[EMAIL PROTECTED] (nothanks) wrote in message news:<[EMAIL PROTECTED]>... > [EMAIL PROTECTED] (Penley, Julie) wrote in message >news:<[EMAIL PROTECTED]>... > > Mercy, I hope y'all aren't aiming this at the original poster. > Anyways, maybe part of the "problem" is that some fields think > they do need "fancy" statistics and models (you know. > "I NEED P-VALUES, GIMME P-VALUES", etc.) when, IMO, simple > summaries will suffice.
[snip] I recall serving as an external member on a Ph.D. committee. The student had a pretty well designed randomized experiment with only 2 comparison groups. The response variable was continuous and approximately normally distributed. When the student proposed using a t-test to compare means, one of the committee members said something to the effect of: "This is a Ph.D. dissertation, you can't use a t-test. You have to use something more sophisticated like ANOVA." Ouch. I cringed, bit my tongue, and tactifully defened the student's proposed method of analysis. A perceived need for statistical sophistication and lengthy results sections shouldn't drive research questions. Some important research questions can be correctly addressed using very simple methods of analysis. To refer back to the original question, if a 2 x 3 ANOVA is the appropriate analytical procedure, then it's the correct procedure. I do agree with those who suggested presenting results exploring the distributional properties of the response variable. The student could certainly present results of diagnostic statistics defending the choice of statistic. Depending on the nature of the research question, the student might test for interaction, test for simple main effects, and use appropriate procedures for multiple comparisons. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
