USC Dornsife/L.A. Times Daybreak National Tracking: 11/7 - Trump +3.2

[The Vorlon]

Who?

This poll is METHODOLOGICALLY a clone of a poll run by the Rand Corporation in 2012.

The Rand poll did VERY well - their final poll has Obama +4, and the actual result was Obama +3.9.

Two questions:

Did Rand just "get lucky" in 2012? (Zogby did well in 1996, and Rasmussen was awesome in 2004, so one swallow does not spring make)

The Rand Corporation is anal retentive/obsessive compulsive on quality control, is USC in the same league?

This is a very interesting, but ultimately unproven polling methodology.

[The Vorlon]

So this actually isn't really a poll and should probably not be recorded as such.

In 2012 The Rand Corporation poll predicted Obama +4, and the actual result was Obama + 3.9

This is a VERY serious methodology.  The Rand corporation has had 32 (!) persons associated with the company win various Nobel prizes for everything from economics to physics.

I have questions if USC can match Rand in terms of anal retentive/obsessive compulsive validation and quality control, but this is indeed a VERY serious poll.

[The Vorlon]

Dubious methodology. Questionable randomness.

This poll is different, but most things touched by the Rand Corporation are well outside the box.

For the record... In 2012 Rand predicted Obama +4.0% - The actual was +3.9$

It could have simply been beginner's luck, but the Rand corporation is a VERY serious research organization - Staff members at Rand have won 32 (!) Nobel Prizes over the years.

[The Vorlon]

lol this is such junk

I can't be the only one thinking that it is pretty much statistically impossible for them day after day, with such a small n of respondents, to have results that only vary a point or two. There should be more variation with such a small sample.

It's 5 Day Rolling.

7 day rolling actually....

[The Vorlon]

It never fails to amaze me how statistically illiterate people here are when all we do all day is analyze polls.

They real should teach statistics, or at least math and critical thinking in the schools.

Why did they stop doing that BTW?

[The Vorlon]

Dubious methodology. Questionable randomness.

A factual point of information:

The Rand Corporation (or the "Bland Corporation" for you Dr. Strangelove fans) ran a poll based upon essentially identical methodology in 2012 and came within 0.1% of predicting the actual margin of victory.  Rand had Obama +4.0, the actual was Obama +3.9

The Rand Corporation has also had 32 members of it's staff win the Nobel Prize in various disciplines over the years.  This may be an internet based poll but it's about as far from John Zogby as you can get.  RAnd is not a bunch of underfunded lightweights to put it very mildly.

The methodology is new, and unproven over a long term series of races,  so I would personally put this poll under the "worth watching, but unproven" category - but that being said this is a VERY serious and legitimate effort to get it right.

[The Vorlon]

I mean, people are now trying to claim RCP are right wing hacks. RCP!

In fairness, RCP is a Republican propaganda machine.

RCP is conservative, by and large. Their aggregate, however, is not biased.

It picks and chooses polls, yeah it is just as biased.

538's "Nowcast" says Hildebeast would win the popular vote by 0.2%
The RCP average has Trump up by 0.9%

1.1% is noise, not bias.

[The Vorlon]

The internals are legitimately bad (dive in and look) whereas Reuters aren't.

I'm not sure about now, but before Reuters had ridiculous Democratic registration advantages. Like D+15 at one point I think. They're dreadful.

That's why I said the internals are fine. Do the internals 'add up' to an accurate national Clinton v Trump share? No. That's in the weighting,

Unless you have a %$%#&ing  HUGE sample, the internals simply cannot add up.

You balance your sample to the national data points (age, race, education, rural, urban, suburbs, etc) but the regional samples are simply too small to be balanced.

To make an extreme illustration if you did a "poll" of 8 people in the US there should be 1 african american (Blacks are about 12.2% of the population in the US) - So if you broke the US into 4 zones of 2 people each, one zone would have to be 50% black (wrong!) and the other 3 zones would be 0% black (wrong!) - but the overall sample would be ok.