Rasmussen Tracking Poll [Obama vs McCain]

[Alcon]

Wow, you Democrats were right! The Obama nomination clinching bump was HUGEEE!!!

47-45 lead to 48-45 lead in the Rasmussen tracking poll.  AMAZING!!!

I know I'm not a Democrat, but anyone who expected an overnight 15-point bump is kidding themselves.  Why would Obama get the support of disaffected Clintonites overnight?  I imagine that won't really happen meaningfully into the one-on-one campaign gets going.

Not saying it will be 15 points, btw.

[Alcon]

It will probably take 3-4 days for any "rally effect," June 15-16 should be the most telling.

I honestly think the "returning to the ideological norm" bump has the greater potential than the "rally effect," assuming rally also includes excitement about having a Dem nominee.  I'm not sure this is going to be a sudden bump.  Sam and I seem to be the only one who think this though.

[Alcon]

Wtf happened with Obama's favourables. That jump is a bit insane in such a short period of time.

I was expecting a gradual increase as Clinton people started flowing back to Obama...instead apparently it was almost overnight.   It's nice to see that I, again, overestimated the sincerity of the American voting public.

The bump among unaffiliated voters seems less likely to last, though.  Not that it matters.  If Obama wins Democrats by the same clip McCain wins Republicans, McCain would have to do stellar among independents to win.

And it still seems hard to imagine that all of the Clinton spite voters have returned home...the immediacy of this bump baffles me.

[Alcon]

45% (+1) trust McCain most when it comes to economic issues and managing the economy while 42% (+2) prefer Obama.

On national security issues such as the War in Iraq and the War on Terrorism, 49% (-2) have more trust in McCain while 41% (+4) prefer Obama.

Then what the heck are they voting on?

[Alcon]

Alaska is going to be within a few points (3 at most, 1 at least, if what I'm hearing is correct - probably McCain +2)

[Alcon]

Alaska is going to be within a few points (3 at most, 1 at least, if what I'm hearing is correct - probably McCain +2)

4 to be correct.

Where are you seeing that?  Is the poll out?  I thought the extrapolation was 43-41.

[Alcon]

Alaska is going to be within a few points (3 at most, 1 at least, if what I'm hearing is correct - probably McCain +2)

4 to be correct.

Where are you seeing that?  Is the poll out?  I thought the extrapolation was 43-41.

Ah, Alcon, I don't quite know how to tell you this (you do seem to have trouble understanding basic concepts), but, extrapolating on survey results is stongly frowned upon in the survey research industry.

Extrapolating cross-tabs from the same survey, however, is not frowned upon.  I just managed to forget about the Undecided sample, which evidently broke for McCain

[Alcon]

It's not extrapolation; I was using the wrong phrase.  It's using sub-samples to calculate other samples.

CARL, this is essentially what I did.  Here's an example:

Male (51%): Obama 51%, McCain 48%
Female (49%): Obama 54%, McCain 45%

Now, 51% of the sample is male and 49% is female.

Thus, the Obama sample is (0.51*0.51)+(0.49*0.54) = 0.5247 = 52% Obama

And the McCain sample is (0.51*0.48)+(0.49*0.45) = 47% McCain

I realize this introduces some rounding error, but when the sub-sample is from the same poll, it does give you a very close approximation of the results -- unless you screw up the math by forgetting Undecideds, which I did.  This should work in any instance where you have samples that constitute the entirety of the poll, and then results for what you're trying to determine for those samples.

You can feel free to tell me where I'm wrong on the math, but I do not believe I am.    And I haven't ever deleted a post and claimed I didn't make it, so you're really just archiving your inability to use the QUOTE tag.

[Alcon]

Alcon,

First, extrapolation for "undecideds" according to the proportion of "decideds" is very, very wrong!

It assumes that the "undecideds" are reflective of the decideds (generally false).  The undecideds are generally those with lower SES than the decideds.  Also, the undecideds several months prior to an election are not only different from the decideds at that time, but they also differ from the undecideds in the closing days prior to an election.

It also runs counter to experience.  You may have heard that the general rule is that when there is an incumbent involved in a race that is essentially even among the decideds, the undecideds will generally support the challenger.

No, man, I didn't do that.  You're misunderstanding.  I accidentally ignored the break-out of Undecideds.

