Rate and Scale Conventions in Quantile Formulas

Why Rate and Scale Both Appear

The exponential distribution is often written using a rate parameter in one reference and a scale parameter in another. The formulas may look different, but they describe the same distribution when the parameters are reciprocals. This distinction matters in quantile calculations: using a scale where a rate is expected, or vice versa, can change a result by a reciprocal factor.

The goal here is to translate the formulas reliably and check the translation against three features: the support, quantiles, and tail probabilities. The parameter’s units provide a useful check: a rate has units inverse to the measurement, while a scale has the same units as the measurement.

The Two Parameterizations

Let \(X\) have an exponential distribution. In the rate convention, the parameter \(\lambda\) is positive, and the density is proportional to \(\lambda e^{-\lambda x}\). In the scale convention, the positive parameter \(\beta\) appears in the exponent as \(x/\beta\). The conversion is \(\beta=1/\lambda\).

Definition: For \(\lambda>0\), the rate parameterization has density \(f(x)=\lambda e^{-\lambda x}\) for \(x\geq 0\), and \(f(x)=0\) for \(x<0\). For \(\beta>0\), the scale parameterization has density \(f(x)=\frac{1}{\beta}e^{-x/\beta}\) for \(x\geq 0\), and \(f(x)=0\) for \(x<0\). The parameterizations are equivalent when \(\beta=1/\lambda\), or, equivalently, \(\lambda=1/\beta\).

Both densities are positive for \(x\geq0\) and zero for \(x<0\), so both have support \([0,\infty)\). Substituting \(\beta=1/\lambda\) into the scale density gives \(\frac{1}{\beta}e^{-x/\beta}=\lambda e^{-\lambda x}\), showing that the two density formulas agree at every \(x\).

Worked Example: Rewriting a Density Using the Reciprocal Parameter

Suppose a waiting time \(X\), measured in minutes, has rate \(\lambda=0.5\) per minute. Find its scale parameter and write its density in both conventions.

The reciprocal scale is \(\beta=1/\lambda=1/0.5=2\) minutes. The rate form is \(f(x)=0.5e^{-0.5x}\) for \(x\geq0\), and zero otherwise. The scale form is \(f(x)=\frac12 e^{-x/2}\) for \(x\geq0\), and zero otherwise.

To check the match, \(\frac12 e^{-x/2}=0.5e^{-0.5x}\). The support is \([0,\infty)\) in either form. The units also agree: \(0.5\) has units per minute, whereas \(2\) has units minutes.

Distribution, Quantile, and Tail Formulas

For \(x\geq0\), integrating either density from \(0\) to \(x\) gives the CDF. The quantile \(Q(p)\), for \(0<p<1\), is defined by \(Q(p)=\inf\{x:F(x)\geq p\}\). For the exponential distribution, the CDF is continuous and strictly increasing on its support, so \(Q(p)\) is the unique \(x\geq0\) satisfying \(F(x)=p\).

Theorem: If \(X\) has exponential rate \(\lambda>0\) and scale \(\beta=1/\lambda\), then for \(x\geq0\) and \(0<p<1\), \(F(x)=1-e^{-\lambda x}=1-e^{-x/\beta}\), \(P(X>x)=e^{-\lambda x}=e^{-x/\beta}\), and \(Q(p)=-\frac{\ln(1-p)}{\lambda}=-\beta\ln(1-p)\). For \(x<0\), \(F(x)=0\) and \(P(X>x)=1\).

The density is zero below zero, so \(F(x)=0\) for \(x<0\). For \(x\geq0\), integrate the rate form and evaluate the antiderivative:

F(x)=\int_0^x \lambda e^{-\lambda t}\,dt =\left[-e^{-\lambda t}\right]_0^x =1-e^{-\lambda x}.

Using \(\beta=1/\lambda\) gives \(e^{-\lambda x}=e^{-x/\beta}\), so the scale form of the CDF is \(1-e^{-x/\beta}\). For \(x\geq0\), the event \(X>x\) is the complement of \(X\leq x\), hence

P(X>x)=1-F(x)=e^{-\lambda x}=e^{-x/\beta}.

To derive the quantile, solve \(F(x)=p\) for \(x\). Since \(0<p<1\), the quantity \(1-p\) is positive and its logarithm is defined:

p=1-e^{-\lambda x} \quad\Longleftrightarrow\quad e^{-\lambda x}=1-p \quad\Longleftrightarrow\quad -\lambda x=\ln(1-p) \quad\Longleftrightarrow\quad x=-\frac{\ln(1-p)}{\lambda}.

Substituting \(\beta=1/\lambda\) yields \(x=-\beta\ln(1-p)\). The solved value is nonnegative because \(\ln(1-p)<0\), and strict increase of the CDF makes it the unique quantile. These calculations establish each formula in both conventions. \(\square\)

Sometimes the question specifies an upper-tail probability rather than a CDF probability. If \(0<\alpha<1\), the threshold \(c\) satisfying \(P(X>c)=\alpha\) follows by solving \(e^{-\lambda c}=\alpha\). Thus \(c=-\ln(\alpha)/\lambda=-\beta\ln(\alpha)\). This is the same threshold as the quantile \(Q(1-\alpha)\), since \(1-(1-\alpha)=\alpha\).

