Think of it as another form of poll goosing.
As an example of how news media can take a piece of information and make a false statement out it, consider CBC’s online version of the story about a poll released Friday by the same company that polls for the provincial government’s energy corporation.
“Liberal support in free fall” screams the headline.
The first sentence is less dramatic:
A public opinion poll released Friday suggests that Newfoundland and Labrador's Liberals have lost even more ground leading into the last half of the Oct. 11 election campaign.
There’s even a graphic that uses the numbers from the news release. They show a drop of five percentage points in decided Liberal support, according to the poll.
The only problem for CBC is that the headline and the lede are false.
The combined margin of error for this poll and the one before it is more than the five point drop shown in the report numbers in the two polls.. Therefore, the actual numbers for the Liberals fall within a range of 4.5 or 5 points above or below the figures given.
This is why polls with such large margins of error tend to be useless for most meaningful purposes. And for detecting trends, you’d have to see a huge drop between polls – like more than 10 points - in order to get something that could conceivably be called a significant change.
What would free fall look like?
Well, certainly a hell of a lot more than what is shown. 10 points or more would be a likely candidate for such dramatic language, especially over the course of a mere 10 days or so.
It’s also interesting that while CBC mentioned a relationship between MQO and advertising company M5, they didn’t mention that MQO is also Nalcor’s pollster because it is owned by M5. The advertising company is Nalcor’s agency of record.
CBC also said MQO was “affiliated” with M5. That’s not even close to correct either.
According to the provincial registry of companies, the same three men are the only directors of M5 (above), MQO (right)and all the companies within the M5 Group.
MQO is owned by M5.
That’s factually correct.
“Affiliated”?
That would be misleading bordering on deceptive.
- srbp -
Related:
4 comments:
Eric Grenier posted at threehundredeight that:
"The controversy erupted when the Liberals questioned the credibility of the poll. But in my experience, when a party starts saying that the numbers in a poll are cooked it is more of a sign that the party itself is cooked. Standard operating procedure is usually to claim that "I don't pay attention to the polls."
Many polling firms employ people with former ties to political parties and governments, and governments are often clients. It's just how it goes, and it seems unlikely that a parent company would sabotage its polling firm's credibility by putting out numbers which can be shown to be false on election day. How easily people are ready to believe that professionals are capable of lying, cheating, and breaking every ethical code in the book for marginal partisan gain is, frankly, depressing."
Thoughts?
In the first paragraph, Eric sets up the straw man and then proceeds to demolish it.
The second paragraph makes it clear he doesn't understand the issues in this particular case.
MQO is the pollster for Nalcor, a fact the company did not disclose in any of its releases.
They also refused to disclose whether or not they are also working for the provincial government or the Tory party at the same time they are doing this political polling.
Eric's comments don't touch on those points. Maybe he didn't know. maybe he just brushes it off with a pfft.
Certainly my comments - when the first poll appeared - have not been about something as trivial as the firm having a couple of ex-partisans on staff. That's part of the strawman and frankly it's a pretty lame argument on Eric's part.
As for the the bit about putting out numbers that are a crock, CRA did it 2007. It wouldn't be the first time a firm released crap in this province, claimed it was gold and had it reported as platinum.
Marginal partisan gain? That's an amazingly simplistic comment that again shows he is basing his comments more on assumptions than evidence.
Interesting position to be in for some who trumpets the accuracy of his own analysis based on evidence.
"As for the the bit about putting out numbers that are a crock, CRA did it 2007. It wouldn't be the first time a firm released crap in this province, claimed it was gold and had it reported as platinum."
Are you referring to CRA excising the "Unsure/Don't Know" group among those being polled when publishing the results?
If so, is that all the problem is? Granted that is misleading and certainly problematic enough, but the spectre of MQO actually manipulating the raw data seems to be what's suggested by some when people raise the spectre of MQO having previously worked for NALCOR and the PC Party.
Is the accusation also that even among decideds, Liberals are actually getting higher than 13%, but the number is being suppressed by M5 to placate major clients?
First of all, this is not a new issue for me. I've been raising it since 2007.
As for what I am critiquing, here goes:
The question is firstly an ethical one related to the failure to disclose a real or perceived conflict of interest in both CRA and MQo's case.
For the record, they are CURRENTLY working for Nalcor and the prov Gov. You've got a number of points in your comment wrong.
That's enough of a problem up front and it is significant. The incident in 1989 (apparently the use of old polling data) demonstrates the extent to which some people areprepared to do what Eric Grenier apparently thinks would never happen.
The second element is the media one. They simply report the results and report them badly or wrongly. They then support or engage in tons of wild-assed speculation based on inaccurate or useless figures.
The third element is the specific point about CRA in 2007. CRA got the numbers for the Tories dramatically wrong, full stop. I'd have to go back and double check the exact figures but as I recall, they had the Tories at 61% of eligible voters even after you deconstructed the UND torque.
The actual result was 43%.
The, there was UND/Will not vote number that was 38% and not 18% as CRA showed.
In order to be off by that much you have some serious problems in the methodology.
At least one other pollster - Telelink - doesn't appear to have the problem. In 2007 they were one percentage point off the actual PC result and UND/will not vote number.
Post a Comment