Proposal: Enable `contrast_ylim` for matplotlib by adjusting `gardner_altman_adjustments`

[ShumpeiDate]

Summary

The contrast_ylim argument currently has no effect when using matplotlib. A previous discussion (#128) mentioned that supporting contrast_ylim was considered not feasible at that time due to automatic calculation of contrast limits.

What I found

While reviewing the implementation, I noticed that the gardner_altman_adjustments function always computes its own contrast axis limits and does not appear to prioritize a user-specified contrast_ylim.

I understand that the current design may have historical or technical reasons behind it.
However, I wonder if it might be beneficial to allow a user-provided contrast_ylim to override the automatically computed limits when explicitly set.

Proposed solution

I would like to propose adding logic such that:

• If contrast_ylim is provided by the user, it takes priority.
• Otherwise, the current automatic behavior remains unchanged.

Below is a simplified example of how the logic could be applied:

if effect_size_type in ["mean_diff", "median_diff"]:
    if contrast_ylim is not None:
        # Respect user-specified y-limits.
        contrast_axes.set_ylim(contrast_ylim)
    else:
        # Current automatic behavior.
        rawdata_ylims = np.array(rawdata_axes.get_ylim())
        if current_effsize > 0:
            rightmin, rightmax = rawdata_ylims - current_effsize
        elif current_effsize < 0:
            rightmin, rightmax = rawdata_ylims + current_effsize
        else:
            rightmin, rightmax = rawdata_ylims

        contrast_axes.set_ylim(rightmin, rightmax)

        # The remaining lines follow the existing implementation.
        og_ylim_contrast = rawdata_axes.get_ylim() - np.array(control_group_summary)
        contrast_axes.set_ylim(og_ylim_contrast)
        contrast_axes.set_xlim(contrast_xlim_max - 1, contrast_xlim_max)

Expected behavior

When contrast_ylim is given:

• The contrast axes would respect the user-defined y-limits.
• This would provide behavior consistent with the plotly implementation, which already supports contrast_ylim.

Actual behavior

Currently:

• In Gardner–Altman plots, the contrast_ylim argument is not taken into account.
• The axes limits are determined solely by automatic calculations.

Additional context

If maintainers feel that this direction is reasonable,
I would be happy to submit a Pull Request including:

• The proposed patch
• Tests reflecting the updated behavior

Thank you for considering this, and please let me know your thoughts.