0
$\begingroup$

If I am given two repeatable events -

  • event A - that I am told has a probability of 0.5% to produce result X
  • event B - that I am told has a probability of 6% to produce result X

and given that the events have a cost associated with them, can I approximate how many times I would expect to have to perform event A to get result X with the same probability as performing event B once?

ie. how many times would I expect to have to do event A to get a probability of 6% of producing result X, to find out if the cost of doing event A that many times works out more or less expensive than just doing event B once?

Apologies if this has been asked before, but I don't know much about probability and what to search for, the other questions of this form all seem to want to know how many times to repeat event A to get a probability of 100% with some confidence level.

$\endgroup$
  • $\begingroup$ Since each run of A has a chance of producing X, running A multiple times may produce multiple X. Is this a problem? $\endgroup$ – Stephan Kolassa Jun 26 '19 at 13:35
  • $\begingroup$ @StephanKolassa no, that would not be a problem. I'm approaching it from a perspective of a value proposition, would it in theory be better value to attempt Event A however many times, or just to attempt Event B once. Or alternatively, what cost should I assign to Event B to make it theoretically a better value proposition than doing Event A however many times. If that makes any more sense? :) $\endgroup$ – Klors Jun 26 '19 at 14:18
3
$\begingroup$

@Dave's answer is right, but I think you can approach it in a simpler way. Think instead of the probability of NOT getting result $X$, which is $0.995$ if you try once, $0.995^2$ if you try twice.... and $0.995^n$ if you try $n$ times.

In short, you only have to solve $0.995^n = 1-0.06 = 0.94$ With $n=13$ you get under $0.94$ for the first time (93.6% chance)

Finally, there is no way to guarantee that event $A$ will produce result $X$ for any number of trials, since $0.995^n$ is always greater than zero (it actually makes sense if you think about it)

$\endgroup$
  • $\begingroup$ This was my first thought, but I like phrasing it in terms of expected value because that makes explicit that we're talking about having a 6% chance on average. It does concern me, however, that this methods gives 13 and mine gives 12...continuity vs discreteness issue? $\endgroup$ – Dave Jun 26 '19 at 14:10
  • $\begingroup$ The result was "12.something" with that something being small, but since $n$ must be an integer, I rounded up the result. The 6% chance is a 6% chance every single time you repeat the experiment, there is no average there $\endgroup$ – David Jun 26 '19 at 14:45
  • $\begingroup$ I agree with rounding up to 13. What I'm wondering is why my approach results in exactly 12, while the calculation you did results in 12.something. $\endgroup$ – Dave Jun 26 '19 at 15:24
  • $\begingroup$ Could you please provide further details about how you get 12 as the answer? $\endgroup$ – David Jun 26 '19 at 15:29
  • $\begingroup$ N = 0.06/P = 0.06/0.005 = 12 $\endgroup$ – Dave Jun 26 '19 at 15:50
1
$\begingroup$

Assuming that it's not a problem if X happens more than once, you can just reverse the probability and calculate the answer very simply. If event A has a 0.5% chance of producing result X, that means it has a 99.5% chance of not producing result X. How many times do you need to multiply 99.5% (i.e. 0.995) by itself, to get down to 94% (0.94) ? Let t be the number of times:

$0.995^t = 0.94$

$log(0.995)t = log(0.94)$

$t = \frac{log(0.94)}{log(0.995)}$

$t = \frac{-0.0268721464}{-0.00217691925}$

$t = 12.3441171957$

Since t needs to be a whole number, you set it equal to 13. Now you can check your answer:

$0.995^{13} = 0.93691469288$

That means if you repeat event A 13 times, you have a 6.3% chance ($1-0.93691469288$) of result X happening at least once.

$\endgroup$
  • $\begingroup$ Correct, having it happen more than once is not a problem. Thanks $\endgroup$ – Klors Jun 26 '19 at 14:26

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.