diff --git a/Project.toml b/Project.toml
index e18179f..59111d2 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "Lighthouse"
 uuid = "ac2c24cd-07f0-4848-96b2-1b82c3ea0e59"
 authors = ["Beacon Biosignals, Inc."]
-version = "0.14.18"
+version = "0.14.19"
 
 [deps]
 ArrowTypes = "31f734f8-188a-4ce0-8406-c8a06bd891cd"
diff --git a/src/metrics.jl b/src/metrics.jl
index 0eeadf0..f5333c7 100644
--- a/src/metrics.jl
+++ b/src/metrics.jl
@@ -70,9 +70,9 @@ function binary_statistics(confusion::AbstractMatrix, class_index::Integer)
     actual_positives = sum(view(confusion, :, class_index))
     actual_negatives = total - actual_positives
     true_positives = confusion[class_index, class_index]
-    true_negatives = sum(diag(confusion)) - true_positives
     false_positives = predicted_positives - true_positives
     false_negatives = actual_positives - true_positives
+    true_negatives = actual_negatives - false_positives
     true_positive_rate = (true_positives == 0 && actual_positives == 0) ?
                          (one(true_positives) / one(actual_positives)) :
                          (true_positives / actual_positives)
diff --git a/test/metrics.jl b/test/metrics.jl
index 1195985..7e5f3bc 100644
--- a/test/metrics.jl
+++ b/test/metrics.jl
@@ -10,20 +10,26 @@
     @test accuracy(c) == percent_agreement == 3 / 8
     @test kappa == (3 / 8 - chance) / (1 - chance)
     stats = binary_statistics(c, 3)
+    total = sum(c)
+    @test total == 8
     @test stats.predicted_positives == 2
     @test stats.predicted_negatives == 6
     @test stats.actual_positives == 2
     @test stats.actual_negatives == 6
     @test stats.true_positives == 1
-    @test stats.true_negatives == 2
+    @test stats.true_negatives == 5
     @test stats.false_positives == 1
     @test stats.false_negatives == 1
     @test stats.true_positive_rate == 0.5
-    @test stats.true_negative_rate == 1 / 3
+    @test stats.true_negative_rate == 5 / 6
     @test stats.false_positive_rate == 1 / 6
     @test stats.false_negative_rate == 0.5
     @test stats.precision == 0.5
     @test stats.f1 == 0.5
+    @test stats.true_positives + stats.true_negatives + stats.false_positives +
+          stats.false_negatives == total
+    @test stats.actual_positives + stats.actual_negatives == total
+    @test stats.predicted_positives + stats.predicted_negatives == total
 
     labels = rand(StableRNG(42), 1:3, 100)
     hard_label_pairs = zip(labels, labels)