Conditional Laws for Exponential Waiting Times

Why Elapsed Waiting Time Matters

Suppose a system has been operating for several hours without a failure, or a customer has waited without receiving service. A natural question is whether that elapsed time changes the expected wait from now on. For an exponential waiting time, the answer is no: conditional on the event not having occurred yet, the residual waiting time has the same distribution as the original waiting time. This property is called memorylessness.

The result depends on the exponential model, not merely on the fact that a waiting time is continuous or nonnegative. In applications, it is therefore important to state the assumed distribution, rate, and time unit before using a conditional waiting-time formula. The methods here are aligned with the SOA VEE Mathematical Statistics syllabus.

Exponential Waiting Times and Residual Time

Definition: A waiting time $T$ has an exponential distribution with rate $\lambda>0$ if, for $t\geq 0$, $$ P(T>t)=e^{-\lambda t}, \qquad F_T(t)=P(T\leq t)=1-e^{-\lambda t}. $$ Its density is $f_T(t)=\lambda e^{-\lambda t}$ for $t\geq 0$, and zero otherwise. The rate $\lambda$ is measured in inverse time units; for example, a rate of $0.2$ per hour corresponds to a mean waiting time of $1/0.2=5$ hours.

If $s\geq0$ is the elapsed time and $T>s$, define the residual waiting time as $R_s=T-s$, conditional on survival beyond $s$. The conditional statement matters: $R_s$ describes how much longer the wait lasts, given that it has already lasted at least $s$.

Theorem: If $T$ is exponential with rate $\lambda$, then for $x\geq0$, $$ P(T-s>x\mid T>s)=e^{-\lambda x}. $$ Thus, conditional on $T>s$, the residual time $T-s$ is again exponential with rate $\lambda$. In particular, $$ E[T-s\mid T>s]=\frac{1}{\lambda}, \qquad E[T\mid T>s]=s+\frac{1}{\lambda}. $$

Because an exponential distribution is continuous, $P(T=s)=0$. Consequently, using $T\geq s$ instead of $T>s$ makes no difference to these conditional probabilities when $s$ is fixed. The event must still specify that the waiting time has not ended by $s$.

Why the Memoryless Formula Holds

The key is that the exponential survival probability factors in a way that removes the elapsed time. The conditional probability is a ratio: the numerator is the chance the wait exceeds both the elapsed time and an additional $x$; the denominator is the chance it exceeds the elapsed time.

Write the conditional survival probability.
For $x\geq0$, $T-s>x$ is equivalent to $T>s+x$, so the conditional probability is $P(T>s+x)/P(T>s)$.

Substitute the exponential survival function.
Since $P(T>t)=e^{-\lambda t}$, the ratio is $e^{-\lambda(s+x)}/e^{-\lambda s}$.

Cancel the elapsed-time factor.
The ratio simplifies to $e^{-\lambda x}$, which is exactly the survival function of an exponential random variable with rate $\lambda$.

P(T-s>x\mid T>s) =\frac{P(T>s+x)}{P(T>s)} =\frac{e^{-\lambda(s+x)}}{e^{-\lambda s}} =e^{-\lambda x}.

The conditional density of the residual time is therefore $\lambda e^{-\lambda x}$ for $x\geq0$. Its expectation can be justified using the tail-integral formula for a nonnegative random variable: its mean is the integral of its survival probability. Hence

E[T-s\mid T>s] =\int_0^\infty e^{-\lambda x}\,dx =\left[-\frac{1}{\lambda}e^{-\lambda x}\right]_0^\infty =\frac{1}{\lambda}.

Since $T=s+(T-s)$ on the event $T>s$, linearity of conditional expectation gives $E[T\mid T>s]=s+1/\lambda$. The expected total waiting time, given survival to $s$, includes the $s$ units already elapsed. The expected remaining waiting time does not.

Using the Conditional Law

For a residual time $R_s=T-s$ conditional on $T>s$, the most useful forms are

P(R_s\leq x\mid T>s)=1-e^{-\lambda x}, \qquad P(R_s>x\mid T>s)=e^{-\lambda x} \quad (x\geq0).

