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

[nirvanayoda]

The problem with trying to figure out each day is that you have to make an assumption about what the day that rolled off contained.  I tried to do this for Gallup and Rasmussen tracking polls in 2008 and realized that it was sort of pointless because the poll started with multiple days, so it was impossible to tell exactly what day rolled off.

[nirvanayoda]

USC/LA Times national tracking poll (through 10/2) Trump +4.6%
Trump: 47.0% (+0.1)
Clinton: 42.4% (+0.2)

It's worth noting that the sample has six of the seven debate days in it now (with the seventh day being debate night). There really is no significant statistical change outside of basic noise with this survey. (Trump netted 1.1% over the week.)

Trump was at 46.2 before the debate (9/26) and is at 47 now. Clinton was at 42.7 before the debate and is at 42.4% now. This poll, at least, has the debate as a non-event, which is in step with the polling after most debates.

And is out of step with all other polling after this debate

Not necessarily. UPI hasn't shown that much of a net shift (pro-Trump, actually). I also believe Trump closed the gap with Reuters. There was slight movement to Hillary with the People's Pundit Daily results.

The NOLA sample seems to be pro-Clinton, as does Rasmussen, Morning Consult and Fox News.

PPP polled the race, but doesn't have a good pre-debate poll for comparison.

I was shocked that NBC, CBS, CNN and ABC didn't release polls over the weekend.

I think there is one reason to assume that Seriously? is right about the lack of a bounce following the debate based on the LATimes/USC poll, but a couple more reasons to push back.

I do think that the LAT Poll, as a panel, has the potential to smooth out the effects of differential non-response bias. That may explain a good deal of what we're seeing in the other polls. Panels, in theory, should be immune to that. However, knowing that this panel must lose participants through attrition, then new participants have to be added. This must be in part responsible for the wiggle in this poll compared to what you might expect with a more traditional panel. Opting in may be adding a layer of differential non response. New participants are more likely to be pro-Trump as pro-Hillary panelists drop out (rather, they were up to this point).

Additionally, the weights include 2012 vote. That would necessarily mute some swing based on enthusiasm as well. Finally, we know that this poll is showing much more minority support for Trump than nearly all the others. I don't want to unskew, but that is worth noting as well.

In short, I think that the daily internet panels aren't the most likely candidates to show a big bump, but then, that doesn't mean they aren't subject to their own issues.

I have to amend what I said after Morning Consult (and to a lesser extent Red Oak) came out this morning. There appears to be a bounce at least in internet polling. It's odd though that the trackers haven't picked up on it.

I think a lot of it has to do with enthusiasm right now for the self-identifieds. IIRC Morning Consult was as low as D+3 and is at D+9 right now.

That would make sense.  People who are part of a panel that is regularly called may just answer the call out of habit, whereas people who are randomly called may decline the call for lack of enthusiasm.  However, it's probably also safe to assume that those same people may not vote for lack of enthusiasm as well.  I think it's clear that Clinton has a lead now and that not much, aside from astonishing Wikileaks information or an unbelievably poor debate, will change that.

[nirvanayoda]

10/3 Trump +1
Trump 45% (-2)
Clinton 44% (+1)

What happens once Trumpy loses this one?  What unskewed Truth-O-Meter will he be tweeting about on a daily basis?

The numbers that Mike Wells for Northern Assembly posted and that you quoted aren't real.  That said, your question still is a good one, as it's likely that he will lose the lead in this one too sooner or later.

[nirvanayoda]

10/4 Trump +2
Trump 46% (+1)
Clinton 44% (=)

Why do you keep posting fake numbers?

[nirvanayoda]

10/4 Trump +2
Trump 46% (+1)
Clinton 44% (=)

Why do you keep posting fake numbers?

8/23 Approximate Numbers

Clinton 52% (+4)
Trump 38% (-2)

I've decided that I would start using the formula to figure out the approximate numbers for 8/22.

Clinton 48%
Trump 40%

What formula is that? +8 is a bit extreme, probably closer to +5 or +6.

Yes. We can actually calculate, what the last day approximately shows (~ 300 sample):

Trump today:
(6*46.7 + Trump*1)/7 = 46.6 (7 day rolling)
=> Trump = 46.6*7 - 6*46.7 = 46%

Clinton today:
(6*40.6 + Clinton*1)/7 = 41.7 (7 day rolling)
=> Clinton = 7*41.7 - 6*40.6 = 48.3%

But since the sample is so small (~300) the MOE is YUUUGE!

It's not perfect, but it gives you some idea.

I might be missing something, but doesn't this fail to account for the oldest poll dropping off the average in exchange for the newest one? That would change the average of the previous polls.

I'll try and come up with a new formula that takes into account the days dropping off.

 

My sincere apologies. I didn't realize you were calculating daily results.

[nirvanayoda]

10/4 Trump +2
Trump 46% (+1)
Clinton 44% (=)

Why do you keep posting fake numbers?

They're not fake numbers, they are the computed 1-day tracking average for the 7-day tracker.

Yeah, I realized that after he posted again. It's interesting, assuming that the numbers are correct.

[nirvanayoda]

That is a big jump.  Did the Illinois guy get back in?

Not yeat, as I understood.

A strong Clinton day probably fell off along with a decent Trump day rolling on.