home
NBC/WSJ Poll: Oversampling | 51 comments (51 topical)

Readers Must Login or Create an Account to Comment

If they're competent pollsters (5.00 / 5) (#1)
by andgarden on Wed Mar 26, 2008 at 10:02:00 PM EST
they'll have corrected for the over-sampling with weighting. I'm sure they must have.

The reason to over-sample is so that you get a lower margin of error for the subgroups.

Think of it this way: it's very easy to take a big digital picture and make it smaller, but if you try to blow up a small (low resolution) picture, it will look terrible.

Yes (5.00 / 4) (#6)
by Steve M on Wed Mar 26, 2008 at 10:08:15 PM EST
It seems like standard polling procedure to me.  The oversampling is specifically to find out what a particular subgroup says; surely they correct for the oversampling with respect to the overall poll sample, unless they're as incompetent as John Zogby.

[ Parent ]
Wouldn't it increase the overall error? (5.00 / 1) (#61)
by ineedalife on Thu Mar 27, 2008 at 06:54:31 AM EST
After weighting doesn't that essentially convert their 700 person survey to a 602 person survey, and therefore increase the overall error rate? And don't they still have a 7% error rate within the AA population since,  after the over-sampling, it is only 175 people? It seems like they burned themselves both ways.

[ Parent ]
I'm not a poling expert (none / 0) (#62)
by andgarden on Thu Mar 27, 2008 at 07:02:32 AM EST
but I don't see why additional error would be introduced by over-sampling a particular subgroup.

[ Parent ]
It doesn't (none / 0) (#63)
by tek on Thu Mar 27, 2008 at 07:43:14 AM EST
seems that it would skew the poll because the AAa are all Democrats and all for Obama.

[ Parent ]
Should (none / 0) (#64)
by tek on Thu Mar 27, 2008 at 07:43:53 AM EST
read:  Seems it DOES.

[ Parent ]
Not if you reweight in the final results (5.00 / 1) (#65)
by andgarden on Thu Mar 27, 2008 at 07:46:14 AM EST
Which is obviously what this poll did.

[ Parent ]
I Understand Weighting And How That Works (none / 0) (#71)
by flashman on Thu Mar 27, 2008 at 10:39:52 AM EST
But what I don't understand is how they get the "margin of error" result.  As I remember it, the error depends on the sample size, and these polls are typically conducted using only 1k interviews, or so.  How do they get ~3% error from such a puny sample?

[ Parent ]
Margin of error (none / 0) (#79)
by muffie on Thu Mar 27, 2008 at 01:47:43 PM EST
doesn't really depend on the population size, at least when the population is large.  If the population of the US was 100 million or 1 billion, the poll would have almost exactly the same margin of error.

Basically, we suppose the popular preference is fixed.  We can model this preference as a weighted coin, which lands heads some unknown percentage of the time.  If you do 1000 coin flips, you'll have a pretty good idea of the odds of getting heads vs. tails.

[ Parent ]

NBC/WSJ Poll: Oversampling | 51 comments (51 topical)

Readers Must Login or Create an Account to Comment

  • Premium Ads

  • Blog Ads

  • Contribute To TalkLeft

    donate to TalkLeft