If e.g. 100 people can rate a subject either 0 or 1, than the dispersion (e.g. standard deviation) among the 100 raters is potentially largest for a mean of 0.5 (50 people rate 0, 50 people rate 1) while it is smallest (actually 0) at both extremes (i.e. when the mean is 0 or 1). The relation between the mean (ranging from 0 to 1) and the associated standard deviation (ranging from 0 to 0.5) follows a curve. Here some R-code to illustrate what I mean:

N <- 100
res_list <- list()
for(i in 1:(N-1)){
  N1 <- i
  N2 <- N-i
  x <- c(rep(0,N1),rep(1,N2))
  res_list[[i]] <- c(N1=i-1,N2=N-i,sd=sd(x),mean=mean(x))
res_df <- as.data.frame(do.call(rbind,res_list))

plot(res_df$mean,res_df$sd,xlab="mean",ylab="standard deviation")

Mean vs. standard deviation

Is there a mathematical function that describes exactly that relationship (independent from N)? Or is there a special term referring to this mean-vs-sd relationship for censored (0-1) data?

  • $\begingroup$ See Variance of a bounded random variable. Also FYI "bell shaped" is typically used to describe a normal distribution, I would probably just call your curve a parabola. $\endgroup$
    – Andy W
    Commented Mar 22, 2015 at 14:04
  • 2
    $\begingroup$ @AndyW variance would be a parabola, but sd vs mean will be a semi-circle (or a semi-ellipse, if you don't use the same scaling on both axes). $\endgroup$
    – Glen_b
    Commented Mar 22, 2015 at 14:10

1 Answer 1


Suppose the mean of $X$ is $\mu$ and $0 \le X \le 1$. As you state, the variance is maximized if $P(X=1) = \mu$ and $P(X = 0) = 1 - \mu$. In that case the variance of $X$ is

$$\begin{align}E((X - \mu)^2) &= P(X=1)(1 - \mu)^2 + P(X=0)\mu^2 \\ &= \mu(1-\mu)^2 + (1-\mu)\mu^2 \\ &= \mu(1-\mu)\end{align}$$

The standard deviation is just the square root: $\sigma = \sqrt{\mu(1-\mu)}$.

  • $\begingroup$ For completeness: we call distribution of such $X$'s a Bernoulli distribution. $\endgroup$
    – Tim
    Commented Oct 5, 2015 at 13:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.