Connecticut Senate Race
August 23rd, 2006 | by Dr. Forbush |Here’s an interesting set of data from the Connecticut Senate race. Real Clear Politics does a good job gathering polling data from multiple sources. In general this results in better results. What they do with the data is average the surveys together. In this case, however this was a mistake.
There are three polls in this survey. The Quinnipiac survey taken 08/10 - 08/14 shows Lieberman with a 12 point lead. The other two later surveys, American Res. Group 08/17 - 08/21 and Rasmussen 08/21 had similar results showing Lieberman with a 2 point lead. The average is slightly dishonest saying that Lieberman had a 5.3 point lead. It should be quite clear that the Quinnipiac survey is an outlier.
Of course, you can go look at the data yourself.
It looks like this is shipping up to be a close race.
2 Responses to “Connecticut Senate Race”
By Matt Browner-Hamlin on Aug 23, 2006 | Reply
I actually really disagree with this. Quinnipiac has the largest sample size by far (1083 vs 790 vs 550) and is a CT based polling outfit. The long sample duration (4 days) might make it less than accurate, but my guess is that Lieberman’s lead is more like 5-7% among all likely voters than 2%.
I don’t think it’s good to get hung up on the actual numbers 12 vs 5 vs 2… Quinnipiac makes more sense in the context of trends - last month the QPoll had Lieberman up 20 points in the 3 way. Now it’s down to 12. I like that trend much more than small polls showing a statistical tie simply because it has proven to be true. The polls that showed Lamont with 13-15 point leads in the primary were proved wrong and the Q-Poll’s smaller lead proved right.
Look, you know I’m about as hardcore a Lamont supporter as there is out there. I’m ecstatic to see polls showing a tie. If more polls show Lamont even or ahead in the three way, I can even hold out hope that Lieberman will drop out. But more realistically I’ll take the steady positive trend that is evidence of the fact that the more CT voters of all political stripes get to know Lamont, the more they like him.
By Dr. Forbush on Aug 23, 2006 | Reply
I apprieciate you pointing out the sample number, but what does this really mean?
Well, a sample of 1000 has an error of +/- 3%. How do they get the error? It is a formula given by 1/sqrt(N) where N is the sample number. So, we can say that the errors on the samples are:
550 —> +/- 4.2%
790 —> +/- 3.5%
1083 —> +/- 3.0%
So, you can see that there isn’t much difference in the sample sizes and the results. You might even come to the conclusion that 5% is within the error. The point however is that the Quinnipiac poll was taken a week earllier. So, perhaps with all of the support of the “major” Democrats Lamont is winning over supporters quickly. Also, the last two surveys were taken at about the same time and have a total sample of 1340 and a similar result. With the total sample of 1340 there is an error of only +/- 2.7%…