[R] Johnson-Neyman-procedure in R

Leif Engqvist lengqvist at evolution.uni-bonn.de
Wed Nov 3 14:38:51 CET 2004


I was wondering if anyone could please help me with some simple 
questions regarding ANCOVA and the assumption of homogeneity of slopes.

The standard design of ANCOVA assumes the homogeneity of regression 
coefficients of the different groups. This assumption can be tested 
using the factor × covariate interaction, which should subsequently be 
removed. However if this assumption is not met the interaction term 
shouldnÂ’t be removed, but then the test for group differences only tests 
for differences in intercept. This is in many cases not what was 
initially intended.

Instead, what one wants is perhaps to determine values of the covariate 
at which the groups differ. IÂ’ve seen a description of the 
Johnson-Neyman procedure (in Huitema (1980)), which allows to determine 
the so called regions of nonsignificance, which sounds a lot like what I 
want. The problem is I have very seldom seen it used (at least in my 
field of work), but unequal slopes is a common problem. (Searching the 
R-help archives, however, didnÂ’t give me a single match.) My first 
question is therefore if the Johnson-Neyman procedure is a recommendable 
technique. My second question is then of course if somebody knows how to 
perform it in R (also for more complex models than a two-groups, one 
covariate model).

Hope someone can help me

Wit regards

<>Leif Engqvist

More information about the R-help mailing list