Long before elections, poll numbers are fabricated to steer public opinion in particular directions. Pollsters work for pay just like everyone else.
As election day approaches, pollsters incrementally adjust their numbers to more closely match the coming reality. This restores their credibility for the next election cycle.
One of the simplist methods to accomplish this is adjusting the sample distributions, by demographics, geographics, affiliations, etc., to achieve the desired outcomes. A small change in sample distribution can produce a more pronounced change in the reported numbers, just as manipulating shadow puppets just a little bit can drastically change the projected shadow.
Great post. Thank you.