Site Meter

Saturday, October 13, 2012

Polls and Reporting: nitpicking meets grasping at straws


All aggregators are doing funny things by rounding first then subtracting.  This is a comment on Singiser, but it could be on any aggregator.

In the Gallup Tracker likely voter subsample, Romney is up 1.2%  (this link points to numbers which will change)  If one must round one should round to 1 not 2.




The standard report rounds Romney's lead up to 2 %
"NATIONAL (Gallup Tracking): Romney 49, Obama 47 (LV); "



In the IBD/Tipp tracker likely voter sample Obama is up 0.7.  If one must round one should round to 1 not zero.


IBD/TIPP 2012 Presidential Election
Daily Tracking Poll
Day 4: Oct. 12, 2012
Obama: +0.7Obama 46.4% | Romney 45.7%


Read More At IBD: http://news.investors.com/special-report/508415-ibdtipp-poll.aspx#ixzz2998NIqGc



The standard report rounds Obama's lead down to 0%
"NATIONAL (IBD/TIPP Tracking): Obama 46, Romney 46"

The two average Romney up 0.25% not 1%

I often read "percents don't add to 100 because of rounding".  But somehow this is not OK when subtracting rather than rounding.

update: Singiser notes that (with the standard silly rounding) Romney's national poll lead for the day declined from 1.3% to 1%.  He notes that this is totally statistically insignificant and not worth wasting thousands of pixels to criticize as I do below.


It matters to people silly enough to try to back out daily results from trackers.  To be maximally silly I will use the rounded numbers released yesterday (actually day before yesterday here in Rome).

Gallup 10/11 minus Gallup 10/5 (silly rounded) comes up as a shocking Romney up by 1.4% (shockng as 10/5 should have been a great day for Romney) not an even more shocking 7%.

IBD/TIPP 10/11 minus IBS/TIPP 10/6 (silly rounded) comes out as Obama up 10.2 not 6.

The estimate for a normal day vs the two dread days after the debate is 4.4%  with half silly rounding not -0.5% with all silly rounding.  With silly rounding (both days for both pollsters) it looks as if the debate effect was permanent not a bounce. With silly rounding for the second latest day and not the latest day, it looks as if the debate effect was temporary.  This is all silly, because it is torturing the poor numbers to try to make them say something they don't know, but the point estimates are not small.

No comments: