EDIT: above is assuming that you can win more than once if k > 1. If not, the math is more complicated than I can be bothered to try to format in an HN post, but the principle should hold.
EDIT: above is assuming that you can win more than once if k > 1. If not, the math is more complicated than I can be bothered to try to format in an HN post, but the principle should hold.
My expected return may go down if the person in front of my wins (I have to split with more people, so each winner gets less) - but my chances of winning don't change.
I think the form of lottery I (and others) are referring to is the one where there are N unique tickets and you have a 1/N chance of matching those N numbers. Therefore, there is no dependence between different tickets and the likelihood of matching the winning numbers.
The takeaway is that "distributed uniformly at random" is not a great assumption at all when inputs are generated by a process. It's one of the relatively few things I did learn in my graduate algorithms class.