• by Bob Garfield , Featured Columnist, October 8, 2012

Don’t think too ill of me. I Google myself.

It’s not (only) vanity; it’s a Listenomics thing. But the fact is I do that stuff… and more. In addition to self-Googlation, I have alerts for blog mentions, tweets, news items and everything else on the Web involving me: my books, my shows, my columns, my real estate transactions, my sex tapes, my inner demons, etc.  

Take my word for it. It’s a mixed blessing:

“Well Bob, I've clicked on the headline, read the article, and am now commenting it. Happy? But what a petulantly smug, petty and badly written piece it is.”

Yeah. Mixed blessing and, as it turns out, mixed signals. Because positive or negative, you can’t take the accumulation of Internet sentiment at face value. Research from KD Paine & Associates has documented what has always been vaguely apparent: much of the human sentiment expressed online is as sentimental as an actuary, and a fair chunk of it is not human at all.

Paine, a social media and public relations research firm, harvested a year’s worth of posts, tweets and comments on three highly charged areas of health policy debate: genetically modified foods, high-fructose corn syrup and vaccinations. Researchers analyzed 301,497 items, both coding them for positive/negative sentiment and tracing them back to their source. The hypothesis was twofold: that the majority of posts would come from those who do not comment often, but that a minority of commenters would nonetheless have a disproportionate impact on overall sentiment.

The assumptions were borne out, and then some. Not only did Paine find that only 10% of the commenters generated more than a third of the commentary, but a mere 1% of the commenter universe generated 15% of the noise.

I say noise because the 1% are not individuals motivated by concern about issues. They are robots, pay-per-click sites, and content farms and make social media accounts.

Does that mean 15% of social sentiment is irrelevant?

No, it’s worse than that. Because 73% of the dubious posts were coded as “negative,” that means the 15% skews sentiment data downward. Furthermore, because the experiment was run in the subject areas of GMO, corn syrup and vaccinations -- where emotions (if not necessary intellectual rigor) run high -- it is reasonable to assume that there was a higher percentage of genuine, organic commentary in this sample than you’d see for less fraught subjects…such as brands.

In other words, the study’s designers arguably stacked the deck against their own hypothesis. In other words, in social media as a whole, the robot factor is probably significantly worse. And therefore, if you are careless in knowing what you are looking at, you can be misled. You can be misled into thinking you are hated more than you are actually hated. And you can be misled into thinking you are more beloved than you are actually beloved.

For instance, here is a tweet that caught my eye:

Nelson Mertz ‏@personagfavo The Chaos Scenario (Paperback): What happens when the old world order collapses and the Brave New World is unpre... http://amzn.to/wJTU1p

Well, gosh, Nelson -- thank you so much for noticing my last book! And for passing that along! It is so nice to see that three years later, my work still moves people to share.

But then I checked out Nelson’s @personaogavo profile, and what I discovered was a person (or something) following zero others on Twitter but closing in on his own 75,000^th Tweet. Oh, and here were the three right before the one about my book:

Rice Macaroni - 16 oz: If you love Pasta, you won't want to miss out on these. Our Gluten Free Pasta lines are a... http://amzn.to/HNYdBA

Nurturme Plump Peas 2Boxes of 8pk: "Nurturme is 100% natural and gluten free baby food. Nurturme is convenient dr... http://amzn.to/HNYdBs

7-8-9 Chair: The tambino 789chair's geometric, unconventional design engages kids with visual or alternative lea... http://amzn.to/HNYd4s

You’ve more or less got to assume Mr. Mertz, or the robot Mertz, is an affiliate marketer vacuuming up every SKU on Amazon and tweeting them willy nilly. But perhaps I digress. The question is, if SKUers are skewing, whom to pay the most attention to? Well, according to Paine, it is neither the phony 1% or the 56% majority of infrequent commenters.

“The richest, most authentic conversations,” says the report, “took place among medium-volume posters (13-24 posts in the 12-month time frame). These posts tended to be more positive than those from the high-volume or low-volume post groups. These posts occurred mostly on forums and blogs. These people were heavily engaged in specific activities, such as body-building or cooking.”

Key word, I believe: engaged.

The moral of the story is: averages can mislead. When analyzing social-sentiment data, the friendly tweet can steer you wrong, but beware even more of the simple mean.