Hi All,
Nice to meet you today. Hope you enjoyed class. To share broadly our responses to some of the questions asked after class and clarify expectations for HW 1, see below:
- We use the idea of a “target trial” or ideal RCT to help in the design, conduct, analysis, and interpretation of observational studies, i.e., cohorts. In asking you to think about the “target trial” that might help motivate the study on Leisure Time Physical Activity & Cancer Incidence, the goal was to think about how/when cohort studies and RCT’s differ, and appreciate the implications of those differences. Sometimes, the ways in which a cohort can’t or does not emulate a target trial, helps us identify potential biases or threats to validity. Sometimes, it may mean that the RCT and cohort address different specific research questions, and can’t be directly compared (as Dan mentioned in class, we can’t exactly do a RCT of leisure time physical activity).
- The ”identifiability criteria” – exchangeability, consistency, and positivity –are specific terms to describe what needs to be fulfilled for a cohort to emulate a “target trial” or idealized randomized experiment. The extent to which a cohort does not meet these criteria reflects the potential for bias. Note, we often acknowledge that there is a spectrum of how well a cohort can meet these criteria (or not meet them). This topic will be covered in Dr. Graff’s Causal Inference lecture in weeks 2 (recorded) and 3 (live).
- For this week’s HW, you do not need to use or apply the “identifiability criteria” when we ask you to compare a cohort to a RCT (though we assume most are familiar with “exchangeability” / “confounding”). The In-class exercise was planned as a light prelude for Dr. Graff’s later lecture. For those interested, Dan plans to discuss this more during Office Hours on 1/14.
In addition to HW 1, please be sure to watch the recorded portion of Dr. Van Blarigan’s lecture on Matching.
Have a good week!