Model binary outcomes with resistant weighting and diagnostics. Inspect residuals, leverage, odds, and threshold behavior. Turn noisy observations into steadier statistical guidance for decisions.
Use supplied coefficients and case values to evaluate binary probability, odds, residuals, leverage, and resistant observation weights for a single modeled record.
| Case | Observed Y | Age Index | Risk Score | Exposure | History Count | Comment |
|---|---|---|---|---|---|---|
| 1 | 1 | 2.4 | 1.1 | 0.8 | 3.2 | Moderate leverage case |
| 2 | 0 | 1.5 | 0.4 | 1.9 | 1.1 | Low predicted event chance |
| 3 | 1 | 3.1 | 1.8 | 1.0 | 4.7 | Strong positive signal |
| 4 | 0 | 0.9 | 2.7 | 2.4 | 0.6 | Potential outlier pattern |
| 5 | 1 | 2.0 | 0.7 | 1.3 | 2.2 | Balanced central case |
Use these rows as a template for coefficient testing or manual observation review. Map your own field names into the four predictor labels above.
Robust logistic procedures use reduced weights for large standardized residuals. That protects coefficient updates from highly unusual observations while preserving regular cases.
It evaluates one observation using supplied logistic coefficients, predictor values, and a resistant weighting rule. The page reports probability, class assignment, residual diagnostics, and weighted score contributions.
No. It does not estimate coefficients from a dataset. It applies diagnostic and weighting logic to a single case after you provide coefficients from prior modeling work.
Huber weighting is useful when you want moderate protection against unusual cases without fully discarding them. Large residuals are downweighted smoothly instead of being cut off sharply.
Tukey biweight is more aggressive. Observations beyond the cutoff receive zero weight. Use it when strong outlier resistance matters more than preserving influence from extreme cases.
Leverage adjusts residual standardization. A case with high leverage can distort coefficient updates even when its raw residual looks moderate, so standardized diagnostics give a fairer warning signal.
The weighted score shows each variable’s contribution to resistant estimation at that case. Values near zero imply little pressure on coefficient updates, while large values suggest stronger local influence.
This version is designed for four predictors plus an intercept to keep the interface readable. You can expand the repeating predictor blocks in the file if your project needs more fields.
A very low weight means the case is being strongly discounted because its standardized residual is extreme under the chosen tuning rule. Investigate data quality, specification issues, or genuine rare behavior.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.