Trump went into the GOP debate last night with a roughly 20% poll standing. Everyone will tell you to ignore polls early in this race, they never predict the outcome of a primary or a general election. That, however, is a non sequitur. We do not look at early polls to predict the distant future. We look at them to help understand the present, and to get a handle on what might happen over the next few weeks. The meaning of the polls shifts quite a bit before the first primaries, then they meaning of the polls has to be re-evaluated after every primary. At some point the re-evaluations start to return an end result like “Candidates A and B are in a horserace” or “Candidate A is the clear leader.” After that, you can get caught on a boat with your mistress, or you can be killed, and that can change things, but not much else does. Democrats believe in the Dark Horse but no one has ever captured one to my knowledge. But up until that point, polls are useful, and meaningful, if done scientifically, but no, the fact that they don’t predict an outcome over a year in advance is not a surprise and does not mean they don’t have interest or utility.
But what about unscientific polls?
Well, they are not scientific and thus not worthy. However, over the last few hours, several non-scientific polls, and in this case I mean internet polls where anybody who happens on a site can vote, have come out asking who won last night’s GOP primary.
If a bunch of unscientific polls that all return the same result become scientific, or at least, believable? That is a hypothesis I’d like to test with the current polling. It seems to me that if informal web based polls from across a spectrum of political orientation (of the site, not the poll clickers … we don’t know who the poll clickers are) all show similar results, then they might mean something. So, here is the hypothesis. If several informal polls show a very similar result, we expect to see that result reflected in the first scientific polls that come out.
I got poll results from the following sources (shown in order from left to right on the charts):
Sadly MNSNBC had a poll but it was fairly useless in the way it was conducted. Also, HT Politics had a poll with similar results as those above, but I found it after I’d made the graphs.
Trump was a clear winner in these polls.
Trump’s numbers ranged over several points, but are always higher than everyone else, and approached or met 50%. One hypothesis predicts that formal, scientific polls should have Trump as the front runner. Another hypothesis predicts that Trump’s numbers in a scientific poll should be between about 40% and 50%, give or take a few points.
The gaggle of low numbers is difficult to even see on this graph, so I made a second graph with everybody but Trump:
Here we see what looks to me like two tiers. Walker, Christie, Bush, Huckabere and Paul are all really low, while Cruz, Kasich, Carson and Rubio are all relatively high. Note how variable Cruz’s numbers are. But aside from Cruz, just as is the case with Trump, the results are fairly similar across the polls.
One hypothesis would then be that Walker will be shown as dead last in upcoming proper post debate polls. One could produce a number of other hypotheses as well, but it could get messy. Let’s try this hypothesis. Upcoming proper post debate polls will have a rank order statistically like this:
An additional hypothesis should probably be made, that the rank order for all the non-Trump candidates will be as shown. (This avoids the problem of having such a large magnitude of difference between the first and second rank).
There is one poll that I know of that was conducted by pollsters. It is by One America News Network, a conservative news agency that bills itself as “credible” (which is funny, why would you have to say that if you were that?)
If we take this poll by itself, most of the above suggested hypotheses are smashed. Here is the result of the poll questions “who won the debate” and “who lost the debate.”
This poll asked questions of “Republican poll participants.” It shows Ben Carson beating Trump, and a lot less spread between leader and others than the on line polls indicated. Also, very few people thought Scott Walker, who was a big looser in the on line polls, had lost the debate. Generally, the rank order between this poll and the on line polls is different.
Reading the reporting of this poll, it looks a lot like a shill for Ben Carson. Details of the methodology are as follows:
Gravis Marketing, a nonpartisan research firm, conducted a random survey of 904 registered Republican voters across the U.S. Questions included in the poll were focused only on the top ten GOP candidates that participated in the 9 PM ET debate. The poll has an overall margin of error of +/- 3%. The polls were conducted on August 6, immediately following the GOP debate using interactive voice response, IVR, technology. The poll was conducted exclusively for One America News Network.
I should add that the agency reporting the poll is owned by the company that commissioned the poll. Gravis, the pollsters, are used at Real Clear Politics. So I’m on the fence about the legitimacy of this poll and eagerly await other results.