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

[Mallow]

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....

So the "July 25th" result is really centered on July 22nd. That explains why the convention bounce didn't apparently start until after the convention.

[Mallow]

Clinton convention bump, HUUUGGEE!!!

You do realize that 1: this is a seven-day running average (as has been discussed in this thread already), and 2: because of that fact, Trump's convention bump in this poll didn't even BEGIN until AFTER the convention was over, and by analogy, Clinton's shouldn't begin until tomorrow's tracking numbers, right? No?

[Mallow]

Clinton convention bump, HUUUGGEE!!!

You do realize that 1: this is a seven-day running average (as has been discussed in this thread already), and 2: because of that fact, Trump's convention bump in this poll didn't even BEGIN until AFTER the convention was over, and by analogy, Clinton's shouldn't begin until tomorrow's tracking numbers, right? No?

Also the poll itself is just sh**t. 

I'm not convinced of that. Perhaps it has a bias, but that doesn't mean it is useless. It seems to capture trends pretty well (if late).

[Mallow]

Clinton convention bump, HUUUGGEE!!!

You do realize that 1: this is a seven-day running average (as has been discussed in this thread already), and 2: because of that fact, Trump's convention bump in this poll didn't even BEGIN until AFTER the convention was over, and by analogy, Clinton's shouldn't begin until tomorrow's tracking numbers, right? No?

Also the poll itself is just sh**t. 

I'm not convinced of that. Perhaps it has a bias, but that doesn't mean it is useless. It seems to capture trends pretty well (if late).

Not really.  It is a sham compared to what they had four years ago.  Shouldn't have launched it until after the conventions at the least.

"It's a sham because I don't believe assumptions X, Y, and Z" reeks of "unskewing". I'm not prepared to make such a bold statement about any poll just yet.

[Mallow]

Clinton convention bump, HUUUGGEE!!!

You do realize that 1: this is a seven-day running average (as has been discussed in this thread already), and 2: because of that fact, Trump's convention bump in this poll didn't even BEGIN until AFTER the convention was over, and by analogy, Clinton's shouldn't begin until tomorrow's tracking numbers, right? No?

Also the poll itself is just sh**t. 

I'm not convinced of that. Perhaps it has a bias, but that doesn't mean it is useless. It seems to capture trends pretty well (if late).

Not really.  It is a sham compared to what they had four years ago.  Shouldn't have launched it until after the conventions at the least.

"It's a sham because I don't believe assumptions X, Y, and Z" reeks of "unskewing". I'm not prepared to make such a bold statement about any poll just yet.

No, because they won't release the cross tables of the poll.  Why should I trust something just based on the numbers without seeing who they are polling.

The "detailed data" tab has sample sizes. 

Yeah I pretty much tuned out of the conversation when he said that...

[Mallow]

They're doing a poll every day now?

This poll is trash

LA Times? Why? Because TRUMP is leading?

Well the poll is contradicting nearly every other poll taken since it became a Clinton/Trump election and the demographics of the United States just wouldn't allow for a Trump margin to be this large.

There seems to be a pretty substantial R bias in their sample, but the trends seem reasonable if you account for the lag (it's a 7-day running average, so the "7/29" number is an average of 7/23-7/29).

[Mallow]

So basically Clinton's bounce is 2 points so far, and we've got another 6 days before they completely capture the bounce.

Yep. Trump's full bounce wasn't registered until 6 days after the end of the RNC. That suggests Clinton's will have to wait 'til the 7/28-8/3 number released on August 4th.

[Mallow]

Nate Cohn pointed out that it looks like this pollster is weighting it's results to conform to 2012's 51-47 Presidential outcome, which partly explains the big Republican slant since some people in most polls tend to say they voted for the winner when they did not.

This pollster is using the same modeling that RAND used in 2012 when they nailed the election with pretty good accuracy. This is a test to see if it they were just lucky or good. One would assume the same "voted for the winner" in 2008 bias for the 2012 cycle.

Obviously, the jury is still out on this type of polling, generally.

Once again, this is not a poll. It is a pre-selected panel, which as Nate Cohn pointed out, is weighted to conform to 51-47 2012 results. This thread and panel poll is meaningless.

I don't think it's meaningless at all. Indeed, this method is probably a stronger indicator of trends than standard polling, since the sample does not change. One just needs to be aware of the potential built-in bias inherent in the sample.

[Mallow]

Nate Cohn pointed out that it looks like this pollster is weighting it's results to conform to 2012's 51-47 Presidential outcome, which partly explains the big Republican slant since some people in most polls tend to say they voted for the winner when they did not.

