Gamma Distribution
Gamma distributions: state the parameterization before calculating
A gamma random variable is positive and often represents the time until a specified number of events in a Poisson process, or another accumulated positive quantity.
In the shape-rate convention, shape alpha and rate beta give mean alpha/beta and variance alpha/beta squared. In the shape-scale convention, the scale theta is 1/beta, giving mean alpha theta and variance alpha theta squared.
The shape controls the form: small shapes can be strongly right-skewed, while larger shapes are more symmetric. When alpha is a positive integer, the distribution is an Erlang waiting-time distribution for that many arrivals from a constant-rate Poisson process.
This interpretation requires independent increments and a constant event rate; it is not a generic story for every positive dataset.
Rate has units inverse to the measured quantity and scale has the same units as the quantity. Writing the parameterization and units before substitution prevents the common rate-scale inversion error.
A gamma variable uses shape alpha=6 and rate beta=0.75 per minute. What is its mean?
A report gives a gamma rate beta=0.4 per gram. What scale theta describes the same distribution?