Looking for a mixed-anova with 3 factors

[romquentin]

Hello,
Thanks for the great package !
I am looking for a mixed-anova with 3 factors (2 within, 1 between). Is there a way to do that with pinguoin?
Would you know another python package if not?
Thanks a lot!
Romain

[sappelhoff]

Hi @romquentin, it's unfortunately not possible right now, see: #136 #179

I personally used rpy2 and from there the R package "afex". See a code snippet below.

In that snippet I also install R packages from Python ... but I'd recommend installing them via conda (conda-forge channel) instead. you can also install R in your conda env via conda - it's very convenient that way.

But of course if pingouin had that feature, it'd be even nicer :)

import rpy2.robjects as ro
import rpy2.robjects.packages as rpackages
from rpy2.robjects import pandas2ri, StrVector
from rpy2.robjects.conversion import localconverter

# install packages
packageNames = ("afex", "emmeans")
utils = rpackages.importr("utils")
utils.chooseCRANmirror(ind=1)

packnames_to_install = [x for x in packageNames if not rpackages.isinstalled(x)]

if len(packnames_to_install) > 0:
    utils.install_packages(StrVector(packnames_to_install))

# convert pandas DF ("tmp") to R data.frame
with localconverter(ro.default_converter + pandas2ri.converter):
    r_from_pd_df = ro.conversion.py2rpy(tmp)

r_from_pd_df.head()

afex = rpackages.importr("afex")

## see: https://cran.r-project.org/web/packages/afex/afex.pdf
model = afex.aov_ez(
    id="subject",
    dv="correct_choice",
    between=["stopping", "yoked_to"],
    within="sampling",
    data=r_from_pd_df,
    check_contrasts=True,  # 3-way interaction is unaffected by this, "True" is recommended
    type=3,  # using so-called type 3 sums of squares
    print_formula=True,
)

print(model)

[romquentin]

Very cool ! Thanks a lot @sappelhoff !!

[raphaelvallat]

Thanks @sappelhoff for the code snippet! I'm leaving this issue open as I'm sure it will benefit other people.

For more complex ANOVA models, I also strongly recommend using JASP. I don't have the time to implement such models right now, so unfortunately I don't think it will be implemented in Pingouin in a near future.

Thanks,
Raphael

[raphaelvallat]

Closing this issue but feel free to reopen!