Well, we are talking about basic calculations with just a few assumptions put in. As such, I would like to see a research worthy report on the subject.

In my number, I considered that in a group, a subset of the people will have a tendency to do sexual assault. If we only view the assaults as male -> female, than a 95%/5% conference will have 1.9 more such individuals with such tendencies compared to a equal sized 50%/50% conference. I also assume that the number of assaults will have a direct correlation with the number of people that a tendency towards sexual assaults.

The risk of any individual female participant will then be the number of assaults divided with the number of females at the conference. In the 95%/5% case, there will be a only a 1/10th of female participants and as such, you get 10*1.9 = 19 increased risk vs a 50%/50%. I rounded it to 20.

Your data point about the time each male spend with someone of the opposite sex is an interesting data point, but I do not know if there is a correlated to the number of sexual assaults and the time an assailant spends with the opposite sex. It need to be above zero, but beyond that I do not know. It would be effected by behavior theory of predators, and if they actively go out of their way to seek victims or not.

I see where our calculations differ. I'm assuming an assaulter can only assault one of the K < conference size people he meets, while you are assuming he can assault anyone anywhere.

You are also measuring assaults/female as opposed to assaults/human. That seems a little odd to me - if the number of assaults doubles, but the number of females triples, is that a good thing?

The individual risk for each female participant do go down if the number of female participants goes up. I would not categorize it as either bad or good, just an aspect of the nature of risk.

In the end through, I only trust my own numbers as a vague hint about what could be. There is just too many assumptions I make and factors to consider (such as group behavior, predator behavior, and so on). I hope a researcher, research student, or foundation will get interested in this topic so the answer to the question can be made once and for all.

I don't think you are calculating the same thing. Your calculations are that an assult will happen, whereas belorn was calculating for a specific woman, whether she would be the victim.

I calculated the probability of an individual human being assaulted.

Let's use your equation with k = 20, alpha = 0.5 and beta = 0.1. This gives the probability of a person being assaulted as => 1 - pow(1 - alpha, kbeta(1 - alpha)) => 1 - pow(0.5, 20 * 0.1 * 0.5) => 1 - pow(0.5, 1) => 0.5

There's a 50% chance that a person will be assaulted. Divide by the number of women (10) gives a probability of 0.05 that a particular woman will be assaulted.

Now let's do the same thing, but changing alpha to 0.05 1 - pow(0.95, 20 * 0.1 * 0.95) => 1 - pow(0.95, 1.9) => 0.09

However as there is only one woman , that 0.09 probability is all her. In other words her odds of being assaulted have just about doubled.

You've calculated the probability that _some_ individual will be assaulted. The probability that a given female will be assaulted though would be (I think):

P = 1 - \sum_{m=0}^K \choose{m}{K} (1-beta)^m (1-alpha)^m alpha^{K-m}