Using marginal standardisation to estimate relative risk without dichotomising continuous outcomes

Table 1 Simulation results for model diagnostics when α₁ = 0

		ε~Normal(0, 1)	ε~Logistic(0, 1)
Model	Percentiles	Proportion of rejection at 5% significance level	Proportion of rejection at 5% significance level
Normal Linear Model ^a	NA ^c	0.053	0.822
Probit Model ^b	0.075	0.037	0.043
	0.15	0.039	0.032
	0.3	0.041	0.042
	0.5	0.033	0.038
	0.7	0.028	0.039
	0.85	0.04	0.044
	0.925	0.038	0.029
Logistic Linear Model ^a	NA ^c	0.052	0
Logit Model ^b	0.075	0.039	0.032
	0.15	0.036	0.039
	0.3	0.036	0.038
	0.5	0.038	0.034
	0.7	0.034	0.034
	0.85	0.042	0.052
	0.925	0.028	0.041

^aLilliefors corrected Kolmogorov-Smirnov test were used to test whether residuals from normal linear model had a normal distribution, and Kolmogorov-Smirnov test was used to test whether residuals from logistic linear models had a logistic distribution. ^bPregibon link test was used to test whether probit or logit link was appropriate. cThe same normal (or logistic) linear model is applied across different threshold values.

ISSN: 1471-2288