estimate_coefficient method returns different variable types based on if condition

[christopher-wild]

Describe the bug
The estimate_coefficient method in estimators.py can either return a list of pandas Series or a list of numpy floats based on which if condition is met.

Expected behavior
Ideally only Series would be returned from this method.

Additional context
Unit tests will need fixing as well as they rely on both data types. This change will facilitate the tidying of the method.

[christopher-wild]

The other return from the method unit_effect also changes type between Series and numpy float

[jmafoster1]

Please could the return types for the other estimate_* methods be checked too?

[christopher-wild]

Would the preference for all estimate_* functions to only return pd.Series for the confidence intervals and the effect?

[jmafoster1]

Yes I think so

[christopher-wild]

You okay if we merge #264 as that is meant to be limited to the Pasty interaction terms? Then I'll make a new PR for this issue, and release a version that contains both PRs?

[jmafoster1]

Works for me. Just reviewing the code now

[christopher-wild]

I may actually hold off on merging this one, if we make all the estimate_* returns consistent, we won't need some of the conditions the #264 PR introduces.

So I'll work on the same branch just to make life simple and not introduce code that we will immediately remove.

[jmafoster1]

Agreed, now I've looked at the chances.
This is a bit unrelated, but if we can get the model from Patsy, we can have it keep track of the adjustment set from the formula as well, which might help with #238