Removing the (random) intercept
Posted: Sat Aug 15, 2015 4:06 pm
Hi
I would like to have some feedback on some alternative parameterizations: one using an intercept and one omitting it. My question is a mixture of both Mlwin and R2mlwin.
My model: I want to estimate the level of economic orientation in journals during three eras. I have therefore constructed three dummies: Era_1890to1920 + Era_1921to1985 + Era_1986to2014, which cover all my observations.
The model I want to estimate (preferably in R2mlwin):
# removing the intercept
(Model1 <- runMLwiN(OrgSocThy~ Era_1890to1920 + Era_1921to1985 + Era_1986to2014 + (Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: gives This gives warning about random params has gone –ve definite. But it gives sensible numbers.
[see attached file for output]
But I have these two alternative parameterizations:
# Keeping the intercept as is with two dummies
(Model2 <- runMLwiN(OrgSocThy~ 1+ Era_1921to1985 + Era_1986to2014 + (1+ Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: This gives warning about random params has gone –ve definite
[see attached file for output]
# keeping the 1 in, 1+ Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal)
(Model3 <- runMLwiN(OrgSocThy~ Era_1 + Era_1921to1985 + Era_1986to2014 + (1+ Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: This gives warning about random params has gone –ve definite
[see attached file for output]
My question
1. I know that by removing the intercept one can directly estimate the values of the three eras. (dummies), but why is model 1 giving sensible numbers where as models 2 and 3 are not. They are just alternative parameterizations of the same model?
2. Moroever, in model 3, what does it mean when I keep the constant in (1 + Era_1986to2014|journal)? I am not entirely sure, but I assume that when I remove the intercept I should also remove the “1” in the randome part. Please, correct me if I am wrong?
Thanks in advance
Adel
I would like to have some feedback on some alternative parameterizations: one using an intercept and one omitting it. My question is a mixture of both Mlwin and R2mlwin.
My model: I want to estimate the level of economic orientation in journals during three eras. I have therefore constructed three dummies: Era_1890to1920 + Era_1921to1985 + Era_1986to2014, which cover all my observations.
The model I want to estimate (preferably in R2mlwin):
# removing the intercept
(Model1 <- runMLwiN(OrgSocThy~ Era_1890to1920 + Era_1921to1985 + Era_1986to2014 + (Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: gives This gives warning about random params has gone –ve definite. But it gives sensible numbers.
[see attached file for output]
But I have these two alternative parameterizations:
# Keeping the intercept as is with two dummies
(Model2 <- runMLwiN(OrgSocThy~ 1+ Era_1921to1985 + Era_1986to2014 + (1+ Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: This gives warning about random params has gone –ve definite
[see attached file for output]
# keeping the 1 in, 1+ Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal)
(Model3 <- runMLwiN(OrgSocThy~ Era_1 + Era_1921to1985 + Era_1986to2014 + (1+ Era_1890to1920 + Era_1921to1985 + Era_1986to2014|journal) + (1 | Article_ID), estoptions = list(EstM = 0, debugmode =F, resi.store=T, x64=F, optimat=T), data = df.eo.sorted))
# Result: This gives warning about random params has gone –ve definite
[see attached file for output]
My question
1. I know that by removing the intercept one can directly estimate the values of the three eras. (dummies), but why is model 1 giving sensible numbers where as models 2 and 3 are not. They are just alternative parameterizations of the same model?
2. Moroever, in model 3, what does it mean when I keep the constant in (1 + Era_1986to2014|journal)? I am not entirely sure, but I assume that when I remove the intercept I should also remove the “1” in the randome part. Please, correct me if I am wrong?
Thanks in advance
Adel