Augmented Inverse Probability Weighting (AIPW) #30
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The two methods only differ by a bit and are both referred to as "AIPW", so they were consolidated to a single class (with the IP-weighted-residual-correction being the default one, as it is slightly more general [applies to multiple treatments] than the Glynn and Quinn version.
1. Add masked-weight-matrix which is a method from Bang and Robins 2005 [e(∆, V; β , φ1, φ2)] 2. Add signed-weight-vector, which is from the erratum for Bang and Robins (2008): https://doi.org/10.1111/j.1541-0420.2008.01025.x
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Completing PR #28.
Glynn and Quinn (2010) present a version of AIPW that is different from the one described in Kang and Schafer (2007, section 3.1).
The two methods are comparable, but slightly differ from one another:
This might be beneficial as it will rely less on outcome-prediction of extreme-propensity data points - those lacking covariate overlap and more prone to model extrapolation. However, since it also rely on opposite-group, it is limited to a binary-treatment setting.
Since both versions are frequently referred to as "AIPW", and since code overlap a lot, it is better to consolidate them to a single AIPW class.
Additionally, following up on the changes in #28,
It appears Bang and Robins 2005 has a 2008 errata describing the IP-feature as needing a negation to the control group (matching with Hernán and Robins book Fine Point 13.2, and aligning with the ATE-targeted version of TMLE).
Additionally, a masked-IPW matrix (described as
e(∆, V; β , φ1, φ2)in B&R) is also added.These two were previously commented out, as they seem to have bigger variance (less-efficient) than the other methods on some simple simulated datasets, but they are theoretically justifiable, so they are added anyway.