Audience measurement, Nielsen's validity, and surveying

A couple weeks ago I got in a debate with someone on the internet about Nielsen ratings. He said something to the effect of: "of course these are basically accurate; after all, the advertisers and networks rely on these for billions of dollars of expenses". I responded that comparable amounts of money were involved in the mortgage crisis - and he responded that that was "entirely different" and "involved collusion". I thought this was naive, but it sparked my interest in the topic.

I set about to learn more about ratings. According to Nielsen's website, it measures 40% of the world's TV viewing, and in the United States it has 25,000 meters. I didn't really understand what that means and tired of Nielsen's website quickly, so I went to my next source: Google Scholar. There I found an interesting, but rather lengthy Master's dissertation by Seles (Audience Research for Fun and Profit: Rediscovering the Value of Television Audiences). Seles referred me to the book Ratings Analysis, but I hadn't finished the first chapter before I wandered to the Wikipedia page - where I should have started, and which covers the gritty process details decently - and read this passage (referenced to a USA Today article):

The number of U.S. television households as of 2009 is 114,500,000.^[12] As a result, the total number of Nielsen homes only amounts to 0.02183% of the total American television households, meaning that 99.97817% of American households have no input at all into what is actually being watched. Compounding matters is the fact that of the sample data that is collected, advertisers will not pay for time shifted (recorded for replay at a different time) programs.

Further, these sampled households self-select into this group. These are people who apparently care less on average about privacy. I'm not sure if they are being paid, if they are that would suggest they are also lower income. Seles points out that digital set-top boxes allow for an enormous amount of datamining which is much more valid than Nielsen ratings, but a confluence of factors have not led to these being fully exploited. As an odd sidenote, this 2004 Wired article reports that at that time, Nielsen was using diaries from samples of 500 to assess viewing. Diaries from 500 people, in the 21st century! It's worth noting that television networks should not be necessarily bothered by inaccurate data so long as the data overall does not underestimate viewers; it's only the advertisers that are screwed.

I've commonly heard people say that television advertising is more expensive than online advertising, but a 2009 eMarketer report abstract puts tv advertising at 13 cents/hour and online advertising at 17/cents hour. This suggests that advertisers are willing to pay up for the relatively rich and interactive data (e.g., click throughs and purchases). It's a topic which I need to become familiar with as I look into the media sector for investments, but like all these sectors it's mysterious at first. I've been looking at television networks (content owners) as a value alternative to the relatively "hot" area of Netflix, although I haven't gotten deep into it as of yet.

In my investigation into this topic, I ran across Reg Baker's The Survey Geek blog. This is an apparently honest, critical look into the survey industry which lays out the issues which I've suspected lie in the area: telephones increasingly difficult to access, surveytakers unresponsive, self-selection bias, etc. I'm still not clear on how the political surveys are accessing cellphone only users like myself. UPDATE: as I finish a survey on 9 Nov 2012, I have to say I didn't expect the surveys to be so terrible or tedious. There's really a lot of self-selection bias. I got $30, but not for sending it in (at least it's not biased towards poor people). Clearly this is only targeting really old-school cable; apparently the intelligent set-top boxes still haven't rolled out. Since I didn't have cable, I plugged my Hulu Plus shows but probably for no reason.