Lognormal Distribution

What is the lognormal distribution?

The lognormal distribution is a continuous probability distribution for positive values. It is useful when a quantity changes by multiplication or percentage growth rather than by adding fixed amounts. If a random variable X follows a lognormal distribution, then log(X) follows a normal distribution

What are some examples of the lognormal distribution?

Examples of the lognormal distribution include:

When should I consider using the lognormal distribution?

Consider the lognormal distribution when values must be positive and the data are right-skewed: most observations are relatively small, with a few much larger values. It is especially appropriate when the logarithms of the observations are approximately normally distributed.

The lognormal distribution is often used in reliability analysis when damage accumulates over time, such as cracks growing until a part fails. You can also model a lognormal response as a function of other variables using methods such as penalized regression or generalized linear models.

Characteristics of the lognormal distribution

The lognormal distribution is always positive and usually right-skewed. Its long right tail allows occasional values much larger than the typical observation. Taking logarithms makes the distribution symmetric and normal, which can simplify analysis.

The distribution has two parameters $\mu$ and $\sigma$. They are the mean and standard deviation of log(X), not of X itself. Increasing $\mu$ shifts the distribution toward larger values, while increasing $\sigma$ creates more spread and a longer right tail. The mean is larger than the median unless $\sigma$ is zero.

Model parameters

$\mu$, the location (mean of log(X))

$\sigma$, the scale (standard deviation of log(X))

Density function $f(x)=\frac{1}{x\sigma}\,\phi\!\left(\frac{\log(x)-\mu}{\sigma}\right)$, where $\phi$ is the normal density function, $x>0$, -$\infty<\mu<\infty,\;\sigma>0$
Mean $e^{\mu+\sigma^2/2}$
Median $e^\mu$
Variance $e^{2(\mu+\sigma^2)}-e^{2\mu+\sigma^2}$

The graph below shows lognormal distributions with the same mean and different values of sigma.

lognormal-pic1.png

Example of a lognormal random variable

Suppose a process produces parts whose heights follow a lognormal distribution. The median height is 3.0 cm, and the scale parameter is $\sigma$ = 0.1. Parts are acceptable only when their heights are between 2.0 and 4.0 cm. What fraction of parts will be scrapped?

Let X represent part height. For a lognormal distribution, the median is $e^\mu$, so $\mu$ = log(3.0) = 1.0986. Therefore, X follows a lognormal distribution with parameters $\mu$ = 1.0986 and $\sigma$ = 0.1. We find the probability that X is below 2.0 cm or above 4.0 cm. Equivalently, we find the probability between 2.0 and 4.0 cm and subtract it from 1.

$P(X<2 \text{ or } X>4)=1-P(2<X<4)$

$\begin{aligned} &= 1 - \int_{2}^{4} f(x)\,dx \\[6pt] &= 1 - \int_{2}^{4} \frac{1}{0.1\sqrt{2\pi}\,x} \exp\!\left(\frac{-(\log(x)-1.0986)^2} {2(0.1)^2} \right)\,dx \\[6pt] &= 0.0020\end{aligned}$

The calculation gives a scrap fraction of about 0.20%. Most of that risk comes from parts above 4.0 cm; parts below 2.0 cm are extremely unlikely. In practical terms, about two out of every 1,000 parts would be expected to fall outside the specification limits.