Given a binomial random variable with probability of success p and probability of failure (1-p). What is the expected number of failures given that a success was observed within k trials and then trials were halted?

For example, if k=3 then the observations that qualify would be:

Failure, Failure, Success

Failure, Success


Note that the following observations do NOT qualify even though the success occurs within the first k trials because the trials not halted after the first success:

Failure, Success, Failure

Success, Failure, Failure

Failure, Success, Success


And of course these observations do NOT qualify because there are no successes in them at all:

Failure, Failure, Failure

Failure, Failure

My solution:

In the example for k=3, I think we can calculate the expected number of failures for k=3 in the following way. First, calculate the total probability of observations that qualify:

$$Total Conditional Probability = TCP = P(Failure, Failure, Success) + P(Failure, Success) + P(Success) = p(1-p)^2 + p(1-p) + p$$

Then, we can answer the question by calculating the expected number of failures for the all observations that qualify and normalize it by our probability space (TCP):

$$(P(Failure, Failure, Success)*NumFailures(Failure, Failure, Success) + P(Failure, Success)*NumFailures(Failure, Success) + P(Success)*NumFailures(Success))/TCP$$

$$(P(Failure, Failure, Success)*2 + P(Failure, Success)*1 + P(Success)*0)/TCP$$

$$\frac{(p(1-p)^22 + p(1-p)1 + p0)}{TCP}$$

We then repeat this procedure using a general k. The formula for TCP is:

$$TCP = \sum_{i=0}^{k-1}{(p(1-p)^i)}$$

And the expected number of failures for general k:


I would also like know if there is a way to simplify this formula, assuming it is correct. Thank you.


A binomial random variable counts the number of "successes" among a fixed number of trials $n$, without regard to the order in which the outcomes of those trials occur.

Since you stop observing trials after the first success is observed, a more appropriate probability model is a geometric random variable, which in your case would count the number of failures before the first success is observed. This is given by $$\Pr[Y = y] = p (1-p)^y, \quad y \in \{0, 1, 2, 3, \ldots\}.$$ Your question then amounts to $$\operatorname{E}[Y \mid Y \le k-1];$$ this is the conditional expectation of the number of failures given that there are at most $k-1$ such failures (which implies that the first success occurs by the $k^{\rm th}$ trial). To calculate this, note $$\Pr[Y \le k-1] = \sum_{y=0}^{k-1} \Pr[Y = y] = \sum_{y=0}^{k-1} p(1-p)^y = 1 - (1-p)^k.$$ Thus $$\operatorname{E}[Y \mid Y \le k-1]\Pr[Y \le k-1] = \sum_{y=0}^{k-1} \operatorname{E}[Y \mid Y = y]\Pr[Y = y] = \sum_{y=0}^{k-1} y p(1-p)^y = \frac{1-p - (1-p)^k(1-p+kp)}{p}.$$ Therefore, $$\operatorname{E}[Y \mid Y \le k-1] = \frac{1}{p} + k - 1 - \frac{k}{1-(1-p)^k}.$$ So for $k = 3$ and $p = 1/5$, we have $52/61$.


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.