Investigating the relationship between predictability and imbalance in minimisation: a simulation study

Table 5 Summary of observations from the four-treatment case

	Predictability at P= 1.0, %	N^a	Recommended P-value	Predictability at this P-value, %	Reduction in predictability, %	P-value at which imbalance occurs
One variable
Two categories (equal prevalence)	24 to 25	100	0.7^b	24	0	0.6
		≥200	0.7 ^b	25	0	-
Two variables
Both with 2 categories (equal prevalence)	45	100	0.9	44	1	0.8
		200	0.7	41	4	0.6
		≥300	0.7	41	4	-
Three variables
All with 2 categories (equal prevalence)	58	100	1.0	58	0	1.0
		200	0.7	49	9	0.6
		300	0.7	49	9	0.5
		≥400	0.7	49	9	-
Four variables
All with 2 categories (equal prevalence)	67	100	1.0	67	0	1.0
		200	0.8	59	8	0.7
		300	0.7	53	14	0.5
		≥400	0.7	53	14	-

^aThe categories of N are dependent upon the point at which imbalance is observed (the value of probability of assignment P). ^bFor one prognostic variable the reduction in predictability is so small as the probability of assignment P is reduced that the recommended P- value is 1.0.

ISSN: 1745-6215