Actually, what I mean is that by running a poll on a hockey website you are choosing your audience very specifically for people who already enjoy hockey,and are possibly (I'd say probably) more likely to see Don Cherry in a more positive light. You get results which are skewed by the polling sample. In other words, it loses a lot of it's value.
Of course there is no assumption that this poll is useful for anything really, so fire away and have fun with it. Just realize that it's pretty much useless for anything outside of gauging opinions here at Dobberhockey.
I have no idea what your link (as interesting as it is and worthy of discussion elsewhere) has to do with the problems of using highly specific population samples for polling.