Conventional linear model to log-normal distribution

86 Views Asked by At

Let $T_{i}$ be a random variable describing the survival time for the $i$th sample.

This random variable may be modelled using a conventional linear model

$T_{i} = e^{X'_{i}\beta + \epsilon_{i}}$ (1)

where $\epsilon_{i}$ is the error term for the ith sample.

It's logarithm form is

$\log(T_{i}) = X'_{i}\beta + \epsilon_{i}$

In a text I am looking at, it says,

'if the $\epsilon_{i}$ are normally distributed, then, one obtains a log - normal model for the $T_{i}$'.

From Wikipedia, https://en.wikipedia.org/wiki/Log-normal_distribution

it is mentioned that for a log - normal random variable, X, this random variable may be described as

$X = e^{Z\sigma + \mu}$(2)

I fail to see the connection between (1) and (2). Any help is appreciated.

where $\sigma$ is the standard deviation, $\mu$ is the mean an $Z$ is the standard normal variable.

1

There are 1 best solutions below

0
On BEST ANSWER

From $X = e^{Z\sigma + \mu}$ you have that $\ln(X) = \mu + \sigma Z$. Now, $$\text{E}(\ln(X)) = \text{E}(\mu + \sigma Z) = \mu$$ because $\mu$ and $\sigma$ are constant and $\text{E}(Z)=0$.

Also, we have that $$\text{Var}(\ln(X)) = \text{Var}(\mu + \sigma Z) = \sigma^{2} \text{Var}(Z)= \sigma^{2}$$ because $\text{Var}(Z)=1$ and because of the properties of the variance.

From the fact that $Z$ is a standard normal, and $X$ is a linear function of $Z$, we have that $\ln(X)$ is a normal random variable.

So, from $X = e^{Z\sigma + \mu}$ with $Z$ a standard normal, you have that $\ln(X)$ is a normal random variable with mean $\mu$ and variance $\sigma$. You can also do this in reverse, obtaining the former statement from the latter.