CNN Sucks.

[theweaselking]

Their original graph:


Note the sampling error: +-7%- and the difference, 8%

A more accurate version:

Comments

[harald387.livejournal.com]

Poll: Examine graph one.

Tell me if you see the problem.

Problems.

I'd like to see -that- distributed instead of one of those stupid 'what anime sterryotipe r u' surveys.

-K

[ceara.livejournal.com]

Even the second doesn't have error bars. And the scale of the first one would be OK if it did have error bars, but at that scale they'd look hideously inaccurate. I'd probably do a scale somewhere in between, so that the 8% difference looked significant-ish, but the 7% margin of error was clearly delineated.

I think I was a science student for too long.

[desdenova.livejournal.com]

Published opinion polls never show error bars. Margin of Error does not make for easily-digestible newsertainment.

[ceara.livejournal.com]

Yeah, I know. Doesn't stop me wanting 'em, though.

[ex-cerebrate131.livejournal.com]

I'd be a lot more OK with the scale of the first one if it hadn't been published before in How to Lie With Statistics, p. 61. After all, it's only the vastly small percentage of us who've been science students for too long, or at least at all, who will even *notice* the scale.

[publius1.livejournal.com]

Let's just cut to the chase here: the reason that both graphs are absolutely worthless is that CNN commissioned a poll that had a 7% MOE.

A 7% MOE is a 14 point spread.

This is a worthless poll.

[sivi-volk.livejournal.com]

Hah! I remember learning in Grade 5 how changing the scale of a graph could allow you to skew results.

Good to see the folks at CNN were paying attention.