POLL- Kerry leads Bush 43-37

[PA-DEM]

The polls are all over the place.

April 16, 2004
April InsiderAdvantage/Creators Syndicate poll

InsiderAdvantage
John Kerry over the past month has dramatically turned things around in his race for president against George W. Bush, according to a new poll from InsiderAdvantage/Creators Syndicate. Kerry has the support of 43 percent of respondents in our latest survey, while Bush has 37 percent. In mid-March, Bush led Kerry 46 percent to 41 percent. The poll also asked Americans their views on several other, related issues.



http://www.insideradvantage.com/stories/2004/apr/13/2080479.shtml

[MidwestTransplant]

1. "500 Americans"

don't make a great sample unless we get can get 70% of people to vote!

[Jackpine Radical]

3. Whether n=500 or n=100k makes less difference

than how the sample is drawn.

The size of the sample determines only the reliability of the poll, i.e. the size of the sampling error. The nature of the sample determines the poll's validity, i.e. the degree to which the poll measure what it purports to measure.

[MidwestTransplant]

7. Right, that was my point

500 "Americans" is not a good sample. They should be registered or likely voters.

[ProfessorGAC]

10. Not Quite Accurate, JPR

First, the overall sample size is desgnated as k, not n. The subgroup under consideration from within the sample set is designated by n. So, it would k = 500 or k = 100k. In nondiscrete data, if one had 100,000 data points (k = 100,000) and we looked at the average of each 100, then we'd have 1,000 averages where k = 100k and n = 100. Just picking a nit, there.

More critical to your point:

Due to the nature of randomness, and the need, with a small sample, to use quasirandom techniques, including stratification, the larger the sample, the less inherent error in each strata and the lower the likelihood of strata overlapping, or gaps being created.

So, when a very large sample is employed, the likelihood that the nature of the sample is inappropriate falls by the natural log of the proportional difference.

In the case of your example, with the difference being 200x, then the power of the analysis would increase by nearly 93% relative, and the likelihood of missing the intended population measure would fall by ln(200) or about fivefold.

So, sample size, in and of itself, enhances the nature of the sample that you correctly assert is the most important factor.

The problem is, k = 100,000 datasets are VERY expensive surveys. So, we're stuck with the realities of life in social statistics.
The Professor

[AngryAmish]

13. Oww! Stats headache. Oww! Stop it!

I thought I left that stuff long ago. I'll stick to batting average.

[malachibk]

15. Sort of...

While the "nature of the sample" (that is, how representative the sample is of the group the poll intends to tap into -- all voters, all likely voters, all White voters, whatever) is central to the internal validity of the poll, sample size is also crucial. (Selection Bias = skewed sample)

The larger the sample, the stronger the "Power" of the poll which means the pollster is more likely to be able to detect a statistically significant difference if one does indeed exist (this is called "avoiding a Type 2 error". Larger sample = smaller margin of error because the confidence intervals will be tighter.

Reliability is something else. If the poll, given to the same sample over and over again yields the same result, it is a reliable measure.

Sorry for the "lesson", but I'm an epidemiologist.

[DarkPhenyx]

2. I'm going to say this once, and only here.

The only poll that means anything is the one in November. Everything else is complete crap and generally useless for anything.

[wabeewoman]

4. Yep and I'm going

to keep the faith that the American people are smarter than it seems and won't vote for a continuation of the last 4 years. That is really what it is about. Do you want 4 more years of this?

[DarkPhenyx]

5. Nopr! Definately not.

I'm not as hopefully as you are, and my belief in the collective intellignece level of the American voter isn't as strong. Still, I definately do NOT want four more years of this crap.

[blm]

6. Watch the doom and gloomers avoid this thread.

They only bother to post about bogus polls like Gallup and spread defeatism.

[NewYorkerfromMass]

8. Kerry's gonna kick shrubby's ass

you can take that to the bank! right Rummy?

[HFishbine]

9. Yippee! It's in the bag! What a relief!

I'm so optimistic, I think I'll stay home and leave it up to the rest of you. Congratulations President Kerry!

[Feanorcurufinwe]

11. Can't seem to find that middle ground?

lol

[slinkerwink]

12. whoa....37 percent?

[NewYorkerfromMass]

14. undecideds at 15%, and according to Dick Morris

85% go to the challenger. :)