Fox News Promotes Batch of Agenda Polls

Conservative Tree House ^ | 01/08/2016 | Sundance

[Greetings_Puny_Humans]

[...]

While these might seem like favorable polls for Trump and/or Cruz, it doesn't matter when the sample data is intentionally manipulated to produce a desired result.

When looking at any polling methodology two things are first reviewed. The raw numerical data of the sub-sets (men, women - ages, races, etc.) how many are there, and the Margin of Error (MOE) for each category.

Any polling data sample that is possibly over/under stated, and therefore requires the polling result to be modified to fit the pollsters "assumptions", carries a higher margin of error (MOE).

[...]

♦ If you were to poll gender, there is only two possible sub-sets: Men or Women. If you poll 1000 respondents and get 510 women (51%) and 490 men (49%) you would have a perfect sample of the general population. [Which is 51/49 women/men] Again, your MOE would be -0-%

However, if you were polling for a historical assumption of voting: 53% women voting -vs- 47% male voting (which is traditional turnout) with the above data you would have a MOE of +/- 4%. Why, because you are polling 51% female and need 53% female for your assumption – that’s 2%; and the reverse is true for the male sample (polling sample 49, traditional turnout 47). 2% for each sample leads to a combined MOE of +/- 4%

Hopefully that *roughly* explains how it works.

So, with that in mind when you look at the Fox poll data what sticks out? First, the raw number within the sub-sets is hidden, they don’t show it. That’s a big red flag.

[Greetings_Puny_Humans]

Rest at link

[SubMareener]

Interesting analysis. I don’t think this is going to sway the outcome this time around. There is a larger player in the game and He always gets His way in the END.

[tallyhoe]

Yea Ted’s pretty good at that.. It’s the ground game that wins!!!

[nopardons]

Don’t count your chickens before they’re hatched.

[BigEdLB]

Overtime Politics releases IA next week. I wonder if CBS or CNN will have something. What happens other than a bunch of suicide watches on Fox if Trump wins IA by 10? Will, Rove, Kraauthammer, York, Meegyn, Goldberg, etc etc will have to be watched carefully.

[chopperman]

I don’t think margin of error is figured that way.

I believe that actual sampling can be weighted to compensate for the proportionality problem he mentions.

I think the MOE error is more related to how small the sampling of the poll is versus how large the expected voting population would be.

[BigEdLB]

Neil Cavuto has been probably upsetting management at Fox. “Trump support is understated” and “If folks are willing to wait 2 hours for a rally, they are probably willing to spend two hours at a caucus”

[wildandcrazyrussian]

This guy obviously has never studied statistics, and is misleading by inartfully explaining it. I only read the excerpted paragraphs, but that is sufficient to pick out one main error. As you say, Margin of Error is an function of sample SIZE, calculated as 1/square root(number), and has nothing to do with whether or not the sample really is random, biased, etc.

Basically, "assuming that the sample was randomly selected and an attempt has been made to eliminate the most obvious sources of bias, there is a 95% probability that the selected subset is within MOE% of the true results if one were to poll the entire population" -- which still leaves 5% probability due to chance that the results could be considerably different. False precision is the enemy of understanding.

[PghBaldy]

I agree the author is being sloppy, but he has a MAJOR point: the MOE of the subsets are hinky. I looked at them - some as high as 9.5%

The Iowa poll is worthless.

[FrankR]

Fox News is dead to me.

[Caipirabob]

Wow...good thing I longer have access or pay for FAUX news.

One America News...via subscription on Roku and other devices.

[LS]

There are a couple of "statisticians" on Twitter like Jay Cost and Adrian Gray (both of whom said, or strongly implied, that Romney would win) and they are constantly playing these little statistical games. They have been trying, for ex,, to claim that the big leads for Trump represent stupid people and that there are fewer college grads involved than in previous "good" elections like 2008 and 2012.

I am definitely no statistician. But I can look at a poll. The last two Overtime Politics polls for MO and FL polls, for example, don't give education but they do give income. So I think it's reasonable to use $45,000-64,000 and $65,000 and up as ---not perfect, but good---proxies for college educated. Well, Trump utterly dominates these categories in these polls.

[parksstp]

Yeah, let the liberal Trump supporters show up at the caucus and spew the same leftist garbage they spew here “He can win. I don’t care he’s not a conservative. I don’t care about any positions he has or substance. He’s gonna build a wall and make America Great. Oh, and Cruz sucks and is just copying him”.

In a Caucus setting? I guarantee you that Cruz supporters won’t tolerate that crap and will embarass the hell out of the Trumpsters. You think Jeffrey Lord looks bad on CNN when he’s gotta blush over the dumb crap Trump says? You ain’t seen nothing yet.

Cruz: At least 35% in IA, maybe 40% or more on Caucus night.

Trump: He’ll be lucky to hit 27. More likely 25% or less.

[HoosierDammit]

It’s the 2016 version of the “unskewed polls” movement of 2012. Instead of “look at the party breakdown!”, the rallying cry this year is “look at the internals!”

Pseudoscience explained using real statistical terms is still pseudoscience.

[chopperman]

I think election polls should at least have a sampling size of 1000 and only be of likely voters. Otherwise, the pollsters are just lazy and should be ignored.

[tallyhoe]

What do you mean?? Don’t count your Chickens before their hatched?? I wasn’t counting chickens..