Correction of the morning
The folks downtown manage to get almost everything wrong in this correction:
On Thursday's front page, a chart showing which candidates were supported by black voters in Michigan's Democratic primary should not have said the survey's margin of sampling error was 4 percentage points. It was higher.
1) Avoid speculating about what should or shouldn't have happened. Corrections are about what was or wasn't wrong, not about what should have been.
2) Unless it's needed for clarity (as it sometimes is), avoid repeating the original error in the correction. [Update: As The Ridger notes below, Bill Walsh has a good take at Blogslot explaining the opposing view on this issue.]
3) On the other hand, a correction really, really ought to give the correct information -- not just say "it was higher."
This is a particularly interesting correction for a couple of other reasons. One, the item that's being corrected isn't necessarily "wrong." It's easy to make the margin of sampling error for that subgroup come out at 4 points, as long as you don't mind abusing the confidence level.* Two, in this case, the error doesn't affect the point of either the graphic or the story (which mentions a number of other subgroups with no indication of their size). Black voters were significantly more likely to report voting uncommitted than to report voting for Clinton.
Three and most fun, the correction doesn't have anything to do with the story's bizarre hed:
If Clinton gets the
nod, she must woo
some black voters
Even though seven out of 10 African Americans did not vote for Sen. Hillary Clinton in Michigan's Democratic presidential primary, analysts and black voters don't see it as a sign of irreversible trouble for Clinton if she becomes the party's nominee.
If you hang on until the 23rd graf (past all the anti-Clinton primary voters who say they'd vote for her in the general election anyway), you get to the opinion that might support the hed:
Todd Shaw, an assistant professor of political science and African-American studies at the University of South Carolina, said Michigan's primary results are an indication that black voters are apprehensive about Clinton.
If Clinton does receive the nomination, recognizing that a significant segment of the black electorate does not support her, she'll have to court those voters as if she's just beginning to campaign, he said.
Interesting conjecture, but -- given what Real People seem to be saying -- it needs a lot of hedging before it can stand in a hed.
* Today's quiz: Calculate the margin of sampling error at 95% confidence for this subgroup of 225 voters in which 30% said "Clinton" and 68% were uncommitted. How much does it differ from the maximum margin of error for the subgroup?
On Thursday's front page, a chart showing which candidates were supported by black voters in Michigan's Democratic primary should not have said the survey's margin of sampling error was 4 percentage points. It was higher.
1) Avoid speculating about what should or shouldn't have happened. Corrections are about what was or wasn't wrong, not about what should have been.
2) Unless it's needed for clarity (as it sometimes is), avoid repeating the original error in the correction. [Update: As The Ridger notes below, Bill Walsh has a good take at Blogslot explaining the opposing view on this issue.]
3) On the other hand, a correction really, really ought to give the correct information -- not just say "it was higher."
This is a particularly interesting correction for a couple of other reasons. One, the item that's being corrected isn't necessarily "wrong." It's easy to make the margin of sampling error for that subgroup come out at 4 points, as long as you don't mind abusing the confidence level.* Two, in this case, the error doesn't affect the point of either the graphic or the story (which mentions a number of other subgroups with no indication of their size). Black voters were significantly more likely to report voting uncommitted than to report voting for Clinton.
Three and most fun, the correction doesn't have anything to do with the story's bizarre hed:
If Clinton gets the
nod, she must woo
some black voters
Even though seven out of 10 African Americans did not vote for Sen. Hillary Clinton in Michigan's Democratic presidential primary, analysts and black voters don't see it as a sign of irreversible trouble for Clinton if she becomes the party's nominee.
If you hang on until the 23rd graf (past all the anti-Clinton primary voters who say they'd vote for her in the general election anyway), you get to the opinion that might support the hed:
Todd Shaw, an assistant professor of political science and African-American studies at the University of South Carolina, said Michigan's primary results are an indication that black voters are apprehensive about Clinton.
If Clinton does receive the nomination, recognizing that a significant segment of the black electorate does not support her, she'll have to court those voters as if she's just beginning to campaign, he said.
Interesting conjecture, but -- given what Real People seem to be saying -- it needs a lot of hedging before it can stand in a hed.
* Today's quiz: Calculate the margin of sampling error at 95% confidence for this subgroup of 225 voters in which 30% said "Clinton" and 68% were uncommitted. How much does it differ from the maximum margin of error for the subgroup?
Labels: corrections
posted by fev at 10:53 AM
3 Comments:
I find this whole concept very bizarre. As someone whose candidate rarely makes it to the election, I can say I've never been tempted to either vote for the other party or not vote at all. If you prefer Obama to Clinton, you're not likely to vote for Huckabee or Romney, are you? I mean, surely not.
Also, interestingly, just last month Blogslot had a post called "Repeat the Error! Repeat It, Already!" I guess there's a difference of opinion, huh?
I need to go back and put in a link to that. Bill has a nicely reasoned (as usual) explanation of why the conventional wisdom about corrections isn't necessarily a mandate from the gods. I'm in the camp that thinks repeating the error creates a risk (however slight) of confusion, and that when you can be specific about the nature of the error -- in the case at hand, "understated the margin of sampling error at 95% confidence" would do fine -- there's no need to introduce that risk. Bill is a fearless debunker of hoary grammatical unwisdom and always worth listening to, but I think he's overcorrecting on the corrections.
I entirely agree on the party-shifting thing (and a state lawmaker in the story puts it quite similarly to you). That just isn't the way electorates work. Makes the hed seem even more like an editorial, doesn't it?
I'm still trying to work up a good all-in rant about the framing of the primary process, but the semester keeps intruding.
Post a Comment
<< Home