UGBA 88 LECTURE 4 POTENTIAL OUTCOMES FRAMEWORK 2023/2024 already passed

UGBA 88 LECTURE 4 POTENTIAL OUTCOMES FRAMEWORKCausal inference compares WHAT? - correct answer Potential outcomes Comparisons of observed differences between treated and control units are often contaminated by _______ bias. - correct answer Selection What ELIMINATES selection bias between treatment and control groups? - correct answer Random assignment What does Yi represent? Y1i? Y0i? - correct answer The observed OUTCOME of interest for unit i; Y1i= potential outcome for unit i with treatment; Y0i = potential outcome for unit i WITHOUT treatment What does Di represent? Di=1? Di=0? - correct answer The indicator of treatment for unit i; Di=1 means i has the treatment; Di=0 means i does not have the treatment At the most basic level, what is a TREATMENT EFFECT, and how is this represented in notation? - correct answer The difference between 2 potential outcomes; Y1i - Y0i. Y1i - Y0i also represents what? - correct answer The CAUSAL EFFECT OF TREATMENT on outcome Y for unit i What is the fundamental problem of causal inference? - correct answer Each individual unit i either takes the treatment (and has the corresponding outcome) or they do not; WE CAN ONLY OBSERVE ONE POTENTIAL OUTCOME, we cannot observe their counterfactual What is the KEY TO CAUSAL INFERENCE? - correct answer A valid counterfactual; to answer a causal question, you need to at least guess what the counterfactual is. Can we observe a treatment effect for any one person? - correct answer No What does E[Xi] mean? - correct answer Population AVERAGE for random variable X What is the notation for AVERAGE TREATMENT EFFECT? - correct answer E[Y1i - Y0i] = E[Y1i] - E[Y0i] What is a conditional expectation? How is it written according to potential outcomes framework? - correct answer The average for a random variable where the population meets some subset; E[Xi | Di = d (some condition)] How would you interpret this conditional expectation: E[Y0i | i is left-handed] - correct answer Average outcome for the non-treatment group among left-handed people What is the Average Treatment Effect on the Treated (TOT)? - correct answer E[Y1i - Y0i | Di = 1] What is the Average Treatment Effect on the Untreated? (TOU) - correct answer E[Y1i - Y0i | Di = 0] How come, in both TOT and TOU we were able to do Y1i - Y0i even though we can only observe one potential outcome for any given individual? - correct answer Because E[Y1i - Y0i] in this context is looking at the AVERAGE of the outcomes among a group of people! In practice, when we are measuring average outcomes for treatment and control groups (E[Yi | Di=1] - E[Yi | Di = 0]), what is a major concern about inferring anything from the difference in their outcomes? - correct answer You always have to be worried that Di=1 group and Di=0 group are not the same (they are systematically different), and that something else than just whether or not someone was given the treatment is contributing to the difference in the average outcomes. What is the potential outcomes framework for the selection bias equation? If the equation sums to 0, then what can we conclude about selection bias? - correct answer E[Y0i | Di=1] - E[Y0i | Di =0] ; if this equation sums to 0, then there is no selection bias.