mean centering & variables scaling

[HyunJoo LEE]

hello. I am planning to use a mixed model when writing my paper. I have three questions:

1. When controlling covariates such as age, years of education, and gender, it is said that gender is treated as a dummy variable because it is nominal. What about age or years of education? Can I check centered?

2. When the level 2 independent variable is a group, mean centering cannot be performed, so can it be treated as a dummy variable in this case as well?

3. When I want to add covariates, is it correct to put the nominal variables in factors and the rest in covariates?

thank you for your help
have a good day.

[mcfanda@gmail.com]

1. sure, that is the default. With mixed model is always a good idea to center the covariates (although not necessary)
2. any factor is transformed in contrast variables and their coding is mean-centered (if do not select "dummy" as a coding system)
3. yes

[ertugruluyar]

1. sure, that is the default. With mixed model is always a good idea to center the covariates (although not necessary)
2. any factor is transformed in contrast variables and their coding is mean-centered (if do not select "dummy" as a coding system)
3. yes

Hi, I have this variable with coding "simple" and I am doing linear mixed model analysis, this variable is mean-centered now? If I change it to "dummy", does the result of my analysis change?

The variable

aaaaaaaaa.png (23.16 KiB) Viewed 664 times

[ertugruluyar]

I am confused right now and I didn't quite get this explanation:
https://gamlj.github.io/mixed.html#factors-coding

[mcfanda@gmail.com]

Yes, that' correct output.

[mcfanda@gmail.com]

Please have a look here https://gamlj.github.io/book/glm.html#anova, and here https://gamlj.github.io/rosetta_contrasts.html

[ertugruluyar]

Please have a look here https://gamlj.github.io/book/glm.html#anova, and here https://gamlj.github.io/rosetta_contrasts.html

Thank you so much, I get a bit more now.
For other than the interactions, simple and dummy doing the same thing because "centering around zero = no centering".
But, when doing interactions, (1) if "simple" coding is used for variable called A (and let say the other variable is "B"), in A*B interaction effect:
B's effect on A (meaning B's indirect effect on dependant variable) is calculated using the mean of the B.the mean of the B
(2) if "dummy" coding is used for variable called A (and let say the other variable is "B"), in A*B interaction effect:
B's effect on A (meaning B's indirect effect on dependant variable) is calculated using reference contrast variable (reference dummy variable/ the first one for GAMLj3) of the B.
Am I thinking correctly?

But I couldn't guessing/forecasting that how can I interpret this two situation.

And, also, thank you for the book. Actually, I was found that book but It slipped my mind.

[mcfanda@gmail.com]

Great

[ertugruluyar]

Great

Thank you again and again, have a great day.