Tracking poll addicts -- prepare for some rocky days ahead...

[regnaD kciN]

Today's polls showed absolutely no movement over the results of the day before, with Gallup showing Obama up by 8 and Rasumssen up by 4. Based on an earlier, erroneous read of the data, I jumped to the conclusion that the Obama speech/Palin announcement amounted to a wash. However, that was due to forgetting the nature of a three-day tracking poll, in which the earliest day drops out each day, replaced by the new polling numbers. When you work that in, a completely different picture is painted. For the poll announced today involves data from Wednesday, Thursday, and Friday (WTF -- no pun intended, while the results of the day before were from Tuesday, Wednesday, and Thursday (TWT).

Concentrating on Gallup just for the sake of argument, what were the results of each of those days? Fortunately, one of the diarists at DailyKos posted this breakdown a couple of days ago (I added the days of the week for clarity):

August 20 (W): O 45 M 43
August 21 (T): O 45 M 44
August 22 (F): O 45 M 45
August 23 (S): O 45 M 46
August 24 (S): O 45 M 44
August 25 (M): O 42 M 48
August 26 (T): O 48 M 40 (post-Michelle)
August 27 (W): O 54 M 38 (post-Hillary)

Now, given that we know that both the TWT and WTF samples resulted in an average 49-41 lead for Obama, it's relatively easy to figure out the daily results for the last two days as well:

August 28 (T): O 45 M 45 (post-Bill/Biden and nomination)
August 29 (F): O 48 M 40 (CORRECTED) (post-acceptance speech/Palin announcement)

So, in essence, the main portion of the "convention bounce" really consisted of one day of a huge Obama lead after Hillary's endorsement. The following day, after the nomination vote and the cavalcade of Democratic speakers like Bill Clinton, Kerry, and Biden, the results had gone back to a tie. Friday, after the speech and the Palin announcement, Obama finished solidly ahead, but still not as far up as in Wednesday's sample.

Why is this so important to keep in mind? Because when tomorrow's results are released, the huge post-Hillary day will have dropped out of the results, replaced by whatever the results were today. What to expect? Well, if we took a two-day average of TF, we would find that Obama's apparent lead will have dropped from 49-41 to 46.5-42.5. Since Obama has been known to poll poorly on weekends, it's likely that neither Saturday nor Sunday results will be all that good. It's quite possible that, when tomorrow's results are released, the race will either be tied or within the margin of error. When Monday's results come out (comprising FSS), it's quite possible McCain might be slightly ahead.

While, in ordinary weeks, we could then expect Obama's totals to start climbing throughout the next few days, this is, of course, the RNC's week, and it's likely that they will experience at least a modest bounce of their own throughout the week as they monopolize the airwaves with their message. Then, of course, will come another weekend, probably bringing more week polling for Obama.

So, from my analysis, you should probably expect McCain to take over the lead in the GDTP by Monday, and hold it at least through the Monday thereafter (9/8). It would not surprise me, by that time, to see his lead hit a mirror-image of the 49-41 currently for Obama, or even more -- at least, for one day's results during that time period. But don't panic -- from that point on, you should see McCain's numbers drift back to more normal levels, and the horse-race continue anew.

Or, of course, you can do what the Obama campaign does with the tracking polls: ignore them, and concentrate on state-by-state results. Because that's what's going to determine who gets sworn in next January 20th. (And, of course, work like hell to make sure it's our guy!)

But let's not get caught in the mindset that "everything's O.K., because we have a nice lead in the tracking polls." Those numbers will change, and fairly rapidly, over the next few days. I hope I don't see a whole lot of panicky "the sky is falling!" threads here when that happens.

(ON EDIT: Due to a calculation error in my initial post, I had Friday's results as 48-45, instead of the actual 48-40. Obviously, that would have made the likelihood of a bigger apparent swing to McCain in the coming days much higher. The post has been re-written to correspond to the actual data. Tracking poll addicts will notice that, based on the new analysis, the post-speech/Palin day of polling actually showed a sizable bounce for Obama over the tie of the day before, but some of that may well have been due to an unusual amount of "statistical noise" in Thursday's sample.)

[davepc]

1. K&R

[roguevalley]

29. thanks for doing the polls. they make my head hurt. :)

[FrenchieCat]

2. Fuck the polls period, then and now.

We should all read this and understand it is all fixed...which is why Axelrod told us to ignore that shit...and keep your eyes on what is happening on the ground. That sounds like a much sounder strategy than relying on some corporate polling outfits who can decide what zip codes to call and when.

Here's the piece: http://www.nytimes.com/2008/08/31/opinion/31rich.html?pagewanted=1&_r=1&ref=opinion

[cliffordu]

6. What FrenchieCat said.

:hi:

[Mojambo]

3. "I hope I don't see a whole lot of panicky "the sky is falling!" threads"

Yeah... You're gonna be disappointed.

Good post.

[still_one]

13. Personally, I don't think the republicans will get that much of a bounce from the convention

[melody]

4. Of course what's important is the trend, not single sets on any day

The trend is +Obama all the way. Whereas Kerry and Bush numbers twined around each other, Obama is consistently tied or ahead. There has been scant deviation from that. There is plenty of reason to be more than optimistic, especially given all the ammunition they have against McCain.

[MadBadger]

5. Based on your numbers, shouldnt August 29th be 48-40 Obama?

45+45+38 comes to 42.67 while McCain was at 41 today.

[regnaD kciN]

9. OOPS...!

You're right. I must have punched in the wrong number on my calculator. :dunce:

That does change matters somewhat. I'll post a correction above.

[akangel2008]

7. Don't worry

polls mean nothing. Once people get a REAL sniff of the new skirt with zero experience it will be points in our corner! She may wear a skirt and be pretty but she's one extremist nut and the voters will see her for what she is; a right wing hack & a bad mother.

[parabolabear]

8. this is the first time i've ever followed polls such as this, so i have a few questions -

how is the information gathered? how accurate are sites/polls such as http://electoral-vote.com/? where can state-by-state polls be found?
there is a FAQ section on that page and i've read it, but i...don't understand!

sorry if they seem like basic questions. this is the first election i've ever closely followed.

[endthewar]

10. Use pollster.com

It is the best source. First, it has pretty much EVERY poll out there at your fingertips. They have an electoral map too. You can click on individual states to see all of the polls for that state.

[parabolabear]

11. thanks!

[endthewar]

12. Welcom to DU!

:patriot: :hug:

[still_one]

14. I prefer ReadlClearPolitics.com /nt

[endthewar]

19. It has a right-ward tilt to it, in my opinion

[still_one]

21. The poll results don't, and they break it done according to states

which is the REAL picture of what is happening

http://www.realclearpolitics.com/epolls/2008/president/us/general_election_mccain_vs_obama-225.html

http://www.realclearpolitics.com/epolls/maps/obama_vs_mccain/?map=10

[Sugarcoated]

15. Thanks for the tip

According to Polster, in the state by state Obama's looking pretty damn good!

[endthewar]

16. Agreed. They haven't updated it recently though.

But you can click on the individual states and judge for yourself. That's what I like most about it.

[Sugarcoated]

22. Check out Alaska!

Look at the graph and the way it's trending.

[regnaD kciN]

23. I suspect that Alaska is now safely in the Republican column...

...and post-Palin polls will show it as such.

[eshfemme]

18. I prefer fivethirtyeight.com myself

I'm not sure about electoral-vote.com since I don't really use it but I know certain polls have biases. 538 actually has state polls and also follows congressional races. The guy behind it basically compiles all the available polls and averages them out, which helps to prevent biases. He also explains his reasoning behind his formula or results. He apparently accurately divined the results for 2004 although I don't know if that's true since I didn't know about his site then. He used to work with baseball statistics so that's where he got his experience dealing with numbers.

As I understand it, polls are usually done by telephone (traditional land lines) and everything is done assuming the ones answering are being truthful. You really will need a basic understanding of statistics/surveying. A poll's accuracy is usually determined by the size of the survey group (if you interview only 5 people, it's obviously less accurate than if you surveyed 1000 people), the quality of the questions (due to push polling which is essentially asking leading questions that can consciously or subconsciously direct the responder to a certain answer) and the margin of error (the percent that the actual survey percentage could fall over or under so 45% with a margin of error of 2% means that the actual result could really be either 43% or 47%). There's all sorts of other questions too like how the survey group is selected (there is an issue because there are differing standards amongst pollsters as to how to select for likely voters from amongst registered voters) and the personal bias of the pollster as well as other factors like daily events. Polls tend to reflect current and more recent events so they don't tend to mean anything unless they are analyzed over the long term-- this is where you will start hearing the word "trend" and why you see all those line graphs or plotting graphs (the ones with the dots). I hope that I'm being fairly accurate about this as this is my personal understanding of it. Hope this helps!

