Hillary Gets the Edge From the Professional Pollsters, But Gets Crushed In On-Line Polls.
Credit:
Buzzfeed is a news aggregator with a left bias. Like many websites, it has a presidential poll: Who Are You Voting For In The U.S. Presidential Election? Here are the results, which are pretty funny
Buzzfeed says of it own poll “Update: This poll no longer accurately reflects the opinions of real people. Please remember to vote in the upcoming election.” 15 million for Trump, 70K for Hillary? I thought it a joke (it wasn’t) but someone obviously hacked the poll. Still, Hill was running well behind ‘don’t know’ and ‘somebody else’.
Some people don't like on-line polls at all, see for instance Chuck Todd clashes with Trump aide over 'bogus' online polls
CNN/ORC did a ‘scientific poll’ right after the 1st presidential debate that Hillary won 62% to 27%. But CNN admitted that the 521 voters included in the sample leaned 26% toward Clinton before the debate – that is they over-sampled Democrats by that amount. So, that poll doesn’t say much except that it’s child’s play to bias a poll.
Most on-line polls went strongly for Trump, with sample sizes that dwarfed the ‘scientific’ poll. Some examples:
Drudge: 82% Trump, 1.1 million votes
Time: 53% Trump, 1.7 million votes
CNBC: 67% Trump, 1.2 million votes
CBS NY: 55% Trump, 119K votes
Slate: 54% Trump, 42k votes
Smaller polls from local TV websites reflected similar results, with a couple pro-Hillary results. See, e.g. The Daily Mail
The spinmeisters are out telling us that Trump lost the debate badly and that his polls numbers will be adversely affected by 2-5 points. They dismiss with a sniff the ‘unscientific’ Internet polls for lots of make-weight arguments.
I’m not so sure. Surely the Buzzfeed poll was hacked but the others cited above probably weren’t. Even if you throw out Drudge as an outlier and average the remaining four (all of which websites have a pro-Hillary bias) the result is 57% for Trump. That’s how Real Clear Politics does it with ‘scientific’ polls, and RCP is considered the gold standard.
The LA Times/USC national daily tracking poll is a new experimental survey that touches base with about 2650 people everyday. Here’s the result up until today
It shows Trump with a 5.6% lead after the debate, just at the limit of the error bars. People claim the LA Times poll is ‘flawed’ for some reason.
In truth, ‘scientific’ polls are really “push-polls”, that is the result is intended to influence an election not measure it. This is done by turn-out models built into a poll. Pollsters guess how each demographic will vote and the percentage of the electorate they will represent. So, the poll results can be predicted without taking the poll. Polls only start to reflect the ground truth as election day nears—pollsters try to earn bragging rights by calling the election by removing bias from their turnout models. Every four years, the cries go out near election day that “the polls are tightening”. That’s what I like about the LA Times tracking poll—the pool isn’t adjusted over time.
In summary, an average of relatively ungamed Internet polls with upwards of a million participants might have as much validity as a poll like CNN/ORC consisting of 524 people that over samples one party by 26%. Does the theory of large numbers have meaning?
Internet polls are telling us something, and should be given more credence as the voice of the people. Of course, the last thing the pollsters, politicians and media want to hear is the voice of the people.