Friday, October 19, 2007

Polls and surveys -- this one's better

With election coverage simmering, I guess we'll see plenty of examples of news people writing about political poles, uh, let's make that polls! Now comes Eugene Robinson, op-ed* columnist for The Washington Post. Today Robinson's column explores the significance of recent presidential preference poll results for Democratic candidates Hillary Clinton and Barack Obama.

Here's the part I want us to look at. I think he does a much better job than the blog, from another major metro paper, we looked at yesterday:
Obama, of course, would be the first African American president, which would be equally historic. And the predictable notes of caution that must be inserted into any column about Clinton's campaign -- that the polls in Iowa show a close three-way race, that anything can happen in New Hampshire, that Obama can rebound in South Carolina, that Howard Dean was leading in the polls four years ago, that nothing matters until someone, somewhere, actually casts a vote -- are more than cover-your-behind caveats, they're real. There's plenty of time for the whole nature of the Democratic race to change.

But some of the numbers are stunning. A national CNN poll released Wednesday showed that among registered Democrats, 68 percent of African American women said Clinton was their likely choice for the nomination while only 25 percent backed Obama. By contrast, Obama led Clinton -- 46 percent to 42 percent -- among African American men.

The CNN sample of black voters was small, meaning those numbers are not precise. But even taking into account the margin of error, the poll reports an unmistakable tendency that clearly works in Clinton's favor: To the extent that African American voters are taking identity into account as they ponder the presidential race, women are considering not only race but also gender. Democratic strategist Donna Brazile, who ran Al Gore's 2000 campaign, has called it "the 'sistah' vote."
A couple of points I'd make about it:

Robinson is cautious about all the polls. And he's especially cautious about the CNN poll he was analyzing. Notice how he points out the small sample of black voters? He's aware of the margin of error, too, which has an inverse correlation to sample size. In English that means the smaller the sample, the higher the margin of error.

And Robinson cites the CNN story. Links to it, too, on the website. Political junkies who live in Washington and read his column in the print edition will know how to find it. So I think that lets him off the hook a little -- he doesn't have to go into methodology because he isn't writing the main story. His analysis is kind of like a sidebar to CNN's mainbar.

Now let's look at the original CNN report. Here's the lede:
WASHINGTON (CNN) -- Sen. Hillary Clinton's lead over Sen. Barack Obama, her chief rival for the Democratic presidential nomination, is growing among African-American voters who are registered Democrats, and particularly among black women, a poll said Wednesday.

Sen. Hillary Clinton is the top choice of African-American Democrats, a new poll suggests.

Among black registered Democrats overall, Clinton had a 57 percent to 33 percent lead over Obama.

That's up from 53 percent for Clinton and 36 percent for Obama in a poll carried out in April.

The 26-point difference between black women and men underscores the fact that the nation's vote is divided not only by race, but also by gender, said CNN political analyst Bill Schneider. "Black women don't just vote their black identity," he said. "They also vote their identity as women."

Among white registered Democrats, Clinton drew 49 percent support, versus 18 percent for Obama and 17 percent for former Sen. John Edwards of North Carolina, the latest poll found.

The question had a sampling error of plus-or-minus 6.5 percentage points.
I think that's well done. It reports the head-to-head numbers. It offers, and attributes, a little bit of analysis. And it notes the high margin of error near the beginning of the story.

Plus it does something I think is important when you're writing -- or editing -- copy with a lot of numbers. It makes comparisons. See how Wednesday's poll is compared to one in April? If you're working with a story on a new city tax rate, compare it to last year's. Is it up or down? How much? A new city budget is adopted. What was last year's? Readers want to know stuff like that.

One reason the margin of error is so high here is because the analysis discusses survey findings among smaller groups -- or subsets -- of the main sample. It's worth looking at the way CNN throws the numbers around, at one point even noting results on one "question had a sampling error of plus-or-minus 6 percentage points." How can that be? Because the subset for that question was "white registered voters." A smaller subset of the overall sample.

At the very end, CNN appends this graf:
he CNN/Opinion Research Poll of 1,212 adult Americans was carried out by telephone Friday through Sunday. The sample included interviews of 307 blacks and 762 non-Latino whites.
That's the way it ought to be done.

Footnote.* An "op-ed" column is called that because in the print edition (also known as "dead tree format" because it's on paper), the "op-ed page" is opposite the editorial page. It's another word for an opinion column, which traditionally run on the op-ed page.

No comments: