The Causal Modeling Web Site from the
University of Pennsylvania Causal Group
This web site provides talks and SAS macros on our work on
causal modeling to go beyond standard approaches in analyzing
mechanisms of complex interventions in randomized trials through
causal analyses of treatment non-adherence, mediators and post-
randomization moderators or effect modifiers that identify groups
of participants for whom complex interventions are more effective.
Such analyses may lead to better personalized treatments.
We provide:
Randomized controlled trials (RCTs) are a key method for
understanding causation. Within RCTs,
The traditional methods to study mediation and moderation make
a number of key assumptions such as "sequential ignorability" or
no unmeasured confounding after randomization, which is not
needed with our causal modeling approaches. Our introductory
papers give a fuller introduction.
This research is due to many collaborators, including Dylan Small,
Marshall Joffe, Kevin Lynch, Mark S. Cary, Robert Gallop, Julia Lin,
Mike Elliott, Jen Faerber, Peter Yang, and Rongmei Zhang
University of Pennsylvania Causal Group
This web site provides talks and SAS macros on our work on
causal modeling to go beyond standard approaches in analyzing
mechanisms of complex interventions in randomized trials through
causal analyses of treatment non-adherence, mediators and post-
randomization moderators or effect modifiers that identify groups
of participants for whom complex interventions are more effective.
Such analyses may lead to better personalized treatments.
We provide:
- Technical papers on causal modeling
- Introductory lectures and talks, and
- SAS macros for implementing the models
Randomized controlled trials (RCTs) are a key method for
understanding causation. Within RCTs,
- mediation is the study of how the intervention affects an
intermediate variable that then leads to an outcome, while - moderation is study of how post-randomization variables
early in the intervention modify the effects of the latter
stages of the intervention on subsequent outcomes.
The traditional methods to study mediation and moderation make
a number of key assumptions such as "sequential ignorability" or
no unmeasured confounding after randomization, which is not
needed with our causal modeling approaches. Our introductory
papers give a fuller introduction.
This research is due to many collaborators, including Dylan Small,
Marshall Joffe, Kevin Lynch, Mark S. Cary, Robert Gallop, Julia Lin,
Mike Elliott, Jen Faerber, Peter Yang, and Rongmei Zhang
CausalModels.org
 | |  | |  | |  |