A sampling of bias

Hot Air:

This is what propaganda looks like:

Americans overwhelmingly support substantial changes to the health care system and are strongly behind [72%] one of the most contentious proposals Congress is considering, a government-run insurance plan to compete with private insurers, according to the latest New York Times/CBS News poll.

Bruce Kesler points out that — in traditional NYT/CBS fashion — the sample is badly skewed:

According to the actual poll data, of the 73% of respondents who said they voted in 2008 only 34% voted for McCain and 66% for Obama. The actual vote was 48% McCain.

This is a good example of why reading a poll is as much art as science, because the first problem is the percentage who say they voted in the 2008 election. In reality, no more than 62% of eligible voters cast ballots last year. Accordingly, the poll has sampled a lot of adults who were ineligible to vote… or, as often happens, respondents lied about voting. In such cases, the lie tends to skew in favor of the winner.

Does that mean the sample might be more valid than Kesler suggests? Not in this case. In this poll, the sample identified as 27% liberal, 37% moderate, and 29% conservative. In contrast, last week’s Gallup Poll showed Americans identify as 21% liberal, 35% moderate, and 40% conservative.

But wait… there’s more. The same NYT/CBS poll previously published more information about this very sample, showing that 16% was temporarily out of work, and another 10% was not in the market for work. Democratic pollster Stanley Greenberg tells us that those who have been unemployed within the past year (or have an immediate family member in that category) are the most supportive of a government takeover of the US healthcare system. So a sample with much higher unemployment than the national average tells us something about the skew here also.

...

There are enough questions about the sampling in this poll to at a minimum require further explanation by the Times and their polling company. Perhaps they failed to get enough Republicans because they were at work or were not interested in participating, but that is no excuse for not adjusting the sample to reflect the percentage of McCain or conservative voters and also reduce the number of unemployed to the current stats.