|
|
|
|
|||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
GENTLEMEN, TAKE OFF YOUR HATS
In the Range War between Advertising and the Public Interest, Sampling Error Rides Again.
By Erwin Ephron
|
|
Advertising unfortunately is a “black hat” to many. Like a saloon gambler, it seems smooth and slippery and plays with other people’s money. “Public interest” is a white hat. The poor honest sheriff trying to keep Dodge City safe. But both hats share a human weakness. They might palm a card or two to win
a game. The ongoing range war between the Council on Alcohol Marketing and
Youth (CAMY) and Beer, Wine and Spirits advertisers is an example from our
own Wild West. Does Alcohol Advertising Overexpose Youth?
But a quick look at the data -- CAMY always supports its conclusions with data -- shows the number of ads is not a good measure of the volume of advertising. The increase is from lower ratings as these advertisers, like many others, shift dollars to cable. The losing card buried by CAMY is while there are more ads, GRP delivery to the 12-to-20 group has not increased. Now that’s not the kind of hand a white hat should be playing. The second charge, that there are more ads overexposing youth, is serious and requires us to deputize a statistician. The 30% Threshold As of January 2004, the alcohol industry has agreed not to buy media where the underage audience is greater than 30% of the total. CAMY reports that in full year 2003, there was a dramatic increase in the number of TV ads exceeding this threshold, as if Alcohol was cranking out its messages to youth before the deadline. But wait a minute. CAMY also reports the percent of ads exceeding the 30% threshold was much greater for national Cable (13.3%) than it was for national Broadcast (2.5%), which makes no sense. The programs selected are similar and they are bought to the same demo target. This suggests another hidden card, a bias in the CAMY analysis. Sampling Error Can Bite Our research deputy tells us two things about statistical error: 1) the smaller the rating the larger the error as a percent of the rating. 2) The sampling error goes in both directions. It is as likely to overstate as to understate the true value. And that’s the source of the CAMY bias. Cable has smaller ratings compared to broadcast by a magnitude of ten, i.e., 0.4 versus 4.0. That means the relative error on estimates of percent composition 2-to-20 (the 30% threshold) is much larger for Cable than it is for Broadcast.[1] And that a larger number of alcohol ads on Cable (versus broadcast) will be reported as exceeding the 30% threshold because of sampling error even when their “true” composition is less than 30%. This explains the unusual pattern shown by the data, cable more guilty than broadcast in exceeding the threshold. Cable’s larger statistical error is not a problem for buying and selling TV because we select programs based upon their average performance. The analysis bias occurs when a buyer selects a program where the average 12-to-20 composition is well under the 30% threshold, but some of the individual telecast ratings (about 13%) are reported as exceeding 30% because of sampling error. You Have to Count Both The CAMY analysis is biased because it looks at individual telecast ratings, which are unstable, and counts only the overestimates (and not the underestimates which balance them out to provide the average value). It is the equivalent of the Commissioner of Baseball criticizing pitchers for throwing too many balls without noting that they also throw too many strikes. CPM Guarantees This kind of bias is familiar. The TV networks face a similar bias in CPM guarantees: Sampling error in the data creates both under- and over-delivery. But when they under-deliver it costs them money and when they over-deliver they don’t get credit. Their solution is to guarantee on average performance, not individual telecasts or small demographic targets. The solution for CAMY is to use average performance in cases where statistical error is large enough to distort the numbers. Meanwhile, back at the saloon, the sheriff is under suspicion of stacking the deck. Like I said, it’s getting harder to tell who the good guys are when you don’t look at their hats.
[1] The error calculation shows that given the size of the Nielsen sample, a true 30% composition of viewers 2-to-20 for a 0.4 rating will be reported as a value between 10% and 50%, 95% of the time. - December 8, 2004 -
|
|
|||||||||||||||
|
|||||||||||||||
|
|||||||||||||||
|
|||||||||||||||
It’s getting harder
and harder to tell who the good guys are. Especially when you’re just
a bystander looking at the cards. John Wayne understood this. Even playing
poker, he wore a white hat. 
In the
latest website posting, CAMY accuses the advertisers of Beer, Wines and Spirits
of increasingly overexposing youth (12-to-20) with their TV advertising. And
they show the particulars comparing 2003 to 2002:
Many of us use audience data from the Nielsen sample of people. We know
sample-based measurements have statistical error, which estimates the probable
range of difference between the sample data and a full census of the population
being sampled. 
