I participated in the Nielsen's in the summer of '98, filling out a diary for one week of shows watched, and I've been fascinated by the Nielsen's ever since. At the time, an employee at my local TV news told me that for the week in question, the Nielsens send out 1,000 requests to each major Metropolitan area (mine is Pittsburgh) and that roughly only half respond.
So I thought it was cool that moi and approximately 500 people got to tell the Nielsen's what nearly a million people in the Pittsburgh metro area were watching.
The Nielsens are notoriously unreliable measurements of TV viewing. First off, you simply can't get within a statistically acceptable margin of error when sampling so few people in the first place. Second, the price to sample more people is astronomical, because it's not like a regular poll where you just make phone calls and ask people a few questions; you have to tally those diaries by hand. And third, people do an awful job of keeping accurate diaries. They lie about what they watch, saying they watched a show they actually missed just because they want it to stay on the air; they'll say they didn't watch a show they really did, just because it stars a person they don't like, etc. Even Nielsen's "people meters," electronic boxes that sit on top of the TV set in some family's homes in major markets, are only marginally better, because every member of the family has to use a remote control to "log in" and "log out" every time they enter or leave the room; they often forget, and often simply lie to the "people meter" the same way they lie in the diaries. And Nielsen completely ignores all out-of-home viewership: bars, offices, dorms, etc. In short, Nielsen's numbers are total crap by 21st century statistical standards.
In fact, they're so bad that all the major networks have either quit subscribing to the Nielsen service at one point or another, or at least threatened to, and tried to create their own measurement systems. The problem is that nobody's ever been able to come up with anything better that isn't so expensive as to be cost-prohibitive. So they invariably come slinking back to Nielsen. They suck, but if they didn't exist, the networks and advertisers would have nothing to base advertising prices on at all. Thus, everyone just keeps on playing the game, since the only alternative is a total collapse of advertiser-supported television.
(The general trends, of course, are true regardless: Even with such a bad system as Nielsen's, Fox News would not come in first place every single day if they weren't truly watched by more people than CNN and MSNBC. The question is, does FNC get 1.5 million viewers at, say, 9 pm? Or is it actually more like 5 million? Or only 500,000? Nobody really knows.)