Given an exponential distribution $\lambda e^{-\lambda t}$ for $t \geq 0$, the mean is $1/\lambda$. I'm wondering how one would compute the mean if the process starts at a certain time and we start observing it at a time $T$ later. Intuitively given the exponential distribution is memoryless I think the mean would be $1/\lambda$ from the time after we start observing (so the mean would occur $1/\lambda$ time units after T). If this is correct, I'm not sure how to prove this; what would the integral of the expectation would look like? Is it the same as the one we perform if we start observing at t = 0?


If you have some probability function $P(t)$ for $t\geq0$, then the mean value of $t$ is given by

$$\left<t\right>=\int_0^\infty tP(t)\text{d}t$$

Because you are shifting the start time, you need to adjust the normalization constant; you find this by solving

$$A\int_T^\infty \text{e}^{-\lambda t}dt = 1$$

where $A$ is your new normalization constant. We find that $A$ is given by

$$A = \lambda \text{e}^{\lambda T}$$

so your distribution becomes

$$P(t)=\lambda\text{e}^{\lambda (T+t)}$$

thus the integral you need to compute is

$$\left<t\right>=\lambda\text{e}^{\lambda T}\int_T^\infty t~\text{e}^{-\lambda t}~\text{d}t=T+\frac{ 1}{\lambda}$$

so your intuition is correct, this is just shifted forward in time by T.

  • $\begingroup$ Given the exponential distribution is memoryless (P( t > r + s | t > r) = P(t > s)), why does it make sense that the mean does not remain the same relative to the point in time where the observation is recorded? $\endgroup$
    – Yandle
    May 6 '19 at 3:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.