Until a few weeks ago I too hadn't come across the Johnson-Neyman procedure, so I appreciate your comments. While I agree that it is no replacement for further investigation of what 'correct models' may be better suited to the data, I think it did prove useful for the application I was working on.
A key point in this case was that only a small number of outcomes being investigated didn't show HOS. The J-P procedure was used for those to try and keep a consistent framework for all analyses in the study. A literature search showed a small number of examples using J-P in recent years in the subject area. For the author these were pretty important points. Perhaps it can be considered a kind of descriptive analysis for those scenarios with homogeniety of slopes isn't met? I found a really useful reference to be Bradley E. Huitema's 'The Analysis of Covariance and Alternatives' (1980, Wiley) which discusses a number of alternatives to ANCOVA both from the perspective of choosing the most appropriate model and of what is available when assumptions are not met. Kylie. Rich Ulrich wrote: > On Tue, 23 Sep 2003 06:31:36 GMT, Kylie Lange > <[EMAIL PROTECTED]> wrote: > > > Further to Donald's comments about investigating the nature of the > > dependent-covariate interaction is the Johnson-Neyman technique for > > calculating the region of significance. This will tell you at what > > cut-points the regression slopes of your treatment groups change from being > > not significantly different to significant. > > "... at what cut-points ... change from being not > significantly different..." > > Don't you have a linear model? > That sounds pretty bogus to me... Off hand, > I don't think of good reasons for helping > the reader to think that an effect works in *lumps*. > > I see from google that the technique apparently dates > back 50 years, and it is used for showing where the > regression lines 'intersect' and are 'not different' in > terms of the CI on the intersection. > For something that old, it is not well known -- I did > not know that name or that it was legitimate until google > gave me those hits. > > From my brief glance, I don't think it is anything that > I will use or recommend. (Am I being unfair? is this > important to someone?) > > The inference drawn from the 95% CI on > the intersection-of-regression lines is *cute* but > I don't think you can read it that strongly, as a fair point. > Also, a point about the technicality: Does the technique > get applied *only* in the case of disordinal interactions, > or is it also used when the lines do not cross? > > - I think that one thing that affects me here is that I > tend, rather strongly, to regard 'interactions' as being > a failure to find the proper elements to model. That is, > if the definitions were right, we'd see main effects; > while the definitions are wrong, we should be rather > calm and quiet about our pronouncements. > > > > > This then allows you to put values on the regions that Donald described > > where group A > group B, group A < group B etc. > > > > There is SPSS syntax for the J-P technique available at > > http://support.spss.com/answernet/details.asp?ID=19193 which in turn was > > developed from SAS code (reference given). > > [ ... ] > > -- > Rich Ulrich, [EMAIL PROTECTED] > http://www.pitt.edu/~wpilib/index.html > "Taxes are the price we pay for civilization." . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