Worked Example: Comparing an 80th Percentile in Both Conventions

Let the waiting time have rate \(\lambda=0.5\) per minute, equivalently scale \(\beta=2\) minutes. Find the 80th percentile in both forms and check that the CDF at the result is \(0.8\).

Here \(p=0.8\), so \(1-p=0.2\). In rate form, \(Q(0.8)=-\ln(0.2)/0.5=\ln(5)/0.5\approx1.609438/0.5=3.218876\) minutes.

In scale form, \(Q(0.8)=-2\ln(0.2)=2\ln(5)\approx3.218876\) minutes. The two calculations agree.

For a direct check, the CDF at this value is \(1-e^{-0.5(3.218876)}=1-e^{-1.609438}=1-0.2=0.8\). The value lies in the common support \([0,\infty)\).

Worked Example: Matching Tail Probabilities at a Fixed Time

Suppose \(\lambda=0.5\) per minute and \(\beta=2\) minutes. Find the probability that the waiting time exceeds 4 minutes using both tail formulas.

Rate form: \(P(X>4)=e^{-\lambda x}=e^{-0.5(4)}=e^{-2}\approx0.135335\).

Scale form: \(P(X>4)=e^{-x/\beta}=e^{-4/2}=e^{-2}\approx0.135335\).

The formulas give the same probability because \(\lambda x=x/\beta\) when \(\lambda=1/\beta\). The CDF check is \(F(4)=1-e^{-2}\approx0.864665\), and \(1-F(4)\approx0.135335\), as required.

Worked Example: Finding a Threshold for a 5% Upper Tail

A lifetime has rate \(\lambda=0.2\) per day, so its scale is \(\beta=1/0.2=5\) days. Find the threshold \(c\) for which \(P(X>c)=0.05\), then verify it in both conventions.

Using the rate, \(c=-\ln(0.05)/0.2=\ln(20)/0.2\approx2.995732/0.2=14.97866\) days. Using the scale, \(c=-5\ln(0.05)=5\ln(20)\approx14.97866\) days.

At this threshold, \(e^{-0.2c}=e^{-2.995732}=0.05\); likewise, \(e^{-c/5}=e^{-2.995732}=0.05\). The threshold is also the 95th percentile because \(Q(0.95)=-\ln(1-0.95)/0.2=-\ln(0.05)/0.2\).

How to Check a Conversion

A reliable conversion check compares more than the appearance of one formula. Start by confirming the reciprocal relationship, then compare the support, the expression in the exponential, and the resulting quantile or tail probability. The exponents must be dimensionless: if \(x\) is measured in minutes, then \(\lambda\) must be measured per minute, while \(\beta\) must be measured in minutes.

Identify the convention.
A density with \(\lambda e^{-\lambda x}\) uses rate; a density with \(\frac{1}{\beta}e^{-x/\beta}\) uses scale.

Take the reciprocal.
Set \(\beta=1/\lambda\) or \(\lambda=1/\beta\), keeping the units consistent.

Check a feature.
Compare support and then evaluate the CDF, a quantile, or a tail probability in both forms.

Common Mistakes and Why the Convention Matters

A frequent mistake is to copy the right formula but insert the wrong kind of parameter. For example, if \(\lambda=0.5\) per minute, then the median is \(\ln(2)/\lambda\approx1.3863\) minutes. Substituting \(0.5\) into the scale formula \(-\beta\ln(0.5)\) would give about \(0.3466\) minutes, but \(0.5\) is not the scale: the scale is \(2\) minutes. Using \(\beta=2\) correctly gives \(-2\ln(0.5)=2\ln(2)\approx1.3863\) minutes.

The rate and scale are not two unrelated choices of units: they are reciprocal numerical parameters, with different interpretations. A larger rate means faster exponential decay and shorter typical waiting times; a larger scale means slower decay and longer typical waiting times. This interpretation follows directly from the tail formula. At fixed \(x>0\), increasing \(\lambda\) decreases \(e^{-\lambda x}\), while increasing \(\beta\) decreases \(x/\beta\) and therefore increases \(e^{-x/\beta}\).

Another error is to omit the complement when converting a quantile to a tail threshold. The \(p\)-quantile solves \(F(x)=p\), so its formula contains \(\ln(1-p)\). A threshold with upper-tail probability \(\alpha\) solves \(P(X>x)=\alpha\), so it contains \(\ln(\alpha)\). These formulas agree when \(p=1-\alpha\); they are not interchangeable if the probability levels are left unchanged.

Key takeaway: Rate and scale formulas describe the same exponential law when \(\beta=1/\lambda\). Their common support is \([0,\infty)\); the CDF, quantiles, and upper tails match after this reciprocal substitution. Check both the probability level and parameter units before calculating.

Check Your Understanding

Use the reciprocal relationship and the formulas above to answer each question.

A rate is \(0.25\) per hour. What is the equivalent scale, including its units?
Write the CDF at \(x\geq0\) using the rate parameter, then rewrite it using the equivalent scale.
For rate \(\lambda=2\), find the median in rate form and then in scale form.
If \(P(X>c)=0.1\), express \(c\) using \(\lambda\) and then using \(\beta\).
Explain why using a rate value as though it were a scale can produce an incorrect quantile even when the chosen exponential formula is correct.