General Discussion
Related: Editorials & Other Articles, Issue Forums, Alliance Forums, Region ForumsImpossible results all throughout every single Gravis poll this year
Hey guys, I've missed you all! Grantcart asked me to take a look at the Gravis polls and give my opinion. Well, if you know me, I can't leave a statistical stone unturned (blame my OCD lol), so I looked at them all for this year. Every single poll shows crosstabs with mostly impossible results, since they calculate out to fractions of individuals instead of calculating out to full integers of individuals.
The Logarchism article shows one example of this type of result with mostly impossible results. I take it a little further and illustrate a few more examples below at each link:
https://www.box.com/s/9nfb9uqdaqcyf9jt8v2p
https://www.box.com/s/3jj5pcubo2ayvexqarno
https://www.box.com/s/b0qcnf6b0dbe4nps44ii
Gravis Marketing's polls are absolutely nothing but frauds, in my opinion. It's incredulous to me to see their polls given any weight, or any serious consideration for that matter. Print them out and line the kitty crapper with them, that would be a better use for them.
(Sorry for the links instead of embedding the pictures, I forgot how!)
- Phrig
Panasonic
(2,921 posts)Was wondering if you'll be making an appearance soon...
phrigndumass
(5,809 posts)cthulu2016
(10,960 posts)It sems to me that any weighted poll (which all real polls are) can end up with fractional people.
If your sample has more "other/unsure" responses than your demographic model then wouldn't the weighting of the sample reduce each response from an "other/unsure" to a fraction?
I wouldn't be surprised if Gravis did not poll anyone at all and just made up the numbers. I am satisfied that Gravis is a RW fraud created to influence polling averages.
I am just not sure that fractional people are a major part of the critique of Gravis..
phrigndumass
(5,809 posts)In my illustrations I gave room for fractions up to a tenth for that very reason, weighted polls can end up with fractional people. But only up to a point, and never half a person, or two-thirds or one-third for that matter.
cthulu2016
(10,960 posts)I am not worried about their methodology much either way, since I am not sure they even have a real sample to subject to a methodology.
phrigndumass
(5,809 posts)Couldn't have said it better lol
myrna minx
(22,772 posts)His widgets and analysis were spot on in 2008. We're fortunate to have an old friend back in the fray to calm us (me all down.
phrigndumass
(5,809 posts)jsmirman
(4,507 posts)I see cthulu's point, that obviously, you do a "re-weighting" essentially, to turn your survey respondents into a properly weighted sample (to match the right weights suggested by the demographics).
Could you work us through the math as to what the limits are in a "re-weighting" that can only spit out a "partial person," diminished, at most by a tenth due to re-weighting?
jsmirman
(4,507 posts)grantcart
(53,061 posts)If you followed DU in 2008 you know that Phrig's model was as good as anyones.
I challenge any statistician to disprove his assertions.
phrigndumass
(5,809 posts)I was showing 365 to 173 as late as two days before the election! (btw you helped)
Raster
(20,998 posts)Thank you!
jsmirman
(4,507 posts)can't wait to read when I get back!
LiberalAndProud
(12,799 posts)could be explained away. I'd like to read Gravis' rebuttal to your analysis, but I'm not holding my breath.
And it *is* good to see you here, phrig.
phrigndumass
(5,809 posts)Might turn blue! Great to see you here, too!
reflection
(6,286 posts)Nah, the cat would just look at them and say "That would be redundant."
phrigndumass
(5,809 posts)Doremus
(7,261 posts)I too remember your 2008 posts very well. Glad to see you back.
:
phrigndumass
(5,809 posts)grantcart
(53,061 posts)Back to the top!
myrna minx
(22,772 posts)phrigndumass
(5,809 posts)Hi myrna minx! I think I still have a picture of that fainting couch, maybe I'll try and dig it out later, lol ...
It's terrific to see you here, too!
myrna minx
(22,772 posts)[IMG][/IMG]
It really is wonderful to see you. It hasn't been the same without out you!
phrigndumass
(5,809 posts)Break out the bon-bons and drape yourself elegantly over the arm! You have mad decorating skills, sister!
I had to prioritize this year, so unfortunately election-tracking was out. But this election cycle hasn't been the same without seeing you lots and lots!
myrna minx
(22,772 posts)Hey, every statistician needs a Gloria Swanson type to drama up their threads.
phrigndumass
(5,809 posts)Waiting For Everyman
(9,385 posts)Someone should check their ID's.
phrigndumass
(5,809 posts)Zombies don't have ID's, only a thirst for blood and flesh. Just ask a zombie for his ID and I'll bet he looks at you funny and eats your brain.
OldDem2012
(3,526 posts)Spazito
(50,165 posts)Your work in 2008 was fascinating and very informative, I remember it well. Thanks for this analysis and I totally agree with you when you state your opinion that "Gravis Marketing's polls are absolutely nothing but frauds".
I do not understand why Nate Silver continues not only to use them but, at times, gives them a very high rating. It makes me much more sceptical of his analysis than I otherwise would be.
phrigndumass
(5,809 posts)Thanks for your kind words. Perhaps Nate Silver believes in the law of averages, since there are a couple D-leaning pollsters out there. What he might not understand, though, is that there are far more R-leaning pollsters than D-leaning, making the law of averages unworkable here, unless you weed out some of the R-leaning and balance them. I believe Nate Silver's results are skewed to the right this year, just by looking at them.
Spazito
(50,165 posts)based on whether the poll he is using is understood to be one biased toward either the repubs or the Dems. His use of this sketchy poller is, in itself, an outlier and definitely causes, imo, his results to be less trustworthy than they might.
Using Gravis Marketing makes me think, right or wrong, Nate is focusing on making the numbers work for him rather than letting the numbers speak for themselves.
phrigndumass
(5,809 posts)"making the numbers work for him rather than letting the numbers speak for themselves"
Exactly!
MuhkRahker
(104 posts)Thanks for your work phrigndumass!
phrigndumass
(5,809 posts)68 Rex
(81 posts)Spazito
(50,165 posts)It is quite educational and eye-opening as to the polling done, the owner, etc.
Well worth the read, imo.
MuhkRahker
(104 posts)So KICK!