Dear alle,
I have an lme model with four independent variables. There is a significant interaction between two of the variables (context and group). Each of the variables has 3 levels. My question relates to the reading of the Value and p-value in the model: Does the main effect represent an average of all the values for all the subjects at that particular level (for exemple context N in the first model below) as compared to the average of all the values for all the subjects at the reference level (context F that is included in intercept)? Or does it show the values for the third group (gr.av) that doesn't figure in the interaction between group and context? How do I interpret the interaction between context and group? Only two of three contexts and two of three groups are shown in the interaction in the model.
Value Std.Error DF t-value p-value
(Intercept) 7.048984 0.03735696 2308 188.6926 <.0001
contextN 0.022821 0.00891043 2308 2.5612 0.0105
contextNN 0.122887 0.01707241 2308 7.1980 <.0001
målf.type -0.055151 0.02529478 2308 -2.1803 0.0293
gr.iav 0.056431 0.03122954 2308 1.8070 0.0709
gr.ns -0.156082 0.02279682 2308 -6.8467 <.0001
position -0.080792 0.02467325 2308 -3.2745 0.0011
contextN:gr.iav -0.024463 0.01049197 2308 -2.3316 0.0198
contextNN:gr.iav -0.020950 0.00739759 2308 -2.8320 0.0047
contextN:gr.ns 0.015571 0.00651959 2308 2.3883 0.0170
contextNN:gr.ns -0.012505 0.00412531 2308 -3.0312 0.0025
position:m.lf.type 0.051293 0.02468307 2308 2.0781 0.0378
In the second model below I have tried to leave out one of the main effects (context). This gives me the interaction between all three groups and two of the contexts. I suppose that for each group the value of each of the contexts is given as compared to the value of the third context (contextF) for that same group and that this third context is included in the intercept?
Value Std.Error DF t-value p-value
(Intercept) 7.048984 0.03735697 2308 188.6926 <.0001
m.lf.type -0.055151 0.02529478 2308 -2.1803 0.0293
gr.iav 0.056431 0.03122954 2308 1.8070 0.0709
gr.ns -0.156082 0.02279682 2308 -6.8467 <.0001
position -0.080792 0.02467325 2308 -3.2745 0.0011
gr.av:contextN 0.031713 0.01299862 2308 2.4398 0.0148
gr.iav:contextN -0.017213 0.01649266 2308 -1.0437 0.2968
gr.ns:contextN 0.053963 0.01650482 2308 3.2695 0.0011
gr.av:contextNN 0.156342 0.01855375 2308 8.4264 <.0001
gr.iav:contextNN 0.114442 0.01978861 2308 5.7832 <.0001
gr.ns:contextNN 0.097878 0.01871174 2308 5.2308 <.0001
position:m.lf.type 0.051293 0.02468307 2308 2.0781 0.0378
Does it mean anything whether the variable is a between or within subjects variable?
I suppose a logistic regression model would have to be read the same way as the lme model?
Is there a better way to get S+ to show the interaction for each level of the variables?
Any help would be appreciated. Thank you.
Lise Hedevang
Ph.d.-studerende
Institut for Sprog, Litteratur og Kultur, Fransk
Aarhus Universitet
Jens Chr. Skous Vej 5, bygn. 1461, 425
DK - 8000 Århus C
(+ 45) 89 42 64 34
romlh@hum.au.dk
http://person.au.dk/romlh@hum
|
|