interaction effect?

noetsi

Fortran must die
#25
I have to run a federally mandated model, that is with 41 mandated variables not counting any interactions. I think a MLM is the ideal, but since I have not done a MLM in a long time I will stick with the interaction at first. MLM is not going to run with 41 control variables I am pretty sure, but I can try.

I am not sure I am sophisticated enough to do a simulation :p which would have to be in R because we don't have PROC IML. I am a long way from knowing the R code (I use it for basic stuff, I mean basic coding not basic stats).
 
#26
I have to run a federally mandated model, that is with 41 mandated variables
I find this fascinating. A federal authority ha mandated a special model. [@Noetsi is talking about a state in the USA of course.] Can such a system work? That is the model. But what about the estimation method? Does it have to be OLS? Or could you use LASSO? Or anything else?

MLM is not going to run with 41 control variables I am pretty sure, but I can try.
Why would it not?
In a fixed model you have:

Y = X*beta+ eps

But in a mixed model:
Y = X*beta + Z*v + eps

If you just add a few variables (the Z*v part) I guess that it would be possible to estimate that. Agree?
 

noetsi

Fortran must die
#27
Because the method commonly won't work with that many. The system does not converge. This is a software issue, but one generic to all MLM
 

noetsi

Fortran must die
#28
Getting to details (this is not from my project which I am just starting)

loss′=7.8–9.4hours–.08effort+.39hours∗effort

How do you interpret the .39 slope for the interaction term hours*effort. What is it actually telling you. The impact of hours goes up with each unit of increased effort (by .39)?

This is about simple effects. It is often recommended when you have interaction.

In the equation from above they test the simple effect of hours at level 2 and effort at level 30. They get a value of about 10 for the simple effect. Does that mean the value of the DV goes up by 10 when you move from 2 hours one unit? I don't really understand what simple effect slopes show.

Unlike regular regression I don't think simple effects show the impact of a one unit change

The set of regression coefficients are multiplied by their corresponding values, and the resulting products are summed to form the linear combination. As an example, we will show how to estimate the predicted loss for a subject who averages 2 hours of exercise per week at an effort level of 30.