What is the difference between the borderline group and borderline regression methods?
The borderline group method sets the pass mark as the mean checklist score of candidates rated borderline by examiners. Borderline regression instead fits a regression of checklist scores on global ratings across all candidates and predicts the score at the borderline point. Regression uses the full dataset, so it is more stable with small cohorts.
Both methods belong to the same family: examinee-centered standard setting for OSCEs, in which examiners score each candidate on a checklist and assign a global rating (e.g., fail / borderline / pass / good / excellent). The difference is what happens to those data afterwards.
Borderline group method
The borderline group method takes only the candidates whose global rating was "borderline" at a station and sets the station's pass mark at the mean (or median) checklist score of that group. The logic is direct: these are the performances examiners genuinely couldn't classify as clear pass or clear fail, so their average score is a natural cut point.
Its weakness is statistical. The cut score rests entirely on the borderline subgroup, which may be a handful of candidates — or none — at a well-taught station. Small groups make the cut score volatile from year to year, and a few atypical borderline candidates can move it substantially.
Borderline regression method
Borderline regression fits a linear regression of checklist scores on global ratings using every candidate at the station, then takes the predicted checklist score at the borderline grade as the pass mark. All the data anchor the line, so the estimate is markedly more stable, and the regression's R-squared doubles as a station quality metric (Pell, Fuller, Homer & Roberts, 2010). The mechanics are covered in what is the borderline regression method?
Side-by-side
Borderline group | Borderline regression | |
|---|---|---|
Data used | Borderline-rated candidates only | All candidates |
Cut score | Mean/median score of borderline group | Predicted score at borderline grade |
Stability with small cohorts | Poor — may have few or zero borderline candidates | Better — whole cohort anchors the line |
Statistical assumptions | Minimal | Linearity between global rating and score; sensitive to outliers and range restriction |
Quality by-products | Little | R-squared, residual diagnostics per station |
Computation | Trivial | Simple regression, easily automated |
Which should you use?
For most institutions running OSCEs with typical cohort sizes, borderline regression is the default recommendation in the current literature and the more defensible choice: it wastes no data, produces auditable diagnostics, and is easy to automate in modern assessment software. The borderline group method remains a reasonable, transparent fallback when regression assumptions are clearly violated, or as a cross-check on the regression-derived standard.
Whichever you choose, both depend on examiners giving genuine, independent global judgments — which makes examiner training part of the standard-setting method, not an optional extra. For the wider menu of methods, including test-centered options, see OSCE standard-setting methods compared.