This pollster is using the same modeling that RAND used in 2012 when they nailed the election with pretty good accuracy. This is a test to see if it they were just lucky or good. One would assume the same "voted for the winner" in 2008 bias for the 2012 cycle.

Obviously, the jury is still out on this type of polling, generally.

Once again, this is not a poll. It is a pre-selected panel, which as Nate Cohn pointed out, is weighted to conform to 51-47 2012 results. This thread and panel poll is meaningless.

I don't think it's meaningless at all. Indeed, this method is probably a stronger indicator of trends than standard polling, since the sample does not change. One just needs to be aware of the potential built-in bias inherent in the sample.

It is indeed meaningless when used as a poll. That's what this thread is all about. We don't have a "trends" thread. And when looked at from the guise of a "poll" it shows a head heat. So, yes, this is meaningless when looked at as a poll.

That's also not true. A poll that is consistently biased in favor of one candidate, but is otherwise very precise (that is, less noisy than others), is actually significantly more useful than a poll that has no bias but very low precision (swings wildly due to noise). If that's the case with this particular poll, then it is very meaningful as long as you sufficiently account for the bias.

[Mallow]

A good piece by Nate Silver:
http://fivethirtyeight.com/features/election-update-leave-the-la-times-poll-alone/

[Mallow]

A good piece by Nate Silver:
http://fivethirtyeight.com/features/election-update-leave-the-la-times-poll-alone/

I'd be fine with it except for the fact that it's throwing off the RCP average.

The RCP average has always been prone to bias in its polling averages, because of the way they choose which polls to include and which polls not to include. As nice as it would be to have a simple polling average that was not prone to bias (though I think that's what 538's "polls-only" forecast attempts to represent), it's simply not what RCP does (nor do they claim to). We should always keep that caveat in mind when analyzing their polling average.

[Mallow]

A good piece by Nate Silver:
http://fivethirtyeight.com/features/election-update-leave-the-la-times-poll-alone/

Ahaha. Nate is saying EXACTLY what I've been saying for WEEKS.


Me ↓↓↓

Clinton +1 is laughable. I thought the LA Times is a liberal newspaper.

You mean that liberal newspapers use a liberal methodologies in their liberal polls?

It is not really a poll. They use the same members all over again. It means that such polls will have an inbuilt bias (it seems like it favors 5-6 pps towards Trump), but it will probably better on "catching trends" (but with 7 day lagging).

Nate ↓↓↓

You must be logged in to read this quote.

Red hacks on this forum make me believe that Trump is not doing well enough with uneducated

Not to get into a pissing contest, but since we're handing out kudos to ourselves, I'll point out that this has been my position for (actual) weeks (your quote is from six days ago). See the conversation that begins here, on July 29th.
https://uselectionatlas.org/FORUM/index.php?topic=240856.msg5187774#msg5187774

And I'm a proud "red hack"

[Mallow]

I'm sexy and I know it!

Meaning, you're Shy Trump supporter!

As a reasonable human being, if forced with the choice of being shot dead or voting for Trump, I would choose to be shot dead.

[Mallow]

I'm sexy and I know it!

Meaning, you're Shy Trump supporter!

As a reasonable human being, if forced with the choice of being shot dead or voting for Trump, I would choose to be shot dead.

Words hurt, you know!

I do! Which is precisely why I could never support Trump.

[Mallow]

Despite it's immediately obvious shortcoming(s), this poll is actually useful.

That's what I've been trying to say this whole time

It's the classic illustration of precision vs. accuracy. The poll seems like it may be highly precise, and consistently inaccurate. It's like trying to hit the center of the dart board, but always being two inches down and to the left. But as long as you know that the bias is consistent, you can just correct for that bias and then you have highly valuable, useful information.

[Mallow]

Yes, the poll serves a broad purpose and has shown the trends - I just don't want it included in aggregates when it's SO off.

I think it needs to be "bias-corrected" before being included in an aggregate, but then only if you bias-correct all polls in a similar manner. I agree that simply "throwing it in the average" when it clearly has such a strong bias is not the way to go.

[Mallow]

Despite it's immediately obvious shortcoming(s), this poll is actually useful.

That's what I've been trying to say this whole time

It's the classic illustration of precision vs. accuracy. The poll seems like it may be highly precise, and consistently inaccurate. It's like trying to hit the center of the dart board, but always being two inches down and to the left. But as long as you know that the bias is consistent, you can just correct for that bias and then you have highly valuable, useful information.

I tend to agree, but the fact that the recent shift towards Trump was driven by a huge swing among African Americans in this panel undercuts the "precision" argument a bit.

