multilevel or logistic regression ???

[Ellen-18]

Dear All,

I have a dataset with 13 000 students in adult education.  I divided the
learners (on the basis of their personal characteristics) in typical and
atypical learners.  Now, I want to compare these 2 groups.

I sampled institutions and within those institutions, students completed a
questionnaire. For each institution, around 10 students compeleted the
questionnaire.  I created codes so I can detect which student belongs to
which school.

Now I want to know how students perceive their learning process (there was
a Lickert scale with 15 items about this in the questionnaire) and look at
the characteristics of the learning institutions.  I want to know if the
typical versus atypical learners perceive their learning process in
another way and link these perceptions with the characteristics of the
learning institutions in which they are enrolled.  With other words : does
the educational institution matters in perceiving your learning process ?

My question is : which technique is most adequate for this?  On the one
hand, I think it is best to use a logistic regression so that I can see
which characteristics belong more to typical learners and which to
atypical learners (0 versus 1).  On the other hand, my students are nested
in schools and this is the basis of multilevel analysis ?

Can you give me your opinion?

Many thanks!
Ellen

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Hector Maletta]

Ellen,
Yours is a false dilemma.
Multilevel models are defined by the fact that coefficients at lower levels
depend on variables at higher levels. They are often linear, but they can be
non linear (e.g. logistic). Logistic regression is one particular functional
form for a relation between a dependent variable (which is a binary variable
in ordinary logistic regression, or may be a categorical or ordinal variable
in more complicated logistic models) and several predictor variables. In
theory, you may have a multilevel model when the functions are logistic. I
do not know which software you may use for that, but my point here is that
the horns of your dilemma pertain to different orders: one is about the
hierarchical or non-hierarchical nature of your model, the other is about
the mathematical form of the prediction function.

Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Ellen
Sent: 22 April 2008 08:40
To: [hidden email]
Subject: multilevel or logistic regression ???

Dear All,

I have a dataset with 13 000 students in adult education.  I divided the
learners (on the basis of their personal characteristics) in typical and
atypical learners.  Now, I want to compare these 2 groups.

I sampled institutions and within those institutions, students completed a
questionnaire. For each institution, around 10 students compeleted the
questionnaire.  I created codes so I can detect which student belongs to
which school.

Now I want to know how students perceive their learning process (there was
a Lickert scale with 15 items about this in the questionnaire) and look at
the characteristics of the learning institutions.  I want to know if the
typical versus atypical learners perceive their learning process in
another way and link these perceptions with the characteristics of the
learning institutions in which they are enrolled.  With other words : does
the educational institution matters in perceiving your learning process ?

My question is : which technique is most adequate for this?  On the one
hand, I think it is best to use a logistic regression so that I can see
which characteristics belong more to typical learners and which to
atypical learners (0 versus 1).  On the other hand, my students are nested
in schools and this is the basis of multilevel analysis ?

Can you give me your opinion?

Many thanks!
Ellen

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD