April Purple Poll: CO Tie; VA O+2; OH O+5; FL R+2

[Bacon King]

I think I understand what this "Purple Poll" is trying to do; this shouldn't be looked at as a weird assortment of useless multistate polls, but like a sorta-national poll that only polls states that have a chance of swinging the election, rather than the entire country. The state groupings with the silly names would thus be analagous to the regional subsamples of a true national poll.

Here's the number you should really take from the poll, the "mini-national" result of the twelve swing states polled.

Obama v. Romney matchup
Obama: 48%
Romney: 44%

However, delving deeper in to the PDF here, there are a lot of things that do look pretty bad. For starters, it looks like they really messed up on partisan identification- Obama does significantly worse among independents than he does overall for every single issue/character question they asked. For example, Obama wins the question, "would be a stronger commander in chief" among all voters 47%-44%, but among independents, Romney wins 45%-39%. Considering how consistent this skew is, they either overpolled Democrats or called a lot of very liberal Republicans.

And while I'm no expert on this topic, their margins of error appear to be totally bogus. To get their stated MoE of 4.1% in each of the states, they would have needed to poll 571 people in all four states- 2284 in all. That's impossible in itself, because the entire poll only had 1705 respondents. In addition, they claim the same MoE of 4.1% for each of the four regions- already impossible, because it'd certainly be lower for the "Southern Swing" due to the 1148 people supposedly polled in Florida and Virginia- but that would mean they'd need to poll another 571 people in "The Heartland,"; that brings the total they would have needed to poll up to 2855 people at minimum in order to reach their stated accuracy, and even then that's without a single respondent from five of the states.

With only 1705 respondents, there's actually much higher error here. If you assume they polled the same number of people in each state (142), that'd be a margin of error of plus or minus 8.2% for VA/FL/OH/CO.

If you assume they polled the four highlighted states more than the other eight, then that'd mean the entire poll isn't a random sample and everything except the four state results are entirely useless (and they would actually have a 5-7% MoE anyway). I guess "The Heartland" would still be possibly random, I guess, just with an insane MoE as well.


Ultimately, it seems this pollster is utter garbage. Even setting aside the inane tri-state polls, and whatever anomaly happened for their partisanship metric, this company is releasing polls that at best have insane error margins (and lying about it!) and at worst make such a poor attempt at randomized data collection that these results might as well be made up.

So, yeah, this is a meaningless junk poll through and through; this Purple Strategies group is certainly trying its hardest to be the just as respected as Zogby/ARG/Strategic Vision.

(edit: damn you adderal, why did I just spend an hour looking in to this, I am supposed to be writing an essay right now)