Here's what I did, with fake numbers.  I took the Begich/Stevens break-down and who they voted for (with fake numbers):

Stevens (50%): McCain 83%, Obama 17%
Begich (45%): Obama 88%, McCain 12%

With that, I got the example total of:

McCain: (0.50*0.83)+(0.45*0.12) = 46.9%
Obama: (0.50*0.17)+(0.45*0.88) = 48.1%

As I said, there's rounding error involved here, since the Obama/McCain breakdown among Begich and Stevens supporters was presented in whole percentages.  However, my big mistake was forgetting the Undecided sample, which I'll arbitrarily say was 62% McCain, 38% Obama.  That adds:

McCain: 0.469+(0.05*0.62) = 50.0%
Obama: 0.481+(0.05*0.38) = 50.0%

The concept was fine, I just messed up the execution.  I introduced extra error from discluding undecideds to the trivial errors related to rounding.

Further, what you are attempting to do is statistically invalid.  While the MoE for a total poll may be x.  If you try to extrapolate based on subsamples, you drastically increase your MoE.

No, again, you don't understand.

The MoE for male+female in a poll (and Stevens+Begich+Undecided) is the same as the poll itself.  It contains the same number of respondents, unless I underestimate the number of eunuchs in Alaska.  Therefore, the MoE is identical.  The only source of error is rounding (as long as the sub-samples add to 100% of the sample)

[Alcon]

Where do you get the idea that a subsample of a survey has the same MoE as the total sample?

This is why some polls interested in a particular group use an oversample so that they can get some statistically valid data from the subsample.

You really need to take a course in survey research.

You really should only be glib when you're not full of it.  The subsamples do have larger MoEs.  When you add them together, they do not.  They're no longer subsamples -- they're the entire sample, just combined after having been separated.

Male + Female = 700 respondents
Whole survey = The same 700 respondents

Unless there are:

1) Fewer responses
2) Different responses

...it's the same MoE (same respondent count and respondents), and valid..

In this case, it's neither, so it's a valid calculation.

[Alcon]

Gustaf explained what I did correctly.  I was not making any assumptions about undecideds or extrapolating data.  I was doing simple percentage addition.  Here is one last attempt at explaining it to you, CARL:

52% of a sample is men.  40% like carrots, 60% do not.

48% of a sample is women.  62% like carrots, 38% do not.

100% of the sample is either male or female.  Thus, the sample size is identical to the overall sample and has the same Margin of Error.

What I did is calculate the number of people who like carrots.  Now, 52% of the overall sample is male and 40% of them like carrots.  So, approximately 20.8% (40% of 52%) of the overall sample is men who like carrots.  For women who like carrots, that comes out to 30.0% (60% of 48%).  That means that about 50.8% (with rounding error) of the sample is ether men or women who like carrots.  Since there are no other categories than men or women, we know that 50.8% of respondents in the poll liked carrots.

My error was not including a third group--people who were undecided in the Senate race--in my numbers.  Again:  the "Undecideds" were undecided in the Senate, not Presidential race.  I was not allocating Presidential undecideds to one candidate or another.  Do you understand, now?

[Alcon]

Alcon attempted (improperly) to allocate the undecideds.

How many times do I have to explain that I didn't allocate Presidential undecideds?  Here, let me try one more time:

I didn't allocate Presidential undecideds.

I allocated people who were undecided in the Senate race, but were decided in the Presidential race.  I allocated decided voters who were undecided in another race BUT decided in the presidential.  You can't possibly be this slow in understanding.  Read this time.

Here is one last example.  Again, theoretical numbers.  On the left is the respondent's opinion in the Senate race; on the right, Presidential.

Stevens (50%): McCain 85%, Obama 11%, Undecided 4%
Begich (45%): Obama 86%, McCain 9%, Undecided 5%
Undecided (5%): Obama 40%, McCain 36%, Undecided 24%

Here's a graphical representation:



So, 5% of voters are undecided in the Senate race.  We can't allocate those for the Senate race, but we can allocate those who are decided in the Presidential.

- 40% of Senate undecideds are decided for Obama.  40% of 5% means 2.0% of the overall sample are "undecided Senate voters decided for Obama."

- 36% of Senate undecideds are decided for McCain.  35% of 5% means 1.8% of the overall sample are "undecided Senate voters decided for McCain."

- 24% of Senate undecideds are undecided Presidentially, too.  24% of 5% means 1.2% of the overall sample are "undecided Senate voters undecided in the Presidential race."

