Performs a series of tests to ensure that assumptions for logistic regression are met, with optional detailed feedback if any tests fail.
Arguments
- glm_model_results
Results from a binomial Generalised Linear Model (GLM), as produced by
stats::glm().- confint_fast_estimate
Boolean (default =
FALSE) Use a faster estimate of the confidence interval? Note: this assumes normally distributed data, which may not be suitable for your data.- details
Boolean (default =
TRUE) Show detailed feedback for any failed tests?
Examples
# Load the Titanic dataset
df <- datasets::Titanic |>
dplyr::as_tibble() |>
# convert aggregated counts to individual observations
dplyr::filter(n > 0) |>
tidyr::uncount(weights = n) |>
# convert character variables to factors
dplyr::mutate(dplyr::across(dplyr::where(is.character), as.factor))
# Perform logistic regression using `glm`
lr <- stats::glm(
data = df,
family = binomial,
formula = Survived ~ Class + Sex + Age
)
# Check the model for logistic regression assumption violations
check_or(lr)
#>
#> ── Assumption checks ───────────────────────────────────────────────────────────
#>
#> ── Summary ──
#>
#> ✔ The outcome variable is binary
#> ✔ The predictor variables are not highly correlated with each other
#> ✔ The outcome is not separated by predictors
#> ✔ The sample size is large enough
#>
#> Your model was checked for logistic regression assumptions in the following
#> areas:
#>
#> Binary outcome:
#> The outcome variable was checked for containing precisely two levels.
#>
#> Multicollinearity:
#> The `vif()` function from the car package was used to check for highly
#> correlated predictor variables.
#>
#> Separation:
#> The `detectseparation()` function from the detectseparation package was used to
#> check for complete or quasi-complete separation in the data.
#>
#> Sample size:
#> A rule of thumb was applied, requiring at least 10 events per predictor
#> variable and at least 10 events per level of categorical variables to ensure
#> sufficient data for reliable estimates.
#>
#> ✔ These tests found no issues with your model.
#>