3
$\begingroup$

In a prior post, I developed an "unfolded" gamma distribution generalization of a normal distribution as an example of how to relate a gamma distribution to a normal distribution. This yielded

$$ \text{ND}(x;\mu,\sigma^2,a) = \dfrac{a e^{-2^{-\frac{a}{2}} \left(\frac{1}{\sigma }\right)^a \left| x-\mu\right| ^a}}{2 \sqrt{2} \sigma \Gamma \left(\frac{1}{a}\right)} \,,$$

where the mean is $\mu$, the variance is $\sigma^2$, and the shape is $a>0$, where $a=2$ for an ordinary normal distribution.

This appears to be a different distribution from the generalized error distribution

$$\text{GED}(x;\mu,\alpha,\beta)=\frac{\beta}{2\alpha\Gamma(1/\beta)} \; e^{-(|x-\mu|/\alpha)^\beta}\,,$$

where $\mu$ is the location, $\alpha>0$ is the scale, and $\beta>0$ is the shape and where $\beta=2$ yields a normal distribution. It includes the Laplace distribution when $\beta=1$. As $\beta\rightarrow\infty$, the density converges pointwise to a uniform density on $(\mu-\alpha,\mu+\alpha)$.

Questions Are these distributions the same? If so, how does one convert between them? If not, and $\text{ND}(x;\mu,\sigma^2,a)$ is a different distribution, what are its other properties, for example, does it reduce to a distribution other than a normal distribution?

$\endgroup$
3
  • 1
    $\begingroup$ Assuming the domain is $x\in\mathbb R$, comparing the parts of the two densities that depend on $x$ shows immediately that the role played by $\mu$ is the same in both (it's the location parameter) and $a=\beta$ (the shape parameter) and $\sigma=\alpha$ (the scale parameter). $\endgroup$
    – whuber
    Commented Mar 5, 2018 at 13:53
  • $\begingroup$ @whuber Close, but when one lets $\beta=a=1$, then $\sqrt{2} \sigma=\alpha$. There is a bit more to this, will explore further. $\endgroup$
    – Carl
    Commented Mar 5, 2018 at 18:47
  • 1
    $\begingroup$ I'm having trouble reading the "ND" expression. If it means the argument of the exponential is $$-2^{-a/2} \sigma^{-a}|x-\mu|^a=-\left(\frac{|x-\mu|}{\sigma\sqrt{2}}\right)^a,$$ then the scale factor must be $\sigma\sqrt{2}$ and everything works out. $\endgroup$
    – whuber
    Commented Mar 5, 2018 at 19:32

2 Answers 2

2
$\begingroup$

The result from Whuber's comment in an answer:

It seems like both expressions can be related to the function below

$$f(x;c_1,c_2,c_3) = c_1 e^{-c_2 \vert x - \mu \vert^{c_3}}$$


GED: $\frac{\beta}{2\alpha\Gamma(1/\beta)} \; e^{-(|x-\mu|/\alpha)^\beta}$

$$\begin{array}{rcl} c_1 &=& \frac{\beta}{2\alpha\Gamma(1/\beta)} \\ c_2 &=& \left(\frac{1}{\alpha} \right)^\beta \\ c_3 &=& \beta \end{array}$$

ND: $\dfrac{a e^{\left(-2^{-\frac{a}{2}} \left(\frac{1}{\sigma }\right)^a \left| x-\mu\right| ^a\right)}}{2 \sqrt{2} \sigma \Gamma \left(\frac{1}{a}\right)} $

$$\begin{array}{rcl} c_1 &=& \frac{a}{2 \sqrt{2} \sigma\Gamma(1/a)} \\ c_2 &=& 2^{-\frac{a}{2}}\left( \frac{1}{\sigma} \right)^a \\ c_3 &=& a \end{array}$$


and you can say

$$\begin{array}{rcl} \mu &=& \mu \\ {\alpha} &=& \sqrt{2}{\sigma} \\ \beta &=& a \end{array}$$

$\endgroup$
1
  • $\begingroup$ (+1) for the help. $\endgroup$
    – Carl
    Commented Mar 10 at 7:10
1
$\begingroup$

Indeed, my notation was bad. Better notation is now used on the post this came from,

The generalized gamma distribution

$$\text{GD}\left(x;\alpha ,\beta ,\gamma ,\mu \right)=\begin{array}{cc} & \begin{cases} \dfrac{\gamma e^{-\left(\dfrac{x-\mu }{\beta }\right)^{\gamma }} \left(\dfrac{x-\mu }{\beta }\right)^{\alpha \gamma -1}}{\beta \,\Gamma (\alpha )} & x>\mu \\ 0 & \text{other} \\ \end{cases} \\ \end{array}\,,$$

implies that

$$ \begin{align} \text{GND}(x;\mu,\alpha,\beta) &= \frac{1}{2}\text{GD}\left(x;\frac{1}{\beta},\alpha,\beta,\mu \right)+\frac{1}{2}\text{GD}\left(-x;\frac{1}{\beta},\alpha,\beta,\mu \right)\\ &= \frac{\beta e^{-\left(\dfrac{\left|x-\mu\right|}{\alpha }\right)^{\mathrm{\Large{\beta}}}}}{2 \alpha \Gamma \left(\dfrac{1}{\beta }\right)}\\ \end{align} \,,$$

is a generalization of the normal distribution, where $\mu$ is the location, $\alpha>0$ is the scale, and $\beta>0$ is the shape and where $\beta=2$ yields a normal distribution. It includes the Laplace distribution when $\beta=1$. As $\beta\rightarrow\infty$, the density converges pointwise to a uniform density on $(\mu-\alpha,\mu+\alpha)$.

See the original post for the logic behind this.

$\endgroup$

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.