Capitol Hill Blue Forums Political Issues Campaign 2016

Anyone who has understood signal analysis knows there is a phenomenon called "transient noise". Polls are subject to this effect. I would advise that we give some time for the dust to settle, after the immediate effects of the convention, etc. die out, and then concern ourselves with the polls. And then, only as we draw near the election date - October is probably the best month to begin to take the numbers more seriously.
The volatility of the polls is an intrinsic characteristic of uncertainty, and the higher it is, the less certain the numbers.
Do not, I repeat, do not confound this with the MoE. MoE is a statistical measure of error built into the survey itself. Volatility is an organic quality of the resulting numbers. You can manipulate the MoE. The volatility is almost impossible to manipulate. It is far more relevant to understanding the results of the polls than is the MoE.
At the moment, because of the transient noise, the volatility is very high. This makes the results very uncertain.
So beware of numbers that are changing willy nilly on a daily basis.

And yes, 538 is still the most reliable source for an impartial reading of the numbers.

That is sort of the point I was making, Zeke. And while MoE is supposed to be accurate (statistically speaking) how one determines the "pool" throws the results off more than statistical variation. Many of the polls use such small samples that the MoE makes their results almost meaningless.

I look at the results and then I wait for the SC decision

Indeed. Like many statistical measures, it is subject to bias and manipulation. One needs to understand it and take it with a grain of salt (if at all).

MoE is a product of sample size: The bigger your sample, the smaller your Moe in general. But if your sample all came from a non-representational pool, then your results can be way off. You can sample every single member in that pool and get your MoE down to zero, and your results will still be wrong.

I am more concerned with pollers methodology: How are they reaching the people they poll? If they are still using random land-line calls then their methodology is wildly biased. I don't even answer my land-line any more and I think few people under 50 do or don't even have land-lines.

Actually you can't survey the entire population. But you can increase the sample size, within reason.
But you're right, the quality of the sample is questionable - and most times they don't take the necessary care with that. Which is why Silver's Bayesian method is much better.

This has been a great month for voting rights. Appeals courts have invalidated several different Republican voting suppression laws, and the 4/4 split in the Supreme Court means these decisions will likely stand. The four liberals in the SC are very supportive of voting rights of minorities and poor people. This bodes well for Clinton in the November election.

I was just about to post that. In two states - Wisconsin and North Carolina - I think it will have an immediate impact. I was particularly struck by the unequivocal conclusion of the North Carolina court.

It is good news. And about time the courts start putting the kaibash on this remnant of Jim Crow.

And just to cloud the picture... Clinton leads Trump by 6 points after Democratic confab: Reuters/Ipsos poll. Is this bounce or noise?