Chapter 9 Time Varying Treatments Flashcards
What is the difference between modeling the probability of treatment assignment of time varying treatments as compared with time-invariant treatments?
For time-invariant treatments, removing selection bias with propensity score methods required the identification of confounding variables measured prior to treatment assignment only once. With time-varying treatments, which are treatments that can be repeated at multiple occasions, each individual’s probability of receiving the treatment a certain time depends on the previous treatment history, the previous outcomes, time-invariant covariates, and time-varying covariates.
How does the inverse probability of treatment weight accomplish reduction of selection bias in the estimation of the treatment effect of time-varying treatments?
Selection bias can be controlled by weighting each observation at time t with the inverse of the probability of exposure to the conditions the individual was exposed to by time t.
What is the difference between inverse probability of treatment weights and stabilized inverse probability of treatment weights?
compare two formulas
Why is it advantageous to use stabilized inverse probability of treatment weights instead of inverse probability of treatment weights?
It addresses the problem that the IPTW formula is likely to result in extreme weights.
What is the difference between stabilized inverse probability of treatment weights and basic stabilized inverse probability of treatment weights?
The basic stabilized IPTW does not reduce extreme weights as much as the stabilized IPTW.
What issue with stabilized inverse probability of treatment weights does the basic stabilized inverse probability of treatment weights address?
Basic stabilized IPTW addresses the issue that the stabilized IPTW does not remove the confounding effect of treatment and requires that treatment history is included in the outcome model.
What are the advantages and disadvantages of inverse probability of treatment weights over marginal mean weights through stratification?
Disadvantages of IPTW: MMWS is less likely to have extreme values than the IPTW.
Advantages of IPTW: Inverse probability of treatment weights (IPTW) allow removal of bias due to time-varying and time-invariant covariates, as well as previous history of treatment.
What are two possible outcome models for estimating the effects of time-varying treatments?
weighted least squares estimation with design-based adjustment for clustering effects and generalized estimating equations
What is the role of the choice of correlation structure in generalized estimating equations?
The parameter estimates of the GEE are dependent on the correlation structure of two observations for the same individual.
What is the advantage of using generalized estimating equations instead of multilevel models for estimating the effects of time-varying treatments?
GEEs focus on estimating a nonvarying (or average) coefficient in the presence of clustering, whereas MLMs (HLMs) focus on estimating the aspects of the model that vary by group. GEEs allow more flexibility. GEEs appeal to people who don’t like distributional assumptions, whereas MLMs appeal to people who like generative models.
