Dear Statalist users,
I am estimating a --350 students nested in 30 small schools--.
DV:
My DV is a dichotomous variable indicating whether students pass compulsory secondary education (PCSE) (=1), otherwise (=0).
IVs:
At the individual level, my main independent variable is "Social class--family" (SCF): a categorical variable with four possible values (1= upper class/upper middle-class family, 2= new middle class family, 3= lower class family, 4= unskilled workers)
At the school level, I am investigating the effect of the following variables:
"Public character of schools" (PCS): a categorical value with three possible values (1=Public School, 2= Private, 3=Mixed model)
"Budget of the school" (SB): a continuos variable ranging from 0 to 1.
"Presence of integration policies of immigrants within schools" (IPI): a dichotomous variable indicating presence of such policy (=1), otherwise (=0).
Since I am including a sample of students, I am using population size weights is to compensate the under-representation of certain groups of students ( I am examining all the schools in a small town, but I am including a sample of students belonging to different social classes). For this purpose, I am using the following syntax:
xi: gllamm PCSE i.SCF i.PCS SB i.IPI, i(SCHOOLNUM) link(logit) family(binom) nip(30) pweight(pw) adapt
My main problem is that the weighted model versus the non-weighted model leads to different results, while in the weighted model three variables are significant; in the unweighted model no variable has a significant effect. With the aim to correct this, I would like to use bootstrap to obtain standard errors, but I have not found the way to do it.
Can anyone help me with this?
Thank you very much in advance for your help!!
William
I am estimating a --350 students nested in 30 small schools--.
DV:
My DV is a dichotomous variable indicating whether students pass compulsory secondary education (PCSE) (=1), otherwise (=0).
IVs:
At the individual level, my main independent variable is "Social class--family" (SCF): a categorical variable with four possible values (1= upper class/upper middle-class family, 2= new middle class family, 3= lower class family, 4= unskilled workers)
At the school level, I am investigating the effect of the following variables:
"Public character of schools" (PCS): a categorical value with three possible values (1=Public School, 2= Private, 3=Mixed model)
"Budget of the school" (SB): a continuos variable ranging from 0 to 1.
"Presence of integration policies of immigrants within schools" (IPI): a dichotomous variable indicating presence of such policy (=1), otherwise (=0).
Since I am including a sample of students, I am using population size weights is to compensate the under-representation of certain groups of students ( I am examining all the schools in a small town, but I am including a sample of students belonging to different social classes). For this purpose, I am using the following syntax:
xi: gllamm PCSE i.SCF i.PCS SB i.IPI, i(SCHOOLNUM) link(logit) family(binom) nip(30) pweight(pw) adapt
My main problem is that the weighted model versus the non-weighted model leads to different results, while in the weighted model three variables are significant; in the unweighted model no variable has a significant effect. With the aim to correct this, I would like to use bootstrap to obtain standard errors, but I have not found the way to do it.
Can anyone help me with this?
Thank you very much in advance for your help!!
William
