add x as argument to ppc_error_binned

[behramulukir]

In #358, one of the two tasks that I mentioned was adding x as an optional argument to ppc_error_binned so that it supports covariates in the x-axis. As x argument would be optional, it wouldn't be breaking any existing graphs.

This PR is my work on the first task of #358.

My current implementation passes existing tests without any issue, so I think I am not breaking anything.

• Add x argument
• Add new tests
• Update documentation

Examples

Here are a few examples of it working.

Default version

This default version simply follows the example presented at the PPC errors documentation.

> suppressPackageStartupMessages(library(rstanarm))
> suppressWarnings(example("example_model", package = "rstanarm"))
> formula(example_model)
cbind(incidence, size - incidence) ~ size + period + (1 | herd)
> y <- example_model$y  # matrix of "success" and "failure" counts
> trials <- rowSums(y)
> y_prop <- y[, 1] / trials  # proportions
> 
> # get predicted success proportions
> yrep <- posterior_predict(example_model)
> yrep_prop <- sweep(yrep, 2, trials, "/")
> ppc_error_binned(y_prop, yrep_prop[1:6, ])

Herd

This is the plot of average errors against herd variable.

> herd <- as.numeric(example_model$data$herd)
> ppc_error_binned(y_prop, yrep_prop[1:6, ], herd)

Size

This is the plot of average errors against size variable.

> size <- as.numeric(example_model$data$size)
> ppc_error_binned(y_prop, yrep_prop[1:6, ], size)

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.60%. Comparing base (527c48c) to head (67b863c).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #359   +/-   ##
=======================================
  Coverage   98.60%   98.60%           
=======================================
  Files          35       35           
  Lines        5650     5660   +10     
=======================================
+ Hits         5571     5581   +10     
  Misses         79       79           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jgabry]

At first glance this looks good. I will try it out tomorrow when I have more time.

My current implementation passes existing tests without any issue, so I think I am not breaking anything.

Yeah, the only reason the R-CMD-check github actions are failing is because there's a new argument that isn't documented yet: https://github.com/stan-dev/bayesplot/actions/runs/16175724060/job/45660276284?pr=359#step:6:142.

The unit tests themselves are passing so I agree that this doesn't seem to break anything.