By targeting a definable probabilistic model, it is possible for all research studies to be consistently grounded.   Certainly, problems will arise in extremely low incidence studies where there are an insufficiency of profiled respondents but the idea is clear.   If POD gets big enough, the research community will able to tell its clients what their sampling frame represents.

As panels provide us with profiled individuals we will be able to draw samples from their contributions to check for the consistency of the panels themselves.   We do it now but are dependent on the panel to provide us with samples.  In the future we will be able to draw random samples that are quota controlled from POD and generate our usual consistency analysis.

Panel companies contribute respondents by the thousands which we in turn put through the profile platform.   We pass all the profiling data to the panel and are the gatekeepers of the profiling data base to provide future “profiles on demand” or POD.  POD is very powerful.  It frees the research of a dependency on a single panel and actually creates the ability to choose respondents on an as needed basis.   Some panel companies have an over abundance of a particular respondent type but through POD their strengths can become the online communities strength.   In turn their weakness can be strengthened by relying on the profiled respondents that fill their gap, all from POD.

No matter where respondents are sourced, we put them through the profile questionnaire and then one at a time we distribute them according to the segments that they are allocated to in the proportions dictated within demographic groups.   We keep demographic quotas and behavioral quotas.   The system solicits respondents that are in a profile data base.   We often use respondents from multiple panels representing different sourcing.   The result is that each project sample is a very sophisticated blend of individuals chosen because they fit the model.

We propose that in each country we are studying, a probabilistic sample be drawn by some method other than online panel.  In the United States we interview respondents by telephone.  To overcome some of the problems of telephone we augment the sample with a manually dialed cell phone sample.   The non-cell phone is Random Digit Dialed, eight attempts with one attempt during the day and refusal conversions.  We use a standard thirteen minute profiling questionnaire that generates ten segmentations that we seek:  buying, media, socio-graphic, automotive, grocery, clothing, entertainment, small electronics , large appliance, banking and insurance.   We call the process “behavioral fingerprinting”.

Once we know the distribution of segmentations within a population, we are then able to populate that standard.  The distribution is probabilistic and representative.

No, let’s not stop there. How about Socio-graphics, Media usage, all could be targets alone or together. In fact, that is what we propose. Why not overlay behavioral segmentations over demographics to achieve a new online stability.

Stability? Yes.

If we knew how the segments were distributed in the population we were studying, then we could create a behavioral target.

For those of you who are too young to remember, the Rosetta Stone provided archaeologists with a means of translating ancient languages. Buying behavior is as core to the study of Market Research as language is to archaeology. If you don’t understand it things can be quite difficult.
It would be interesting to compile a list of questions regarding Buying Behavior and use their answers as a way of categorizing respondents. Using segmentation analysis we could arrive at a labeling scheme that might be quite useful. For example, early adopters that would accelerate purchasing decisions using a credit card, who are brand fixated and own many of the hottest “goodies”, might be one category elucidated by our analysis. Those who were price sensitive, price over brand and kept their plastic for today’s essentials, would define a second and vastly different crowd. If we knew what the segments were and how they were distributed in the target population we were studying, then we would have a new set of targets that could be used in combination with demography.

We often write about Ron Gailey. Kind of a pioneer in the issue of respondent tenure and the way in which it impacts on purchasing intent. Respondent tenure is a type of behavior. Those who continue to complete our surveys are behaving in a manner which we certainly approve. We have asked previously if tenure should be the new demographic. That is a quota controlled measure that we must target in our studies. It is not the only kind of behavior that we are interested in controlling but certainly it makes a reasonable candidate.

If the idea of leaving your career to lady luck doesn’t thrill you than perhaps you would consider stabilizing your online samples around a relevant target that is quantifiable. Since we are to some extent, behavioral scientists, how about we peg our bets around relevant behaviors.

Now we have been harping on this issue for years. So I’ll give it another whack. Demography doesn’t cut it alone. The online community does not represent the off line community that well on many areas of interest. But let’s pass on that contentious comment. Let’s just say that the jury is out and the debate is in their hands. Instead let’s just say that we need to anchor to something and demography appears to have a few problems. Behavior might be a reasonable supplement to demography.

You remember the old joke: When does an online study represent the population well?
Answer: When you get lucky.
I recently spoke to a client who swore that online research works extremely well for him. He spoke boldly, “I have never had a train wreck!”
I suggested to him that perhaps he was extremely lucky or he was missing something. He was quite offended.
Then with a lump in my throat I explained. Online research works well when the variables being studied are not influenced by the factors that affect bias in the data set.
The truth is that some product categories are not impacted by the online respondent variables that cause some respondents to stick with online panels and others to attrite. For example there is no obvious reason why a respondent who enjoys completing a few hundred questionnaires a year should prefer Pepsi over Coke. There is a reason why those who refuse to complete surveys at all might choose one cola over another, we know that there are demographic differences between cola choice. However, as long as the panel you choose always sends you those respondents most willing to complete questionnaires you remain safe.
You just gotta be a bit lucky. That’s all.

Imagine a bulls eye five feet across and you win if you hit the target in that big round red center. No problem from a few feet– you score easily. The larger the target, the closer you are, the easier the score. Now we could make it tougher if you were to stand back a few feet and then a few more and so on and so forth. But we could make it impossible if we took away the target.
Completing an online study is like shooting darts without a target. Think about it: what is our target? Oh I remember now, in the recesses of our brain it is the census. But we all know that doesn’t cut it anymore. The online community does a reasonably poor job of representing the offline community. I better not throw that comment into a crowded bar filled with market researchers. So let me recant it for a little while, a very little while.