Diagnostics Reference¶

All regression tasks include a diagnostics sub-object in their mid-artifact output. These diagnostics are computed locally at each site and provide privacy-safe summary statistics (no individual-level data is shared).

Common Diagnostics¶

Field	Description
`diagnostics.vif`	Variance Inflation Factor for each feature -- values > 10 suggest multicollinearity
`diagnostics.residual_summary`	Summary statistics of residuals: mean, std, min, q25, median, q75, max
`diagnostics.cooks_distance`	Cook's distance summary: max, mean, n_influential (count > 4/n threshold)

OLS / Linear Regression¶

Field	Description
`diagnostics.shapiro_wilk`	Shapiro-Wilk normality test on residuals (statistic, p_value)
`diagnostics.prediction_intervals`	Summary of CI and PI widths (ci_width_mean, pi_width_mean, etc.)

Logistic Regression¶

Field	Description
`diagnostics.hosmer_lemeshow`	Hosmer-Lemeshow goodness-of-fit test (statistic, p_value, df)
`diagnostics.deviance_residual_summary`	Summary of deviance residuals
`diagnostics.prediction_intervals`	Summary of predicted probability CI widths

Poisson / Negative Binomial¶

Field	Description
`diagnostics.overdispersion`	Overdispersion ratio and p_value -- ratio > 1 suggests overdispersion
`diagnostics.deviance_residual_summary`	Summary of deviance residuals
`diagnostics.pearson_residual_summary`	Summary of Pearson residuals
`diagnostics.prediction_intervals`	Summary of predicted rate CI widths

Cox Proportional Hazards¶

Field	Description
`diagnostics.proportional_hazards_test`	Schoenfeld residuals test -- per-feature test_statistic and p_value
`diagnostics.deviance_residual_summary`	Summary of deviance residuals

Censored Regression (Tobit)¶

Field	Description
`diagnostics.vif`	Variance Inflation Factor for each feature
`diagnostics.residual_summary`	Residual summary for observed (uncensored) data points
`diagnostics.shapiro_wilk`	Normality test on residuals of observed data
`diagnostics.censoring_summary`	Counts: n_observed, n_right_censored, n_left_censored, pct_censored
`diagnostics.aic`	Akaike Information Criterion
`diagnostics.bic`	Bayesian Information Criterion

Multiple Imputation (MICE)¶

Field	Description
`diagnostics.vif`	VIF computed on first imputed dataset
`diagnostics.residual_summary`	Residual summary from first imputed dataset's OLS fit
`diagnostics.shapiro_wilk`	Normality test on first imputed dataset's residuals

R Tasks¶

All R-based tasks include equivalent diagnostics via the shared r_diagnostics_utils.R module. The output fields follow the same structure as their Python counterparts.