I agree Cactus that with Rasmussen's hard party weight, it's tough to move out very far. The Gallup should give us a better view in the next couple of days.
Paging the Vorlon...
A "Hard Weight" by party Id had advantages, in that is damps out huge changes, but it also has disadvantages in that during to the weighting, strange things can also happen if your party ID weight is wrong:
Here is an example.
Lets say we do a poll of 1000 people, and I will pick numbers that make the math easy to do, and since we are basicaly talking about Rasmussen, I'll use Rasmussen' 39/35/26 weighting formula
1000 people
Sample #1
=> 350 Republicans
315 vote Bush (90% of GOPers)
35 Vote Kerry (10% of GOPers)
=> 260 Independants
130 Vote Bush (50% of indys)
130 Vote Kerry (50% of Indys)
=> 390 Democrats
351 Vote Kerry (90% of Dems)
39 Vote Bush (10% of Dems)
In this sample, since the actual party ID of the sample matches the weighting assumption, the weights do not change the result which in this sample is:
Bush 315 + 130 + 39 => 484 => 48.4%
Kerry 35 + 130 + 351 => 516 => 51.6%
Kerry leads by 3.2%
Now let's say there is a high publicity event in the campaign that a lot of people watch (Say, maybe a convention...?)
And in our sample the following happens:
Among the 390 Democrats:
20 of the Democrats, who were ALREADY supporting Bush say "You know, I am really a Republican... and thus now self identify as GOP in a poll, remember, these folks DID NOT change their vote, just their party ID
In addition, 10 Democrats actually do change their vote, and decide to vote Bush, but still call themselves Democrats
Our sample of what was 390 Democrats now looks like this:
370 Democrats (20 now call themselves GOP, so 390-20)
341 vote for Kerry (351 - 10 who switched vote) (92.16%)
29 vote for BUsh (10 new for Bush minus the 20 party switchers) (7.84%)
Among the 260 Independants, the following happens to the sample:
20 Independants, who we already voting for Bush, and hence have not changed their actual vote, now say "You know, I actually am a republican", and thus self Identify as GOP
Our Independant sample now looks like this:
240 Independants (20 now call themselves GOP)
130 Independants continue to support Kerry (54.16%)
110 Independants continue support Bush (130 original - 20 who now self indentify as GOP) (45.84%)
Note that
even though not a single voting intention actually changed among these 260 people, Kerry nop "leads" by 8.32% among independants, because 20 of the independants now self identify as GOP
And to keep things simple, lets say nothing happened to the original sample of 350 GOPers other than the fact that 390 now self identify as GOP (The original 350 + 20 indys + 20 Dems who switched)
Bush gets his original 315 supports, plus 20 indys + 20 dems who ALREADY supported him but now self Identify as GOP - so the GOP sample now looks like this:
Bush 355 out of 390 => 91.02%
Kerry 35 out of 390 => 8.98%
This is now what the actual UNWEIGHTED support in this sample looks like:
Bush
29 Self Identified Dems
110 Self Identified Indys
355 S3lf Identified GOPers
Total Bush => 494 => 49.4%
Kerry
341 Self Identified Dems
130 Self Identified Indys
35 Self Identified GOPers
Total Kerry => 506 => 50.6%
In reality Bush has indeed gained 2%, narrowing the gap from 51.6/48.4 to 50.6/49.4.
Kerry IN REALITY leads by 1.2% now
But now look what happens when we apply the 39/35/26 Weighting formula:
Bush
GOP support @ 91.02% X .35 weight => 31.86%
Indy Support @ 45.86% X .26 weight => 11.92%
Dem support @ 7.84% x .39 weight = 3.06%
Total Support = 31.86 + 11.92 + 3.06% => 46.84%
Kerry
GOP support @ 8.98% X .35 weight => 3.14%
Indy Support @ 54.14% X .26 weight => 14.08%
Dem support @ 91.16% x .39 weight = 35.55%
Total Support = 31.86 + 11.92 + 3.06% => 53.16%
In REALITY Bush
gained 2%
But because of the weighting to the wrong party ID, this poll shows Kerry's lead
increased by 3.16%
Rather than showing a 2% Bush gain, it showed a 3.12% Kerry gain (!) ie the weighty process made the poll 5.12% worse, not better.
This is the danger of weighting: If you party ID is right your poll can be very accurate, but if your weighting is wrong, it can really blow up (a.k.a. Zogby...
)
During times when party self identification may change (Conventions, Reagan's death etc) weighted polls can behave oddly.
Virtually all of the change in last nights Bot sample was in the Indys - which makes me think what I outlined above may have happened.
Bottom Line...?
It's a blip... Bush is up 4% ish, maybe 6 if you "unweight" the thing for a bit of "froth" due to temporary party ID shifts.
Give the poll a few days to settle down and we will see..
Wait for the Gallup
