Exponential Rate And Scale Estimates

Why Exponential Rate and Scale Need Careful Notation

An exponential model can be written using a rate parameter or a scale parameter. The two forms describe the same family of distributions, but their parameters are reciprocals: a larger rate means a shorter average time, while a larger scale means a longer average time. This makes it easy to use the right method-of-moments idea and still report the wrong estimate by taking the reciprocal in the wrong direction.

For this tutorial, assume \(X_1,\ldots,X_n\) are independent and identically distributed observations from an exponential distribution with support \(x\geq 0\), no location shift, and a positive parameter. The model and measurement units should be identified before calculating an estimate. These estimates are aligned with the SOA VEE Mathematical Statistics syllabus.

Definition: In the rate parameterization, \(X\) has density \(f_X(x)=\lambda e^{-\lambda x}\) for \(x\geq 0\), where \(\lambda>0\). Its mean is \(E[X]=1/\lambda\). In the scale parameterization, \(X\) has density \(f_X(x)=\frac{1}{\theta}e^{-x/\theta}\) for \(x\geq 0\), where \(\theta>0\). Its mean is \(E[X]=\theta\). The parameters satisfy \(\theta=1/\lambda\), or equivalently \(\lambda=1/\theta\).

The rate has units that are the reciprocal of the units of \(X\). For example, if \(X\) is measured in hours, \(\lambda\) is measured in inverse hours. The scale has the same units as \(X\). In this exponential model, the standard deviation is also \(\theta\), but the method-of-moments estimate considered here matches the first raw moment, the mean.

Deriving the Method-of-Moments Estimates

The method of moments equates a population raw moment to its corresponding sample raw moment. For one parameter, the first moment is sufficient here. The sample first raw moment is the sample mean, \(\bar X=\frac{1}{n}\sum_{i=1}^n X_i\). Thus, match \(\bar X\) to the model mean and solve for the parameter.

\text{Rate form:}\qquad \frac{1}{\lambda}=\bar X \quad\Longrightarrow\quad \hat{\lambda}_{\mathrm{MM}}=\frac{1}{\bar X}.

\text{Scale form:}\qquad \theta=\bar X \quad\Longrightarrow\quad \hat{\theta}_{\mathrm{MM}}=\bar X.

Here the subscript \(\mathrm{MM}\) denotes a method-of-moments estimate. The hat indicates an estimate calculated from the observed sample. In both parameterizations, the same observed sample determines the fitted exponential distribution; the numerical parameter differs because rate and scale have reciprocal meanings.

Why the model mean is \(1/\lambda\): For the rate form, integrate the density over its nonnegative support. Integration by parts gives \(\int_0^\infty x\lambda e^{-\lambda x}\,dx=1/\lambda\). In the scale form, substitute \(\lambda=1/\theta\); the same mean is \(1/(1/\theta)=\theta\). Matching this population mean to \(\bar X\) gives the two estimates above.

The reciprocal relationship also explains how the two estimates transform. Since \(\theta=1/\lambda\), applying that transformation to the rate estimate gives \(1/\hat{\lambda}_{\mathrm{MM}}=1/(1/\bar X)=\bar X=\hat{\theta}_{\mathrm{MM}}\). Conversely, \(1/\hat{\theta}_{\mathrm{MM}}=1/\bar X=\hat{\lambda}_{\mathrm{MM}}\). This equality holds for these one-parameter estimates whenever \(\bar X>0\).

Identify the parameterization and units.
Check whether the model uses a rate or a scale, and record the units of the observations.

Calculate the sample mean.
Add the observations and divide by their number, keeping the measurement unit attached.

Match the model mean.
For a rate, set \(1/\lambda=\bar X\). For a scale, set \(\theta=\bar X\).

Transform only if needed.
Use the reciprocal to report the other parameterization, and invert the units at the same time.

