CNN Polls: Ongoing experiment in sample sizes

[Buffy]

Those of you who have taken statistics were probably just as amazed as I was to learn about the tiny samples required to get statistically significant results. Obviously polls are skewed by the sample profile, so the results of "self-selected" samples are always suspect in determining "real" full-population opinions.

 

Nonetheless, something I've been watching for ages is the CNN daily poll. They change the time of day that they change the poll question, and some are left up for longer than others, and some poll questions get 3-10 times more votes than other questions. But one very consistent fact is, if you catch the poll when its just been posted, with say just a few thousand votes, and then compare the percentages to later in the day--when there may be 100-200 thousand votes, the percentages vary by no more than 1-2%! As I said above, yes, the poll is skewed: you'd get different results on the Fox News or Al Jazeera sites, but nonetheless, its wonderful on going proof of this very hard to swallow (yeah, totally counterintuitive) fact of statistics is really true.

 

Keep a watch on it. Do you see other online polls you see that violate this? Others that validate it?

 

Stochasticallly,

Buffy

[Boerseun]

Interesting!

 

I suppose the sample would be biased towards "people who watch CNN", and a few thousand will be a good sample of a hundred thousand.

 

It should differ from a sample taken for the same poll in SkyNews or Al Jazeera, for that matter, because they will be biased towards "people who watch SkyNews", or "Al Jazeera".

 

But a good example nontheless of the validity of (relatively smallish) samples!