical health. Ding et al. (2009) analyse whether depression, ADHD and obesity exhibit adverse effects on academic performance. Their results indicate that there is a negative effect only for girls but not for boys. Within the present study I draw particular attention to methodological issues in order to facilitate causal inference. The selected econometric approach accounts for reversed causality and unobserved heterogeneity. Birth cohort data from the Millennium Cohort Study allows extending the scope of the existing literature to a broad range of health (physical and mental conditions) and developmental (cognitive and noncognitive skills) outcomes and accounting for numerous confounding variables. Additionally, the analysis separately addresses the effect health has on children living in lone parent households compared to children living in two parent families. Reversed causality is ruled out by investigating the lagged impact of health on child development. Besides this there may be observed or unobserved variables that affect both health and developmental outcomes. Thus, the model comprises a development and a health equation which can have variables in common and incorporates equation specific terms representing unobserved characteristics that are allowed to be correlated. The resulting system of equations is estimated by Bayesian methods as these avoid simulation of multivariate integrals. The results suggest that one has to be aware of the striking difference between children living in different household types. Whilst there is roughly no significant effect of physical conditions on cognitive and noncognitive skills among children living with single parents, chronic diseases exhibit significant negative effects on a broad range of noncognitive outcomes. Cognitive skills among children living with both parents are only affected if they are related to spatial thinking. Mental health in terms of hyperactivity and inattention exhibits strong negative effects on cognitive and noncognitive abilities irrespective of household type and child age. In summary, the results are in line with those of Currie and Stabile (2006) and Ding et al. (2009). The analysis highlights the importance of considering unobserved characteristics. Comparisons of Bayesian and conventional maximum likelihood estimates indicate substantial bias and that adverse consequences are underestimated if unobserved heterogeneity is not taken into account. In view of these results human capital policies should consider two important aspects: first, policies intended to foster human capital accumulation should regard health as a certain type of human capital and should, therefore, aim at improving health too. Second, given the intertemporal effects described above, policies should intervene in early childhood. Both claims are in line with the arguments put forward by Heckman (2006) and Cunha and Heckman (2007). Chapter (4) („Does the combination of family and paid work make you ill?“) analyses the health implications of struggling with combining potentially opposing demands from the labour domain and the home domain. Engaging in both domains might result in positive and negative effects on maternal health. According to role accumulation theory (e.g. Sieber, 1974 or Greenhaus and Powell, 2006) women gain from engaging in several roles as they allow compensation for adverse effects in the working domain through positive experiences in family life, for example. In contrast, role strain theory concludes that women suffer from the double burden of combining multiple demands due to role overload (e.g. Verbrugge, 1983) or role conflict (e.g. Arber et al., 1985). There is large number of empirical analyses dealing with this issue or related questions. Most of these studies come from medicine, psychology or sociology and only few have been conducted by economists (e.g. Chatterji and Markowitz, 2012 and Chatterji et al., 2013). The existing literature offers widespread support for the role strain hypothesis as most studies find adverse health implications. This finding holds true irrespective of whether physical (e.g. Krantz and Östergren, 2001 or Emslie et al., 2004) or mental health (for example Roxburgh, 2004, Nomaguchi et al., 2005, Oomens et al., 2007 or Perry-Jenkins et al., 2007) is under consideration. Only few studies find little or even no evidence for negative effects on maternal health (Waldron et al., 1998, Fokkema, 2002, Krantz et al., 2005 and Usdansky et al., 2012). Role accumulation theory receives little empirical support as only Fokkema (2002) and Nordenmark (2004b) find positive effects on maternal health. Chatterji and Markowitz (2012) and Chatterji et al. (2013) provide evidence for negative effects of the double burden on self-assessed health and mental health. This thesis challenges several methodological issues that have not been appropriately considered by the majority of the preceding studies. I use dynamic panel estimators as health production is a dynamic process (Grossman, 1972, 2000). Additionally, dynamic panel estimators and in particular system GMM (Arellano and Bover, 1995 and Blundell and Bond, 1998) estimators allow for endogenous right-hand side variables and account for unobserved individual effects. The analysis focuses on the intertemporal impact of the double burden on self-assessed maternal health in order to avoid reversed causality. As women struggling with the double burden might reduce their hours worked or completely withdraw from the labour market for two years, for instance, the regression models account for potential coping strategies. In contrast to some studies mentioned above, I use comprehensive panel data from the German Socioeconomic Panel covering the years 2001 to 2011. This data provides reasonable sample size and facilitates causal inference. The extent of labour market demands is measured by broad employment categories and the amount of time spent on household chores and child care is approximated by the number of children living in the household. The results at hand broadly support role strain theory as there is robust evidence for negative health implications resulting from the double burden. Women combining part- or full-time work and simultaneously caring for children suffer from worse self-assessed health. This result is robust with respect to the selected estimator and potentially confounding variables. In addition to this, it seems important to consider individual reactions to the double burden as neglecting coping mechanisms leads to an underestimation of adverse effects. Thus, the results confirm major findings from the preceding empirical literature and suggest that policies should carefully address reconciliation of family and work.