Add support for pDigits argument for p-value precision

[vljlangen]

First off, thanks for your work on tableone—it's an incredibly helpful package for producing descriptive statistics in clinical research.

Feature request:
It would be great to have a pDigits argument to control the number of decimal places shown for p-values in the summary table.

It also seems that in certain situations, p-values are now truncated, so that a p-value of 0.0116 becomes 0.011 in the output table—this is not an ideal situation in my opinion, as it may misrepresent the true statistical significance.

Why it matters:
In certain statistical reporting scenarios—especially in academic manuscripts—it’s important to display p-values with more granularity. Without a pDigits option, we need to manually check the results afterward, which defeats some of the automation advantages of using tableone.

Proposed solution:
Introduce a pDigits argument to TableOne that controls the formatting of p-values in the output table.

[tompollard]

Thanks for your kind words @vljlangen.

so that a p-value of 0.0116 becomes 0.011 in the output table

This is definitely wrong and should not happen, I agree. I'll look into this!

[vljlangen]

I am deeply sorry for misinforming you about the truncation issue. It turns out it wasn't about truncation, but rather that tableone seems to run something like:

t_stat, p_value = stats.ttest_ind(var1, var2, equal_var=False)

Whereas in Python, the default behavior of ttest_ind is equal_var=True. That difference explains the small discrepancy I noticed in the p-values, not any rounding or truncation.

Again, I sincerely apologize for the confusion and miscommunication!

On a second note, it would be really awesome to have control over both:

a) the number of digits displayed for p-values, and
b) whether equal_var=True or False is used in the statistical test.

Grateful for your time and help with this!

[tompollard]

No problem, thanks for explaining!

For the number of digits, could we extend the digits argument for this purpose (e.g. to modify columns, you could do e.g. decimals = {'P-Value': 4)), or do you prefer something more direct like: pval_digits = 4?

[vljlangen]

I was actually thinking about the same question. After giving it some thought, I agree that a separate pval_digits argument would probably be the clearest and most intuitive solution in the long run.

In addition, if it's at all feasible, having control over equal_var (with False as the default) would be really useful as well. Maybe even with a small warning or note like:

"Note: the default behavior of ttest_ind in Python is equal_var=True, which might explain differences in p-values between tableone and manually run tests."

That could help avoid confusion for users who are comparing results across different implementations.

Grateful for the thoughtful conversation around this—much appreciated!

[tompollard]

@vljlangen Thanks for raising this issue. Support for the pval_digits argument was added in #191.

You can get the latest version from PyPI (v0.9.3): https://pypi.org/project/tableone/. It will be available through Conda soon.

I'll open a new issue for the equal_var point.