1
$\begingroup$

Consider a variable, say x following the normal distribution N(0,5.99). I want to translate this into a symmetric mixture of two normal distributions such that the mean of x is zero and the variance of x is the same as the one following the normal distribution.

Getting the zero mean is straight forward because one can choose a mixture of say 0.5*N(-2,A)+0.5*N(2,A), but I am stuck with how to get the variance, i.e. A in this case.

Any help on this would be greatly appreciated. Thanks

$\endgroup$
1
$\begingroup$

Take a look at the Wikipedia entry for mixtures, specifically the section on moments. The relevant formula is that the variance of a mixture of $n$ distributions with means $\mu_i$, variances $\sigma_i^2$ and weights $w_i$ is

$$ E\big((X-\mu)^2\big) = \sigma^2 = \sum_{i=1}^nw_i(\mu_i^2+\sigma_i^2)-\mu^2, $$

where $\mu$ and $\sigma^2$ are the mean and variance of the mixture.

In your specific case, you have $\mu=0$, $\sigma^2=5.99$, $w_i=\frac{1}{2}$ and (say) $\mu_i=\pm 2$, and symmetry implies that $\sigma_1^2=\sigma_2^2$. We get

$$ \sigma^2 = \frac{1}{2}(\mu_1^2+\sigma_1^2)+\frac{1}{2}(\mu_2^2+\sigma_2^2) = \mu_1^2+\sigma_1^2, $$

or

$$ \sigma_1^2=\sigma_2^2=\sigma^2-\mu_1^2 = 5.99-4 = 1.99. $$

Here is a picture and a simulation for verification:

mixture

xx <- seq(-5,5,by=.01)
plot(xx,dnorm(xx,0,sqrt(5.99)),type="l",xlab="",ylab="")
new_mean <- 2
new_var <- 5.99-new_mean^2
lines(xx,0.5*dnorm(xx,-new_mean,sqrt(new_var))+
    0.5*dnorm(xx,new_mean,sqrt(new_var)),col="red")

nn <- 1e5
var(c(rnorm(nn,-new_mean,sqrt(new_var)),rnorm(nn,new_mean,sqrt(new_var))))
# [1] 5.996673
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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