How to add customized model statistics in the table?

[cytwill]

Hi, there!

Thanks for this fantastic package! I am writing to ask if there are any easy ways to add other model statistics at the bottom of the table in addition to the default ones. Currently, my tables return observations, R2, R2 within, and SE, I also want to include pseudo R2 and adjusted R2. Curious to hear your thoughts on this.

Best,
Yutao

[s3alfisc]

Hi Yutao, this is indeed possible!

You can pass custom model stats to etable via the custom_model_stats argument (slightly hidden in the docs here):

pf.etable(
    [fit1, fit2, fit3, fit4, fit5, fit6],
    labels=labels,
    custom_model_stats={
        "Number of Clusters": [42, 42, 42, 37, 37, 37],
        "Additional Info": ["A", "A", "B", "B", "C", "C"],
    },
)

I hope this helps!

Best, Alex

[cytwill]

Thanks for your response, Alex! I see this option there, but I am wondering if I can fetch adjusted R2 or pseudo R2 from the fitted models (e.g., fit1, fit2, fit3, ....). so I can manually add these statistics to the output table as you suggested above. I assumed that if these statistics are available, I should be able to get them by calling a function such as fit1.adjusted_r2 or something similar, but I did not find one. I also checked the raw summary table of these models and did not see additional model statistics there, too.

Do you have further suggestions on this (maybe there are some other functions to generate these stats in Pyfixest)? Thanks again!

Best,
Yutao

[cytwill]

Sorry, I think I found some of the entries for these statistics! They are inner instances starting with "_", "adj_r2". At the same time, there is function called get_performance(), which I guess is supposed to return or generate these statistics without the inner "" marker? But this function seems to not work (the function returns None, and there is no instances like "adj_r2" generated), did I get something wrong here?

[s3alfisc]

HI Yutao, all statistics computed via get_performance() have a _ prefix. So you should be able to get the adjusted r2 via Feols._adj_r2 etc. I just realize that this is not documented in the docs for the Feols class, which is something I should update.

get_performance() is an internal function that should not be called by end users - it is triggered for every Feols call automatically. I should likely rename it to _get_performance?

Also, I should probably ditch the _ prefix for all stats that might be interesting to users.

Are there any other statistics you are interested in?

[s3alfisc]

Sorry actually, everything is documented (puh!). See here.

We should point more directly to all attributes in the quickstart section & also mention it in the functional pf.feols() docs.

Would you be up to try yourself at a PR @cytwill ? I think the following should do:

1. In the feols function documentation, point to the Feols estimation class by linking to the docs here and mentioning that attributes can be checked there.
2. Here you could also point to the Feols class docs and mention that all attributes can be seen there. Also, we can fix the typo "Feols class" instead of "Fixest" class. link
3. If you have a lot of breath, you could repeat step 1 for the other estimation classes: Quantreg, GLM, etc.

Let me know what you think, I'd be happy to also do it.

Btw, this is of course related to #959 .