which only has Clinton +2 over Trump in either 2-way or 4-way race.
Looking at the detail for the release of the complete poll, I do not find a breakout of the percentage of the poll that is Hispanic or Black or white, nor a breakout of the distribution by age group, etc. That would be helpful, because while it might be possible to reverse calculate some of that, it is not possible to calculate all. Also, it would be helpful to know the sample sizes and thus the MoE of the various subgroups, since some of that data seems questionable to say the least.
I will also note upfront that in general Quinnipiac seems to understate the percentage of voters that will be of color, particularly the percentage that will be Latino.
Consider just these two items — distribution of Black voters and of Hispanic voters in the 2-way model, Clinton numbers first:
Blacks 91-01
Hispanics 50-33.
Neither seems on target. Yes, it is believable that Clinton would draw 90+% of Blacks, but I would still expect Trump to draw at least 4-5%. But it is the Hispanic numbers that seem really problematic. I doubt anyone thinks either of the numbers nor the margin between the candidates is realistic. That would have Trump doing significantly better among Hispanics than did Romney, and Clinton significantly worse than Obama. In fact, anything showing a maring of less than twice the 17 point margin Q is showing should be more than suspect, imho. It may be, as the folks at Latino Decisions make clear, that to get an accurate reflection of Hispanic attitude you need to use bilingual live interviewers.
But there’s more to question.
The poll shows a 13 point margin for Trump among men, 47-34, while Clinton has a 17 point margin among women, 50-33. For that to convert to a net 2 point margin for Clinton seems to imply an equal number of men and women in the sample. And yet, there has been no recent election in which exit polls or analysis of people actually voting where women have not been the majority. If one presumes a 53-47 female to male distribution, which is what most other pollsters seems to be using.
So what if you took the male female breakouts in Q and did the calculations assuming a 53-47 distribution of male to female? Here are the calculations giving female and Clinton shares first.
Candidate W + M = Total
Clinton 26.5 + 15.98 = 42.48
Trump 15.51 + 22.09 = 37.6
In other words, simply with a proper distribution of men and women, Clinton’s lead would be more than 5 points, similar to some other recent polls.
There is a source of data other than exit polls well worth examining, and you can see it here. You will note that 53.7% of the turnout in 2012 was female. That has been consistent — also 53.7 in 2008 and 53.5 in 2004.
By contrast, the pattern for the groups among which Trump has done best — Whites, especially without college degrees — have been decreasing as a share of the electorate while groups in which he does poorly -people of color, especially Hispanics and people with college degrees — have been increasing.
Thus I examine the data in this particular Quinnipiac poll with more than a little skepticism.
I truly believe that no network should be reporting the result of any poll that does not release ALL data — that includes the breakouts by subgroups, the margins of errors of subgroups, and the actual questions asked and in the order they are asked. Top line numbers can be very distorted by lots of factors, not the least of which is the demographic makeup of the sample as compared to the likely demographic makeup of the electorate.
Bottom line. Don’t panic over this Q poll.
Just saying…...
Wednesday, Jun 29, 2016 · 2:40:21 PM +00:00 · teacherken
It turns out Q does provide the demographic breakdown and weighting model only not in the press release. it can be found here. But I have problems with their model. They presume a 51-49 female to male, when for the past two elections it was 53.7-46.3, and in ‘04 still 53.5 -46.5. They presume a 73% white turnout. While it was 73.7% in ‘12, few other people would predict that number to hold, with the only question being whether it will be at least 70%. Consider that % of vote that was Hispanic has gone from 6 to 7.4 to 8.4 in the last 3 cycles (drawing all my data here). This Q model holds Hispanic at 8%. Given demographic changes and serious voter registration efforts, unless you think voter suppression is going to be even more effective this cycle, it makes no sense to presume an only 8% Hispanic turnout. And it is not just that Trump’s strongest demographic, white men without a college degree, has been going down as a share of the vote, 58.5- 48-44.3 over the last three cycles, in the same time the total number of people in that demographic has declined from 106 million to only 103 million.
I stand on my original assertion that the model being used for this poll is flawed. I do apologize that I had not found the demographic model they were using, but their model does not change the thrust of my argument.
Wednesday, Jun 29, 2016 · 4:13:18 PM +00:00 · teacherken
Both polling averages now include the Quinnipiac.
For Pollster.com it is Clinton 46 — Trump 39
For Real Clear Politics it is Clinton 45.3 — Trump 39.1
The pattern seems to be pretty consistent. And Quinnipiac seems very much an outlier on the lower side. The six other polls for Real Clear Politics have Clinton margins of 12, 5, 5, 10, 4 and 5.
Pollster has 8 other polls on its front page, with Clinton margins of 5, 8, 12, 5, 10, 5, 2, 9. Before you focus on that 2, let me note it is for Morning Consult, which appears twice on this list. I could find no rating for them from Nate Silver.
6/15-20 Clinton + 2
6/24-27 Clinton + 5
The +12 is ABC/WashingtonPost which Silver rates A+ (Dem lean +.6) The +10 is Reuters/Ipsos, which Nate rates at A- (Dem lean +.1) the same as he rates Quinnipiac (Repub lean +.7).