Thursday, December 01, 2005

The anatomy of a poll

Image Hosted by ImageShack.us

CNN headlined their web site with this screaming headline, "Most doubt Bush has a plan for Iraq victory". Upon reading the story, some interesting things come out:
But the CNN/USA Today/Gallup poll released Wednesday night also found nearly six in 10 Americans said U.S. troops should not be withdrawn from Iraq until certain goals are achieved.
Just 35 percent wanted to set a specific timetable for their exit, as some critics of the war have suggested.

Almost surrealistically, the CNN reporter then proceeds to decontruct the headline further by decontructing their own poll:
The poll conducted Wednesday does not directly reflect how Americans are reacting to Bush's speech, because only 10 percent of the 606 adult Americans polled had seen it live and two-thirds had not even heard or read news coverage about it.

Sounds like s a pre-emptive admission to mute criticism from right wing bloggers, who might probably say things like, "that headline was baseless". No worries, CNN went on to tell us why their self-admitted suspect poll contains true value:
But it does indicate the scope of the battle ahead as the Bush administration seeks to regain support for the war among an increasingly skeptical public.

For those scoring at home here's a recap...shorty after a major speech on an Iraq victory plan, CNN headlines their contention 'Bush has no plan for victory' based on their own poll. Then, in the same story they admit only 10 percent of those polled actually watched the speech, and only about 1/3 had any information about it at all. How does one score a foul ball off the foot?

Furthermore:
Asked if the war will make the United States safer from terrorism in the long run, 48 percent said yes and 43 percent no

Say what? The same poll indicated most Americans felt the war will make us safer? Not so fast. CNN made sure we understood that a 5 point margin could be meaningless, since it was "within the poll's sampling error.". What was the sampling error? Five percent.

No comments: