And speaking of the Nielsens, does anyone else wonder how accurately 100,000 households can represent a nation of 300 million people?
Two different samples: One is a national sample, which contains about 9,000 meters inside your TV.
The second is local where the country is divided up into 210 DMAs. This is a combination of meters and diaries.
Have to agree with Mewzilla here. Sure, the ratings were low, but if it had been doing CSI-style ratings and no advertisers had been stepping up, that would have been it. Networks are in the profitmaking business, and both low ratings and unwilling advertisers hit right in the pocketbook.
Fairly accurately, I imagine. Statiscal sampling, if it's done properly, can be done highly accurately.
There are some potential problems. People with Nielsen boxes may cheat, and this could be done for different reasons. First, if the person likes a particular show, he may lie about how many people are watching the show. This could artificially boost ratings. Next, the person may change his viewing habits with the Nielsen box in the room. They may like a particular show, and although they don't feel like watching it (it's a repeat, a better, one time movie is on HBO, etc), they will either watch it (or at least leave the TV on so the box records it) or claim they watched it.
Political polling has a few problems: people don't want to admit they aren't voting for the minority race candidate, for instance.
What sampling can do, though, is allow you to ask a relatively small group of people, and extrapolate that to as if you asked everyone in the country. If there's a black guy running against a white guy, the white guy is usually going to get a few percent more than in the polls, but that's not because the polls were improperly sampled: if Gallup called and reached every household, there would still be the overstating of black support.
A problem that political pollsters have that Nielsen probably doesn't is determining if that person is really going to vote. Nielsen at least knows if you've turned on the TV.
Any poll also needs to make sure that the samples are representative of the population they are seeking to measure. Doing a nationwide poll of anything, but confining your sample to residents of Manhattan, is going to yield wacked out results. Oversampling Republicans or Democrats is going to skew the poll. In telephone polls, there's the risk that you are underpolling certain population segments not likely to take calls from strangers on their Caller IDs.
Wes Mantooth says Nielsens doesn't take into account households with more than two TVs.
My big thrill was listing "La Femme Nikita" and "America's Most Wanted", my favorite shows at the time. I hope it helped keep them both on the air - AMW at one time faced cancellation.
I asked the Nielsen rep, during our phone interview, why they had chosen me, and they said it was completely random. I assumed they got my name from the phone book since they first contacted me by phone, but they wouldn't confirm that.
Later I asked someone in the TV business who said, the Nielsens contact approx. a thousand people a week in each major metro area during Sweeps periods, which last approx. four weeks, and about half of them respond back. So approx. 500 families in my area (Pittsburgh) served as the barometer of taste for several hundred thousand people in our metro area for that particular Sweeps week.
If done properly it can be extremely accurate.
How accurately? About as accurate as any other "push poll"...
the infowarrior