You can see the contribution of Undecided Senate voters to the Presidential options Decided McCain, Decided Obama, and Undecided, below:



You can also see the results of performing the same operation with Decided Stevens and Decided Begich, and then adding to estimate the percentage for the Presidential response (with rounding error) for the overall sample.

So, tell us where I am attributing voters who are Undecided in any race as being Decided in that race, as you claim I am.

[Alcon]

Well, now that we're no longer dealing with "fake" numbers

Actually, no, this is the same concept that I've been trying to explain the entire time that you seem incapable of understanding.  The fact that you think this is a new concept shows that you have been confused this entire time but hurling accusations anyway.

You assume that seventy six per cent of the persons listed as undecided in the Presidential preference question were in fact lying, and that by projecting from results from demographics of the Senate questions onto the Presidential question you can get the results you want.

It doesn't work that way!!!

No.  In those theoretical numbers, 76% of Senate undecideds were decided in the Presidential race, and I allocated those to the Presidential race.  They were NOT undecided voters in the Presidential race.

There's nothing that stops someone from being undecided in the Senate race and decided in the Presidential.  If 25% of voters are undecided in the Senate race, but none of those undecided Senate voters are undecided in the Presidential, you're arguing that 25% would still be undecided in the Presidential.  Obviously, that makes no sense and a few seconds of thought should have told you that.

Seriously though, if you are having this much trouble understanding a relatively simple concept, please do contact the Vorlon.  Maybe he can make you understand.

[Alcon]

First, NOTHING you have done validly indicates the breakdown of the undecided voters in the Presidential contest.

...

Are you still contending that on ONE demographic factor you can project the breakout of the undecideds in the Presidential race?

I'm not allocating undecideds in the Presidential vote, just decideds who are undecided in another race.

Indicating the preferences of undecided voters in the race they're undecided on, is impossible because they are undecided in that race.  Any such extrapolation is arbitrary.  That's what you're (incorrectly) complaining that I'm doing.  I'm not doing that.

Finally, are you contending that the breakout in a poll (not results) in another race by party of candidate has a clear correlation to the breakout of the results in a Presidential poll?

CARL, again, in the simplest way possible, you have the following results:

Male (50%): McCain 60%, Obama 40%
Female (50%): McCain 50%, Obama 50%

The McCain sample is (60% of 50%) plus (50% of 50%), which adds together "males for McCain" and "females for McCain," which gives you "males and females for McCain," which is the same as "voters for McCain."

I did this same operation for the Senate race, except with "Begich voters for McCain," "Stevens voters for McCain," and "undecided Senate voters for McCain."  Together, these represent "voters for Begich, Stevens, or undecided [in the Senate race only], for McCain."  This is the same as "voters for McCain," since it is the entire sample.  These voters were undecided in the Senate, but decided in the Presidential.

What about this operation was logically unsound?  The answer is, "nothing."

Or are you now simply acknowledging what I said from the beginning, that there is currently no valid way of determining the eventual vote of the undecideds in a Presidential race?

You're not wrong that allocating undecided Presidential voters to either McCain or Obama is incorrect.  You were correct about that.  I have never contested this.

You're wrong in that I have done that at any point, which I haven't.  You continue to assert that I have, but have failed to show where for as many posts.  Considering how simple this concept is mathematically, maybe you shouldn't be throwing around "take a statistics class!" snark so much.

[Alcon]

Yet again, you've completely failed to understand what's going on, CARL.

Rasmussen released the Senate side of a Senate/Presidential poll, with cross-tabs that indicated Presidential support by Senate support.  Using the method I described above, these results correctly predicted McCain +4.  I made the mistake of not including the Presidential preferences of voters undecided in the Senate race, which incorrectly came out to McCain +1-3 (with rounding error leeway).  My conceptual methodology was fine; I just made an arithmetic error.

Calling this an "extrapolation" was incorrect, since it implies I was "assuming" something based on trends or arbitrary allocations.  I was just extracting results that already existed in the poll.

As for:

You must be logged in to read this quote.

Where did this BS come from?  I've never said that.  I was using gender as an example of what I'll call a "full-sample subsample set" (FSSS)*, and I haven't even mentioned the Rasmussen national poll.

I really do encourage you to contact the Vorlon, Sam Spade, or any other poll-savvy poster (or one with remotely competent math knowledge), to explain to you that there's nothing wrong with what I did.  The only source of non-human error would be rounding error.

