2016 Postmortem
Related: About this forumGood news: Obama ahead by 2% in OH (Rasmussen)
This is good news because:
1) Obama is ahead despite this being a Rasmussen poll
2) This is a Likely Voter screen
3) Other polls (PPP, Quinnipiac) show Obama's lead shrinking in NM and Virginia, but this did not generalize to Ohio.
http://www.rasmussenreports.com/public_content/politics/elections/election_2012/election_2012_presidential_election/ohio/election_2012_ohio_president
pnwmom
(108,980 posts)These close poll results are starting to drive me crazy.
Marzupialis
(398 posts)MOE doesn't make it a tie. Mark Blumenthal, polling expert, says the expression "statistical tie" is useless, because a candidate who is ahead by any percentage is more likely than the other candidate to end up ahead in the actual results.
So let's stop retending that a 1% or 2% or 3% lead is a "dead heat." Let's leave that up to the pundits.
http://blogs.wsj.com/numbersguy/whats-a-statistical-tie-anyway-234/
Why turn Obama's 2% lead into 0% instead of 4%? Why don't we just say Obama is up by 6"% because the margin of error is for example 4%? Why subtract instead of adding?
pnwmom
(108,980 posts)Last edited Thu Jul 19, 2012, 05:50 PM - Edit history (1)
He couldn't have been talking about the results of a single poll. Perhaps you are confused, and he was talking about a series of polls, that all showed one candidate slightly ahead of the other (but within the margin of error).
I also worked for a major polling organization, and when we reported our results, we were always careful to not pretend that our results had more precision than they actually do. When we reported a result that was within the margin of error, all we could truthfully say was that there was a 95% likelihood that the sample result was within the margin of error, whatever that margin was (which depended on the sample size). So if there was a MOE of plus or minus 3%, then there was a 95% chance that a 49% approval rating (for example) was between 46% and 52%.
Edit to add:
I think I found the statement of Blumenthal's you were referring to. He's objecting to the misconception that "statistical tie" means that two candidates are actually "tied." It just means that the results are too close to call. In other words, we can't say statistically that one candidate is behind or ahead when the results are within the margin of error.
He adds that when a candidate is in the lead in a properly conducted poll, then there is at least a somewhat greater than 50% chance that that candidate is actually in the lead; in other words, depending on the lead and the margin of error, the odds of the result being correct are at least somewhat higher than a coin flip.
http://blogs.wsj.com/numbersguy/whats-a-statistical-tie-anyway-234/
Several polling experts told me they dont care much for the term statistical tie. Its certainly not a term that practitioners use, Nancy A. Mathiowetz, president of the professional group American Association for Public Opinion Research, told me. Mark Blumenthal, editor and publisher of Pollster.com, added, Statistical tie is one of those expressions I wish we could do away with.
Their gripe with the term: It suggests that the statistics indicate there is a tie between the candidates. Instead, its used when statistics cant demonstrate a difference between candidates support. The term is used when we lack the statistical power to know for certain which candidate is ahead, said Mr. Blumenthal, a former longtime Democratic pollster. It doesnt mean we know that they are tied.
A poll is said to show a statistical tie when two candidates numbers fall within the sampling margin of error, a measure of how confident pollsters are in their results. As the number of interviews goes up, down goes the sampling margin of error (called that because it doesnt include other potential sources of error, such as non-response).
SNIP
That said, when a candidate is in the lead in a properly conducted poll, there is a greater than 50% chance that the candidate leads in the broader population. How much greater than 50% depends on the size of the lead and the size of the sampling margin of error.
Hippo_Tron
(25,453 posts)Whether you look at it as 95% confident that Obama is up by 2 +/- 4% MOE or 51% confident that Obama is up by 2 +/- some number that's smaller than 2%, you're still looking at a measurement without a whole lot predictive value.
Add in to that other sources of error: bad questions, bad sampling, not really being able to determine who is going to turn out.
Rassmussen also uses IVR, which doesn't allow them to poll cellphones. Furthermore, it's entirely possible that an 8 year old can pick up the phone and just press the buttons. IVR has no way to actually make sure that they're collecting responses from adults who are registered (let alone likely) to vote.
Finally, with 47%-45%, there's still 8% undecided.
What this poll shows is that Ohio is extremely competitive and either Obama or Romney could win. Trying to determine who has a "lead" or an "edge" from numbers this close and methods this imprecise, is not very useful.
bigdarryl
(13,190 posts)you guys and these polls