For an interval $a<R_s\leq b$, where $0\leq a<b$, subtract the residual survival probabilities:

P(a<R_s\leq b\mid T>s)=e^{-\lambda a}-e^{-\lambda b}.

The elapsed time $s$ is not inserted into these residual-time formulas. It appears in the event that defines the conditioning, but the memoryless property removes it from the residual distribution. If the question instead asks about the total time $T$, translate the event carefully; for example, $T>s+x$ given $T>s$ has probability $e^{-\lambda x}$.

Worked Example: A Hypothetical Service-System Wait

In a hypothetical service system, a customer's waiting time $T$ is modeled as exponential with rate $\lambda=0.2$ per hour. A customer has already waited 3 hours. Find (a) the probability the customer waits no more than 2 additional hours, (b) the expected additional wait, and (c) the expected total waiting time, conditional on having waited 3 hours.

Method and model. Use the conditional exponential residual law. The rate is $0.2$ per hour, so all time results will be in hours. The condition is $T>3$, and the residual time is $T-3$.

(a) Probability of at most 2 additional hours. With $x=2$, $$ P(T-3\leq2\mid T>3)=1-e^{-\lambda x} =1-e^{-(0.2)(2)} =1-e^{-0.4}. $$ Since $e^{-0.4}\approx0.6703$, the probability is approximately $1-0.6703=0.3297$.

(b) Expected additional wait. The exponential mean is the reciprocal of the rate: $$ E[T-3\mid T>3]=\frac{1}{0.2}=5\text{ hours}. $$

(c) Expected total wait. Add the time already elapsed: $$ E[T\mid T>3]=3+5=8\text{ hours}. $$ As a check, the conditional survival probability for an additional 2 hours is $e^{-0.4}\approx0.6703$, so the probability of finishing within those 2 hours is $0.3297$; the two probabilities sum to 1. The common mistake avoided here is reporting 5 hours as the expected total wait rather than as the expected remaining wait.

Worked Example: Residual Time With a Rate in Minutes

For a hypothetical equipment alert, let the time $T$ from alert activation to a specified response be exponential with rate $\lambda=0.3054302$ per minute. Given that no response has occurred after 5 minutes, calculate the expected residual time and the probability of a response within the next 2 minutes. Give the probability to four decimal places and the expected time to four decimal places.

Method. Apply the residual exponential law with the rate expressed per minute. The elapsed time is $s=5$ minutes; it does not change the residual-time rate.

The expected residual time is $$ E[T-5\mid T>5]=\frac{1}{0.3054302} \approx3.2740705 \approx3.2741\text{ minutes}. $$ The probability of a response within 2 more minutes is $$ P(T-5\leq2\mid T>5) =1-e^{-2(0.3054302)} =1-e^{-0.6108604}. $$ Using $e^{-0.6108604}\approx0.5428835$, this is approximately $1-0.5428835=0.4571165$, or $0.4571$ to four decimal places.

Therefore, the conditional mean remaining time is 3.2741 minutes and the conditional probability of a response within the next 2 minutes is 0.4571. As a check, the probability of waiting more than 2 additional minutes is approximately $0.5429$, and $0.4571+0.5429=1.0000$ at the stated precision. The common mistake avoided is rounding the exponential survival probability too early and then reporting an inconsistent subtraction or reciprocal.

Worked Example: Probability of a Residual-Time Window

In a hypothetical monitoring model, the time $T$ until an event is exponential with rate $\lambda=0.25$ per hour. Given that the event has not occurred after 2.5 hours, find the probability that it occurs more than 1 hour but no more than 4 hours from now. Also find the expected total time from the start until the event.

Method. Let $R=T-2.5$ conditional on $T>2.5$. For the window $1<R\leq4$, subtract the residual survival probability at 4 from that at 1: $$ P(1<R\leq4\mid T>2.5) =e^{-(0.25)(1)}-e^{-(0.25)(4)} =e^{-0.25}-e^{-1}. $$

Numerically, $e^{-0.25}\approx0.7788008$ and $e^{-1}\approx0.3678794$. Thus the probability is $0.7788008-0.3678794=0.4109214$, or $0.4109$ to four decimal places.

