Conditional Waiting Times After Elapsed Time

Why Elapsed Time Matters

Suppose a device has been operating for several years without failing, or a customer has already waited several minutes for service. The question is not necessarily how long the full wait lasts, but how much longer the wait is expected to continue. For an exponential waiting time, the conditional distribution of that additional wait has a particularly simple form: it is the same as the distribution of the original wait.

This result applies only under an exponential model. It does not mean that elapsed time is irrelevant for every lifetime or waiting-time distribution. The model and its rate must be specified, and the event being conditioned on must be interpreted correctly.

Definition: Let \(X\) be an exponential waiting time with rate \(\lambda>0\), where \(\lambda\) is measured per unit of time. For a stated elapsed duration \(s\geq 0\), the additional waiting time, conditional on the event not having occurred by time \(s\), is \(W_s=X-s\mid X>s\). Under the exponential model, \(W_s\) has an exponential distribution with rate \(\lambda\), regardless of \(s\).

The condition \(X>s\) means the event has not occurred by time \(s\). Because an exponential variable is continuous, the probability of \(X=s\) is zero. Thus, “the event occurs exactly at time \(s\)” is not the condition used here; the relevant information is that it has not occurred by that time.

Conditional Probabilities and Expected Additional Time

For an exponential variable with rate \(\lambda\), the survival function is \(P(X>t)=e^{-\lambda t}\) for \(t\geq 0\). If the wait has already exceeded \(s\), the probability of waiting more than another \(t\) units is

P(W_s>t) =P(X>s+t\mid X>s) =\frac{P(X>s+t)}{P(X>s)} =e^{-\lambda t},\qquad t\geq 0.

The elapsed duration \(s\) cancels. The conditional survival function is therefore the exponential survival function with the original rate. Equivalently, for \(t\geq 0\), the conditional cumulative distribution function is \(P(W_s\leq t)=1-e^{-\lambda t}\).

This gives several useful calculations. A probability of waiting at most \(t\) more units is \(1-e^{-\lambda t}\). A probability of waiting between \(a\) and \(b\) more units, where \(0\leq a<b\), is the difference between two survival probabilities. The mean additional wait is \(1/\lambda\).

P(a<W_s\leq b)=e^{-\lambda a}-e^{-\lambda b}, \qquad E[W_s]=\frac{1}{\lambda}.

If a question instead asks for the expected total waiting time, conditional on having waited at least \(s\), add the elapsed time back:

E[X\mid X>s]=s+E[W_s]=s+\frac{1}{\lambda}.

The difference between \(E[W_s]\) and \(E[X\mid X>s]\) is important. The first measures time from now until the event; the second measures the full time from the start of the wait.

A Short Justification

The conditional probability formula follows directly from the definition of conditional probability. Since \(s,t\geq 0\), the event \(X>s+t\) is contained in \(X>s\). The exponential survival probabilities therefore give

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

This is the survival function of an exponential variable with rate \(\lambda\). It establishes the conditional distribution of the additional wait, not merely a statement about its mean. Since that conditional distribution is exponential, its mean is \(1/\lambda\). Finally, on the event \(X>s\), the full waiting time is \(X=s+(X-s)\). Taking the conditional expectation gives \(E[X\mid X>s]=s+1/\lambda\).

Worked Conditional Waiting-Time Calculations

Worked Example: A Service Request Still in Progress

In a hypothetical service system, the time \(X\) from the start of a request until completion is exponential with mean 10 minutes. A request is still incomplete after 7 minutes. Find (a) the probability it takes more than another 12 minutes, (b) the probability it completes between 4 and 12 minutes from now, and (c) the expected additional and total waiting times, conditional on the request still being incomplete.

Model and parameters. The stipulated model is \(X\sim\operatorname{Exponential}(\lambda)\), with time measured in minutes. Since \(E[X]=1/\lambda=10\), the rate is \(\lambda=0.1\) per minute. There is no interest-rate or mortality assumption in this waiting-time problem.

Method. Condition on \(X>7\), then use the exponential distribution of \(W_7=X-7\mid X>7\), which has the same rate \(0.1\).

(a) The probability the request takes more than another 12 minutes is

P(W_7>12)=e^{-0.1(12)}=e^{-1.2}=0.3012.

(b) The probability it completes between 4 and 12 minutes from now is

P(4<W_7\leq 12) =e^{-0.1(4)}-e^{-0.1(12)} =e^{-0.4}-e^{-1.2} =0.6703-0.3012 =0.3691.

(c) The expected additional wait is \(1/0.1=10\) minutes. The expected total time from the start, given that the request is still incomplete after 7 minutes, is \(7+10=17\) minutes.

Check. The interval probability is positive and less than \(P(W_7>4)=e^{-0.4}=0.6703\), as expected. A common mistake is to report 10 minutes as the expected total time; it is the expected additional time, while the conditional expected total is 17 minutes.

Worked Example: A Hypothetical Inverter Lifetime

For a hypothetical reliability calculation, suppose an inverter's time \(X\) to failure follows an exponential distribution with rate \(\lambda=0.04\) per year. It has operated for 8 years without failure. Find the probability that it fails between 5 and 18 years from now, and the probability it operates for more than 30 additional years.

Model and assumptions. Use the stated exponential lifetime model, with all times measured in years and \(\lambda=0.04\) per year. Its mean lifetime under this model is \(1/0.04=25\) years. This is a hypothetical model; no interest-rate or mortality assumption is involved.

