st: The difference of -FE, -BE, -RE and -PA in xt- models

[Young Hee Rho <mania1@korea.ac.kr>]

I am currently studing longitudinal models and found that there are many estimating options
in the xt- series. Each of the commands provides -FE (fixed effects) -BE (between effects)
-RE (random effects) and -PA (population averaged).
I have bought two books from Stata Press, Multilevel and Longitudinal Modeling Using Stata
(Rabe-Hesketh et al) and Applied Longitudinal Data Analysis for Epidemiology (Twisk).
I also have read the FAQs in Stata.com about this problem. Still I am not clear about this topic.
What I conceptually know is that -FE is the effects of when an individual changes 1 unit
,how much it changes the response variable. -BE is the effects of when an individual
differs from another individual in 1 unit, how much it changes the response variable.
-RE is essentialy -FE + -BE (it is described as a weighted average). -PA is similar with
-RE, in this case it is the effect of an "average" individual on the "average" response
variable, whereas the -RE is on a "same" individual. This should make the -FE
and -BE not population averaged, but also like -RE, express effects on a "same" individual.
Does what I am thinking is right? It does not match with the FAQs on Stata.com, however
it seems right to me..
** This also may be a duplicate (however it seems that the previous was cut due to unwanted HTML
coding)