Weibull Shape-Scale Parameterizations

Why Weibull Parameterizations Need Care

A Weibull model can be written using a shape parameter and a scale parameter, but the same distribution is also expressed with a coefficient multiplying a power of the variable. That coefficient is sometimes called a rate parameter. The distinction matters: the scale has the same units as the modeled quantity, while the coefficient has units that depend on the shape. Confusing them can produce an incorrect exponent and an invalid probability.

This tutorial defines the Weibull distribution, derives its cumulative distribution and survival functions, and shows how to translate between common conventions. The focus is parameter interpretation and conversion; it does not require an interest-rate or mortality model. In the examples, any modeled lifetime is a hypothetical continuous random variable, with no additional mortality assumption.

Definition: A random variable $X$ has a Weibull distribution with shape $k>0$ and scale $\lambda>0$, written here as $X\sim\operatorname{Weibull}(k,\lambda)$, if its support is $x\geq 0$ and its cumulative distribution function is $$ F_X(x)= \begin{cases} 0, & x<0,\\ 1-\exp\left[-(x/\lambda)^k\right], & x\geq 0. \end{cases} $$ The shape $k$ is dimensionless. The scale $\lambda$ has the same units as $X$.

Density, Cumulative Distribution, and Survival

For $x>0$, differentiating the cumulative distribution gives the Weibull density:

f_X(x)=\frac{k}{\lambda}\left(\frac{x}{\lambda}\right)^{k-1} \exp\left[-(x/\lambda)^k\right].

The density is zero for $x<0$. Its value at the single point $x=0$ does not change probabilities; depending on $k$, the formula's limiting behavior there may be zero, finite, or unbounded. The CDF remains well-defined at zero, with $F_X(0)=0$.

The survival function is the probability of exceeding a specified value. Since $X$ is continuous, $P(X>x)=P(X\geq x)$, so the survival function is the complement of the CDF:

S_X(x)=P(X>x)=1-F_X(x)=\exp\left[-(x/\lambda)^k\right], \qquad x\geq 0.

For $x<0$, $S_X(x)=1$, because the model assigns probability zero to negative values. The CDF and survival function are two descriptions of the same distribution; they are not separate models. Their values must add to one at every $x$.

Why the Cumulative Distribution Has This Form

The CDF can be derived directly by integrating the density. This also shows why the power term and the exponential appear together. For $x\geq 0$, begin with the density and make the substitution $u=(t/\lambda)^k$. Its differential is $du=(k/\lambda)(t/\lambda)^{k-1}\,dt$.

\begin{aligned} F_X(x) &=\int_0^x \frac{k}{\lambda}\left(\frac{t}{\lambda}\right)^{k-1} \exp\left[-(t/\lambda)^k\right]\,dt\\ &=\int_0^{(x/\lambda)^k}e^{-u}\,du\\ &=\left[-e^{-u}\right]_{0}^{(x/\lambda)^k}\\ &=1-\exp\left[-(x/\lambda)^k\right]. \end{aligned}

At $x=0$, the upper integration limit is zero and the result is zero, as required. As $x$ increases without bound, $(x/\lambda)^k$ increases without bound, so the exponential term approaches zero and $F_X(x)$ approaches one. Thus the density integrates to one over its nonnegative support.

Translating Between Common Conventions

A common alternative writes the CDF as $F_X(x)=1-\exp(-a x^k)$ for $x\geq 0$, where $a>0$. Matching exponents gives $a x^k=(x/\lambda)^k=x^k/\lambda^k$. Therefore the conversion is:

a=\frac{1}{\lambda^k}, \qquad \lambda=a^{-1/k}.

In this form, $a$ is an inverse-power coefficient, not the scale itself. If $X$ is measured in hours, then $\lambda$ is in hours and $a$ is in hours$^{-k}$. The alternative density follows by differentiating its CDF:

f_X(x)=akx^{k-1}\exp(-a x^k), \qquad x\geq 0.

Another common convention uses $\beta$ for shape and $\eta$ for scale:

F_X(x)=1-\exp\left[-(x/\eta)^\beta\right].

