I'm developing some game theory models that involve normal distributions, and am not sure how to solve this probability problem:

Suppose there are two normal distributions: $$ X_1 \sim N(\mu_1, \sigma^2) $$ $$ X_2 \sim N(\mu_2, \sigma^2) $$

You know what both distributions are, but that's all you know. I take a draw (call it $s$) from one of the two distributions and show it to you. Given $s$, what is the probability I chose from $X_1$?

Thanks for any help!

Here's of a picture of two overlapping normal distributions in case it's helpful to have a visual:

Two overlapping normal distributions

  • 2
    $\begingroup$ What are the mixing weights? In other words, what is your prior belief that the draw will come from distribution 1 versus distribution 2? $\endgroup$
    – Adrian
    Commented Mar 2, 2016 at 21:34
  • $\begingroup$ I'm not sure. How does that affect the analysis? I guess I'd like to see the case where your prior is that I will pick from $X_2$ with probability 1, and the case where your prior is that I'll pick from either with equal probability (i.e., .5). $\endgroup$
    – sundance
    Commented Mar 2, 2016 at 21:47
  • 2
    $\begingroup$ The mixing weight would alter the prior. Using 0.5 implies you're as likely to have a point from either. You can't tell if that's reasonable from the information in your question $\endgroup$
    – Glen_b
    Commented Mar 3, 2016 at 0:12
  • $\begingroup$ The probability that the draw is from the first distribution is either $0$ or $1$ but no one except you can tell which value is the correct one. $\endgroup$ Commented Mar 4, 2016 at 3:23

3 Answers 3


As Adrian already suggested you need to know the prior probability that $X$ came from each distribution. If $Y$ is an indicator telling us whether or not $X$ came from distribution one and $p_1$ and $p_2$ are the mixing (prior) probabilities then

$$ P(Y = 1 \mid X = x) = \frac{p_1 f_{X \mid Y=1}(x)}{p_1 f_{X \mid Y=1}(x) + p_2 f_{X \mid Y=0}(x)} . $$

All you've specified are the conditional densities $f_{X \mid Y = 1}$ and $f_{X \mid Y = 0}$ but this isn't enough to calculate the probability. You also need to know something about $p_1$ and $p_2$.


The probability is the ratio of the densities.

$ P(s \text{ from }X_1) =\frac{f_{X_1}(s)}{f_{X_2}(s)} $

Intuition. Look at where the distributions cross:

enter image description here

Anything to the left of this line is more likely to come from the distribution with the smaller mean, while anything above this line is more likely to come from the distribution with the larger mean. In the picture you've shown if the value $s$ actually does come from the distribution with the smaller mean, and it falls below $\mu_1$ then we are almost certain it comes from $X_1 \sim N(\mu_1,\sigma^2)$ since the pdf of $X_2$ is basically 0 at this point (i.e. $f_{X_2}(s)\approx 0, s<\mu_1)$

  • 1
    $\begingroup$ As comments and a good reply indicate, this answer implicitly assumes the two distributions are drawn with equal probability. It is crucial to make this assumption clear. $\endgroup$
    – whuber
    Commented Jul 19, 2018 at 12:50

I'll try giving a very low-level answer. If there's no prior reason to believe that s was drawn from X1 or X2, i.e. the value of s is the only information you have, then the probability is directly obtainable from the relative values of the probability density functions evaluated at s. That is, $P(X_1 | s)\propto X_1(x=s) / X2(x=s)$. (The actual probability would require some normalization.)

For example, if s=0 in your example, it's much more likely that your draw was from the solid line distribution (value ~ 0.8) than the dashed line distribution (value ~ 0.1), and the relative probability is approximately 8:1. i.e. It's 8 times more likely that s=0 was drawn from the solid line distribution than the dashed line.

  • 1
    $\begingroup$ You write of odds as if they were probabilities--but they are not. $\endgroup$
    – whuber
    Commented Jul 19, 2018 at 12:51

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.