Arbitron's sampling methodology is FAR more important than Mr or Mrs DePetro's misdeeds

Sid Schweiger
Fri Aug 22 14:22:11 EDT 2008

>>Why is everyone giving Arbitron a free pass on this fiasco? Let's
assume that whoever falsified his or her identity for the purposes of
the survey was smart enough to give an address with a different
"apartment number" for each of the six "people" who received a diary.
That would still mean that Arbitron provided diaries to six "people"
who lived in what Arbitron would have had to believe was one building
(a six-unit apartment building that was, in fact, a single-family
house). It makes no sense that Arbitron should have furnished diaries
to six people who lived in the same building--even if the company
honstly believed that each one lived in a different apartment and thus
that no two were members of the same household.<<

Only true if you accept the premise that Arbitron's selections for measurement households occur at random.  They don't.  The company is privy to a host of statistical data that they wade through before making even one solicitation for diary or PPM placement.  If it turns out that six of their respondents live in the same building, so be it.  Where people live has no statistical correlation whatsoever to whether or not they are part of the sample Arbitron wants.

>>Having six people who live in the same building as part of a total
sample that had to be less than 100 should pretty obviously invalidate
the claim that the sample was statistically representative of the
market population. And BTW, it would make no difference if the survey
had used the PPM or diaries. If Arbitron can send diaries to six
people that it believes live in the same "apartment" building, it can
also send meters to six people who live in the same building. Given
the small size of the sample that is being used to represent the
market population, I very much doubt whether Arbitron could justify
sending meters to six people who live in the same apartment bulding in
New York City, where the sample size would be several times the size
of the Providence sample!<<

No statistical analysis Arbitron could possibly perform can compensate for untruthful answers from their prospective respondents.  In the past Arbitron has demonstrated that their samples, over time, are a fairly accurate broad spectrum sample of the general population.  Correlation to other well-known generalized statistical data (i.e., what % of the population owns a Toyota, what % has two children in the house, what % have a college degree, etc.) is quite high.

>>The problem is deeper than the accuracy of diaries; the problem is what
appears to be a totally invalid sampling method!<<

Again, only if you accept the premise (the incorrect premise) that Arbitron picks its respondents at random.

I don't believe for one second that Arbitron gets it right 100%, 90% or even 50% of the time.  The size of their samples has been a continuing source of conflict with their clients.  But...they're all we have, they try hard to get it right (their methodology takes up about half a dozen pages in the back of every rating book) and billions of dollars of advertising every year ride on its results.  If you can do better, be my guest.

Sid Schweiger
IT Manager, Entercom New England
20 Guest St / 3d Floor
Brighton MA  02135-2040

More information about the Boston-Radio-Interest mailing list