A Break for Senate National Polling Data

Comments

1. fredct says

   November 7, 2006 at 2:58 pm

   Forgive me if my definition of MoE is wrong, but my understanding is 4% is what you could add to one candidate and subtract from the other.

   <

   p>
   So none of these are outside the MoE.

   <

   p>
   But I do agree that I’d rather be on this side of the error, especially considering several polls over the last few days (www.pollster.com) seem to give Dems a small lead in the swing states.

   <

   p>
   Nothing we can do now but wait. Oh, and if you’re home… make calls for Call for Change! (callforchange.org)

   • david says

     November 7, 2006 at 3:05 pm

     explained in full here!

     • fredct says

       November 7, 2006 at 3:29 pm

       I couldn’t read all of tbe comments, but there’s only one thing I wanted to add.

       <

       p>
       The +- x% is only really valid in 2 way races near 50% each. You said that ‘undecides’ were 8, so therefore they’re anywhere between 3 and 13% with 5% MoE. That’s not true.

       <

       p>
       the same logic could mean that with Grace Ross polling at 2% in a 5% MoE poll, that she can be as low at -3%. That’s of course silly. The MoE is somewhat porportional (not saying its linear) to the % of the vote.

       <

       p>
       Someone polling a 50% may have a 5% MoE, but someone polling at 10% would have significantly less. Saying 1% (5% / (50%/10%) ) would be wrong, but it’d be the general idea.

       <

       p>
       Still, great post 🙂

       • weissjd says

         November 7, 2006 at 4:29 pm

         It’s worth keeping in mind that sampling error will tend to look like a bell curve. There will be a lot more polls that are well within the MoE than are right at the edges.

         <

         p>
         On the other hand, most polls use a 95% confidence level. That basically means that their 95% sure that the poll’s sampling error will be within the MoE. So 1 of 20 polls is outside of that. Also, this is random sampling error. It doesn’t account for bias that was introduced by any other means which could include:

         <

         p>
         * Calling at times when you’re more likely to reach voters for one candidate or the other,

         <

         p>
         * one candidate’s supporters are more/less willing to talk to pollsters than the other’s,

         <

         p>
         * you only called land lines and one candidate’s supporters are more likely to use only a cell phone.

         <

         p>
         That’s why the occasional poll is way off. But it’s true that being ahead within the MoE is much better than being behind within the MoE. That goes back to the bell curve…

         • fredct says

           November 7, 2006 at 5:08 pm

           There are plenty of other factors that are non-numerical. All we can really say is the numerical aspects.

           <

           p>
           Btw, found a great chart on the wiki:

           <

           p>
           http://en.wikipedia….
           (top chart applies to 95% confidence interval polls, like nearly all political polling).

           <

           p>
           Just as an example, Tester had a 5% lead in a 4% MoE. So pick 5% on the top, and then 4% MoE on the side, and you get an 89% chance that Tester actually has a lead.

           <

           p>
           Add that to polls in the last week showing 4 and 11 point leads, and a tie. And there’s a pretty good chance that Tester has the lead. But we can never know for sure (the 100% on those charts are rounded).

           <

           p>
           Just to point out another source of error, there’s always the question as to how well their likely voter models predcits those who actually show up to vote.

           <

           p>
           Btw, as yet another knock as the silly topic of ‘statistical dead heat’, that chart shows that in a 4% MoE poll, a 2% lead means a 68.8% chance of actually having a lead. Not overwhelming, but its still not a ‘who knows’ thing.

           • weissjd says

             November 7, 2006 at 5:25 pm

             is a term I really hate. As you say, it’s not true. Because of the bell curve, it’s more likely that the candidate who’s just a bit ahead really is ahead. If the poll’s within the MoE it just means there’s at least a 5% chance that it’s tied. The closer the gap is to the MoE, the smaller that chance is.

             <

             p>
             Another thing to consider is that 95% confidence is considered quite low for some things. For example, most research studies that look for correlations use a 99.5% confidence level. Now that’s confidence!

           • weissjd says

             November 7, 2006 at 5:45 pm

             I really like the charts from wikipedia, thanks for posting that. It illustrates quite well what a lead within the MoE really means.

     • johnk says

       November 7, 2006 at 3:33 pm

   • alexwill says

     November 7, 2006 at 6:00 pm

     for rough use of MoE, is that two candidates near 50% with a difference of the margin of error, there’s a 67% chance the leader is actually winning… one sigma away from each number…