Clinton Regains Lead in Dubious LA Times/USC Daybreak Poll

[louis c]

The LA Times/USC Daybreak Poll is difficult to understand. For starters, it's mostly experimental, using a methodology that has never been used before. Let me see if I can explain it as I think I understand it.

It's a daily tracking poll. It has about 3,600 participants, all broken down into a 400 voter blocks for each day. The strange part about it is, that it's the same 3,600 voters polled straight through, including the same 400 for each day. As you watch the trend line, it seems to move from a pro-Hillary number to a pro-Trump number, over time. I think each block, just through chance, has become more a leaner to one candidate or another. Remember, the voters never change. Every 7 days, a block is sidelined for two days and then re-enters the poll for seven. If you look at the trend line in the link below, it bears out my thesis. Hence, a Trump lead of 2 points yesterday disappeared today, just as it did 11 days ago. The numbers swing wildly from Trump up 7 just after the Republican convention, to Hillary up 5 just days after the Democratic convention. Of the participants, there seems to be a Trump bias, as he always over performs all the national polls. When the average is Hillary up 8 points, she usually only has a 3 point lead here. When the national polls had the race about even just after the Republican convention, as I said, it had Trump up seven.

I used to use this poll to follow a trend line, but it's getting so it can't even be trusted for that. In the end, this is a useless poll, but we should understand it so we can respond when Trumpets refer to it. Today, Clinton has a fraction of a point lead.

Link;
http://cesrusc.org/election/

[Gothmog]

1. This poll has a strange methodology

https://www.washingtonpost.com/news/the-fix/wp/2016/08/22/donald-trumps-new-favorite-poll-is-way-out-of-step-with-the-polling-average-but-why/

[Drunken Irishman]

2. RAND used the same methodology for their poll in 2012.

It was correct. But who knows if that was an anomaly or the sign of things to come. It also wasn't the lone poll showing Obama leading - this is the ONLY poll that has had Trump leading post-DNC.

[standingtall]

5. I think the reason the Rand poll was correct in 2012

is because it had a sample that was representative of the country. The problem with a tracking poll like the LA Times/USC poll is the sample of 3,00 is locked for the duration of the election. If they over or under sampled any group then they are going to be off badly and I suspect they did. Even they say their poll isn't going to have as much movement as other polls. Hard to get a real detailed break down of their internals,but some things were interesting. Among the 18-34 old demographic Clinton leads 51 to 41. There is no way Trump is within 10 points of that demographic. Also even among people who said they planned to vote for Trump don't believe he is going to win. I think they probably over sampled republicans.

[Cicada]

3. It weights for Vote In 2012, a clear flaw

The sample is weighted to accurately sample age, race, sex but also reported presidential vote in 2012. We know from prior polls that more people report they voted for the winner than actually did. So some in the Obama voter group, weighted in number to capture the correct number of prior Obama voters actually voted for Romney. So the sample likely has too many Romney voters than it should. Thus this poll has a bias in favor of those who vote Republican. The magnitude of this bias is probably one or two points.

[former9thward]

6. It always amazes me how many expert pollsters we have at DU.

I will never understand why they don't go out and start their own polling companies since they know all the faults of polls (ones they don't agree with) and what to do to correct them.

[Thrill]

4. Nate says the poll basically gives Trump an extra six points

You have to account for that and then the poll is somewhat useful