Statistical model building: Background “knowledge” based on inappropriate preselection causes misspecification

Table 1 Investigated simulation scenarios

Scenario	Sample size in prior			Variable selection used in prior study
	study
	n₁	n₂	n₃
S1_a	200	200	200	Univariable selection with p=0.05
S1_b				Univariable selection with p=0.2
S1_c				Backward Elimination with AIC
S2_a	500	500	500	Univariable selection with p=0.05
S2_b				Univariable selection with p=0.2
S2_c				Backward Elimination with AIC
S3_a	2000	2000	2000	Univariable selection with p=0.05
S3_b				Univariable selection with p=0.2
S3_c				Backward Elimination with AIC
S4_a	200	200	500	Univariable selection with p=0.05
S4_b				Univariable selection with p=0.2
S4_c				Backward Elimination with AIC
S5_a	200	500	500	Univariable selection with p=0.05
S5_b				Univariable selection with p=0.2
S5_c				Backward Elimination with AIC
S6_a	200	500	2000	Univariable selection with p=0.05
S6_b				Univariable selection with p=0.2
S6_c				Backward Elimination with AIC
S7_a	200	200	2000	Univariable selection with p=0.05
S7_b				Univariable selection with p=0.2
S7_c				Backward Elimination with AIC
S8_a	500	500	2000	Univariable selection with p=0.05
S8_b				Univariable selection with p=0.2
S8_c				Backward Elimination with AIC
S9_a	200	2000	2000	Univariable selection with p=0.05
S9_b				Univariable selection with p=0.2
S9_c				Backward Elimination with AIC
S10_a	500	2000	2000	Univariable selection with p=0.05
S10_b				Univariable selection with p=0.2
S10_c				Backward Elimination with AIC

ISSN: 1471-2288