Joint hypotheses

[LamAdr]

Here is my go at joint hypotheses testing. It seems pymarginaleffects does not support making hypotheses about comparisons/slopes objects yet, and I didn't change that. I made changes to two other files :

1. classes.py
   • added the capability to change the mapping used to change column names at print time
   • added a dataframe header print_head to match R's output
2. model_statsmodel.py
   • get_variables_names() now keeps the variables in the order they appear in the formula. That's useful if users want to specify which variable they're hypothesizing on based on their indices.
   • added the get_df function

One potentially problematic assumption is that the presence of the intercept is verified with len(theta_hat) == len(var_names)+1.

Thanks

[vincentarelbundock]

Thanks! Are the tests hard-coded numbers verified against R?

FYI, I'm not planning to allow hypotheses() to process slopes or comparisons or predictions objects. I regret this design choice in R. Users should just use the hypothesis argument instead of a separate function call.

[LamAdr]

Are the tests hard-coded numbers verified against R?

You mean as in e.g.

hypo_py = hypotheses(mod, joint=[0, 1, 2], hypothesis=[1, 2, 3])

? If so, yes, I generated a csv using the corresponding R call.

I'm not planning to allow hypotheses() to process slopes or comparisons or predictions objects.

In that case maybe we should rename the input obj to model or mod?

[vincentarelbundock]

Let's do exact parallelism in argument names between R and Python, even if it feels a bit weird.

Let me know when I can merge. Looks great!

[LamAdr]

alright, I think you can merge.

[vincentarelbundock]

great thanks!