Table 5.

Key Findings. Effects of measurement heterogeneity on predictive performance in general scenarios of measurement heterogeneity. The scenarios were defined by generating different qualities of measurement across settings using the general measurement error model in Equation (1). Measurements in the derivation set corresponded to the random measurement error model (Equation 1), ie, under ψ _D = 0 and θ _D = 1.0. Using similar logic, all patterns can be translated to differential measurement of cases and noncases (ie, when ψ ₁ ≠ ψ ₀ and/or θ ₁ ≠ θ ₀ and/or ${\sigma }_{ϵ1}^2\ne {\sigma }_{ϵ0}^2$)

		Predictive Performance at Validation
Predictor Measurements at Validation		Discrimination	Calibration‐in‐the‐large	Calibration Slope	Overall Accuracy
Less precise compared to derivation;	${\sigma }_{ϵ(D)}^2<{\sigma }_{ϵ(V)}^2$	Deteriorated	‐	b < 1	Deteriorated
More precise compared to derivation;	${\sigma }_{ϵ(D)}^2>{\sigma }_{ϵ(V)}^2$	Improved	‐	b > 1	Improved
Weaker association with actual predictor value, while
‐ less precise compared to derivation;	θ V < 1.0, ${\sigma }_{ϵ(D)}^2<{\sigma }_{ϵ(V)}^2$	Stronger deterioration	‐	Stronger b < 1	Stronger deterioration
‐ more precise compared to derivation;	θ V < 1.0, ${\sigma }_{ϵ(D)}^2>{\sigma }_{ϵ(V)}^2$	Less improvement	‐	Stronger b > 1	Less improvement
Stronger association with actual predictor value, while
‐ less precise compared to derivation;	θ V > 1.0, ${\sigma }_{ϵ(D)}^2<{\sigma }_{ϵ(V)}^2$	Less deterioration	‐	Less b < 1	Less deterioration
‐ more precise compared to derivation;	θ V > 1.0, ${\sigma }_{ϵ(D)}^2>{\sigma }_{ϵ(V)}^2$	Stronger improvement	‐	Less b > 1	Stronger improvement
Increased by a constant relative to derivation.	ψ V > 0	‐	a < 0	‐	‐