This is the same shape-scale form: $k=\beta$ and $\lambda=\eta$. Symbols alone do not determine a convention. When encountering a Weibull formula, identify the exponent and compare it with $(x/\lambda)^k$ before assigning meanings to the parameters.

Form shown	Shape	Scale	Conversion or interpretation
$1-\exp[-(x/\lambda)^k]$	$k$	$\lambda$	Shape-scale form used here
$1-\exp(-a x^k)$	$k$	$a^{-1/k}$	$a=\lambda^{-k}$, with units $X^{-k}$
$1-\exp[-(x/\eta)^\beta]$	$\beta$	$\eta$	Same shape-scale form with different symbols

Worked Examples

Worked Example: Calculating from Shape and Scale

A hypothetical device has service time $X$, measured in days, modeled by a Weibull distribution with shape $k=2$ and scale $\lambda=10$ days. Find the probability that service time is at most 10 days, and the probability that it exceeds 10 days. No interest rate or mortality assumption applies; the stated Weibull model is the full distribution for this calculation.

Method: use the shape-scale CDF and its complement. Substitute $x=10$ days, $k=2$, and $\lambda=10$ days. The ratio $x/\lambda$ is dimensionless:

\begin{aligned} F_X(10) &=1-\exp\left[-(10/10)^2\right]\\ &=1-\exp(-1)\\ &=1-0.3678794\\ &=0.6321206. \end{aligned}

Thus $P(X\leq 10)=0.6321$ to four decimal places. The probability of exceeding 10 days is $S_X(10)=\exp(-1)=0.3678794$, or $0.3679$ to four decimal places. As a check, $0.6321+0.3679=1.0000$. The common mistake avoided is treating the scale $\lambda=10$ as though it were the coefficient $a$ in $\exp(-a x^k)$.

Worked Example: Converting an Inverse-Power Coefficient

A hypothetical test records the operating time $X$ of a component in hours. Its survival function is $S_X(x)=\exp[-(0.008\text{ hours}^{-3})x^3]$ for $x\geq0$. Identify the shape and scale, then find the probability that operating time is greater than 5 hours but no more than 10 hours. No interest rate or mortality assumption applies.

Method: match the exponent to the inverse-power form, then use survival probabilities. Comparing $\exp(-a x^k)$ with the supplied expression gives $k=3$ and $a=0.008\text{ hours}^{-3}$. Convert the coefficient to scale:

\lambda=a^{-1/k} =(0.008\text{ hours}^{-3})^{-1/3} =5\text{ hours},

because $0.008=1/125$, and $125^{1/3}=5$. In shape-scale notation, $X\sim\operatorname{Weibull}(3,5\text{ hours})$. For the requested interval, continuity gives $P(5<X\leq10)=S_X(5)-S_X(10)$. Evaluate each endpoint:

\begin{aligned} S_X(5)&=\exp[-0.008(5)^3]=\exp(-1)=0.3678794,\\ S_X(10)&=\exp[-0.008(10)^3]=\exp(-8)=0.0003355,\\ P(5<X\leq10)&=0.3678794-0.0003355=0.3675439. \end{aligned}

The interval probability is $0.3675$ to four decimal places. The exponents are dimensionless: the coefficient is in hours$^{-3}$, while $x^3$ is in hours$^3$. A common error is to call $0.008$ the scale; it is the inverse-power coefficient, and its units confirm that it cannot be a scale measured in hours.

Worked Example: Identifying Parameters from a Density

A hypothetical continuous time $X$, measured in weeks, has density $f_X(x)=0.5x\exp(-x^2/4)$ for $x\geq0$, and zero otherwise. Identify its Weibull shape and scale, then find $P(1<X\leq2)$. There is no interest-rate or mortality assumption in this distributional calculation.

Method: match the exponential term and the density coefficient. In the inverse-power form, the exponential term is $\exp(-a x^k)$. Here it is $\exp(-x^2/4)$, so $k=2$ and $a=1/4\text{ weeks}^{-2}$. The scale is

\lambda=a^{-1/k}=(1/4\text{ weeks}^{-2})^{-1/2}=2\text{ weeks}.

