linear model to logistic

linear_to_log.r

n <- 1000

nsim <- 100
res1 <- rep(0, nsim)
res2 <- rep(0, nsim)
res3 <- rep(0, nsim)

for(i in 1:nsim)
{

	x <- rbinom(n, 2, 0.5)
	y <- 0.2 * x + rnorm(n)

	cc <- as.numeric(y > quantile(y, runif(1, 0.5, 0.99)))

	prev <- sum(cc) / n
	adj <- prev * (1 - prev)


	mod1 <- coefficients(glm(cc ~ x, family="binomial"))
	mod2 <- coefficients(lm(cc ~ x))

	res1[i] <- mod1[2]
	res2[i] <- mod2[2] / adj
	res3[i] <- prev
}

library(ggplot2)

plot(res1 ~ res2)





a <- rnorm(10000)
b <- rnorm(10000)


png("pred.png", width=1500*0.7, height=500*0.7)
par(mfrow=c(1,3))
c <- b * sqrt(0.01) + a * sqrt(0.99)
plot(b ~ c, xlab="Genetic prediction", ylab="True phenotype", main="Prediction accuracy rsq = 0.01")

c <- b * sqrt(0.3) + a * sqrt(0.7)
plot(b ~ c, xlab="Genetic prediction", ylab="True phenotype", main="Prediction accuracy rsq = 0.3")

c <- b * sqrt(0.8) + a * sqrt(0.2)
plot(b ~ c, xlab="Genetic prediction", ylab="True phenotype", main="Prediction accuracy rsq = 0.8")
dev.off()