* - Something like male + female, or Begich + Stevens + Undecided, that adds up to 100% of the sample.

[Alcon]

First, my apologies on one respect, i.e. I have never seen a pollster release the Senate side of a poll prior to the Presidential side.

Your insanely limited semi-apology is accepted.

It appeared to me that you were projecting from the existing national poll.  Your use of fake numbers and assumptions made what you are now asserting very unclear.

Second, there is still a minor problem, i.e. the hidden assumption that those voting in the Presidential election are the same as those voting in the Senate election.

While the "roll-off" typically is modest (1 to 2%), it can be substantial (watch what happens in Montana this year).

It was not unclear.  I explained it four or five times, and you repeatedly misunderstood even the basics of what I was doing, while being snotty about it.

The sample sizes for the Senate and Presidential were the same.  And even if they were different, I'd be producing a poll with a few different respondents, but the same sample size and margin of error.  If there are 500 Presidential responses, it's just as valid as a poll, even if it isn't officially released as the Presidential poll.

Not that it matters; I'm pretty sure that Rasmussen either discards respondents who refuse any answers, or combines these into Undecided or Other -- hence why all of their polls are sample size n=500, even in the Presidential cross-tab of the Senate poll.

I'm going to assume, by "fake numbers and assumptions," you meant "clearly-labeled examples that I didn't understand."  Let me point out that I never mentioned the Rasmussen national poll, and clearly stated what I actually used.  So, you're the only one here that made incorrect extrapolations and fake assumptions.

[Alcon]

You guys were right! This election is over!

Who exactly are "you guys"?  I've heard maybe four people say the election was over, ever, and two of them were pessimistic Republicans.

[Alcon]

What´s even more interesting: I don´t get how Obama is ahead by only 1% overall, if he´s getting 44% of the white vote, to McCain's 50%. Kerry got 41%, Bush 59%. Obama's getting higher shares among African-Americans and Hispanics. Is Rasmussen's sample 90% White ? Because on Election Day it will be 73-75% ...

That's a good point.  If the sample is 25% non-white, Obama would have to be leading by a barely 3:2 margin among non-whites.  Kerry's margin was more like 5:2, even in the ill-adjusted exit poll.

Scott Rasmussen is cooking up the numbers so he can make his weekly appearance on the Sean Hannity show to flaunt off his numbers to Hannity's base.

That, on the other hand, is not a good point!

Only in Bizarro World where Kerry won and Mypalfish is Secretary of State.

And that is not a good joke!

[Alcon]

Let me suggest a slightly more different breakdown by race/ethnicity:

Race/ethnicity                    % of Voters                    Percentage for Obama                    % of Voters

White (nonhispanic)               76                                         41                                             31.16

Black (nonhispanic)                12                                         95                                             11.40

Hispanic (all races)                   8                                         67                                               5.36

Asian (nonhispanic)                  2                                         56                                               1.12

All others (nonhispanic)            2                                         56                                               1.12

Total                                     100                                                              50.16

That's a realistic breakdown, but doesn't help us reconcile the poll at all.

[Alcon]

So, you called and found out what leaner method they're using?  What is it? 

[Alcon]

At 3:00 p.m. Eastern, new data will be released for key Senate races in Nebraska...

Fingers crossed guys!!!

[Alcon]

I'm sorry, but that is bull, and people ask me why I say Rasmussen is lean toward the republicans.

What's bull about it?  It's essentially supported by Gallup.  There's not really any proof that it is bull, even if the proof that it isn't is somewhat thin.

[Alcon]

More political analysis should contain the phrase "circle jerk."  Wait, no it shouldn't.

My inclination is that McCain has an easier road ahead with the debates -- he mostly has to stop sneering and condescending, easily his least attractive public quality.  Obama's "um" probably seems to be something he won't solve.

I've seen McCain's Arizona Senate debates, and they're unsurprisingly unimpressive.  More than anything, I expect the debates to be pretty boring and without any genuinely strong wins.  McCain has the ability to deliver an occasional devastating monotone line, and Obama can sometimes go into Teleprompter mode, but beyond that...eh.

[Alcon]

McCain: 54% favorable; - 44% unfavorable (-2, +3)
Obama: 53% favorable; - 45% unfavorable (-1, +1)

The U.S. must have had a crappy Thursday.

[Alcon]

No Rassy polls today?