Tuesday, October 07, 2008

Bwa Ha Ha

Lorie pointed out that CBS has a new poll, so I went and had a look, and it’s funnier than the comics. Sure, I generally like CBS because they have a consistent methodology, and because they always publish their questions and party breakdown. This allows me to reverse their calculations and see how they got their results.

First off, it’s more fun if you play along at home, so here’s the poll address.

OK, now let’s keep it simple and start with page 14, the last page. Let’s read together the last two sentences in that poll:

”Every registered voter is included in the likely voter model, and is assigned a probability of voting, which is used to calculate the likely voter results. The sum of these probabilities is the effective number of likely voters.”


In those two sentences, the folks at CBS are admitting that they MADE UP THEIR STATISTICS ON LIKELY VOTERS!!! There are no real voters who were asked if they were likely to vote in this poll – not even one! Remember, these polls are how the media tells us what they claim we are thinking, and here at the end is a disclaimer saying that at least a portion of their major findings is fabricated. Just something to keep in mind as we move along, nothing to see here, move along.

Also on page 14, just above that comedy about CBS making up their ‘likely voter’ database, is the list of respondents. Let’s mark that, and total up the participants and their proportions:

...................... Unweighted ....... Weighted
Total Respondents .... 957
Registered Voters .... 875 ................ 821
Republicans (RV) ..... 248 (28.3%) ...... 233 (28.4%)
Democrats (RV) ....... 320 (36.6%)....... 311 (37.9%)
Independents (RV)..... 307 (35.1%)....... 277 (33.7%)

Why is this important? Because let’s go look at page 8 and look at the support by party affiliation:

...................Obama ......... McCain
Republicans........ 7%............ 86%
Democrats ......... 85%........... 10%
Independents ...... 39%........... 44%

CBS released a poll headline, remember, saying Obama was leading McCain 47% to 43% among Registered Voters. So let’s apply the math of the weighting and see how they did:

Republicans: 7% at .284 = 1.99%
Democrats: 85% at .379 = 32.22%
Independents: 39% at .337 = 13.14%
Total : 1.99 + 32.22 + 13.14 = 47.35%

Republicans: 86% at .284 = 24.42%
Democrats: 10% at .379 = 3.79%
Independents: 44% at .337 = 14.83%
Total: 24.42 + 3.79 + 14.83 = 43.04%

OK, now we have to stop here and ask if those number make sense. Well, for one thing, I like that CBS tells us the exact unweighted numbers. It tells us, for example, that 89.55% of the people they reached said they were registered to vote, and we know that the poll was done over the weekend by telephone, and that the total respondent pool was below 1,000 people. Looking at the detail on page 8, we also know that urban areas figured heavily in the poll (look at the white vote, and the conservative/moderate/liberal breakdown of support). By the way, the claim at the bottom of the page by CBS that the poll has an overall margin of error of only “three percent” is not statistically true, as there are less than 900 valid respondents; the actual MOE would be 3.5 percent.

Why do I think the actual numbers are better for McCain? Let’s look again at those specific supports by party affiliation and reline them.

BASE SUPPORT (Republicans for McCain, Democrats for Obama)
McCain 86%, Obama 85%, McCain +1

COUNTER-PARTY SUPPORT (Republicans for Obama, Democrats for McCain)
McCain 10%, Obama 7%, McCain +3

McCain 44%, Obama 39%, McCain +5


Well, there’s one way that Obama comes out on top and CBS told you on page 14. But what happens if we use the historical party affiliation from, say, the last federal election in 2006? Let’s see:

Democrats: 38.5%
Republicans: 35.5%
Independents 26.0%

Democrats: 85% at .385 = 32.73%
Republicans: 7% at .355 = 2.49%
Independents: 39% at .260 = 10.14%
Total : 32.73 + 2.49 + 10.14 = 45.36%

Democrats: 10% at .385 = 3.85%
Republicans: 86% at .355 = 30.53%
Independents: 44% at .260 = 11.44%
Total: 3.85 + 30.53 + 11.44 = 45.82%

Well within the margin of error, to be sure, but a very different picture, also plainly to be seen.

The lessons are the same as always: Pay attention to the details, note trends within polls rather than aggregating disparate poll reports, and never forget that every media outlet is selling a story, and every poll is bought by someone for a specific purpose. The election is something far different from any poll, and we must not let ourselves be dismayed or misled by what anyone tells us in the media.

1 comment:

rightwingprof said...

Aggregated data are masked data.