This dissertation examines cognitive skill returns to different features of education sys-tems in three independent research articles. It concentrates on interventions in early childhood, which is the age when children's cognition is particularly malleable. Each of the three articles makes an independent contribution to the economics of education li-terature. Chapter 2 estimates the medium- and long-run effects of attending universal center-based child care. While short-run benefits of child care attendance especially for children from low socio-economic backgrounds are well-established in the literature, causal evi-dence on long-run outcomes is still patchy. The article fills a gap in the literature by focus-ing on a number of edu¬cational outcomes, most of which have not been causally studied before. These include second¬ary track choices, grade retentions, cognitive skill outcomes as well as aspirations towards fur¬ther education. The study draws on information from the German Socio-Economic Panel (SOEP), a large representative household survey that provides annual information on children's child care ca¬reers as well as rich personal background data. For identification, it exploits an amendment to the German “Child and Youth Welfare Act” (Kinder- und Jugendhilfegesetz) in 1992. The re¬form established a legal right to a heavily subsidized half-day place in child care for all children from the age of 3 until the beginning of primary school. The reform was meant to especially improve the situation for 3-year-old children who had been hit hardest by the prevalent situation of undersupply at the time, as places were often assigned by age. Exploiting the fact that the expansion in child care supply was staggered across counties for arguably exogenous reasons, the causal effect of one additional year of child care attendance is estimated in an instrumental variables framework where the level of regional child care supply serves as the excluded in-strument. The results indicate that German language grades in adolescence are positively influ-enced by longer child care attendance. The effect is particularly strong among weaker stu-dents as reflected by a sizably reduced likelihood of obtaining one of the three worst grades. There is also evidence for increased educational aspirations, again most notably at the lower margin where students decide between pursuing a vocational degree upon completion of high school or not pursuing any further degree. Taken together, the results corroborate previous evidence that child care attendance is most beneficial for disadvantaged children. This strengthens the case for public funding of child care centers as a means to tackle inequalities in child development. Chapter 3 focuses on the largely unanswered question what classroom actions by teachers are effective in conferring cognitive skills upon students. Specifically, the chapter assesses the effectiveness of primary school teachers' intentions to increase their students' engage¬ment with the course content via different teaching practices. These practices include summa¬rizing key messages, relating lessons to students' daily lives, use questioning, encouraging stu¬dents, giving praise, and bringing interesting materials to class. The analysis is based on a unique dataset that combines information from the 2011 waves of the Trends in International Mathematics and Science Study (TIMSS) and the Pro-gress in International Reading Literacy Study (PIRLS) for a representative sample of Ger-man fourth-graders. TIMSS is an achievement study that tests student competencies in mathematics and science, while PIRLS is dedicated to reading. The year 2011 marks the only occasion so far that the two studies have been sampled together. For this reason, test scores are observed in three different subjects for each student. This provides extra variation for the estimation of within-student between-subject models that form the empirical basis of this chapter. The main results indicate that engaging teaching practices have beneficial effects on students from low socio-economic backgrounds only. Averaged across subjects, it is estimated that a one-standard-deviation-increase on a composite scale measuring the use of potentially engaging teaching practices raises test scores by 4.6 percent of a standard deviation of the test score distribution. In subject-specific analyses based on correlated random effects mod¬els, the detected effect is largest in reading. With regard to policy implications, it is important to note that the benefits for chil¬dren from low socio-economic backgrounds are not offset by significantly lower achievement among other students. Therefore, it is concluded that greater use of engaging teaching practices in primary schools can – very much like longer child care attendance – serve as a vehicle to tackle inequalities in children's cognitive skill development. Chapter 4 is dedicated to the estimation of class size effects in primary school. Class size is one of the key policy levers in education, as teachers' salaries account for the bulk of educa¬tional expenditures in many countries. As of yet, however, there is no academic consensus about the effects that class size reductions or increases generally have. In particular, effect sizes are typically much larger in experimental studies than in quasi-experimental setups. Chapter 4 is an attempt to reconcile these research findings. The chapter shows how grade retentions of poorly performing students give rise to an up¬ward bias in class size estimates that are based on within-school variation in cohort size over time. The bias, which depresses the typically negative class size effects toward zero, is produced by a negative mechanical relationship between initial cohort size and the share of previously retained students in the same cohort in higher grades. The existence of this compositional effect finds empirical support in administrative data on school enrollment for all primary schools in the German states of Saarland and Saxony. It is also shown that the resulting bias can be easily corrected by controlling for previous grade retentions. Performing this correction, class size effects are esti¬mated in a dataset that covers four entire cohorts of students in Saarland who participated in state-wide centralized exams in language and mathematics at the end of grade 3. Instrumenting class size in grade 3 by predicted class size based on imputed cohort size, the results indicate that test scores are increased by around 1.9 and 1.4 percent of a standard deviation for each one-student decrease in class size. However, class size effects seem to be highly non-linear. Whereas language and math test scores are increased by 4.8 and 3.8 percent of a standard deviation in larger classes of more than 20.5 students for each student less, respectively, no effect is found in classes with fewer students. Significant heterogeneities are also observed with regard to student background. Again, disadvantaged students (those with insufficient German proficiency or a learning disability) benefit disproportionately strongly. Finally, evidence is found for a de¬creased likelihood of grade repetitions in smaller classes. The results of Chapter 4 provide first causal evidence of significant class size effects on test scores in Germany. They suggest that class size reductions to increase student achievement should be targeted at larger classes. Conversely, class size in small classes may be increased up to a threshold of around 20.5 students per class with no adverse effects on achievement.