Polling Data Breakdown

[mezz]

You dems want some bad news? Of course you do... your identity would be in doubt if it weren't for negativity in your lives.


The election is just a few days away and lots of folks are watching the polls. One of the easiest places to get a good snapshot of the polls has been Real Clear Politics which collects the data from various polling agencies and pulls them together in one site. One of their more popular screens has been their general election tracking poll which presents data from different polsters over the past week and averages them together to give the viewer a look at the daily pulse of the polls at a quick glance. So how accurate is this 'glance' at the poll data? Well, potentially not very.

Before proceding I wish to disclaim that I am any kind of statistician (or English professor for that matter) so my terminology may be a little off here but in laymans terms and with a laymans statistical analysis (which you'll hopefully find as simple to follow as I do) here are my findings.

There are some obvious deficiencies built in to the simple way in which RCP presents their averaging data. Without getting into any debate on the merits of which methodologies are used by the various polsters, nor entering into a dissection of their accuracy history, let's assume they are all equal in merit (that's what people do when they glean data from the RCP General Election Weekly Average screen anyway). So simply looking at the data presented on the page, (See here: http://www.realclearpolitics.com/epo...bama-1171.html) for each listed polster they provide the date, sample size, the polster's stated margin of error, then the poll results. The top line is devoted to RCP's average which in today's screen sits at Obama +0.3. This RCP average is simply an averaging of all the polling data without any weighting. The sample size and the MoE are simply provided as footnotes and not taken into account (or weighted) in the average.

Looking a little more closely, we can ignore the polster provided MoE which if you look at it will basically indicate a lower MoE for a higher sample group. This makes sense, but the MoE is a rounded estimate and isn't weighted into the averages by RCP anyways. So if we want a more accurate averaging from the data being provided by RCP we can pretty easily apply a weighted mean to come up with more accurate data.

So looking at today's (Nov 2) RCP General Election screen we can see the average sample size is 1232.2. So taking each poll sample we can get the deviation (+ or -) from the average sample size. Taking that deviation and multiplying by the average (1232.2) will give us a % deviation factor to then be multiplied by the difference (+ or - for O or R) for each poll to give us a weighted result of the difference for each poll. Average these together and we get a far different number than the simple unweighted RCP average.

RCP Average = Obama +0.3
Weighted Average described above = Romney +5.8

Put that in your pipe and smoke it dems... or just attack the messenger as per usual. I'm comfortable with it and it's part of the reason I think we're going to clobber you dimwits on Nov 6th.