Worked Estimates in Both Parameterizations

Worked Example: Repair Times Recorded in Minutes

A hypothetical maintenance team records five repair times, in minutes: 10, 14, 16, 20, and 30. Assume the times are an independent sample from an exponential model with zero location. Find the method-of-moments estimates of the scale in minutes and the rate per minute. State the corresponding rate per hour.

Method: match the sample mean to the exponential mean. First calculate the total and sample mean:

10+14+16+20+30=90\text{ minutes},\qquad \bar X=\frac{90}{5}=18\text{ minutes}.

In the scale form, the model mean is \(\theta\), so \(\hat{\theta}_{\mathrm{MM}}=18\) minutes. In the rate form, the model mean is \(1/\lambda\), so

\hat{\lambda}_{\mathrm{MM}}=\frac{1}{18} =0.055555\ldots\text{ per minute} \approx 0.05556\text{ per minute}.

There are 60 minutes in an hour. A rate of \(1/18\) per minute is \(60/18=3.333\ldots\) per hour, or approximately \(3.333\) per hour. The scale expressed in hours is \(18/60=0.3\) hours, and its reciprocal is \(1/0.3=3.333\ldots\) per hour.

Check: \((0.055555\ldots\text{ per minute})(18\text{ minutes})=1\), and the fitted mean of 18 minutes equals the observed sample mean. The common mistake avoided is reporting 18 as the rate: 18 is the scale estimate, not the rate estimate.

Worked Example: A Scale-Parameter Model in Hours

A hypothetical monitoring device has four observed operating times, in hours: 0.5, 1, 1.5, and 3. The model is stated in scale form. Find the scale estimate in hours and minutes and the equivalent rate estimate per hour and per minute.

Method: estimate the scale from the sample mean, then take its reciprocal for the rate. The total operating time is

0.5+1+1.5+3=6\text{ hours}.

Therefore,

\bar X=\frac{6}{4}=1.5\text{ hours},\qquad \hat{\theta}_{\mathrm{MM}}=1.5\text{ hours}.

Convert the scale to minutes: \(1.5\times60=90\) minutes. The rate per hour is the reciprocal of 1.5 hours:

\hat{\lambda}_{\mathrm{MM}}=\frac{1}{1.5} =\frac{2}{3} =0.6667\text{ per hour}.

Since one hour is 60 minutes, this is \((2/3)/60=1/90=0.01111\ldots\) per minute, or \(0.01111\) per minute. The units confirm the conversion: \(90\) minutes multiplied by \(1/90\) per minute equals 1.

Check: The rate per hour multiplied by the scale in hours is \((2/3)(1.5)=1\). The common mistake avoided is treating a scale-form model as if its parameter were the rate: the scale estimate is 1.5 hours, and its reciprocal is the rate.

Worked Example: Reading a Rate Given Per Day

A hypothetical model for the waiting time \(X\) between events uses the rate parameterization, with time measured in days. Six recorded waiting times are 1, 2, 2, 3, 4, and 6 days. Estimate the rate per day and the equivalent scale in days. Then determine whether the observed sample mean agrees with the fitted model mean.

Method: use the reciprocal of the sample mean for the rate. The observations sum to

1+2+2+3+4+6=18\text{ days}.

With \(n=6\), the sample mean is \(\bar X=18/6=3\) days. Thus the rate estimate is

\hat{\lambda}_{\mathrm{MM}}=\frac{1}{3} =0.3333\text{ per day}.

The equivalent scale estimate is the reciprocal of this rate:

\hat{\theta}_{\mathrm{MM}}=\frac{1}{\hat{\lambda}_{\mathrm{MM}}} =\frac{1}{1/3}=3\text{ days}.

The fitted model mean is \(1/\hat{\lambda}_{\mathrm{MM}}=1/(1/3)=3\) days, which agrees exactly with the sample mean of 3 days. Equivalently, the fitted mean in scale form is \(\hat{\theta}_{\mathrm{MM}}=3\) days.

Check: \((0.3333\text{ per day})(3\text{ days})\) is approximately 1; using the exact rate \(1/3\), the product is exactly 1. The common mistake avoided is using \(3\) per day as the rate simply because the sample mean is 3 days. The units and reciprocal relationship show why the rate is \(1/3\) per day.

