Big Lizards:Blog:Entry “Swervy Surveys”

« Weapon of Mass Media Deception | Main | Is Paris Burning? »

November 2, 2005

Swervy Surveys

Hatched by Dafydd

The latest surveys from various sources are still mixed -- but now they're virtually incomprehensible. As always, Daniel Weintraub leads the way with his Sacramento Bee-blog California Insider.

The disappointment is SurveyUSA, which consistently had all four of the Schwarzenegger initiatives up; now it has Proposition 73 (Parental Notification Before Abortion) winning by 55-44, 77 (Redistricting Reform) losing by 44-53, and 74 (Teacher Tenure Reform), 75 (Paycheck Protection), and 76 (State Spending Limitation) all dead even.

One caveat: SurveyUSA uses short, punchy descriptions of the ballot initiatives. But they did a "split ballot" on Proposition 76 this time and noticed that the longer their description was, the worse the initiative did: with the original description of 36 words, the measure ties 49-49; with a longer description of 51 words, Proposition 76 loses by 42-56; and with the third description of 54 words, it loses by a whopping twenty-five percent: 36-61!

Bear in mind, these descriptions are all read to the respondents over the telephone (by a recorded voice); they do not have them in written form before them. It's entirely possible that the longer a description that the respondents have to listen to on the phone, the more lost they become -- and when they get lost, they tend to vote No. Interestingly, SurveyUSA noted that the longest description they read is similar to the descriptions read by the other two phone pollsters, the Field Poll and the PPIC poll... both of whom showed strikingly similar results for Proposition 76: losing by 28% and 32%, respectively.

By contrast, a new Knowledge Networks Poll conducted by the Hoover Institute of Stanford University shows a kinder picture, with only Proposition 76 (State Spending Limitation) losing, by 10 points (45-55). The other initiatives are all winning.

Once again, the...

Table 1: Justifiably World Famous Big-Lizards Vote-At-a-Glance Table

Survey USA (in bold); Stanford Knowledge Net Poll (in italics)

Prop 73: Parental Notification Before Abortion
55 yes, 44 no (2% undecided) lead: +11
58 yes, 42 no (0% undecided) lead: +16
Prop 74: Teacher Tenure Reform
49 yes, 50 no (1% undecided) trail: -1
53 yes, 47 no (0% undecided) lead: +6
Prop 75: Paycheck Protection
50 yes, 49 no (2% undecided) lead: +1
64 yes, 36 no, (0% undecided) lead: +28
Prop 76: Limit State Spending Growth
49 yes, 49 no (2% undecided) Tie: 0
45 yes, 55 no (0% undecided) trail: +10
Prop 77: Redistricting Reform
44 yes, 53 no (3% undecided) trail: -9
55 yes, 45 no (0% undecided) lead: +10

(The Knowledge Net Poll is a "forcing poll," where respondents do not have the option of voting "no opinion" or "undecided;" just as with the real ballot, they can vote Yes, vote No, or skip the question -- in which case they are not counted for that question. That is why all questions have 0% undecided above.)

I have one reason to be somewhat skeptical of the Knowledge Net Poll -- and one reason to consider it more reliable:

My back of the envelope calculations indicate the poll may have oversampled Republicans.

Assuming that roughly the same people voted on each question, taking the first three questions, turning everything into matricies and inverting, I ended up calculating 42% Republicans, 25% Independents, and 34% Democrats. Now, bear in mind that the initial sample was probably closer to the state party percentage; it's possible this simply represents a "side prediction" that turnout will be higher among Republicans. But I'm still a bit worried, and this makes me somewhat skeptical about the results.

Therefore, I recalculated all the question results using two alternative turnout models; see Table 2 below.

On the other hand, unlike the other polls, the Knowledge Networks Poll is not a telephone poll. Instead, respondents vote on a simulated paper ballot, just as they would vote in the voting booth.

The ballot descriptions are there in front of them, and they can read and reread as necessary, rather than just listening over the phone. This makes it a better simulation of the actual vote than a telephone poll.

As in the 2003 California Recall Election Surveys conducted by Stanford University and Knowledge Networks, the wording from the actual ballot was presented to survey respondents. Therefore, there was not provided to respondents a category to capture “undecided” or “not sure” responses. Respondents were forced, as in the actual ballot, to make a “vote” decision or to skip a ballot question. Also, as in an actual vote decision, the interview simulated conditions in the ballot box in that an interviewer did not administer the ballot questions; the questionnaire is self-administered.

(Daniel Weintraub opines that during the 2003 California Recall election, both the SurveyUSA poll and the Knowledge Networks Poll "pretty much nailed it.")

Here is a table contrasting what you get with the original mix, as above (in bold); then after correcting the percentages of Republicans, Independents, and Democrats to 37.5%, 25%, and 37.5% (equal numbers of Rs and Ds, italics); and then finally correcting again to flip the percentage of Rs, Independents, and Ds to 34%, 25%, and 42% (ordinary Roman text):

Table 2: Knowledge Networks Poll, Three Turnout Models

Prop 73: Parental Notification Before Abortion Robust Lead
58 yes, 42 no -- lead: +16
56 yes, 44 no -- lead: +12
55 yes, 45 no -- lead: +10
Prop 74: Teacher Tenure Reform Shaky Lead
53 yes, 47 no -- lead: +6
50 yes, 50 no -- Tie
48 yes, 52 no -- trail: -4
Prop 75: Paycheck Protection Robust Lead
64 yes, 36 no -- lead: +28
61 yes, 39 no -- lead: +22
59 yes, 41 no -- lead: +18
Prop 76: Limit State Spending Growth Loser
45 yes, 55 no -- Trail: -10
41yes, 59 no -- trail: -18
39 yes, 61 no -- trail: -22
Prop 77: Redistricting Reform Good Lead
55 yes, 45 no -- lead: +10
52 yes, 48 no -- lead: +4
50 yes, 50 no -- Tie

I define an initiative as having a "robust lead" if it wins in all three turnout models; it has a "good lead" if it wins in two of the three; it has a "shaky lead" if it leads only in the first; and one initiative is a "loser," losing in all three turnout models.

So that's my story, and I'm sticking to it!

Hatched by Dafydd on this day, November 2, 2005, at the time of 12:42 AM

Trackback Pings

TrackBack URL for this hissing: http://biglizards.net/mt3.36/earendiltrack.cgi/181

Comments

The following hissed in response by: California Conservative

The most important poll is on Nov 8th. We're covering the initiatives here, and encouraging debate in an open post here.

Go Arnold!

The above hissed in response by: California Conservative at November 2, 2005 12:25 PM

The following hissed in response by: KarmiCommunist

After what happened in Texas with Redistricting, the Dems will turn *PURE* **** if they lose that in California.

[Edited to remove Godwinism. -- the Mgt.]

The above hissed in response by: KarmiCommunist at November 2, 2005 3:16 PM

The following hissed in response by: KarmiCommunist

OK...in order to avoid editing, humble me will let the Milwaukee Journal Sentinel speak for themselves:

In losing a woman, the court with Alito would feature seven white men, one white woman and a black man, who deserves an asterisk because he arguably does not represent the views of mainstream black America.

Think of Justice Clarence Thomas* and that he "deserves an asterisk" behind his name, or perhaps a "Star of David"...

Karmi

The above hissed in response by: KarmiCommunist at November 2, 2005 5:16 PM