Comments on: Morning Briefing for June 20, 2012

By: lineholder

lineholder — Wed, 20 Jun 2012 21:04:17 +0000

And I do mean thanks! I had tried to figure it out but I’ve never seen any type of double-weighting done on samples. Decision matrices…yes, but that usually involves realigning subsets mid-stream in the process.

Plus, this poll actually has two data subsets included in the poll that they might have been used for weighting age stats…one of the two being a subset that simply states “over or under age 65″ and the other being a more comprehensive breakdown of age divisions.

It’s hard to tell with some of these polls whether they are simply reporting stats or whether there is an effort to utilize stats for the purpose of influencing public opinion. I’m still very much so in the learning curve on differentiating between the two.

By: tnfriendofcoal101368

tnfriendofcoal101368 — Wed, 20 Jun 2012 20:40:09 +0000

When I take my dog, Sydney out into the backyard for her daily run, if any neighbors are about at that time Sydney barks at them. A neighbor once asked me “why don’t you teach her better”. My response was “she’s a dog, she barks, can’t blame her for a being a dog” which is why she only goes out to use the bathroom and her daily run for exercise otherwise she is closed indoors. Likewise as Sydney can not be blamed for being a dog, Paultard nutjob can not be blamed for being a Paultard nutjob…they can only be locked up somewhere so as not to cause harm.

By: tnfriendofcoal101368

tnfriendofcoal101368 — Wed, 20 Jun 2012 20:33:51 +0000

I’d really be interested in how they did it…just guessing but they probably assigned weights to the repondents and then took the weighted average i.e. if the sample was of percentages were
20% Caucausian old (CO)
10% Latino old (LO)
10% African old (AY)
30% Caucasian young (CY)
10% latino young (LY)
15% African Young (AY)
2.5% Other Young (OY)
2.5% Other Old (OO)

now say I compared to the general population which is
35% CO
5% LO
5% AO
25% CY
15% LY
10% AY
3% OY
2% OO

Now to remove my sample bias, I’d assign linear weights to each of my sample subsets coming up with the following formula….

35/20*CO+5/10*LO+5/10*AO+25/30*CY+15/10*LO+10/15*AY+3/2.5*OY+2/2.5*OO

Basically, I define the subsets based on age and race, then assign weights based General Population Size/Sample Population Size*Subset response. I am sure there are other ways but I generally go for the simpliest to explain (and my job is data analysis target vs actual performance – so a polling analyst will probably have a different idea).

By: gekster

gekster — Wed, 20 Jun 2012 19:05:46 +0000

We will change the rules so we can win.

Sounds like liberals, don’t they.

Now what were all you Paulbots saying about the RP supporters?

By: PowerToThePeople

PowerToThePeople — Wed, 20 Jun 2012 18:41:45 +0000

and each and every Paultard who is involved in this lawsuit and every single one who pulls any shenanigans at the convention should be removed the convention and banned from every returning.

By: lineholder

lineholder — Wed, 20 Jun 2012 18:20:25 +0000

http://politicalticker.blogs.cnn.com/2012/06/19/rnc-calls-frivolous-lawsuit-by-ron-paul-supporters/

By: lineholder

lineholder — Wed, 20 Jun 2012 18:19:44 +0000

Consider this an “inquiring minds want to know” question, tnfriendofcoal

How does the process work? When a poll has two data elements that are used for weighting purposes…how does that work?

I’ve seen single-element weighting. And I’ve seen decision matrices used after applying something along the of a Pareto analysis for the purpose of weighting an option that might be considered highly-significant in a decision-making process.

But how does using two data elements work on a poll like this?

By: tnfriendofcoal101368

tnfriendofcoal101368 — Wed, 20 Jun 2012 18:04:25 +0000

Let me counteth the ways (I am in the data analysis business)…
1) it is wildly inconsistent with every other poll, even by known hacks like PPP – in the data business we believe in consistency – the electorate will not suddenly change in the manner Bloomberg describes not in the real world. If I had been the lead analyst for Bloomberg…I’d have put another poll in the field
2) The data shows inconsistency – 27% Protestant but 31% evangelical…33% Republican/31% evangelical. A majority of the people describe themselves as other on the question of Protestant or Catholic (which tells you in and of itself they oversampled Democrats).
3) Given both the external and internal inconsistency, the data analyst should have thrown the poll out and ordered another one put in the field but as Al Hunt (who is the person who would make that decision) said on Morning Joe: “I believe the poll”. Believes it because he wants too.

By: lineholder

lineholder — Wed, 20 Jun 2012 16:59:40 +0000

http://www.bloomberg.com/news/2012-06-20/obama-leads-in-poll-as-voters-view-romney-as-out-of-touch.html

By: lineholder

lineholder — Wed, 20 Jun 2012 16:56:17 +0000

Yes, I know….weekend polls tend to lean D.

Poll demographics show 38D/26I/33R.

According to internals, the poll is weighted by race and age. I know what a weighted sample is, but I’ve never actually seen sampling that is double-weighted, so to speak.

How does that work? Is there are way to tell by looking at the poll results how the weights were determined or what impact it had on the results displayed in the poll?

http://media.bloomberg.com/bb/avfile/rQyA68BW5P20

Any and all input appreciated. Thanks.

By: commonsenseobserver

commonsenseobserver — Wed, 20 Jun 2012 14:26:20 +0000

http://weaskamerica.com/2012/06/20/ia-mi/

I can’t find the internals though.