The density coefficient provides a second check. The Weibull density coefficient in this form is $akx^{k-1}$, which here becomes $(1/4)(2)x=0.5x$, matching the supplied density. Therefore $X\sim\operatorname{Weibull}(2,2\text{ weeks})$. Now calculate the interval probability from the CDF, or equivalently from survival:

\begin{aligned} P(1<X\leq2) &=S_X(1)-S_X(2)\\ &=\exp[-(1/2)^2]-\exp[-(2/2)^2]\\ &=\exp(-0.25)-\exp(-1)\\ &=0.7788008-0.3678794\\ &=0.4109214. \end{aligned}

The probability is $0.4109$ to four decimal places. The check is that the identified parameters reproduce both the exponential term and the density's leading coefficient. A frequent mistake is matching only the power $x^2$ and overlooking the factor $1/4$, which is what determines the scale.

Common Parameterization Traps

When a problem supplies a Weibull formula rather than naming its parameters, first rewrite its exponent in a recognizable form. This avoids relying on symbols that may have been assigned differently. The following checks are useful:

Check the power. The power of $x$ is the shape $k$ in both $(x/\lambda)^k$ and $a x^k$.
Check the full exponent. In shape-scale form it is $(x/\lambda)^k$; in inverse-power form it is $a x^k$. Match the entire expression, not just one symbol.
Check units. The exponent must be dimensionless. A scale has units of $X$; $a$ has units of $X^{-k}$.
Check endpoints and complements. For nonnegative $x$, $F_X(0)=0$, $S_X(0)=1$, and $F_X(x)+S_X(x)=1$.

For example, if $x$ is measured in months and an exponent is written as $-0.04x^2$, the coefficient has units of months$^{-2}$. The shape is 2, and the scale is $0.04^{-1/2}=5$ months. The scale is not $0.04$, nor is it the coefficient's reciprocal without taking the appropriate root.

Key takeaway: In the shape-scale convention, $F_X(x)=1-\exp[-(x/\lambda)^k]$ and $S_X(x)=\exp[-(x/\lambda)^k]$. In the inverse-power convention, the same distribution has exponent $-a x^k$, where $a=\lambda^{-k}$. Identify the exponent and check its units before calculating.

Check Your Understanding

For each problem, identify the parameterization before calculating. Report probabilities to four decimal places.

A hypothetical duration in days has Weibull shape $k=2$ and scale $\lambda=6$ days. Find $P(X\leq3)$. (Answer: $0.2212$)
A nonnegative measurement $X$, in days, has CDF $F_X(x)=1-\exp(-0.04x^2)$ for $x\geq0$. Identify its shape and scale, including units. (Answer: shape $2$, scale $5$ days)
A hypothetical lifetime in months has shape $k=1.5$ and scale $\lambda=12$ months. Find $P(6<X\leq12)$. (Answer: $0.3343$)
A time $X$, in hours, has survival function $S_X(x)=\exp[-(1/625\text{ hours}^{-4})x^4]$. Convert to shape-scale form and find $P(X\leq2.5\text{ hours})$. (Answer: shape $4$, scale $5$ hours; probability $0.0606$)
A hypothetical duration $X$, in weeks, has density $f_X(x)=(x/18)\exp[-(x/6)^2]$ for $x\geq0$. Identify its Weibull shape and scale, then find $P(0<X\leq6)$. (Answer: shape $2$, scale $6$ weeks; probability $0.6321$)

Form shown	Shape	Scale	Conversion or interpretation
\(1-\exp[-(x/\lambda)^k]\)	\(k\)	\(\lambda\)	Shape-scale form used here
\(1-\exp(-a x^k)\)	\(k\)	\(a^{-1/k}\)	\(a=\lambda^{-k}\), with units \(X^{-k}\)
\(1-\exp[-(x/\eta)^\beta]\)	\(\beta\)	\(\eta\)	Same shape-scale form with different symbols

What You'll Learn