Don't these races always appear to "tighten" when the pollsters all switch to a likely voter screen?

[pnwmom]

Instead of registered voters?

And those screens tend to favor older voters, whites, and Republicans.

So everything will hinge on turnout, which Hillary has known all along.

[tinrobot]

1. Because they make assumptions about who is likely to vote.

...and in may areas of the country, voter suppression makes it much easier for rich white people to vote.

[Blue Idaho]

2. That's a very good point...

I agree with you. This year I think we have the added problem of the M$M deciding to,use two different scales to judge Trump and Sec. Clinton. For example - we hear endlessly about Sec. Clinton's "transparency problem" while no one seems to Be at all bothered by Trump's refusal to release his taxes or a meaningful health history. Who has the transparency problem?

[democrattotheend]

3. Yes

The same thing happened to Obama in 2008 in early September. The national and then the state polls tightened and we started panicking. It was a combination of the Republican convention bounce and the switch to likely voter screens.

I think it's really dishonest the way some media outlets compare this month's poll to last month's without being clear that they are comparing apples to oranges. Honestly, I wonder if the likely voter screen and the switch to it is designed to make the polls tighten to ensure a horse race.

[meow2u3]

4. "Likely voters" doesn't take into account newly registered voters

and voters likely to vote in Presidential elections, but not in off-year elections.

[PoindexterOglethorpe]

10. That's correct, and I'm under the impression

that a noticeable percentage of young voters who voted for Obama in 2008, didn't vote in 2012, and so that means they would also not be considered likely voters either. If large numbers of them vote at all (hopefully in large numbers for Clinton) they are likewise not on the radar, given their unlikely status.

[MohRokTah]

5. It's the quadrennial Repugnant heart flutter and Democratic FREAKOUT.

Happens every four years after Labor Day.

[tandem5]

6. There appears to be no rigor for the methodology of likely voter weighting unlike the approaches

used to gather raw polling data itself. Likely voter weighting (if we are to assume honest intentions) is used to give the end user an absolute figure from a single data point (candidate x is about y% points ahead of candidate z). It is far more useful to use multiple data points to perceive trend. In this case it is not about saying candidate x is ahead of candidate y but only that support for x has appeared to strengthen or weaken over candidate y. Any post collection weighting can be applied, however it is critically important that the weighting (or at the very least the methodology for weighting) never be changed throughout the entire process or else any comparison between data points would be rendered nonsensical.

[Coyotl]

7. The polls have a lot of margoin of error, so don't give any one of them much weight.

Look at the long term averaging and trends.

[pnwmom]

8. But when you switch from registered to "likely" voters, you're comparing apples to oranges --

thus, no trend.

[a kennedy]

9. Than you......I wondered if that made a difference.......so thank you very much. eom

[Coyotl]

12. Of course you need to use the same kind of poll, goes without saying. Both have MOE, so use trends.

No matter which ones you follow, trends and averaging helps increase predictive accuracy. It does not decrease methodological flaws, but it may balance some of that out. One has to know a lot about the polls individually to judge their merits and understand what the trends are telling you, especially if you seek to improve predictive utility.

[DeminPennswoods]

11. NYT on LVs

This was from back in Jan, but it's a pretty good summation of how polls decide on "likely voters".

http://www.nytimes.com/2016/01/31/upshot/why-this-is-the-iowa-poll-that-everyones-waiting-for.html?_r=0