Technical Discussion of Robo-Rasmussen Polls

[Fmr. Gov. NickG]

From what I hear, Scott Rasmussen was mortified by the 2000 results and has recalibrated his polls to include a larger Democrat component than used in 2000.

I have emailed back and for with Scott a fair bit and while he has never (despite repeated attempts) told me the EXACT weighting he is using, reading between the lines and using the data he had published I am 99.9999% certain he is weithing at +3 to the Democratic side.

My best guess is 37/34/29 Dem/Rep/Ind, or in might be 38/35/27, but I am pretty sure it is the first one.

+3 Dem seems about right to me.  Do we know yet if he is weighting these state samples independently by Party ID?  Considering that Democrats in the South are much more likley to vote for Bush than Dems anywhere else, this could make a difference.  

[Fmr. Gov. NickG]

While you have established that the Democrat component of Scott's sample is slightly higher than you originally suspected, did you ever get the data on the black percentage of the sample?

If it is a high as I suspect, then Scott is overstating (compared to what my models suggest for the 2004 turnout) that Kerry is a point or two higher, and Bush a point or two higher.

If this is the case, then a Rasmussen tie, would actually mean a Bush lead of two to four points.

Also, as you have previously correctly noted, summer polls tend to favor Democrat candidates.  I have gotten home twice in the past couple of months and found messages on my answering machine for a robo poll from a local TV station.  

in his National sample Rasmussen stratifies for race at:

White 81%
Hispanic 7%
Black 12%

Actual 2000 turnout based on US Census bureau estimates was:

White + Asian 81.2%
Hispanic 5.4%
Black 11.6%

(does not = 100% due to "other")

The Hispanic number is a crapshoot - who knows in 2004?

Is Rasmussen not counting Asians?  They are one of the fastest growing minorities, and are now a large enough group that they should be counted on their own.  

I expect the % of black and especially Hispanic voters to be higher than in 2000, simpy because their % of the population has increased.  Not much higher, but Rasmussen's figures look about right, assuming the Asian thing is somehow explained.

[Fmr. Gov. NickG]

While you have established that the Democrat component of Scott's sample is slightly higher than you originally suspected, did you ever get the data on the black percentage of the sample?

If it is a high as I suspect, then Scott is overstating (compared to what my models suggest for the 2004 turnout) that Kerry is a point or two higher, and Bush a point or two higher.

If this is the case, then a Rasmussen tie, would actually mean a Bush lead of two to four points.

Also, as you have previously correctly noted, summer polls tend to favor Democrat candidates.  I have gotten home twice in the past couple of months and found messages on my answering machine for a robo poll from a local TV station.  

in his National sample Rasmussen stratifies for race at:

White 81%
Hispanic 7%
Black 12%

Actual 2000 turnout based on US Census bureau estimates was:

White + Asian 81.2%
Hispanic 5.4%
Black 11.6%

(does not = 100% due to "other")

The Hispanic number is a crapshoot - who knows in 2004?

Is Rasmussen not counting Asians?  They are one of the fastest growing minorities, and are now a large enough group that they should be counted on their own.  

I expect the % of black and especially Hispanic voters to be higher than in 2000, simpy because their % of the population has increased.  Not much higher, but Rasmussen's figures look about right, assuming the Asian thing is somehow explained.

The US census lumps them in as "Asian/Pacific" and they are about 2.5% or so.

The more or less vote like hispanics.

In 2000 Hispanics broke for Gore 61/38, Asians broke 62/37 for Gore.

Rasmussen does not take Asians into account directly, but is a bit high on both Blacks and Hispanics which should "about" compensate.

Asian is a very broad term however.

Japanese vote most Republican, while Koreans vote Democratic.

(Unrelated, but if you want to see real racial division, get the Japanese talking about the Koreans or the other way around) Lumping all asians together is just wrong, but at 2.5% they are still pretty small to be dividing up the pie as it were.

Does this mean he is not polling any Asians or people who refuse to disclose their race?  His White+Black+Hispanic %'s add up to 100%.

Also, I don't see how lumping Asians together is any worse than lumping all Hispanics together....Cubans and Puerto Ricans, for example.