The national Nielsen ratings are divined from 22,000 households, a relatively small sample.

But now other measurement services are collecting TV data from millions of households, which is closer to a census than a sample.

“Nielsen continues to produce a measurement system that works well for the broadly based programs but less so for niche programing with smaller audiences,” says Brian Wieser, senior analyst at independent investment research firm Pivotal Research. As audiences continue to fragment, larger samples are required to get reliable ratings for lightly viewed TV programs and channels. This is especially key in that advertising buys often aim at only a narrow slice of audience, such as women ages 18-34.

Providing TV audience data from a large pool, Rentrak generates its database from viewers in more than 8 million TV households, with 19 million TV sets. “We’re providing a more granular number than has been available before,” says Bruce Goerlich, chief research officer at Rentrak. “The huge size of our database allows us to do things you can’t really do with samples,” such as marrying purchasing data with viewing by postal ZIP code in a privacy-protected manner, or even focusing in on lists of households.

After Nielsen reported a sudden drop in audience for Nickelodeon, the kids channel’s owner, Viacom, maintained that set-top box viewing data from other services showed steadier ratings. CEO Philippe Dauman told a November earnings conference call that “evaluating independent set-top box data … shows meaningfully different viewership trends.”

As advertisers use a broader array of audience data, expect more differences to bubble to the surface.