Date:         Fri, 13 Jul 2007 19:14:51 +0200
Reply-To:     nicola.baldini2@UNIBO.IT
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Nicola Baldini <nicola.baldini2@UNIBO.IT>
Subject:      Re: predict organiz. behavior from a sample of individual opinions
Comments: To: "Zack, Matthew M. (CDC/CCHP/NCCDPHP)" <mmz1@CDC.GOV>
In-Reply-To:  <392FF8243BA9634084F5AC5EF07B5CDF01DF0421@LTA3VS002.ees.hhs .gov>
Content-Type: text/plain; charset="us-ascii"

Many thanks for your suggestions. Singer's paper provided a wonderful introduction, both theoretical and practical, to multilevel analysis. However, it left me with one doubt.
In Singer's paper, as well as in many examples which I found on the web using SAS or Stata, the student-level outcome is predicted using student-level and school-level covariates.
Is it insane to ask SAS, e.g. to predict school-level number of students passing the math exam (or their percentage on the number of enrolled students) using information about gender and socio-economic status?
If no, can I just use a school-level y or should I modify the usual syntax (and how)?

Nicola

At 08.15 11/06/2007 -0400, you wrote:
>Organizational-level responses that are not simple aggregations (sums,
>means) of individual-level responses would include responses
>within or particular to the organization (for example, departments
>within a university) like the following:
> 1. kind of decision-making:  hierarchical, consensus, anarchical, etc.;
> 2. presence/absence of a program for graduate students and
>post-doctoral fellows;
> 3. collaboration with commercial interests;
> 4. etc.
>
>You can perform multilevel modelling using PROC MIXED (cf., attachment),
>PROC GLIMMIX, or possibly PROC NLMIXED.  You can
>specify separate random effects for all and only members of a specific
...