September 01, 2007

Research or Baloney?

For a brief period I was a science teacher. I didn’t actually have a license to teach science but because of a teacher shortage you could say I was drafted.

My three years in the classroom taught me a lot about science and the scientific method. Most of all, it gave me a healthy respect for the difference between a fact and an opinion.

Scientists are constantly struggling to establish factual knowledge of the world. Their efforts are relentlessly reviewed and tested by their peers. In fact, in scientific terms, nothing is ever truly a “fact”. They learn to be satisfied with degrees of certainty.

Does the Earth revolve around the sun? We have an extremely high degree of certainty that it is does. But if someone were to produce an experiment tomorrow that conclusively contradicted this, science would have to alter its beliefs. That’s the difference between science and most other disciplines. Politicians, for example, are loathe to alter their beliefs, regardless of the outcome of their actions. Science is always skeptical and always prepared to re-evaluate.

As an admirer of the scientific process, I have a less than sterling opinion of the research methodologies employed in the advertising industry. As a matter of fact, I believe most of them are based on very suspect suppositions and very flimsy science. Most of the research we conduct and conclusions we draw would be laughed out of any respectable lab in the country.

Worst of all is the current trend toward on-line research. On-line research has some major advantages. It is often less expensive than standard methods and also quicker to yield results. However, as currently practiced, it is fatally flawed.

"We're perpetuating a fraud," is what Simon Chadwick has to say. Mr. Chadwick is former head of NOP Research in the U.K. and is now principal of Cambiar, a Phoenix consultancy. "(On-line) surveys tend to poll the same people over and over.” In fact, a study done by ComScore Networks indicated that one-quarter of one percent of the population provides about one-third of all on-line responses. This means that instead of getting one vote, each of these respondents is getting the equivalent of 128 votes.

We are getting the same people responding over and over again to earn points so they can win a toaster. Or as Mr. Chadwick calls them, "professional respondents who go hunting for...dollars”.

What’s so terrible about professional respondents, you might ask? Pulitzer Prize winning New York Times science writer Natalie Angier says: “Nothing tarnishes the credibility of a sample like the desire to be sampled.... a good pollster will hound and re-hound the very people who least want to cooperate.”

So not only are these people ridiculously over-represented, they are the wrong people.

"It's like the hole in the ozone layer," said Shari Morwood, VP-worldwide market research at IBM in an article in Advertising Age. "Everyone knows it's a growing problem. But they just ignore it and go on to the next project."

Kim Dedeker, VP-consumer and market knowledge at P&G, describes an example in which online and mail surveys came up with diametrical results. "If I only had the online result.... I would have taken a bad decision right to the top management," she said. In another case, two surveys conducted a week apart by the same online researcher yielded completely different recommendations.

Furthermore, most of these on-line researchers don’t validate their samples. They don’t know who is responding. It could be my daughter using my computer saying she’s me. Or saying she’s you for that matter. And if all that weren’t enough, many of them don’t limit responses. I can log in as five different people and respond five different times. Or fifty. Or a hundred and twenty-eight.

Another lovely bit of hokum they perpetrate is the “degree of confidence”. They tell us that their results are accurate with a “95% degree of confidence.” However, they never quite tell us what it is that they’re confident about. Is it that, in general, a study with this many legitimate respondents will be statistically valid 95% of the time? Or is it that their interpretation of subjective data will be 95% accurate (by the way, no one’s interpretation of subjective data is 95% accurate) Or is it something else?

Let’s give them the benefit of the doubt for a minute and say that their sample is legitimate (which is highly unlikely) and that they are brilliant people who can interpret data almost flawlessly. Let’s take a look at what “95% degree of confidence” means under the best of circumstances. Once again we’ll turn to Ms. Angier from her book The Canon.

Here’s an example she gives. You go for an HIV test. You test positive. The test is said to be 95% accurate. This means you have a 95% chance of having the HIV virus, right? Not even close.

What it means is that 95% of the time people who have the HIV virus will test positive. But it also means that 5% of the time people who do not have the HIV virus will test positive.

Now let’s say you live in a town with 100,000 people. Fortunately, the HIV virus is very rare and only appears in 1 person out of 350. So in your town of 100,000 people, this means that there will be about 285 people with the HIV virus (100,000 divided by 350).

But if we tested all the people in your town, we would get about 5,000 positives (remember, 5% of the time people who do not have the virus will test positive) and almost all of these 5,000 “positives” would be false.

In fact when you do the math, after testing positive not only is there not a 95% chance you have the virus, there is about a 5% chance you have it. And an almost 95% chance you don’t have the virus.* So much for a “95% level of confidence.”

We advertising and marketing people are drowning in opinions and starving for facts. But we have to be very careful about distinguishing between the two. In the advertising world, research is no different from creative work. Some of it is very good, some of it is worthless and dangerous.

* To figure out the accuracy of the result, you divide the total number of true positives you’d expect from your sample (95% of 285, or 271) by the total number of true and false positives (5,257) and you wind up with a probability of having the HIV virus is actually about 5.2%, not 95%. If you can’t follow the math, and you don’t trust me, don’t worry. You can trust Ms. Angier, she has a Pulitzer Prize. All I have is a stupid website.


No comments: