bashtage
Multi-way clustering standard error with AbsorbingLS
Hi,
Since PanelOLS supports maximum two-way fixed effects and two-way clustering (docs here), and I looked around and found AbsorbingLS that could be an option for multiway fixed effects.
I was reading docs on AbsorbingLS but it doesn't show whether it allows multi-way clustering standard errors. If not, will it be implemented in the future?
No plans to implement 3 or higher-way effects.
Thank you for your reply. And do you know how to implment higher-way effects by other python module? Is there such module exist?
I don't think there are any for 3 or higher way. Of course, if some of the effects are low dimensional (e.g., time in years in most panels), then you can just include these are regressors and still have 2 other effects.
I am also interested in Panel OLS with 3+ effects (possibly including 2+ time dimensions) and standard errors clustering.
No plans to implement 3 or higher-way effects.
Out of genuine curiosity, is it due to technical constraints (e.g., difficulty to refactor the underlying assumption of ≤2 indices in the pd.DataFrame, or drastically different algorithm to implement—I'm afraid I'm quite alien to underlying theory here), or lack of bandwidth on your end?
if some of the effects are low dimensional, then you can just include these are regressors and still have 2 other effects.
To make sure my understanding is correct: Are you here suggesting to OneHot encode the 3rd+ categorical variable (say, country) that we want to use as fixed effect, and add the resulting country_A, country_B, etc. columns to the list of exogenous variables of the model (and simply ignore their coefs when reporting the results of the model)?
