Thursday, March 11, 2004
SUSA Pennsylvania poll is skewed
Kos reader explosiveliberal points out that the most recent Survey USA poll of Pennsylvania, showing Kerry ahead 49-47, is actually skewed in favor of Bush. This is how the voters in the sample voted in 2000:
Note: this data includes only the 700 out of 802 people polled who 1. voted in 2000 and 2. remember who they voted for. There were also 92 people polled who didn't vote in 2000 and an additional ten people who voted in 2000 but didn't remember who they voted for.
In 2000, Pennsylvania went 50.6% to Gore, 46.4% to Bush, and 3.0% other. A balanced sample of 700 people who voted in 2000 would therefore consist of 354 people who voted for Gore, 325 who voted for Bush, and 21 who voted for someone else. To see what would have happened had SUSA polled a balanced sample of voters, we scale the Bush column of the table by 325/366, the Gore column by 354/300, and the "other" column by 21/34:
The totals in the rightmost column are what we're looking for. A balanced sample of 700 people who voted in 2000 would have preferred Kerry over Bush 375 to 305. Finally, we need to add back the people who didn't vote in 2000 or forgot who they voted for. These 102 people broke 52 for Kerry, 43 for Bush and 7 undecided. So the total balanced sample prefers Kerry over Bush 427 to 348. That's 53% Kerry, 43% Bush. Not bad!
Back to the Odd Hours main page
Bush in 2000 -366 respondentsAnd this in a state that went for Gore 51-46! What would the results have looked like with a balanced sample? Here's the raw data:
Gore in 2000 -300 respondents
| Bush in 2000 | Gore in 2000 | Other in 2000 | Total | |
| polled Bush | 301 | 26 | 11 | 338 |
| polled Kerry | 54 | 267 | 19 | 340 |
| polled undecided | 11 | 7 | 3 | 21 |
| total | 366 | 300 | 34 | 700 |
Note: this data includes only the 700 out of 802 people polled who 1. voted in 2000 and 2. remember who they voted for. There were also 92 people polled who didn't vote in 2000 and an additional ten people who voted in 2000 but didn't remember who they voted for.
In 2000, Pennsylvania went 50.6% to Gore, 46.4% to Bush, and 3.0% other. A balanced sample of 700 people who voted in 2000 would therefore consist of 354 people who voted for Gore, 325 who voted for Bush, and 21 who voted for someone else. To see what would have happened had SUSA polled a balanced sample of voters, we scale the Bush column of the table by 325/366, the Gore column by 354/300, and the "other" column by 21/34:
| Bush in 2000 | Gore in 2000 | Other in 2000 | Total | |
| polled Bush | 267 | 31 | 7 | 305 |
| polled Kerry | 48 | 315 | 12 | 375 |
| polled undecided | 10 | 8 | 2 | 20 |
| total | 325 | 354 | 21 | 700 |
The totals in the rightmost column are what we're looking for. A balanced sample of 700 people who voted in 2000 would have preferred Kerry over Bush 375 to 305. Finally, we need to add back the people who didn't vote in 2000 or forgot who they voted for. These 102 people broke 52 for Kerry, 43 for Bush and 7 undecided. So the total balanced sample prefers Kerry over Bush 427 to 348. That's 53% Kerry, 43% Bush. Not bad!
© 2004 Odd Hours
Reproduction permitted provided Odd Hours or the author of the quoted post is credited.