Series format for EffectEstimate

[Le-Mao]

I've been trying to extend the CTF by processing the output of EffectEstimate (and its surrounding classes), and I notice pd.Series is used for each value and cis.
As far as I can tell though, there's no instance of EffectEstimate in the CTF where multiple values are used in the same Series attribute here - I'm not even sure what that would mean.

Is there some situation where multiple values would be used by the same EffectEstimate? Or am I missing something about the way the class is used that justifies the use of pd.Series here? /genq

(if this is actually not a needed/intended part of the system structure then I'd also like to know that)

Thanks :)

[jmafoster1]

Hi @Le-Mao.
Firstly, thank you for your interest in the framework and sorry for the delay! I did not receive any notification of your question.
We use pandas Series objects here because treatments (and other covariates) can sometimes be non-numeric (e.g. binary or category values). In such cases, the estimators return one coefficient for each possible treatment value, e.g. if variable X can take values red, blue, and green, and red is the "control" value, then we get a coefficient for X[blue] AND X[green]. In the vast majority of cases, using the series objects is indeed unnecessary and a little bit annoying, but it is needed to make sure we have a single standard structure for returning outputs that also works in this case.

[Le-Mao]

Interesting! That makes a lot of sense - My intuition would initially have been that you'd just use Enums with integer conversions for that, but now I think about it, that would enforce a constant numerical mapping for all causal effects from the same variable.

[jmafoster1]

Yes, exactly. Some categorical variables (e.g. colour) have no natural ordering, and we do not want to enforce one (especially for protected category data like ethnicity or marital status), so this is just the "least worst" option we could think of. We are, of course, open to suggestions if you can think of a better way :)