30212

WiSe 19/20: Causal inference with observational data

Ehlert / Holtmann

Comments

Researchers, policy-makers and practitioners alike are often interested in causal questions: Do comprehensive schools provide better learning opportunities for low-SES? Do minimum wages destroy jobs? Is parental divorce harmful for children? Do norms about gender roles influence women’s and men’s employment behaviour? However, identifying causal effect is often difficult because it requires distinguishing accidental association from causation. Therefore, experiments with random allocation to a treatment and a control group are often seen as the gold standard for identifying causal effects. However, especially in Sociology we often do not have randomized trials for the questions we would like to answer. They might be expensive, unethical, impractical or untimely. For example, assigning families to divorce at random to study its effects is (and should not) be possible. Therefore, we have to study those questions using observational data, for example from surveys. In such data sets, the treatment of interest is distributed beyond the control of the researcher and possibly connected to many other measured and unmeasured factors that are also connected to the outcome of interest (confounding). In the seminar, we will discuss several strategies to overcome this issue and to identify causal effects with observational data. We will cover strategies that use “natural” experiments such as the instrumental variable approach and the regression-discontinuity approach. Other strategies we cover use longitudinal data to account for endogeneity in observational data such as the difference-in-differences approach and panel data techniques. We will also cover matching methods and sibling and twin designs. We explain the main idea and intuition behind these strategies and illustrate them with real-world examples. Participants should already have working knowledge of basic statistics and multiple regression to be able to follow the course.

Learning goals:

Understand the problems involved in identifying causal effects.
Understand the intuition behind several strategies to draw causal inferences from observational data. Evaluate the strengths and weaknesses of each strategy.
Assess critically the strategies different research papers use to identify causal effects.
Being able to apply the strategies learnt to your own research questions.