Date: Thu, 6 Mar 1997 08:21:13 -0600
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
Organization: Deja News Usenet Posting Service
Subject: Re: Help with PROC MIXED
In article <email@example.com>,
firstname.lastname@example.org (David Scott) wrote:
> I am trying to model some data on the amount of fodder produced at a number
sites on a number of farms
> over some years. I have monthly figures for about 3 years. I have some fixed
effects that are
> straightforward, but include an effect for month. I am having difficulty with
getting the right variation
> structure into PROC MIXED.
> What I would like is that for the same farm I have a correlation between the
fodder production for different
> sites at the same time, and that there is also an autoregressive correlation.
So far I am only modelling one
As I understand your problem, you have 36 months of data, probably with
seasonal and short term autocorrelation , perhaps being influenced by
level shifts or time trends or unusual values.
This data exists for S sites and for F FARMS.
The solution to your problem requires the tools of Pooled Cross-Sectional
Time Series where you have two distinct sections (site and farm) and time
with 36 consecutive equi-spaced readings. I am not aware of any pre-canned
utility to solve this issue and you are very wise to question the simple
procedures you mentioned.
AUTOBOX (http://darkstar.icdc.com/~autobox) has been programmed to allow
estimation for one section but not two. I believe that multiple runs
could be done to estimate the various elements of the ANOVA table. The
problem that you have is identifying and incorporating the form of the
autocorrelation for each site/farm combination. AUTOBOX can help you
perform this task but I think that this problem requires serious
attention from a trained statistician comfortable with modern time series
methods (Transfer Functions,Intervention Detection,etc.). Automatic
Forecasting Systems (AFS) is a possible source for this kind of
Your problem can't be answered simply with off-the-cuff analysis but I
believe that we have a tool that could be useful. I don't believe that
any other software or statistical programming language can adequately
deal with your problem.
Please feel free to contact me by phone to discuss this further.
David P. Reilly
Automatic Forecasting Systems Inc.
-------------------==== Posted via Deja News ====-----------------------
http://www.dejanews.com/ Search, Read, Post to Usenet