Yeah, I was saying that under the assumption that the results were consistently biased. If they're not, that's a different story. Of course, the small sample size for AA's could be the reason for the funky-looking crosstabs. It's still possible that the overall results are precise, but the specifics when it comes to racial breakdown are wonky. That's why caution is always advised when trying to "unskew" based on crosstab data.

[Mallow]

It's obvious this poll is complete junk. Not sure why the LA Times would ruin its reputation by continuing to publish it

The poll has a pro-Trump bias of several points; even the LA Times has acknowledged this.  But the trend information is potentially useful.

As of last month I would have agreed with you. But over the past several weeks, this poll has not trended in a manner consistent with almost every other national poll (tracking or otherwise), so it's harder to justify that claim.

[Mallow]

Clinton actually did lead in this poll once before, but not by more than a fraction of a point, iirc.

As far as I could tell, she has led by more than her current 1.3 points in this tracker at least three times:

USC/LA Times national tracking poll (through 8/8)
Clinton: 45.1 (+0.1%)
Trump: 43.4 (-0.4%)

USC/LA Times national tracking poll (through 8/10)
Clinton: 44.8 (-0.1%)
Trump: 43.3 (-0.2%)

USC/LA Times national tracking poll (through 8/23) Clinton +2.1%
Clinton: 45.3% (+1.1%)
Trump: 43.2% (-0.8%)

That's 1.7, 1.5, and 2.1. She also led by 1.1 in September:

USC/LA Times national tracking poll (through 9/7) Clinton +1.1%
Clinton: 44.5% (+0.4)
Trump: 43.4% (-0.4)

It will show as Clinton +2 due to rounding.

[Mallow]

Here's the link:
http://www.latimes.com/politics/

Clinton leads by 0.9% today, 45.0/44.1

Looks like I missed Clinton's biggest lead back in August. August 12: Clinton +4.7 (46.3/41.6)

That being said, the numbers from mid-August on the site don't match up with Seriously's numbers from earlier in this thread. Did they change their formula?

[Mallow]

Desperate L.A. Times going back and changing their old numbers to avoid egg on their face on election day!  Sad!

Before yet another myth on this board becomes accepted truth, can anyone point to specific dates that don't match?

I spot checked the mid-August dates with Clinton's biggest lead and the latimes.com graph still matches exactly the Seriously? posts.  Take care not to mis-read the other poster who was reverse-engineering some single day numbers as the supposed tracking numbers.

See my replies on the previous page (replies numbered 643 and 646). Unless I'm misreading something (very possible), none of the four dates I quote have an apparent corollary on the LA Times site. Five examples are...
Date___LA Times___Seriously?
8/8____45.0/43.8___45.1/43.4
8/10___44.9/43.5___44.8/43.3
8/14___46.3/41.6___45.6/42.0
8/23___44.2/44.0___45.3/43.2
9/7____44.1/43.8___44.5/43.4

Most of the margins are similar, but some show a decent discrepancy (8/14: Clinton +4.7 vs. +3.6; 8/23: Clinton +0.2 vs. +1.9).

Is it possible the numbers on the LA Times site are daily and not 7-day averages, and Seriously? is calculating the 7-day averages manually? Or is the discrepancy caused by something else (change in methodology? Different source?). I am not suggesting Seriously? made up numbers by any means, I just want to know why there's an apparent discrepancy.

[Mallow]

The Seriously? numbers you cited are exactly what is currently reported for those dates on USC's poll website.  Nothing is changed:

http://cesrusc.org/election/

What website are you looking at for LA Times results?  Your dates seem to be for the polling date that is one day before what is reported on the USC website.

Odd that I didn't see that. I swear I checked "one day off" in both directions and it didn't match before, but you're right, the numbers I posted are "one day off". It's because I'm using this source:
http://www.latimes.com/politics/

That source seems to label the days one day later than your source. At least it makes sense now. Thanks for the clarification!

[Mallow]

Latino Decisions ‏@LatinoDecisions  14m14 minutes ago
If you need a really good laugh in these final days - the infamous USC/LAT tracking poll has Latino vote at +3 Clinton. Final LD has it +63

Wowza... would this alone be enough to account for basically the entire discrepancy between this tracker and the major polls showing a 2-4 point Clinton lead right now?

[Mallow]

Latino Decisions ‏@LatinoDecisions  14m14 minutes ago
If you need a really good laugh in these final days - the infamous USC/LAT tracking poll has Latino vote at +3 Clinton. Final LD has it +63

Wowza... would this alone be enough to account for basically the entire discrepancy between this tracker and the major polls showing a 2-4 point Clinton lead right now?

Bump. Does anyone know?