Table.

Variation in the Occurrence of Alcohol-Related Discussions Attributable to Differences between Patients versus Differences between Physicians

	Variables included in model	Sum of Squared Residuals (SSR)	Difference in SSR from base model	Approximate % Variation in Discussions
Model 1 (base model)	Regression with constant term but no additional predictors (determines total variance)	343.0	….
Model 2	Adding only measured physician characteristics (e.g., age, sex, specialty)	340.7	• (343.0 − 340.7) = 2.3 • 2.3 / 343.0 = 0.7% from measured physician characteristics	0.7% + 3.8% = ~4.5% due to all differences between physicians (measured + unmeasured)
Model 3 (physician fixed-effects model)	Adding only unmeasured physician characteristics	326.8	• (343.0 − 326.8) = 13.2 • 13.2/343.0 = 3.8% from unmeasured physician characteristics
Model 4	Adding measured physician characteristics and measured patient characteristics	281.9	• 343.0 − 281.9 − 2.3 / 343.0 = 17.1% from measured patient characteristics	17.1% + 78.4% = ~95.5% due to all differences between patients (measured + unmeasured), plus measurement error
Remainder of variation (from Model 5)*			• (343.0 − 2.3 − 13.2 − 58.8) / 343.0 = 78.4% from unmeasured patient characteristics + error

^*(Total variation – component from measured physician characteristics − component from unmeasured physician characteristics − component from measured patient characteristics) = component from unmeasured patient characteristics + error