From September 1 to October 13, the Reuters/Ipsos poll has been gradually and incrementally oversampling Democrats and undersampling Republicans as a percentage of likely voters:
Note that this isn’t showing a secular change in voter party identification year-after-year. It shows an oversampling bias for one party over a period of weeks. The number of likely voters did not suddenly become 18% more Democrat in six weeks time.
If you can’t see the graph clearly, click here for a zoom-in version.
Why would polling outfits rig their results to favor Hillary “thecunt” Clinton? My theory is the simplest explanation: