How can I choose mean for centering age in growth model?

[Gayoung Choi]

Hi.
I use growth model for examining the relationship between discrimination and mental health among older adults.

In my model, I used centering age(grand mean centering). However, I have an issue about centering age.

Here are my code.

*age centered
summarize age, meanonly
gen c_age = age - r(mean)

I think upper code is very common code for centering age. But this r(mean) refers the grand mean of the data set.
So I think I have to use the mean of sample included my model.
For example,

mixed cesd ($indi $health i.infm_v2 i.earlyret##i.fm_v4)##c.c_age || pid: c_age, cov(un)
gen sample==e(sample)

sum age if sample ==1
*average mean of age in sample is 65

gen c_age = age - 65

However, if I use the sample mean as an grand mean, the coefficient and significance is differed.
So my question is...

Can I use the grand mean of raw data set? or Should I use the grand mean of sample (only included model)?


If you have an idea about solution, Please share your thoughts.
Thank you for reading my text!

Best,
Gayoung

[Carlo Lazzaro]

Gayoung:
I guess you're centering age for interaction purposes.
In this respect, I would recommend you to take a look at Richard Williams 's teaching notes on this topic (Interaction effects and group comparisons).
Personally, I would try both the approaches and compare the resulting coefficients.
As a second thought, I would advise you to follow the "tribal rules" on force in your research field.

[Maarten Buis]

You can use any value you want. Some coefficients will differ, because they mean something different, though the underlying model remains unchanged. As long as you interpret your coefficients correctly, it does not matter that the coefficients are different; the meaning will be the exact same.

I usually don't use the mean (grand or otherwise), but a nice round number within the range of the data. If I compare my results with that from somebody else using a different dataset, then the means will differ, but say 60 will be 60.

[Gayoung Choi]

Dear Carlo Lazzaro and Maarten Buis,

Thank you for answering my post!
I have an insight about my problem from your answering.
According to Maarten Buis, I can choose the mean that I need!

Thank you!
Gayoung