# Transformation of random variables and joint distributions

Given a variable $$y_i$$, normally distributed with 0 mean and $$σ_y$$ standard deviation

$$y_i$$ ~ NormalDistribution[0,$$σ_y$$ ]

I want to obtain with Mathematica:

1. The distribution of: $$x = \bar{y} = \frac {\sum_{i=1}^ny_i}{n}$$

2. The joint distribution of $$(x,y_i )$$

• What have you tried? For example, have you seen the documentation on TransformedDistribution and ProbabilityDistribution?
– JimB
Mar 24, 2019 at 16:58
• @JimB . I tried this TransformedDistribution[Sum[y, {i, n}]/n, y \[Distributed] NormalDistribution[0, \[Sigma]y]]. The result is NormalDistribution[0, \[Sigma]y]. However, the correct result should be NormalDistribution[0, \[Sigma]y / Sqrt[n]] Mar 24, 2019 at 17:42
• You need to "index" the variable y or else Mathematica thinks it is a single variable.
– JimB
Mar 24, 2019 at 22:04

I don't know how to get Mathematica to get the joint distribution explicitly for a general value of $$n$$ but here is how one can easily see the pattern to figure out the general solution.

First the distribution of the mean:

marginalDistribution = TransformedDistribution[Sum[y[i], {i, n}]/n,
Table[y[i] \[Distributed] NormalDistribution[0, \[Sigma]], {i, n}],
Assumptions -> \[Sigma] > 0]
{#, marginalDistribution/.n->#} &/@Range[2,10]


$$\begin{array}{cc} 2 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{2}}\right] \\ 3 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{3}}\right] \\ 4 & \text{NormalDistribution}\left[0,\frac{\sigma }{2}\right] \\ 5 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{5}}\right] \\ 6 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{6}}\right] \\ 7 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{7}}\right] \\ 8 & \text{NormalDistribution}\left[0,\frac{\sigma }{2 \sqrt{2}}\right] \\ 9 & \text{NormalDistribution}\left[0,\frac{\sigma }{3}\right] \\ 10 & \text{NormalDistribution}\left[0,\frac{\sigma }{\sqrt{10}}\right] \\ \end{array}$$

So we see that the marginal distribution of $$\bar{y}$$ is

NormalDistribution[0, σ/Sqrt[n]]


The joint distribution of $$\bar{y}$$ and, say, $$y_1$$ is given by

jointDistribution = TransformedDistribution[{y[1], Sum[y[i], {i, n}]/n},
Table[y[i] \[Distributed] NormalDistribution[0, \[Sigma]], {i, n}]]
{#, jointDistribution /. n -> #} & /@ Range[2, 10] // TableForm


$$\begin{array}{cc} 2 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{2} \\ \frac{\sigma ^2}{2} & \frac{\sigma ^2}{2} \\ \end{array} \right)\right] \\ 3 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{3} \\ \frac{\sigma ^2}{3} & \frac{\sigma ^2}{3} \\ \end{array} \right)\right] \\ 4 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{4} \\ \frac{\sigma ^2}{4} & \frac{\sigma ^2}{4} \\ \end{array} \right)\right] \\ 5 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{5} \\ \frac{\sigma ^2}{5} & \frac{\sigma ^2}{5} \\ \end{array} \right)\right] \\ 6 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{6} \\ \frac{\sigma ^2}{6} & \frac{\sigma ^2}{6} \\ \end{array} \right)\right] \\ 7 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{7} \\ \frac{\sigma ^2}{7} & \frac{\sigma ^2}{7} \\ \end{array} \right)\right] \\ 8 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{8} \\ \frac{\sigma ^2}{8} & \frac{\sigma ^2}{8} \\ \end{array} \right)\right] \\ 9 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{9} \\ \frac{\sigma ^2}{9} & \frac{\sigma ^2}{9} \\ \end{array} \right)\right] \\ 10 & \text{MultinormalDistribution}\left[\{0,0\},\left( \begin{array}{cc} \sigma ^2 & \frac{\sigma ^2}{10} \\ \frac{\sigma ^2}{10} & \frac{\sigma ^2}{10} \\ \end{array} \right)\right] \\ \end{array}$$

So the general distribution is a multivariate normal

MultinormalDistribution[{0, 0}, {{σ^2, σ^2/n}, {σ^2/n, σ^2/n}}]


The general form of the joint density function can then be found with

FullSimplify[PDF[MultinormalDistribution[{0, 0}, {{σ^2, σ^2/n}, {σ^2/n, σ^2/n}}], {y, ybar}],
Assumptions -> {σ > 0, n > 1}]


$$\frac{n e^{-\frac{n \left(n \text{ybar}^2+y^2-2 y \text{ybar}\right)}{2 (n-1) \sigma ^2}}}{2 \pi \sqrt{n-1} \sigma ^2}$$

• Anyway, I like your answer! I'll have to look at it to understand (not obvious (to me) that this would be the solution).
– mjw
Mar 24, 2019 at 22:38
• @mjw Good. Answers should always be scrutinized and challenged if desired.
– JimB
Mar 24, 2019 at 22:40
• Nice! In addition to trying to understand the technique, I checked the marginal integrals. Looks great!
– mjw
Mar 25, 2019 at 1:13

Here is the distribution of $$x=\overline{y}$$ (Part I of your question):

a[n_] := Table[y[k] \[Distributed] NormalDistribution[0, \[Sigma]], {k, 1, n}];
p[n_] := TransformedDistribution[Sum[y[k]/n, {k, n}], a[n]];


Now

x \[Distributed] p[5] (* n=5, for example *)


The result is

x \[Distributed] NormalDistribution[0, Abs[\[Sigma]]/Sqrt[5]]

• I am not sure, but shouldn't be n instead of 5 here TransformedDistribution[Sum[y[k]/n, {k, 5}], a] ? And what if I want to leave n, without assigning a value to n? Thanks @mjw Mar 24, 2019 at 21:13
• Oh yes, you are right! I started with 10 and changed to five as I was trying it out. I'll fix it ... Thanks!
– mjw
Mar 24, 2019 at 21:18
• Let's go with five because it is clearer. The result is NormalDistribution[0,\[Sigma]/Sqrt[5]]. Not sure why Mathematica puts Abs[] around $\sigma$. Obviously, $\sigma>0$.
– mjw
Mar 24, 2019 at 21:22
• Yes, sure it is clearer. Do you have any idea of how can I use n as a parameter, without assigning a value to n? Mar 24, 2019 at 21:24
• a[n_] = Table[y[k] \[Distributed] NormalDistribution[0, \[Sigma]], {k, 1, n}];
– mjw
Mar 24, 2019 at 21:25