Posted by Herb on February 13, 2002 at 11:34:10:
In Reply to: Analyzing B2B Research posted by kevin on February 12, 2002 at 04:08:20:
Kevin,
My BtoB experience is in researching foodservice operators for manufacturers, and I have faced the problem you describe many times. Just a few thoughts on numbers.
First of all, sometimes numbers of respondents is not terribly relevant, particularly in BtoB. Let me illustrate. If you can get McDonald's to feature one of your products, it really doesn't matter what the rest of the world thinks about it. The ONE person (or few) who makes the decision for McDonald's is all that counts. This is not necessarily an uncommon occurence in BtoB.
Secondly, very small samples can often be very revealing, too. It depends on the variability in the population. If the population is absolutely homogenous, a sample of one is entirely adequate to represent the entire population.
You can think about this in this way, even for populations with some or a lot of variation: If you know nothing about the population, and you interview one person in great detail, you will learn a lot about the population. When you interview the second person, they will add to your store of information about the population, but the amount of new learning will ordinarily be a small fraction of what you learned on your initial foray. The third interview will extend and modify your learning from the first two, but, again, incremental gains in learning are diminishing.
Continuing this process, by the time you have interviewed 10 representative people, you will have had the opportunity to learn very nearly all that there is to know about this population.
The truth of what I have said here accounts, in a broad way, for the success of focus groups, and can be used in your thinking about BtoB.
What this does NOT adequately address is variability in the population. The determination of how many people you need to interview is driven by how large the variations in the population are, and how fine the distinctions you wish to make. For this reason, most of us prefer a cell size of 100 to 150 respondents. With "reasonable" sampling, at these numbers you are reaching the area of practically negligible statistical improvements with increasing sample size. But even a cursory examination of the statistics shows little increase in reliability in going from 50 to 150 respondents (even though costs may double or triple).
But there is an increased margin of comfort, PLUS the possibility of slicing and dicing the data into smaller chunks of 50 to 75 per cell in cross-tabs and banners.
And if you are willing to look at the results respondent by respondent, and consider variability and what is REALLY going on in the population, you may find value in cells of even 10 to 20 respondents.
Been there and done that in foodservice research (a highly fragmented BtoB business.)