A common misconception among students of probability theory is the belief that the sum of two normally distributed (http://planetmath.org/NormalRandomVariable) random variables is itself normally distributed. By constructing a counterexample, we show this to be false.
It is however well known that the sum of normally distributed variables X,Y will be normal under either of the following situations.
The statement that the sum of two independent normal random variables is itself normal is a very useful and often used property. Furthermore, when working with normal variables which are not independent, it is common to suppose that they are in fact joint normal. This can lead to the belief that this property holds always.
Another common fallacy, which our example shows to be false, is that normal random variables are independent if and only if their covariance , defined by
is zero. While it is certainly true that independent variables have zero covariance, the converse statement does not hold. Again, it is often supposed that they are joint normal, in which case a zero covariance will indeed imply independence.
We construct a pair of random variables X,Y satisfying the following.
We start with a pair of independent random variables X,ϵ where X has the standard normal distribution and ϵ takes the values 1,-1, each with a probability of 1/2. Then set,
If S is any measurable subset of the real numbers, the symmetry of the normal distribution implies that ℙ(X∈S) is equal to ℙ(-X∈S). Then, by the independence of X and ϵ, the distribution of Y conditional on ϵ=1 is given by,
It can similarly be shown that ℙ(Y∈S∣ϵ=-1) is equal to ℙ(X∈S). So, Y has the same distribution as X and is normal with mean zero and variance one.
Using the fact that ϵ has zero mean and is independent of X, it is easily shown that the covariance of X and Y is zero.
As X and Y have zero covariance and each have variance equal to 1, the sum X+Y will have variance equal to 2. Also, the sum satisfies
In particular, this shows that ℙ(|X+Y|>2)=0. However, normal random variables with nonzero variance always have a positive probability of being greater than any given real number. So, X+Y is not normally distributed.
This also shows that, despite having zero covariance, X and Y are not independent. If they were, then the fact that sums of independent normals are normal would imply that X+Y is normal, contradicting what we have just demonstrated. Generated on Sat Feb 10 12:04:17 2018 by LaTeXML
If you're seeing this message, it means we're having trouble loading external resources on our website. If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.
If you're seeing this message, it means we're having trouble loading external resources on our website. If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.
5.3.2 Bivariate Normal DistributionRemember that the normal distribution is very important in probability theory and it shows up in many different applications. We have discussed a single normal random variable previously; we will now talk about two or more normal random variables. We recently saw in Theorem 5.2 that the sum of two independent normal random variables is also normal. However, if the two normal random variables are not independent, then their sum is not necessarily normal. Here is a simple counterexample: ExampleLet $X \sim N(0,1)$ and $W \sim Bernoulli\left(\frac{1}{2}\right)$ be independent random variables. Define the random variable $Y$ as a function of $X$ and $W$: \begin{equation} \nonumber Y = h(X,W)=\left\{ \begin{array}{l l} X & \quad \textrm{if }W=0 \\ & \quad \\ -X & \quad \textrm{if }W=1 \end{array} \right. \end{equation} Find the PDF of $Y$ and $X+Y$.
Definition But how can we obtain the joint normal PDF in general? Can we provide a simple way to generate jointly normal random variables? The basic idea is that we can start from several independent random variables and by considering their linear combinations, we can obtain bivariate normal random variables. Similar to our discussion on normal random variables, we start by introducing the standard bivariate normal distribution and then obtain the general case from the standard one. The following example gives the idea. ExampleLet $Z_1$ and $Z_2$ be two independent $N(0,1)$ random variables. Define \begin{align}%\label{} \nonumber X&=Z_1, \\ \nonumber Y&=\rho Z_1 +\sqrt{1-\rho^2} Z_2, \end{align} where $\rho$ is a real number in $(-1,1)$.
