NEW POLL: Reuters/Ipsos Clinton +6% nationally

[Foggyhill]

28. Good grief!

Got, 30 years of engineering with a crapload of stats/maths,
a MBA and even a comm degree (includes polling).
Do we have to trade business cards now?

My point has nothing to do with the SIZE of the sample or MOE.
It has to do with how representative the poll is in the first place of the targeted population : likely voters.
How much it actually matches the demos of the population of likely voters.
That's the crux of polling, the hard part, which most so called polling operation fail at.

If you poll a million white men and zero of everything else, you're not going to have a margin of error of 0.0001%.

Most poll's margin of error is meaningless no matter how many people they poll because they're sample is not representative
because of methodological problems. For time, money or whatever reasons, they didn't get a good sample; they're mostly polling the most convenient to access rather than the right people.

University/scientific studies often run into the same difficulty in studying students and other people they have ready access too.

Those small sample shoddy method polls are there to get people talking and have no factual value.
The Clinton campaign poller said as much when he compared these polls to their own internal polls.

In small samples, those methodological errors are often made even worse.

Also, I was comparing this poll to a lot of other polls which have samples of 500 - 600 (tons of them in the last few weeks)
I was clear in my first post about that.
So, that is what my initial point was.

Why are most polls bad? Because there is no incentive to produce good ones, seemingly they all get reported in the same way: good or bad.