Next Time Around, Which Polls Should you Trust?

Next time that a national election or important national issue comes by it might be important who where the batters and who where the strikers on the most recent election of this past november 2012.

I found that the ones closer to the mark or least amount of errors was YouGov. You Gov is a survey outfit. I don’t know if you have heard the name, but reading polls is important to know who puts it out. Obviously the polls put out by the people affected by the poll   (candidates) are the ones you can’t least lean on, particularly if the pool of people poled is small. 

People play around with polls, like for instance if a particular idea or candidate is being favored. You can’t trust on a poll that was done in a couple of republican or democratic states on a national election. You need the biggest pool available and the biggest geographical area. 

"YouGov conducted a three wave "megapoll" during the recent U.S. national elections, providing detailed estimates for the national popular vote and Presidential and Senate races in 27 states (a total of 51 contests). As in the 2008 Presidential and 2010 midterm elections, YouGov hit the national vote split with almost no error using a large national sample. In individual states, the smaller samples produced larger errors, but these estimates were, on average, more accurate than other telephone and Internet polls."

 Below on the first graph you can see without any interpretation how the majors or the ones closest to the mark did.

The polling was done starting 30 days before the election ending on Nov 3.                  adamfoxie*

In Table 1, we summarize the bias (mean error) and magnitude (mean absolute error; similar results are obtained if the root mean square error is used instead) for firms that conducted Presidential polls in at least ten states in the 30 days prior to the election. YouGov had the second lowest absolute bias and the fifth lowest mean absolute error. The average sample size of the YouGov state surveys was 1,562 and the average standard error was 1.26, so this magnitude of error is what should be expected. We do not claim to repeal the laws of probability.

Table 1: Bias and Magnitude of Poll Errors

                                                                                                                     

Pollster	Number of polls	Average Error	Absolute
SurveyUSA	22	0.02
YouGov	54	0.11
Angus-Reid	11	0.27
Grove Insight	17	-0.33
NBC/WSJ/Marist	13	0.33
We Ask America	11	0.35
JZ Analytics	10	-0.42
Public Policy Polling	60	0.49
Quinnipiac	10	0.68
Public Opinion Research	13	1.25
Gravis Marketing	21	1.41
Rasmussen	49	1.61
Pharos Research Group	11	-1.67
ARG	13	2.52

  

Pollster

(Error in Obama Vote,lower numbers mean less error)

		Mean Absolute Error
SurveyUSA		1.16	1.16
YouGov		1.26	1.26
Angus-Reid		1.10	1.10
Grove Insight		0.92	0.92
NBC/WSJ/Marist		1.27	1.27
We Ask America		1.42	1.42
JZ Analytics		1.64	1.64
Public Policy Polling		1.42	1.42
Quinnipiac		1.54	1.54
Public Opinion Research		1.25	1.25
Gravis Marketing		1.41	1.41
Rasmussen		2.03	2.03
Pharos Research Group		2.23	2.23
ARG		2.52	2.52