Gamma Parameterizations And The Mean

Why Gamma Parameterizations Need Care

A Gamma model can be written using a shape and scale or a shape and rate. Both forms describe the same family of positive continuous distributions, but the second parameter has a reciprocal meaning in the two forms. A mistaken substitution can therefore turn a reasonable-looking mean into a value with the wrong magnitude or even the wrong units.

In a Gamma model for a positive response, the mean is especially useful for interpreting the parameters. The key is to first determine which form is being used, then apply the matching mean formula. Do not infer the meaning of a parameter from its name alone: inspect the probability density or the stated units.

Definition: A random variable $X$ has a Gamma distribution with shape $\alpha>0$ and scale $\theta>0$ if its density is $$ f_X(x)=\frac{x^{\alpha-1}e^{-x/\theta}}{\Gamma(\alpha)\theta^\alpha},\qquad x>0. $$ Equivalently, it has a Gamma distribution with shape $\alpha>0$ and rate $\beta>0$ if its density is $$ f_X(x)=\frac{\beta^\alpha}{\Gamma(\alpha)}x^{\alpha-1}e^{-\beta x},\qquad x>0. $$ The two forms describe the same distribution when $\beta=1/\theta$. Here $\Gamma(\alpha)$ is the Gamma function.

The word rate in the shape-rate parameterization means the reciprocal of the scale, with units inverse to those of $X$. It does not, by itself, mean a claim frequency, an annual rate, or the rate of a regression model. Those quantities may also be called rates in other settings, but their definitions and units must be read from the model.

Mean Formulas and Their Justification

For shape $\alpha$ and scale $\theta$, the mean is $\alpha\theta$. For shape $\alpha$ and rate $\beta$, the mean is $\alpha/\beta$. These are not competing formulas: since $\theta=1/\beta$, they give the same value.

E[X]=\alpha\theta=\frac{\alpha}{\beta}, \qquad \theta=\frac{1}{\beta}.

The density makes the rate-form calculation transparent. For $\alpha>0$ and $\beta>0$, integrate $x f_X(x)$ over the positive real line. The Gamma function is defined by $\Gamma(a)=\int_0^\infty t^{a-1}e^{-t}\,dt$, and integration by parts gives the recurrence $\Gamma(\alpha+1)=\alpha\Gamma(\alpha)$. Thus:

\begin{aligned} E[X] &=\int_0^\infty x\frac{\beta^\alpha}{\Gamma(\alpha)} x^{\alpha-1}e^{-\beta x}\,dx\\ &=\frac{\beta^\alpha}{\Gamma(\alpha)} \int_0^\infty x^\alpha e^{-\beta x}\,dx\\ &=\frac{\beta^\alpha}{\Gamma(\alpha)} \frac{\Gamma(\alpha+1)}{\beta^{\alpha+1}} =\frac{\alpha}{\beta}. \end{aligned}

For the scale form, substitute $\beta=1/\theta$, giving $E[X]=\alpha\theta$. The calculation requires the stated Gamma density with positive shape and positive scale or rate. It is not a general formula for an arbitrary positive response merely because that response is skewed.

Read the density or notation.
An exponential term $e^{-x/\theta}$ indicates scale $\theta$; an exponential term $e^{-\beta x}$ indicates rate $\beta$.

Convert if necessary.
Use $\beta=1/\theta$, or $\theta=1/\beta$, and carry the reciprocal units with the conversion.

Calculate and interpret the mean.
Use $\alpha\theta$ in scale form or $\alpha/\beta$ in rate form. Check that the result has the same units as the response.

Worked Parameter Conversions

Worked Example: Claim Severity in Thousands of Dollars

A hypothetical positive claim severity $X$, measured in thousands of dollars, has a Gamma distribution with shape $\alpha=4$ and scale $\theta=0.5$ thousand dollars. Find its mean and express the same distribution in shape-rate form. Identify the rate’s units.

Method. Use the shape-scale mean, then take the reciprocal of the scale to obtain the rate. This avoids treating the rate as though it were another scale.

Mean in scale form. The mean is shape times scale:

E[X]=\alpha\theta =4(0.5\text{ thousand dollars}) =2\text{ thousand dollars} =\$2{,}000.

Convert to rate form. The rate is the reciprocal of the scale:

\beta=\frac{1}{\theta} =\frac{1}{0.5\text{ thousand dollars}} =2\text{ per thousand dollars}.

The rate-form mean confirms the result:

E[X]=\frac{\alpha}{\beta} =\frac{4}{2\text{ per thousand dollars}} =2\text{ thousand dollars} =\$2{,}000.

The units in the last calculation are dimensionless divided by (per thousand dollars), yielding thousand dollars. The rate-form density therefore uses $\beta=2$ per thousand dollars and shape $4$, not a scale of $2$ thousand dollars. The calculation is checked because both parameterizations give the same mean and compatible units.

Common mistake avoided: multiplying $4$ by the rate $2$ would give $8$, but $\alpha\beta$ is not the Gamma mean in the rate parameterization.

Worked Example: Recovering Scale and Rate From a Mean

A hypothetical repair-time variable $T$, measured in hours, is modeled as Gamma with shape $\alpha=8$. Its mean is specified as 6 hours. Find the scale and rate parameters, and check the mean using both forms.

Method. Solve the mean equation for each parameter in turn. Since the shape is dimensionless, the scale must have units of hours and the rate must have units of inverse hours.

Find the scale. From $E[T]=\alpha\theta$:

6\text{ hours}=8\theta \quad\Longrightarrow\quad \theta=\frac{6}{8}\text{ hours}=0.75\text{ hours}.

