Logistic calibration intercept and slope
Fit a recalibration line to grouped predicted probabilities and binary outcome counts.
Your entries stay in this page’s memory and are not submitted to us. Downloads are generated locally. Reloading or leaving the page clears your entries. Do not enter patient information.
Use 3–100 groups, probabilities strictly inside 0–1 and whole event/total counts. Every observation in a row must have exactly the same original predicted probability. Do not substitute a calibration bin’s mean. Rows must be disjoint independent validation observations; do not use training-set predictions.
The starting values are hypothetical. Results update as you edit. Use aggregate or synthetic inputs only.
Calculation results
- Calibration intercept
- 0
- Calibration slope
- 1
- Total validation observations
- 80
How to interpret the output
Binomial maximum likelihood fits logit(observed risk)=intercept+slope×logit(predicted risk). Ideal values are zero and one. This is joint recalibration, not calibration-in-the-large with slope fixed at one. Aggregating identical probabilities preserves this binomial fit, but bin averages for varying probabilities generally do not reproduce the individual-data fit; separation, sparse data and extrapolation can make estimates unstable. No confidence intervals are supplied. Coefficients are rounded to 12 decimal places to suppress insignificant platform-dependent numerical noise.
Method referenceRead the worked guide and interpretation examples →
Source checked 2026-11-06. Read the reproducible calculator validation report.