I have a design question that I'm actually surprised I haven't encountered before. I think I can describe its essence without getting into the background of the study, which may make things simpler to follow.
In a nutshell, I'm running a 2 x 2 x 2 between-subjects design in which the manipulations are contained in a one-page scenario that participants read and respond to. Because the study is so short, and because it is a pain at a small liberal arts school to get 128 subjects (I'm shooting for 16 per cell), it occurred to me that I could have each student be in 2 of the 8 conditions in order to reduce by half the number of participants required.
It actually works out so that if a participant could get the other level of each of the three variables, they would not be overly sensitized to what the manipulations are about. So, for example, the participants who get the combination of variable 1/level A, variable 2/b, and variable 3/a, could also read a second scenario that is complementary to it, namely variable 1, level b, variable 2/a, and variable 3/b.
The scheme works out so that each 1/4 of the participants sees two profiles as follows (with half of each seeing the two profiles in reverse order):
1A, 2A, 3A --> 1B, 2B, 3B
1A, 2B, 3A --> 1B, 2A, 3B
1B, 2B, 3A --> 1A, 2A, 3B
1B, 2A, 3A --> 1A, 2B, 3B
My questions are as follows:
1) What kind of design is this? How would I describe it? It's not really completely between-subjects and not really within-subjects either. The within-subjects element is really by condition rather than variable.
2) How would/could I analyze this? Is there any way to take advantage of the shared variance of having the same subjects be in complementary conditions? If not, is it wrong to analyze this as a completely between-subjects design (essentially treating each participant's responses to the second scenario as if it came from a different person)? I'm not well versed in all the issues/assumptions here about how to partition the variance.
3) Given these issues, would you recommend that I just simplify things by running this as a completely between-subjects design? Typically, the benefits of a within-subjects design are a) more power and b) fewer subjects. I seem to be getting b but not a here, and I suppose I run a slight risk of sensitizing subjects. Given the analysis problem, would it be more prudent to just run twice as many subjects here and not deal with this mess? :-)
Thanks in advance to all the statistical experts who are much more clever than I am :-)
Cheers, Traci
--
\\|||// ( o o ) -------------o00-(_)-00o------------------
Traci A. Giuliano Associate Professor of Psychology Southwestern University Georgetown, TX 78627 [EMAIL PROTECTED] (512) 863-1596;fax (512) 863-1846 http://www.southwestern.edu/~giuliant
--------------------------------------------
--- You are currently subscribed to tips as: [EMAIL PROTECTED] To unsubscribe send a blank email to [EMAIL PROTECTED]
