Also known as Youden’s J statistic or Peirce Skill Score (PSS), Hanssen–Kuipers Discriminant (HKD), or True Skill Statistic (TSS) but I’ve only ever seen the two Youden terms used in therapy research and neither of them is all that common.
Details #
It’s a measure of agreement between two binary (yes/no) measures of the presence of something and typically used to assess the quality of one measure as an alternative to another or to assess the quality of one measure against a gold standard where you know the what the correct answer should be.
That typically comes up in screening: using a binary measure to indicate whether something is present or not. A good screening measure (see the entry about screening) has high sensitivity, i.e. detects a high proportion of “true cases” and has high specificity, i.e. correctly identifies a high proportion of “true non-cases” as such. Youden’s index is rather nicely sensitivity + specificity – 1 and has a theoretical possible range from -1 (incorrect every time) to +1 where the test gives the correct answer every time.
When using the receiver operating curve (or characteristic; ROC) to look for a cutting point on a continuous measure trying to find the best to use in screening or just in designating things/people as “cases” or “not cases” the vertical distance from the point on the ROC curve to the leading diagonal has the value of Youden’s statistic if that point is used as the cutting point.
Try also #
Chapters #
Not covered in the OMbook.
Online resources #
None currently.
Dates #
First created 18.vii.25, tweaked after adding glossary entry for the ROC, 20.vii.25.
