s-news
[Top] [All Lists]

interaction in lme

To: <s-news@lists.biostat.wustl.edu>
Subject: interaction in lme
From: "Puschner, Bernd" <Bernd.Puschner@bkh-guenzburg.de>
Date: Tue, 22 May 2007 11:08:55 +0200
Dear all,

I have a lme model with in.studm = time since beginning of observation,
and sc10m = psychological impairment.

I want to know the interaction of phase1f (factor 0,1) and anf.kosm
(continuous), but not the main effects of theses variables.

When I run

a.lme <- lme(data = alle.pt, random = ~in.studm | code, fixed = sc10m ~
phase1f * anf.kosm, na.action = na.omit) 

I get

Fixed effects: sc10m ~ phase1f * anf.kosm 
                      Value  Std.Error   DF   t-value p-value 
     (Intercept)  0.8930910 0.03638889 1966  24.54296  <.0001
         phase1f -0.1263110 0.01838107 1966  -6.87180  <.0001
        anf.kosm  0.0066864 0.00672424  604   0.99437  0.3204
phase1f:anf.kosm  0.0043765 0.00336473 1966   1.30069  0.1935

but when I run

b.lme <- lme(data = alle.pt, random = ~in.studm | code, fixed = sc10m ~
phase1f : anf.kosm, na.action = na.omit) 

I get

Fixed effects: sc10m ~ phase1f:anf.kosm 
                      Value  Std.Error   DF   t-value p-value 
     (Intercept)  0.8333118 0.03527006 1966  23.62660  <.0001
phase1f0anf.kosm  0.0279751 0.00703025 1966   3.97925  0.0001
phase1f1anf.kosm -0.0009885 0.00684771 1966  -0.14435  0.8852


i.e. two coefficients for phase1f : anf.kosm, but phase1f (defined as
factor) has only two levels (0 and 1). This also happens when I remove
the main effects from a.lme via the "update" function. It does not
happen for an interaction among two continuous variables.

I do not understand why this occurs/where the mistake is. I would be
grateful for any comments/ideas.

Best wishes

Bernd
<Prev in Thread] Current Thread [Next in Thread>
  • interaction in lme, Puschner, Bernd <=