[R-lang] comparing two mixed-effects models

[T. Florian Jaeger]

>
> SS: What I have found is that X interacts with Y, and that interaction
> term is significantly contributing to the model's performance (i.e., the
> difference between fit1 and fit1A (where the interaction parameter is
> removed) is significant).  I took this to mean I cannot individually remove
> X from the model and compare the two fits; therefore, I concluded X is
> significant for fit1.  The same is true for fit2 (i.e. X2 interacts with
> Y).  Please let me know if you see a flaw in my conclusion.
>

that's all fine, but in order to see whether X1 and X2 matter both, you
still have to do the comparison i talked about (but include the interactions
for X1*Y and X2*Y in the model I called superfit in my previous email; i.e.
compare fit1A vs. superfitA and fit2A vs superfitA).

3. could I use the same formula to examine only one "Age" group (removing
> "Age" as a factor in the formula, of course), even if I am going to later
> proceed and re-examine the data for a larger young/old set?
>
>
> I am not sure that I understand what you mean. Nothing prevents you from
> exploring e.g. only data from "young" people, but be aware that whatever
> conclusions you draw out of the examination of that subset of your data, may
> not generalize to the entire sample (and hence not to the population
> represented by the entire sample).
>
> SS: the data shows an effect of "Age" (which has two levels "young" and
> "old"). I wanted to test a subset of the data that included only "young" and
> see whether there is an interaction between X and Y, for them as a subset.
> I was thinking that the result may be generalized to only "young"
> population, and not the entire sample.  Just want to make sure that I am not
> doing something that has problems that I am not aware of.  I hope I was able
> to state the question more clearly.
>

yes, this is ok.  be aware, however, that by splitting the data set you
loose power. so failure to detect the X*Y interaction MAY be due to lack of
power. an additional test you could confirm is a three-way interaction for
the entire data set X*Y*Age. If it is NON-significant that is a further
indication that the X*Y interaction holds for both age groups. finally, I
always advise visualization of the data set (as a way to compare the effects
of X*Y for young vs. old people).

is this useful? I am cc-ing the language list, so that other people can
correct me.

Florian

Thanks very much.
>
> -shabnam
>
>
>
>
>
> _______________________________________________
> R-lang mailing list
> R-lang at ling.ucsd.edu
> https://ling.ucsd.edu/mailman/listinfo.cgi/r-lang
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://ling.ucsd.edu/pipermail/r-lang/attachments/20070509/2b68b926/attachment.htm