R/stdmean.ps.R

Defines functions size.ci.stdmean.ps ci.stdmean.ps

Documented in ci.stdmean.ps size.ci.stdmean.ps

# DGB
## Standardized Mean Difference from Paired Samples

ci.stdmean.ps <- function(alpha, m1, m2, sd1, sd2, n, cor) {
 # Computes confidence interval for a population standardized mean 
 # difference in a paired-samples design
 # Arguments: 
 #   alpha: alpha level for 1-alpha confidence
 #   mj:    sample mean in condition j
 #   sdj:   sample standard deviation in condition j
 #   n:     sample size
 #   cor:   sample correlation
 # Values:
 #   estimate, SE, lower limit, and upper limit for equal variance and
 #   unequal variance methods plus single condition standardizer
 z <- qnorm(1 - alpha/2)
 s <- sqrt((sd1^2 + sd2^2)/2)
 df <- n - 1
 v1 <- sd1^2
 v2 <- sd2^2
 vd <- v1 + v2 - 2*cor*sd1*sd2
 est1 <- (m1 - m2)/s
 se1 <- sqrt(est1^2*(v1^2 + v2^2 + 2*cor^2*v1*v2)/(8*df*s^4) + vd/(df*s^2))
 ll1 <- est1 - z*se1
 ul1 <- est1 + z*se1
 se2 <- sqrt(est1^2*(1 + cor^2)/(4*df) + 2*(1 - cor)/n)
 ll2 <- est1 - z*se2
 ul2 <- est1 + z*se2
 est3 <- (m1 - m2)/sd1
 se3 <- sqrt(est3^2/(2*df) + vd/(df*v1))
 ll3 <- est3 - z*se3
 ul3 <- est3 + z*se3
 est4 <- (m1 - m2)/sd2
 se4 <- sqrt(est4^2/(2*df) + vd/(df*v2))
 ll4 <- est4 - z*se4
 ul4 <- est4 + z*se4
 out1 <- t(c(est1, se1, ll1, ul1))
 out2 <- t(c(est1, se2, ll2, ul2))
 out3 <- t(c(est3, se3, ll3, ul3))
 out4 <- t(c(est4, se4, ll4, ul4))
 out <- rbind(out1, out2, out3, out4)
 colnames(out) <- c("Estimate", "SE", "LL", "UL")
 rownames1 <- c("Equal Variances Not Assumed:", "Equal Variances Assumed:")
 rownames2 <- c("Condition 1 Standardizer:", "Condition 2 Standardizer:")
 rownames(out) <- c(rownames1, rownames2)
 return(out)
}

size.ci.stdmean.ps <- function(alpha, d, cor, w) {
 # Computes sample size required to estimate a population standardized 
 # mean difference with desired precision in a paired-samples design 
 # Arguments: 
 #   alpha: alpha level for 1-alpha confidence
 #   d:     planning value of standardized mean difference  
 #   cor:   planning value of correlation
 #   w:     desired confidence interval width
 # Values:
 #   required sample size
 z <- qnorm(1 - alpha/2)
 n <- ceiling(4*(d^2*(1 + cor^2)/4 + 2*(1 - cor))*(z/w)^2)
 return(n)
}
cwendorf/dgb documentation built on May 3, 2022, 9:35 p.m.