Method. Condition on survival beyond 8 years. The additional lifetime \(W_8=X-8\mid X>8\) is exponential with rate \(0.04\).

The probability of failure between 5 and 18 additional years is

\begin{aligned} P(5<W_8\leq 18) &=e^{-0.04(5)}-e^{-0.04(18)}\\ &=e^{-0.20}-e^{-0.72}\\ &=0.8187-0.4868\\ &=0.3320. \end{aligned}

The probability it operates for more than another 30 years is

P(W_8>30)=e^{-0.04(30)}=e^{-1.2}=0.3012.

Check. The interval probability is less than \(P(W_8>5)=e^{-0.20}=0.8187\), and all time values use the same year unit as the rate. A common error is to treat the first interval as a range of total lifetimes from 5 to 18 years. The question asks for additional years after the 8 already survived, so the conditional interval is 5 to 18 years from now.

Worked Example: Waiting for Service After a Long Initial Wait

In a hypothetical queue, the time \(X\) from joining the queue until service begins is exponential with mean 12 minutes. A customer has already waited 9 minutes and service has not begun. Find the probability service begins within the next 6 minutes, the expected additional wait, and the expected total wait from joining the queue.

Model and parameters. Here \(X\sim\operatorname{Exponential}(\lambda)\), with time in minutes and \(\lambda=1/12\) per minute because \(E[X]=12\). There is no interest-rate or mortality assumption.

Method. Given \(X>9\), the remaining time \(W_9=X-9\mid X>9\) has rate \(1/12\).

Service begins within 6 minutes if \(W_9\leq 6\). Thus,

\begin{aligned} P(W_9\leq 6) &=1-e^{-(1/12)(6)}\\ &=1-e^{-0.5}\\ &=1-0.6065\\ &=0.3935. \end{aligned}

The expected additional wait is \(1/(1/12)=12\) minutes. The expected total wait from joining, conditional on having waited 9 minutes, is \(9+12=21\) minutes.

Check. The probability of service beginning within the next 6 minutes is below 0.5, consistent with 6 minutes being less than the 12-minute mean of the remaining exponential wait. A common mistake is to calculate \(1-e^{-(1/12)(15)}\), which is the probability the original wait is at most 15 minutes, not the conditional probability of service beginning in the next 6 minutes.

Common Traps and a Reliable Setup

Conditional waiting-time questions are often short, but the wording can hide which clock the question uses. Before calculating, identify the original waiting-time variable, the elapsed time, and the amount of additional time being asked about.

State the model and units.
Write the exponential rate \(\lambda\) with its time unit. Convert the mean to a rate using \(\lambda=1/E[X]\), if needed.

Translate the condition.
“Still waiting after \(s\)” means \(X>s\). The remaining wait is \(W_s=X-s\mid X>s\).

Use the remaining-time scale.
For additional time \(t\), use \(P(W_s>t)=e^{-\lambda t}\). For an interval \((a,b]\), subtract the survival probabilities at \(a\) and \(b\).

Answer the requested clock.
The expected additional time is \(1/\lambda\). If the question asks for expected total time from the start, report \(s+1/\lambda\).

Do not use the elapsed time as the additional-time threshold. The event “more than 4 minutes from now” uses \(e^{-4\lambda}\), not \(e^{-(s+4)\lambda}\).
Do not add elapsed time to a probability threshold unless the question uses total time. Conditional on \(X>s\), “wait another \(t\)” concerns \(W_s>t\); “total wait exceeds \(s+t\)” is the equivalent event.
Keep units consistent. If \(\lambda\) is per hour, a threshold in minutes must first be converted to hours.
Do not apply this result automatically to other distributions. The elapsed-time cancellation shown above follows from the exponential survival function; a general waiting-time model need not have this property.

Key takeaway: Under an exponential model with rate \(\lambda\), conditioning on having waited longer than \(s\) leaves an exponential additional wait with mean \(1/\lambda\). Use the elapsed time \(s\) only when converting an expected additional wait into an expected total wait from the start.

Check Your Understanding

For each problem, identify whether the requested quantity concerns additional time or total time before calculating.

A hypothetical sensor's time to activation is exponential with mean 4 hours. It has not activated after 3 hours. What is the probability it takes more than 5 additional hours to activate? (Answer: \(e^{-1.25}=0.2865\).)
A hypothetical shipment's delay is exponential with rate \(0.05\) per day. The shipment is still delayed after 11 days. What is the probability the additional delay is between 4 and 14 days? (Answer: \(e^{-0.2}-e^{-0.7}=0.3221\).)
A hypothetical repair time is exponential with rate \(0.5\) per minute. A repair is still in progress after 2 minutes. What is the expected additional repair time, and what is the expected total repair time conditional on still being in progress? (Answer: 2 minutes additional; 4 minutes total.)
A hypothetical equipment lifetime is exponential with mean 30 months. Equipment has survived 10 months. Find the probability it lasts at least 15 more months and its expected additional lifetime. (Answer: \(e^{-0.5}=0.6065\); 30 months.)
A hypothetical response time is exponential with rate \(0.1\) per minute. A response has not arrived after 7 minutes. Find the probability it arrives between 3 and 9 minutes from now and the conditional expected total response time from the start. (Answer: \(e^{-0.3}-e^{-0.9}=0.3342\); 17 minutes.)

What You'll Learn