We call the above joint distribution for $X$ and $Y$ the standard bivariate normal distribution with correlation coefficient $\mathbf{\rho}$. It is the distribution for two jointly normal random variables when their variances are equal to one and their correlation coefficient is $\rho$. Two random variables $X$ and $Y$ are said to have the standard bivariate normal distribution with correlation coefficient $\mathbf{\rho}$ if their joint PDF is given by \begin{align} \nonumber f_{XY}(x,y)=\frac{1}{2 \pi \sqrt{1-\rho^2}} \exp \{-\frac{1}{2 (1-\rho^2)} \big[ x^2-2\rho x y+y^2 \big] \}, \end{align} where $\rho \in (-1,1)$. If $\rho=0$, then we just say $X$ and $Y$ have the standard bivariate normal distribution. Now, if you want two jointly normal random variables $X$ and $Y$ such that $X \sim N(\mu_X,\sigma^2_X)$, $Y \sim N(\mu_Y,\sigma^2_Y)$, and $\rho(X,Y)=\rho$, you can start with two independent $N(0,1)$ random variables, $Z_1$ and $Z_2$, and define \begin{align} \label{eq:bivariate-conv} \left\{ \begin{array}{l l} X&=\sigma_X Z_1+\mu_X \\ Y&=\sigma_Y (\rho Z_1 +\sqrt{1-\rho^2} Z_2)+\mu_Y \hspace{20pt} (5.23) \end{array} \right. \end{align} We can find the joint PDF of $X$ and $Y$ as above. While the joint PDF has a big formula, we usually do not need to use the formula itself. Instead, we usually work with properties of jointly normal random variables such as their mean, variance, and covariance.Definitions 5.3 and 5.4 are equivalent in the sense that, if X and Y are jointly normal based on one definition, they are jointly normal based on the other definition, too. The proof of their equivalence can be concluded from Problem 10 in Section 6.1.6. In that problem, we show that the two definitions result in the same moment generating functions.
Definition Theorem Let $X$ and $Y$ be two bivariate normal random variables, i.e., their joint PDF is given by Equation 5.24. Then there exist independent standard normal random variables $Z_1$ and $Z_2$ such that \begin{align} \nonumber \left\{ \begin{array}{l l} X&=\sigma_X Z_1+\mu_X \\ Y&=\sigma_Y (\rho Z_1 +\sqrt{1-\rho^2} Z_2)+\mu_Y \end{array} \right. \end{align} Proof. (Sketch) To prove the theorem, define \begin{align} \nonumber \left\{ \begin{array}{l l} Z_1&=\frac{X-\mu_X}{\sigma_X} \\ Z_2&=-\frac{\rho}{\sqrt{1-\rho^2}} \frac{X-\mu_X}{\sigma_X}+\frac{1}{\sqrt{1-\rho^2}}\frac{Y-\mu_Y}{\sigma_Y} \end{array} \right. \end{align} Now find the joint PDF of $Z_1$ and $Z_2$ using the method of transformations (Theorem 5.1), similar to what we did above. You will find out that $Z_1$ and $Z_2$ are independent and standard normal and by definition satisfy the equations of Theorem 5.3. The reason we started our discussion on bivariate normal random variables from $Z_1$ and $Z_2$ is three fold. First, it is more convenient and insightful than the joint PDF formula. Second, sometimes the construction using $Z_1$ and $Z_2$ can be used to solve problems regarding bivariate normal distributions. Third, this method gives us a way to generate samples from the bivariate normal distribution using a computer program. Since most computing packages have a built-in command for independent normal random variable generation, we can simply use this command to generate bivariate normal variables using Equation 5.23. Example Let $X$ and $Y$ be jointly normal random variables with parameters $\mu_X$, $\sigma^2_X$, $\mu_Y$, $\sigma^2_Y$, and $\rho$. Find the conditional distribution of $Y$ given $X=x$.
Theorem Example Let $X$ and $Y$ be jointly normal random variables with parameters $\mu_X=1$, $\sigma^2_X=1$, $\mu_Y=0$, $\sigma^2_Y=4$, and $\rho=\frac{1}{2}$.
Remember that if two random variables $X$ and $Y$ are independent, then they are uncorrelated, i.e., $\textrm{Cov}(X,Y)=0$. However, the converse is not true in general. In the case of jointly normal random variables, the converse is true. Thus, for jointly normal random variables, being independent and being uncorrelated are equivalent.
Theorem Proof. Since $X$ and $Y$ are uncorrelated, we have $\rho(X,Y)=0$. By Theorem 5.4, given $X=x$, $Y$ is normally distributed with \begin{align}%\label{} \nonumber &E[Y|X=x]=\mu_Y+ \rho \sigma_Y \frac{x-\mu_X}{\sigma_X}=\mu_Y,\\ \nonumber &Var(Y|X=x)=(1-\rho^2)\sigma^2_Y=\sigma^2_Y. \end{align} Thus, $f_{Y|X}(y|x)=f_Y(y)$ for all $x,y \in \mathbb{R}$. Thus $X$ and $Y$ are independent. Another way to prove the theorem is to let $\rho=0$ in Equation 5.24 and observe that $f_{XY}(x,y)=f_X(x)f_Y(y)$. |