R/ratio.mean.ps.R

In cwendorf/dgb: DGB

# DGB
## Ratio of Means from Paired Samples

ci.ratio.mean.ps.data <- function(alpha, y1, y2){
 # Compute confidence interval for a ratio of population
 # means of ratio-scale measurements in a paired-samples 
 # design. Equality of variances is not assumed.
 # Arguments:
 #   alpha:  alpha level for 1-alpha confidence
 #   y1:     vector of scores for condition 1
 #   y2:     vector of scores for condition 2
 # Values:
 #   means, mean ratio, lower limit, upper limit
 n <- length(y1)
 m1 <- mean(y1)
 m2 <- mean(y2)
 v1 <- var(y1)
 v2 <- var(y2)
 cor <- cor(y1,y2)
 var <- (v1/m1^2 + v2/m2^2 - 2*cor*sqrt(v1*v2)/(m1*m2))/n
 df <- n - 1
 tcrit <- qt(1 - alpha/2, df)
 est <- log(m1/m2)
 se <- sqrt(var)
 ll <- exp(est - tcrit*se)
 ul <- exp(est + tcrit*se)
 out <- t(c(m1, m2, exp(est), ll, ul))
 colnames(out) <- c("Mean1", "Mean2", "Mean1/Mean2", "LL", "UL")
 return(out)
}

size.ci.ratio.mean.ps <- function(alpha, var, m1, m2, cor, r) {
 # Computes sample size required to estimate a ratio of population  
 # means with desired precision in a paired-samples design
 # Arguments: 
 #   alpha:  alpha level for 1-alpha confidence 
 #   var:    planning value of measurement variance
 #   m1:     planning value of mean for measurement 1
 #   m2:     planning value of mean for measurement 2
 #   cor:    planning value for measurement correlation 
 #   r:      desired upper to lower confidence interval endpoint ratio
 # Values:
 #   required sample size
 z <- qnorm(1 - alpha/2)
 n <- ceiling(8*var*(1/m1^2 + 1/m2^2 - 2*cor/(m1*m2))*(z/log(r))^2 + z^2/2)
 return(n)
}