iplot: labels for x-axis

[mortizm1988]

Hi everyone.
I hope someone can help me further understand how to personalize an iplot.
I'm currently regressing the % of foreign employees across time (fiscal year: "fyear"), for family and nonfamily firms ( "family_control").
My code is as follows:

fit2 = pf.feols("foreign_employees_perc ~ i(fyear, family_control, ref=2018) + roa + log_at +leverage | rut+ fyear", data=international,vcov={"CRV1": "fyear" })

fit2.iplot(
    coord_flip=False,
    yintercept=0,
    alpha=0.05,
    labels=rename_event_study_coefs(fit2._coefnames),
    figsize=[1400, 600],
)
```
My goal is to label the x-axis just with the year (2014, 2015,..., 2022). I played around with the "label" options, but haven't managed.

Thanks!

[s3alfisc]

Hi @mortizm1988 , in principle you should simply be able to pass a dictionary with the names of the coefficients as keys and values the new names. But that doesn't seem to be working for me using pf v0.29:

import pyfixest as pf
from pyfixest.report.utils import rename_categoricals, rename_event_study_coefs
import numpy as np

df = pf.get_data()
fit1 = pf.feols("Y ~ i(f1, X1)", data = df)
fit2 = pf.feols("Y ~ i(f1, X1) + X2", data = df)

pf.iplot(
    models = [fit1],
    labels = {
        np.str_('C(f1)[0.0]:X1'): "new name", 
        "C(f1)[0.0]:X1": 'new name 2',
    }, 
)

So I would say this is a bug?

I can take a closer look at it tomorrow evening.

If this also does not work for you, you can also rename the y or x axis ticks manually, as iplot returns a matplotlib figure:

fig = pf.iplot(
    models = [fit1, fit2, fit3],
    labels = rename_categoricals(fit1._coefnames)
)
type(fig)
# matplotlib.figure.Figure

ax = fig.axes[0]  
ax.set_yticklabels(['Label A', 'Label B', 'Label C']) 
fig

[mortizm1988]

Thanks @s3alfisc for your response.
Indeed, I tried the first approach, buy got the same result:

fit2 = pf.feols("foreign_employees_perc ~ i(fyear, family_control, ref=2018) + roa + log_at +leverage | rut+ fyear", data=international,vcov={"CRV1": "fyear" })

pf.iplot(
    models=fit2,
    coord_flip=False,
    yintercept=0,
    alpha=0.05,
    labels={"C(fyear,cont.treatment(base=2018))[2015] x family_control": '2015',
            "C(fyear,cont.treatment(base=2018))[2016] x family_control": '2016',
            "C(fyear,cont.treatment(base=2018))[2017] x family_control": '2017',
            "C(fyear,cont.treatment(base=2018))[2019] x family_control": '2019',
            "C(fyear,cont.treatment(base=2018))[2020] x family_control": '2020',
            "C(fyear,cont.treatment(base=2018))[2021] x family_control": '2021',
            "C(fyear,cont.treatment(base=2018))[2022] x family_control": '2022'},   
    #plot_backend='matplotlib',
    rename_models={fit2._model_name_plot: "Model 1",
    },
)

I also tried with the matplotlib approach. I notice an error because it requires assigning values to "ax.set_xticks". I did it, but it did not work either.

I hope this has an easy solution.

[s3alfisc]

Hi @mortizm1988 , this is clearly a bug and I opened an issue here #894 If you need to fix this really, really urgently, the code snippet with the error should be this and it is getting called by iplot via _coefplot via coefplot_matplotlib here. I'll try to fix this in the next days but can't make promises (I only work on pf in the evenings and have plans half he evening tomorrow).

[s3alfisc]

Actually, I figured out the solution! You only rename the "categorical" part of the interaction, not the "full" interaction. Here is an example what I mean by this:

import pyfixest as pf
from pyfixest.report.utils import rename_categoricals, rename_event_study_coefs
import numpy as np

df = pf.get_data()
fit1 = pf.feols("Y ~ i(f1, X1)", data = df)
fit2 = pf.feols("Y ~ i(f1, X1) + X2", data = df)

pf.iplot(
    [fit1, fit2], 
    labels = {"C(f1)[0.0]":"var 1"}, 
)

produces

So use labels = {"C(f1)[0.0]":"var 1"}, instead of labels = {"C(f1)[0.0]:X1":"var 1"}`.

Note sure if this is really too straightforward? At the very least, we should update the docs. Would you have expected such behavior @mortizm1988?

[mortizm1988]

Hi @s3alfisc
Indeed, that brings me closer to the expected behavior. Notice that this solution stops working when we specify the category of reference for the estimation. Following your example:

fit3 = pf.feols("Y ~ i(f1, X1,ref=0.0)", data = df)
fit4 = pf.feols("Y ~ i(f1, X1,ref=0.0) + X2", data = df)

pf.iplot(
    [fit3, fit4], 
    coord_flip=False,
    labels = {"C(f1)[1.0]":"var 1"}, 
)

Generates the same label for the first coefficient:

Btw, I would be happy to help updating/extending the docs. I' don't have that much experience doing PRs but I'm using pf a lot, so please let me know if I could be of some help.

[s3alfisc]

Yes, you really need to use the "full" name of the variable, which in this case then becomes C(f1, contr.treatment(base=0.0))[1.0]

%load_ext autoreload
%autoreload 2

import pyfixest as pf
from pyfixest.report.utils import rename_categoricals, rename_event_study_coefs
import numpy as np

df = pf.get_data()
fit1 = pf.feols("Y ~ i(f1, X1, ref = 0.0)", data = df)
fit2 = pf.feols("Y ~ i(f1, X1, ref = 0.0) + X2", data = df)

pf.etable(
    [fit1, fit2], 
    labels = {"C(f1, contr.treatment(base=0.0))[1.0]":"var 1"}, 
    type = "md"
)

which would give

index                                           est1        est2
----------------------------------------  ----------  ----------
depvar                                             Y           Y
----------------------------------------------------------------
var 1 x                                     0.598**     0.632**
                                             (0.225)     (0.216)
C(f1, contr.treatment(base=0.0))[2.0] x   -1.710***   -1.780***
                                             (0.243)     (0.233)

Not sure if this the most straightforward solution - I guess one could argue that one upside is that it's easy to define a "rule", which would be "provide the full name of variable - everything you would get via fit._coefnames , and for interactions, everything up until the interaction sign"?

[s3alfisc]

And awesome to hear that you're using pyfixest a lot! 🌞

If you want to take a stab at clarifying the docs, you could update the description of the labels argument in pf.etable, pf.coefplot and pf.iplot(), which you can find in the attached links:

• etable: link
• coefplot: link
• iplot: link

and maybe add examples how to effectively use the labels arg for all three functions?

I think the easiest way to set up PRs would be to just use the github web UI - first clone the repo and then click on "edit this file" and go through the entire process of opening a PR. There's really nothing that could go wrong - I have to review and approve any changes you make =)

[s3alfisc]

The other option of course is to fork & clone the repo, change locally, push to your fork and then open a PR from there (as with almost everything, LLMs are pretty good at helping you navigate these things)

[mortizm1988]

I sent a PR with the update to the documentation.
Happy to work on the examples (on quickstart.qmd?) after checking if I learned to do the PR correctly, hehehe.

[s3alfisc]

Awesome! I just merged it =)

For the examples: I would maybe add examples to the code examples of the docstrings of the individual functions, e.g. here for etable.

And then maybe you could also update the how to create nice regression tables section with the examples you've added for etable?