The mean residual time is $1/0.25=4$ hours. The mean total time, conditional on survival for 2.5 hours, is therefore $$ E[T\mid T>2.5]=2.5+4=6.5\text{ hours}. $$ As a check, the residual-time probability below 1 hour is $1-e^{-0.25}\approx0.2212$, the probability from 1 through 4 hours is approximately $0.4109$, and the probability beyond 4 hours is $e^{-1}\approx0.3679$; these add to 1.0000 after rounding. The common mistake avoided is using $P(T\leq 4)$ for the upper endpoint, even though 4 here means additional hours, not total hours from the start.

Common Trap: Forgetting to Condition

An unconditional probability of falling in an interval after time $s$ is not yet the probability of that interval given survival to $s$. For $s=3$, $x=2$, and rate $0.2$ per hour, the unconditional probability of $3<T\leq5$ is

P(3<T\leq5)=F_T(5)-F_T(3) =e^{-0.6}-e^{-1} \approx0.5488-0.3679=0.1809.

That value is a probability among all waiting times from the start. To answer the conditional question, divide by the probability of reaching 3 hours without the event:

P(T-3\leq2\mid T>3) =\frac{P(3<T\leq5)}{P(T>3)} =\frac{0.1809}{0.5488} \approx0.3297.

The unrounded calculation gives $1-e^{-0.4}\approx0.3297$, agreeing with the residual-time formula. This distinction is important whenever a problem says “given that it has already lasted” or “given that no event has occurred by time $s$.”

When the Result Applies

The exponential model is appropriate only when its assumptions are suitable for the problem. In particular, the conditional waiting-time result follows from a constant rate $\lambda$: the chance of an event during a short time interval does not depend on how long the system has already waited. Many real processes do not have a constant event rate. For example, a component whose failure likelihood increases with age need not have an exponential lifetime, so its expected residual lifetime may depend on its age.

Use the rate's time unit consistently. A rate per minute gives a mean in minutes; convert either the rate or the time before substituting if the problem uses hours.
Separate residual and total time. The mean remaining time is $1/\lambda$; the mean total time given survival to $s$ is $s+1/\lambda$.
Keep the conditioning event visible. A probability involving $T-s$ is understood given $T>s$, not as an unconditional probability about $T$.
Do not transfer memorylessness to other distributions. The cancellation in the survival ratio is specific to the exponential survival function.

Key takeaway: For an exponential waiting time with rate $\lambda$, conditioning on survival to $s$ leaves an exponential residual time with the same rate. Its expected remaining duration is $1/\lambda$, while the expected total duration is $s+1/\lambda$.

Check Your Understanding

For each hypothetical model, use the stated exponential rate and keep the elapsed time distinct from the additional waiting time.

A dispatch delay $T$ is exponential with rate $0.4$ per minute. Given $T>2$ minutes, find the expected residual delay and the probability of dispatch within 1 additional minute. (Answer: $2.5$ minutes; $0.3297$.)
A sensor event time $T$ is exponential with rate $0.1$ per day. Given that no event has occurred by day 8, find the probability of waiting more than 5 additional days and the expected total time to the event. (Answer: $0.6065$; $18$ days.)
A hypothetical process has an exponential waiting time $T$ with rate $0.25$ per hour. Given $T>3$, find the probability that the residual wait is more than 2 but no more than 6 hours. (Answer: $0.3834$.)
A response time $T$ is exponential with rate $0.2$ per hour. Given $T>4$, find the probability that the total response time exceeds 9 hours and the conditional expected total response time. (Answer: $0.3679$; $9$ hours.)
For a hypothetical delivery process, the expected residual waiting time after survival to day 7 is 6 days. Under an exponential model, find the probability of delivery within 4 additional days and the expected total time from the start. (Answer: $0.4866$; $13$ days.)

What You'll Learn