Find the rate. From $E[T]=\alpha/\beta$:

6\text{ hours}=\frac{8}{\beta} \quad\Longrightarrow\quad \beta=\frac{8}{6}\text{ per hour} =\frac{4}{3}\text{ per hour}.

The parameters are reciprocal:

\frac{1}{\theta} =\frac{1}{0.75\text{ hours}} =\frac{4}{3}\text{ per hour} =\beta.

Checking the mean gives $8(0.75\text{ hours})=6\text{ hours}$ in scale form and $8/(4/3\text{ per hour})=6\text{ hours}$ in rate form. Both checks agree with the specified mean.

Common mistake avoided: using $8(4/3)$ as the mean confuses rate with scale. It also produces a quantity with units per hour rather than hours, so the units expose the error.

Worked Example: Reading a Rate in a Severity Model

In a hypothetical portfolio, a positive repair amount $Y$, measured in thousands of dollars, has a Gamma distribution with shape $\alpha=5$ and rate $\beta=0.4$ per thousand dollars. Find its scale and mean in thousands of dollars and dollars. Explain why the parameter $0.4$ is not an annual frequency rate.

Method. Start from the stated rate-form parameters, take the reciprocal to find scale, then calculate the mean using the rate formula.

Scale.

\theta=\frac{1}{\beta} =\frac{1}{0.4\text{ per thousand dollars}} =2.5\text{ thousand dollars}.

Mean in rate form.

E[Y]=\frac{\alpha}{\beta} =\frac{5}{0.4\text{ per thousand dollars}} =12.5\text{ thousand dollars} =\$12{,}500.

Mean in scale form.

E[Y]=\alpha\theta =5(2.5\text{ thousand dollars}) =12.5\text{ thousand dollars} =\$12{,}500.

The units of the rate formula are dimensionless divided by (per thousand dollars), yielding thousand dollars. The parameter $0.4$ is the reciprocal scale for the distribution of $Y$; its units are inverse thousand dollars. An annual frequency rate would have units such as events per year and would describe a different quantity.

Check. The scale and rate are reciprocals because $2.5(0.4)=1$, with compatible units. Both mean calculations give $\$12{,}500$.

Common mistake avoided: reading the number $0.4$ as an annual claim rate, or using it as a scale, ignores its stated units and the rate-form density.

Parameterization Checks and Exam Traps

When a problem gives only a parameter name such as “rate,” do not assume the convention. Different applications may use different notation. A density or an explicit definition resolves the ambiguity: the coefficient in $e^{-\beta x}$ is the rate, while the denominator in $e^{-x/\theta}$ is the scale. Once identified, the reciprocal relationship and units provide two independent checks.

A common incorrect calculation is to use a rate as if it were a scale. For example, if a Gamma variable has shape $5$ and rate $0.4$ per thousand dollars, the incorrect calculation $5(0.4)=2$ does not give the mean. Besides using the wrong formula, the product has units per thousand dollars, not thousands of dollars. The correction is:

E[X]=\frac{5}{0.4\text{ per thousand dollars}} =12.5\text{ thousand dollars}.

Another trap is converting the numerical value but not the units. A rate of $2$ per thousand dollars is not $2$ per dollar; expressed per dollar, it is $0.002$ per dollar. Either representation gives the same mean when the response is expressed in the corresponding units. If $X$ is measured in dollars, the same example’s shape $4$, scale $\$500$, and rate $0.002$ per dollar give $4/(0.002\text{ per dollar})=\$2{,}000$.

Form used	Exponential term in density	Second parameter units	Mean
Shape-scale $(\alpha,\theta)$	$e^{-x/\theta}$	Same units as $X$	$\alpha\theta$
Shape-rate $(\alpha,\beta)$	$e^{-\beta x}$	Inverse units of $X$	$\alpha/\beta$

Key takeaway: A Gamma shape is dimensionless. The scale has the response’s units, and the rate has reciprocal units. Identify the density form, use its matching mean formula, and confirm that the result has the response’s units.

Check Your Understanding

For each problem, identify the relevant parameterization and check the units as well as the numerical result.

A hypothetical Gamma response measured in hours has shape $3$ and scale $2$ hours. Find its mean and rate. (Answer: Mean $6$ hours; rate $0.5$ per hour.)
A positive hypothetical payment $X$ is measured in thousands of dollars and has shape $7$ and rate $0.25$ per thousand dollars. Find the mean in thousands of dollars and dollars, and find the scale. (Answer: Mean $28$ thousand dollars, or $\$28{,}000$; scale $4$ thousand dollars.)
A Gamma time variable in days has shape $6$ and mean $900$ days. Find the scale and rate. (Answer: Scale $150$ days; rate $1/150$ per day, approximately $0.0067$ per day.)
A hypothetical severity measured in thousands of dollars has shape $2.4$ and scale $0.75$ thousand dollars. Find the mean and rate. (Answer: Mean $1.8$ thousand dollars, or $\$1{,}800$; rate $4/3$ per thousand dollars.)
A Gamma response measured in days has shape $5$, and its rate is $1/3$ per day. Find its scale and mean, and state why the rate is not itself the mean. (Answer: Scale $3$ days; mean $15$ days; the rate is inverse days and the mean is shape divided by rate.)

Form used	Exponential term in density	Second parameter units	Mean
Shape-scale \((\alpha,\theta)\)	\(e^{-x/\theta}\)	Same units as \(X\)	\(\alpha\theta\)
Shape-rate \((\alpha,\beta)\)	\(e^{-\beta x}\)	Inverse units of \(X\)	\(\alpha/\beta\)

What You'll Learn