[Home Page] [Previous Page]
[Old Roundtable Archives] [New Roundtable]
Posted by Herb Sorensen on October 12, 1999 at 17:53:01:
In Reply to: "Rotation" => Orthogonal design posted by John Taylor on October 12, 1999 at 01:48:00:
I would be interested in this type of software also, but I doubt that it exists. From a practical point of view, we usually have negotiated a specific number of interviews with a client (N) and this may or may not be adequate to address all possible rotations.
In some cases the number of theoretical rotations may significantly exceed the total number of interviews (obviously we are not trying to analyze data by segment for each rotation, but are simply trying to assure that all are represented in the test).
For many years I have constructed rotation plans by hand using the basic principles of combinatorial math, and judgment as to the relevance of various rotations. When I do this, at least I know how every rotation is represented, and if their are distortions I am in a position to judge their relevance to the analysis.
Let me give you a simple example that can be handled in another way: Suppose we want 150 people to rank order 27 items. There are very many more combinations and permutations of 27 items taken 27 at a time than 150.
One practical way out is to put the 27 items (or surrogate identifiers) on 27 cards and shuffle them before each interview. Then present the items in the order that comes up with the shuffling.
This results in a "random" order, with presumably 150 different rotations. No one ordinarily asks how representative these rotations are of the multitude that could be mathematically generated.
As a matter of practice, many researchers do not randomize at all, but simply use a fixed list of the 27 items, and then start at a different point (increase by 1) each interview. This ignores relational effects but is judged adequate by many researchers.
Good luck in your search. If you find anything, I'd like to know.
Subject: