Nerve Endings Firing Away

If you're new here, you may want to subscribe to my RSS feed or email alerts. Thanks for visiting!

More in media election coverage bashing, yesterday Chris Matthews was talking about differences in support for Obama and McCain among women or more particularly suburban women. He had a graphic up on the screen with Obama’s photo captioned with “Woman Trouble?” Now what is seen as a part of increased media scrutiny toward media networks, this contention was immediately called into question. For those even remotely associated with statistics and polling, you might be aware of margin of error i.e. the extent of the confidence interval.

You can see this margin of error mentioned in tiny font size at the bottom of every poll they show on TV. Usually this margin is around 3 i.e. we cannot say with confidence a ‘real’ difference exists if this difference is separated by less than 3 points. The sample size or the number of people you interview often determines this margin of error and generally larger the sample, the better it is to draw conclusions from. Coming back to the NBC-WSJ poll that Matthews was using to infer Obama’s diminished women support talked to around 1000 respondents which is pretty healthy sample size assuming the sampling methods are fair, accurate, and tailored to ensure randomness. The margin of error for this poll was cited as 3.1 percent. In fact, according to this poll, it showed Obama had increased his lead (52-33) but among ’suburban women’, the support leaned toward McCain (44-38). Now since women are a subset of the overall sample, the margin of error will increase if you are going to examine their responses and when you specifically look at suburban women, the margin of error increases substantially considering that this group is now a subset of a subset. So would Chris Matthews contention still hold up given this change in technical details?

It turns out that after Huffington Post confronted MSNBC with this new information, they clarified that McCain’s 44-38 lead over Obama was well within the revised margin of error which is 9.34 based on a sample size of 110. So what does this mean regarding infering from the polls?

This means McCain’s 44 percent figure of support among suburban women could actually be as low as 35 percent, while Obama’s 38 percent figure could rise as high as 47 percent - assuming a 95 percent confidence interval (for the stat wonks in the house). Alternatively, McCain could be leading Obama 53-29.

So effectively due to the large margin of error, this difference reveals little and no news organization worth its salt will base any conclusion much less a blatantly misleading graphic on this finding. Generally when candidates are separated by less than 3 points in any poll that lies within a 3-point margin of error, the pollsters make it amply clear that the candidates are tied in a statistical deadheat. So even though the difference between candidates is larger even though the margin of error is also correspondingly larger, the public might be unaware that this supposedly large difference is almost the same as other polls with smaller differences with smaller margins of error. Admittedly the public is not attuned to such finer details but I would expect the media to know better else face allegations of deliberately misleading the public against one candidate. I hope that’s not what you are doing Mr. Matthews, is it?

by Patrix | on Friday, June 13th, 2008 at 4:00 pm | Media, Politicking . You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.

Article Tags >> deception | margin of error | Media | poll | statistics