Philadelphia Radio Ratings: November 2014

[davideduardo]

Polling and elections are about the only time there is a real-world check on survey/polling data.

The best example and the largest real world check is the ACS data, which is sample based, versus the decennial Census, which is, of course, a true census.

Many believe the ACS data is more accurate than the Census. In any case, the differences between the ACS reuslts and the Census are minimal, and the cost for the American Community Survey are much lower.

Additional proof that polling is usably accurate comes from consumer research ranging from sales estimates done by major package goods manufacturers all the way down to radio station format searches. If such polls did not provide accurate predictions, companies would not be spending hundreds of millions of dollars on them each year.

Yes, the pollsters usually come very close. That's because they've gotten very good at fudging the data. When polling started, pollsters for a decade or two were the punchlines of jokes. They elected Landon and Dewey. But they studied their data and learned how to apply windage. They learned for example that Southerners were less likely to actually show up and vote, so they gave less weight to those responses. There were a large number of factors that ultimately had to be weighted but with experience (and the availability of computers) they could make accurate forecasts. And polls are actually forecasts, more than true surveys.

Weighting for actual behaviour is not "fudging". It is applying reality to the specifications of a poll.

When media report the so-called "margin of error" they generally give only the plus or minus number, which is misleading. The actual margin of error gives a probability: if your sample is 405, your margin of error (no matter how larger the population being studied) is a 95 per cent change of being within plus or minus five per cent. Pollsters in their press releases can make any survey or poll sound good. They could say plus or minus three per cent but then the probability of falling within that margin is say 80 per cent.

If you want to go into standard errors, that is a whole different area and I'll have to figure out how to insert Greek characters in this forum. For the average interested party, it is enough to say that the margin of error is within a certain percentage except in very rare cases when it isn't.

The whole reason why we do polls is to avoid the prohibitive costs of a census. The caveats are well understood by users, who still bitch when the errors fall against them.

I am reminded of something that a former head of Arbitron told me once: At any given time there are three kinds of Arbitron clients. The first group had good books, and they are reminded about what skilled broadcasters they are. The second stayed flat, and say nothing. The third had bad books and publicly state that "something has to be done about Arbitron".

In radio we forget that ratings are "estimates" and have a margin of error. As a key person in the MRC reminded me, "statistics is the only science where 'error' is not a dirty word".

Sampling error is not, as the name may imply, about some mistake in data collection. It's about luck of the draw.

In the case of the census, there is no "luck of the draw" as a census theoretically includes everyone in whatever universe is being studied. In the case of the US Census, there are sampling errors because the theoretical n equals 100% yet the Census Bureau is unable to sample everyone with particularly bad results in transients of any kind and ethnic groups where the fear of any kind of government questioning may be very high.

 

[AM/FMRAINMAN]

I actually do like NJ 101.5 on the weekends. I also love their Saturday Night 80s show, just like the Saturday Night 80s show on WJRZ(Manahawakin). They are both classic hit stations but also play a lot more 80s and a lot less 60s. Is it that WOGL skews to an older crowd then both NJ 101.5 and WJRZ?