Flashcards
potential outcome framework etc 2
Cards in this deck (4)
- 1Potential outcome framework (POF) -Cook, Shadish and Campbell said that it POF is about the magnitude of „causal description" not causal mechanisms/effects of causes not causes of effects and it requires clearly defined causal states. POF assumes that each subject has a potential outcome under both T states with each subj being able to be obs in one T state at any point in time. Therefore, ITE is a non-observable counterfactual, which is what Holland, 1986, labels the fundamental problem of causal inference. Even successive exposure of the same subj to alternative causal states does not allow for the obse of ITE; inference requires assumptions (temporal stability, causal transience)Average T effect (ATE) is non-obs but we can observe the naive estimator of ATE, which might correspond to ATE only under very specific cond regarding the process by which subj are assigned to causal states. Typically, this proc reflects an individual‘s decision to enter one state or another, an outside actor‘s choice to allocate individuals to one state or another, a random allocation planned by the investigator, or some combinations thereof.
- 2Unit homogeneity assumption pertains to individual subj- D is the only variable that affects YIndependence assumption T is independent of potential outcomes and most well-know way to warrant independence is by randomnisation of D
- 3Variations and biases: Obs (as opposed to exp) studies are defined by the lack of control over the T assignment process ( by the infeasibility of rand). To determine the causal effect of D on Y, the first step in obs studies is therefore to investigate the T selection mechanismSUTVA requires that the potential outcomes of subj be unaffected by the exposure to T of other subj; it implies that T effect is independent of ATE
Classical SUTVA violation: evaluation of vaccination programs - 41. The causation should begin with studying the effects rather than the causes of a given effect 2.Effects of causes are always relative to other causes and it takes two to define an effect 3.Not everything can be a cause-attributes of units are never causesCausal inference is impossible without making untested assumptions therefore it gives it an air of uncertainty as Hume argued
Information
- Description
- Flashcards on potential outcome framework etc 2, created by tri_nana on 01/07/2015.
- Created by
- tri_nana
- Views
- 0
- Rating