For instance, assume there are four random variables, Cloud, Mega Man, Ryu, and Terra. Assume their (mean) strength's are 47.6%, 35.5%, 27.5%, and 17%. Thus their realized values are between [0, .952], [0, .71], [0, .55], and [0, .34]. This mimics voter preferences - the model is that for each voter, each character gets a random value between their ranges, and the character with the highest realized value receives the vote.

I thought about this for a bit, but then I found an apparent flaw: let's say Link is at strength 1.0, while another character is at strength .50. That characters getting 25% on Link makes sense, since every time a random number is between .50 and 1.0, Link will always win (50% of the time), and then when Link is below .50, he will win half the time on average (50%*50% = 25%). So, that adds to 75%. But... think about this: let's say Link is at strength 1.0, and he faces a million opponents at a strength of .5 (just pretending). Using that line of thinking, Link -will- get at least 50% of the vote, since he'll always wins with a random number of .5 or higher. That couldn't be true. So, in that case, this line of thinking will always inflate Link's expected percentage.

But, as to attempt to explain Terra's percentage being supposedly too low, that's over my head.

---
"Ah, a party! We haven't had one of those. It could be fun! So...what is a party?"
"Well, you drink punch and eat CAKE! ...I think."