Triage a stacked set of design checks

Applies the standard "what needs a human look" filters to the output of stack_checks and returns only the rows that failed. Elements absent from checks are skipped, so the same call works whether the pipeline ran check_attrition, an attrition_lasso element from estimatrTools::check_attrition_lasso, or neither.

Usage

report_checks(checks, alpha = 0.05)

Arguments

checks

A named list of stacked check tibbles, as returned by stack_checks.

alpha

Rejection threshold for the balance and attrition filters (default 0.05).

Value

An object of class "excheckr_report": a named list of tibbles holding only the failing rows, with an alpha attribute. Elements with nothing to report are present but empty. Has a print method that shows the counts.

Details

The filters are:

out_of_bounds

ybounds rows where in_bounds is FALSE: an outcome outside the expected [0, 1] range.

missing_no_nona

missingness rows with missing values and no imputed companion column: covariates that will silently drop rows from every adjusted model.

nona_still_missing

missingness rows whose imputed companion column still contains NA: the imputation did not take.

balance_joint

balance_joint rows with p_value at or below alpha. A single balance element produced by check_balance(flatten = TRUE) is split on its test column and handled the same way.

balance_covariate

balance_covariate rows with p_value at or below alpha. Expect roughly alpha of these to fail by chance; use summarize_check_pvalues to judge whether there are more than chance would give.

attrition

attrition or attrition_lasso rows flagged at alpha. When the element carries a logical flag column, that column is used instead and alpha is ignored, because the producer already applied a threshold of its own.

See also

Other across-study summaries: plot_check_pvalues(), stack_checks(), summarize_check_pvalues()

Examples

checks <- list(
  ybounds = data.frame(study_id = c("a", "b"), variable = "Y_x",
                       min = c(0, -1), max = c(1, 3),
                       in_bounds = c(TRUE, FALSE)),
  balance_joint = data.frame(study_id = c("a", "b"), p_value = c(0.4, 0.01))
)
report_checks(checks)
#> excheckr triage report (alpha = 0.05)
#>       1  outcomes outside [0, 1]
#>       1  joint balance tests flagged
#> Inspect an element with report$<name>.