Unified PPC/PPD interface via observed argument

[florence-bockting]

Summary

bayesplot currently maintains two parallel function families:

• ppc_* functions: Compare predictive draws against observed data y.
• ppd_* functions: Display predictive draws without observed data.

Conceptually, these families seem to be similar except for the inclusion of y. Maintaining two separate namespaces can make it harder for new users to navigate the package and understand the distinction between the two modules.

Proposed Change

Introduce a unified top-level interface where the presence or absence of observed data is controlled by an additional argument (e.g., observed):

# With observed data (PPC)
pp_dens(y, yrep, observed = TRUE, ...)

# Without observed data (PPD)
pp_dens(yrep = yrep, observed = FALSE, ...)

Relationship to Other Issues

This proposal is related to Issue#433, as both aim to unify the API through high-level wrapper functions. However, while Issue#433 focuses on unifying functions within a specific family (e.g., distributions, intervals, errors), this issue addresses the broader relationship between the PPC and PPD modules.

[ishaan-arora-1]

hey @florence-bockting, If both #433 (unify via type argument) and #434 (unify via observed argument) are implemented together, we'd end up with something like pp_dens(y, yrep, type = "overlay", observed = TRUE).
do you see these as alternatives or complementary approaches.

[tjmahr]

But predictive checking is a different activity than prediction. There is an asymmetry between checking and prediction too because there are more checking function than prediction functions (for example, all the ppc_*_error functions). There is also a bigger chance for silent errors because the argument order depends on `observed`.

[ishaan-arora-1]

hmm, what about a lighter-touch approach. For instance, keeping ppc_* ppd_* as-is but adding a thin pp_check() style dispatcher that routes based on whether y is provided (rather than an explicit observed flag), limited to the 22 functions that actually have both variants? @tjmahr