I'm not sure how much a polsters leaning effect the results. If you ask who you are voting for, the person answers with who they are voting for. still there seems to be a difference doesn't there.
Epic/MRA had DeVos ahead back when only he was doing advertising. This made sense at the time.
I have in the past said the DeVos will be another Feiger. I am still leaning that way to a point. DeVos will likely lose, by how much will be determined on how well the Granholm campaign can define who he really is.
It affects the results because they only poll a few hundred people, and good pollsters will pick a good unbiased sample, whereas bad ones will usually end up with a biased sample.
Plus, some firms can word their questions in such a way as to get the results they want.
EPIC/MRI isn't a horrible firm, but historically they have overpolled Democrats. I believe in 2002 in their final poll they had Granholm winning by about 10 points, and she ended up winning by 4. Hence why I figure she is up by about 3 or so.