[Awsi Dooger]

17. McCain will definitely have the poll lead a week from now

That's the benefit of having the second convention. In 2000, Gore went from several points down for more than a year to a 5 or 6 point edge. In 2004, Bush jumped from a few points ahead to double digit lead in some polls.

The party occupying the White House gets the second convention and it's a major advantage, like the prosecution having the final say to a jury.

It won't mean much. I doubt the betting odds will change, Obama still trading at 60/40.

Here's a guarantee: Once McCain jumps into the poll lead, DUers will scramble for state polls insisting Obama still leads. State polls lag national polls so those still-favorable numbers will be plentiful.

We need the lead in late September, prior to the debates.

[LongDistanceRunner]

20. McCain's Convention will be a footnote.

Obama will hold the lead as Palin crashes and burns.

[regnaD kciN]

25. I doubt Palin will have the opportunity to "crash and burn"...

...at the convention. Everything will be tightly scripted to give her as little impromptu time as possible. In fact, I'd expect that convention's end will find her with decent approval ratings. It's only when she gets out on the campaign trail on her own that she'll go down in flames.

[regnaD kciN]

24. I seriously expect you're right...

...but the margin will be interesting. I would expect McCain to match Obama's eight-point lead at least once during the next week and a half. If, on the other hand, he never breaks a five-point margin during that time, the Republicans are in trouble, big time.

[JCMach1]

26. The Hurricane is going to dominate the news, not the convention

[regnaD kciN]

27. The hurricane will be big news Monday and Tuesday...

...but will start to fade from prominence by Wednesday and Thursday, when McCainiac and Governor Half-Baked Alaska make their speeches. Besides, they may well turn them into a giant NO photo op (if it's safe to go down there by then, of course).

I'm just wondering if they'll try to have McCainiac give his acceptance speech over a bullhorn while standing on a pile of hurricane debris. :eyes:

[Teaser]

28. How did you solve N equations with N+2 unknowns?

Mathematics departments everywhere would be interested in this breakthrough.

[regnaD kciN]

30. I'm not...

If you have the overall three-day tracking results, plus the results of the two days before, it's pretty simple to find the final day's polling. It's just ((total result*3)-result of two days before-result of day before).

[Teaser]

32. no

a 3 day moving average has 3 values. This leads to 3 equations

(d1+d2+d3)/3 =averageday1

(d2+d3+d4)/3 =averageday2

(d3+d4+d5)/3 =averageday3


3 equations, 5 uknowns.

You can't square this circle.

You can make good guesses, and there are sophisticated techniques for making those quesses, but it doesn't look like you've done that here.

[regnaD kciN]

33. You're wrong...

...because, thanks to the DKos article I linked to above, we already have the poll results for the first two days (actually, the first week, but that's irrelevant).

If you have the two previous day's results, all you have to do to get the current day's results is use the formula

current day's results = (three-day average * 3) - (first day + second day)

and then, each day thereafter, you merely need to run the same equation, dropping out the earliest day and shifting each of the others back.

Quite simple, actually.

[regnaD kciN]

31. Well, so much for my initial projections...

Today's Gallup results have Obama up 48-42. While this represents a slight total drop-off, keep in mind that this was the day the huge "post-Hillary bump" of 54-38 fell out of the totals. Plugging the figure into the spreadsheet, it would appear that Saturday's results (normally a bad day for Obama) gave him an even bigger daily lead than Friday, 51-41.

What to expect tomorrow? Well, in tomorrow's total, Thursday's 45-45 tie will fall off the charts, giving a two-day average of 49.5-40.5. Once again, weekends normally show poor results for Obama, so one would expect that result to fall somewhat. However, it would take a genuinely lousy day (McCain winning the day's results by double-digits, say) to bring tomorrow's total down to the MoE.

What would be interesting (but not very likely) would be if Obama's lead were to increase in tomorrow's results. In the days following his convention speech and McCain's choice of Palin, Obama has apparently won both days by +8 and +10, respectively. A continuation of such margins would suggest that either Obama got a considerable convention bounce, or that the Palin pick isn't playing particularly well. After Monday, of course, the Republican convention gets underway, so I would expect a swing in McCain's direction nonetheless.