If you have the probability of the event happening and you want to get the data entries/samples that would successfully result in the event happening, randsample seems like an overcomplication of the problem.
Boiling down your problem as,
Generate 0 and 1 according to the fixed probability which you already have. So, in the output wherever you have 1s, that is the event happening and wherever you have 0s, that is the event not happening. And if you have a population vector for these events, then you can use this binary vector as the index for selecting only those entries where the event happened.
And the code for this is as simple as,
rand(1, n) >= x
where, x = Probability of event happening and n = Number of trials of the event (or the size of your population vector)
