fssem1
commented
5 years ago Priest email from 5-13-2019 From running the Shapiro-Wilks test on the 8 permutations of the 2 rivers, 2 variables, and 2 ages: 4 are normal, 1 is marginally normal, and 3 are not normal. Summary table below:
Of these not normal variables, here's what they look like before transformation with a normal curve overlain (with its own mean, sigma):
Visually, these don't look like severe violations of normality as they aren't bimodal, the means are fairly close to where they "should" be, and the tails match fairly closely. I used Box Cox test on these for optimal transformations; the histograms look very similar with the same number of issues.
test for normality
eda.norm <- function(x, ...) { par(mfrow=c(2,2)) if(sum(is.na(x)) > 0) warning("NA's were removed before plotting") x <- x[!is.na(x)] hist(x, main = "Histogram and non-\nparametric density estimate", prob = T) iqd <- summary(x)[5] - summary(x)[2] lines(density(x, width = 2 * iqd)) boxplot(x, main = "Boxplot", ...) qqnorm(x) qqline(x) plot.ecdf(x, main="Empirical and normal cdf") LIM <- par("usr") y <- seq(LIM[1],LIM[2],length=100) lines(y, pnorm(y, mean(x), sqrt(var(x)))) shapiro.test(x) }