I use the fitlme function in Matlab with 'DummyVarCoding' set to 'effects', which is defined as 'Coefficients sum to 0.'
The lme object lists the coefficients for each level of the fixed factors and tests the hypothesis that they are equal to 0. However, this is done when one of the levels of each fixed factor is used as a reference. What does it say about the coefficient of the reference level? is it 0? Other value?
I am asking because I want to know whether I can interpret the p values given by the model to indicate whether the effect of level X of the fixed factor is significantly different from the effect of the reference level.
Thanks,
Chen
Thank you. This makes a lot of sense.
How do I extend this to a multi-factor model? I have a model that includes several factors (A,B,...) each having several levels (A1,A2,B1,B2...), and also the interactions between these factors. In this case I assume the intercept is for the complete model, but there is still a separate dummy reference level for each factor (and interaction of factors).
So, if A1 is set as a reference level for factor A can I interpret the t and p values reported for A2 as indicating whether A2-A1 is significantly different from zero? (regardless of factors B,C, etc.)