Given a vector of true_values and one or more vectors of test values
(passed to ...), computes sensitivity, specificity, positive predictive
value (PPV), negative predictive value (NPV), and Cohen's kappa.
Examples
# create example data w predictors with different properties:
ex_data <- tibble::tibble(
actual = rbinom(250, 1, .3), # 250 cases, p(outcome) = .3
prediction1 = ifelse(runif(250) <= .05, 1L - actual, actual), # 5% error rate
prediction2 = ifelse(runif(250) <= .15, 1L - actual, actual), # 15% error rate
prediction3 = ifelse(runif(250) <= .35, 1L - actual, actual), # 35% error rate
prediction4 = ifelse(runif(250) <= .15, 1L, actual), # 15% with positive bias
prediction5 = ifelse(runif(250) <= .15, 0L, actual) # 15% with negative bias
)
# testing predicted v actual values
ex_data %>%
accuracy_stats(actual, prediction1)
#> # A tibble: 1 × 7
#> Predictor n Kappa Sensitivity Specificity PPV NPV
#> <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 prediction1 250 0.845 0.901 0.947 0.890 0.952
# can test multiple predictors simultaneously
ex_data %>%
accuracy_stats(actual, prediction1:prediction5)
#> # A tibble: 5 × 7
#> Predictor n Kappa Sensitivity Specificity PPV NPV
#> <fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 prediction1 250 0.845 0.901 0.947 0.890 0.952
#> 2 prediction2 250 0.628 0.840 0.822 0.694 0.914
#> 3 prediction3 250 0.213 0.593 0.639 0.440 0.766
#> 4 prediction4 250 0.828 1 0.882 0.802 1
#> 5 prediction5 250 0.846 0.802 1 1 0.914
# if `include_counts` = TRUE, will also return n of false positives,
# false negatives, etc., as well as and observed and expected % agreement
ex_data %>%
accuracy_stats(actual, prediction1:prediction5, include_counts = TRUE)
#> # A tibble: 5 × 13
#> Predictor n TP FP TN FN pAgreeObserved pAgreeExpected Kappa
#> <fct> <dbl> <int> <int> <int> <int> <dbl> <dbl> <dbl>
#> 1 prediction1 250 73 9 160 8 0.932 0.561 0.845
#> 2 prediction2 250 68 30 139 13 0.828 0.538 0.628
#> 3 prediction3 250 48 61 108 33 0.624 0.523 0.213
#> 4 prediction4 250 81 20 149 0 0.92 0.534 0.828
#> 5 prediction5 250 65 0 169 16 0.936 0.584 0.846
#> # ℹ 4 more variables: Sensitivity <dbl>, Specificity <dbl>, PPV <dbl>,
#> # NPV <dbl>