Conditions, Units, and Common Traps

These formulas apply when the observations are modeled as an unshifted exponential sample with a common positive rate, or equivalently a common positive scale. The method uses the first moment and the sample mean; it does not require choosing between the parameterizations by comparing the data to two different models. A sample mean of zero cannot produce a finite positive rate estimate or a positive scale estimate under these formulas, because \(1/\bar X\) is undefined and \(\bar X\) is not positive. Under the stated continuous exponential model, a sample of exact zeros has probability zero.

Unit conversions are not optional decoration. If the sample mean is measured in days, its scale estimate is in days and its rate estimate is per day. To change from a rate per day to a rate per hour, divide by 24; to change from a scale in days to a scale in hours, multiply by 24. The transformations go in opposite directions because a rate is measured per unit time.

Quantity	Method-of-moments estimate	Units when \(X\) is measured in time
Scale \(\theta\)	\(\hat{\theta}_{\mathrm{MM}}=\bar X\)	Time
Rate \(\lambda\)	\(\hat{\lambda}_{\mathrm{MM}}=1/\bar X\)	Inverse time

A frequent exam error is to remember “exponential estimate equals the sample mean” without checking which parameter is being estimated. Here is the wrong calculation and its correction:

Common mistake and correction: If \(\bar X=4\) hours and the question requests a rate, the wrong calculation is \(\hat{\lambda}=4\) per hour. It uses the scale formula but labels the result as a rate. The correct calculations are \(\hat{\theta}=4\) hours and \(\hat{\lambda}=1/4=0.25\) per hour. Their product is \((4\text{ hours})(0.25\text{ per hour})=1\), as required.

Another useful check is to ask whether a proposed rate and scale have a product of 1 and reciprocal units. A rate in minutes cannot be multiplied directly by a scale in hours without first converting one of them. Keeping the sample mean, parameter, and units together through each step prevents this error.

Key takeaway: For an unshifted exponential model, the scale method-of-moments estimate is the sample mean, and the rate estimate is its reciprocal. Transforming one estimate into the other preserves the fitted mean, provided the units are converted consistently.

Check Your Understanding

For each problem, identify the parameterization and keep the reported units attached to the estimates.

A hypothetical sample of four waiting times, in minutes, is 6, 8, 10, and 16. Find the method-of-moments scale in minutes and rate per minute. (Answer: \(\hat{\theta}_{\mathrm{MM}}=10\) minutes; \(\hat{\lambda}_{\mathrm{MM}}=0.1\) per minute.)
A scale-form exponential model is fitted to observations with sample mean 0.4 seconds. Find the scale and equivalent rate. (Answer: \(\hat{\theta}_{\mathrm{MM}}=0.4\) seconds; \(\hat{\lambda}_{\mathrm{MM}}=2.5\) per second.)
For five observations measured in hours, the total is 15 hours. Find the method-of-moments rate per hour and the equivalent scale in minutes. (Answer: \(\hat{\lambda}_{\mathrm{MM}}=1/3\) per hour; \(\hat{\theta}_{\mathrm{MM}}=180\) minutes.)
A model has rate \(\lambda=0.08\) per day. What scale does this rate correspond to? If a sample from the model has mean 12.5 days, what are its method-of-moments rate and scale estimates? (Answer: The corresponding scale is 12.5 days; the estimates are \(\hat{\lambda}_{\mathrm{MM}}=0.08\) per day and \(\hat{\theta}_{\mathrm{MM}}=12.5\) days.)
A hypothetical sample of waiting times in days is 1, 1, 2, and 4. Find both method-of-moments estimates, and verify their product. (Answer: \(\bar X=2\) days, \(\hat{\theta}_{\mathrm{MM}}=2\) days, \(\hat{\lambda}_{\mathrm{MM}}=0.5\) per day, and their product is 1.)

What You'll Learn