Skip to main content

Table 4 Estimated regression coefficients for nine predictors of dependence in daily activities obtained from a subsample of size N = 275 by applying various versions of tuned ridge regression, ridge regression based on informative priors (IP), ridge regression based on weakly informative priors (WP) or Firth’s logistic regression with intercept-correction (FLIC). Tuning criteria: D deviance, GCV, generalized cross-validation, CE, classification error, RCV50, repeated 10-fold cross-validated deviance with θ = 0.5, RCV95, repeated 10-fold cross-validated deviance with θ = 0.95, AIC, Akaike’s information criterion. Calibration slopes were calculated on a validation dataset of size N = 1539

From: To tune or not to tune, a case study of ridge logistic regression in small or sparse datasets

Method Estimated coefficients Calibration slope
  \( {\hat{\beta}}_0 \) \( {\hat{\beta}}_{\mathrm{age}} \) \( {\hat{\beta}}_{\mathrm{sex}} \) \( {\hat{\beta}}_{\mathrm{BMI}} \) \( {\hat{\beta}}_{\mathrm{APGAR}} \) \( {\hat{\beta}}_{\mathrm{CD}} \) \( {\hat{\beta}}_{\mathrm{fall}} \) \( {\hat{\beta}}_{\mathrm{lonliness}} \) \( {\hat{\beta}}_{\mathrm{health}} \) \( {\hat{\beta}}_{\mathrm{pain}} \)  
D −4.79 0.05 −0.43 −0.25 1.25 0.57 2 0.04 −0.20 0.11 1.09
GCV −4.53 0.05 −0.35 −0.22 1.14 0.54 1.90 0.05 −0.18 0.10 1.16
CE −5.71 0.07 −0.72 −0.36 1.59 0.64 2.25 0.03 −0.25 0.13 0.89
RCV50 −4.85 0.06 −0.45 −0.26 1.27 0.57 2.02 0.04 −0.20 0.11 1.07
RCV95 −4.53 0.05 −0.35 −0.22 1.14 0.54 1.90 0.05 −0.18 0.10 1.16
AIC −5.59 0.07 −0.69 −0.35 1.55 0.63 2.22 0.03 −0.24 0.13 0.91
IP −5.33 0.07 −0.61 −0.32 1.46 0.61 2.16 0.03 −0.23 0.12 0.96
WP −5.89 0.08 −0.78 −0.38 1.65 0.65 2.29 0.02 −0.26 0.14 0.87
FLIC −5.84 0.07 −0.77 −0.37 1.61 0.60 2.16 0.02 −0.25 0.13 0.91