Reply #148: You missed this part that the OP dropped from the EDA link:

[PeaceNikki]

"It is important to understand, however, that these polls could not be designed to control for response bias—a possible differential willingness to respond of Republican vs. Democratic voters. Therefore, the disparities presented in our polls could be the result of 1) Computerized mistabulation (i.e., rigging); 2) Response bias; or 3) Both. We simply have no way of knowing for sure.

When a voter refuses to respond to our poll, we cannot know and adjust for the partisanship of that voter, so we cannot claim that our poll is a representative sample of the voters. To construct such a poll—the kind many are familiar with that is commissioned by the major media and appears on network websites along with the returns in November—requires vastly more data, resources, and funding than we have at our disposal. Our polls were designed for a comparison of raw numbers, not percentages, and in certain circumstances, not achieved here, would be strongly probative stand-alone red flags."