COMPOSITE VARIABLES 1 2 Supplemental Digital Content 1 The derivation of power calculation formula for composite score method 3 4 1 Let denote the effect size of linear model E( , where 5 6 that are standardized by and in the and with least square estimators and . The power calculation formula used by Dupont and Plummer (1998) is 7 (A1) 8 where 9 denote its ( is the cumulative probability of the distribution with ) percentile with right tail probability degrees of freedom and . However, 10 data and may not be useful in the study design stage. One may replace 11 standard deviation of and The variance 13 since can be shown as 15 17 in by can further be shown as and . On the other hand, with covariance Accordingly, the effect size 16 and of , respectively, where 12 14 in (A1) depends on . can be approximated by , and the following power calculation formula can be used in the research design: 18 19 where 20 percentile at 21 The derivation of power calculation formula for Bonferroni correction is the cumulative probability of the standard normal distribution with ( . ) COMPOSITE VARIABLES 22 23 Let denote the correlation matrix between correlation vector between 24 2 and Let , and let denote the denote the bivariate element of the vector be a column vector of standardized covariates from 25 and . Let denote the vector of the least square estimator for the linear model 26 27 Given 28 to 29 method, if we define the effect size for 30 finite sample variance of the least square estimator when the sample size is large. Similarly as is , which is close defined in the composite score as then one can express as 31 32 by the fact that 33 standardization, one can write down the power calculation formula for . Since a p-value for individual is not affected by as 34 35 The power for multiple testing using Bonferroni correction is therefore 36 with 37 50, and 38 39 40 41 . That means, for each . For example, if , then and statistical power equals 0.0876 or 8.76%. , with a sample size n = and for each under Bonferroni correction. Accordingly, the COMPOSITE VARIABLES 42 43 44 45 3 Reference Dupont, W.D., & Plummer, W.D.Jr. (1998). Power and sample size calculations for studies involving linear regression. Controlled Clinical Trials, 19(6), 589-601.