Date:         Thu, 18 Dec 2008 21:42:54 -0800
Reply-To:     stringplayer_2@yahoo.com
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Dale McLerran <stringplayer_2@YAHOO.COM>
Subject:      Re: GENMOD exponentiated linear prediction values
In-Reply-To:  <20081218200547.f3iaecjh2cgokwog@mail.ucla.edu>
Content-Type: text/plain; charset=us-ascii

--- On Thu, 12/18/08, jasonm@ucla.edu <jasonm@ucla.edu> wrote:

> From: jasonm@ucla.edu <jasonm@ucla.edu>
> Subject: Re: GENMOD exponentiated linear prediction values
> To: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
> Cc: "Dale McLerran" <stringplayer_2@YAHOO.COM>
> Date: Thursday, December 18, 2008, 8:05 PM
> I'm interested in three years (ie three adjusted rate
> estimates). What value do I use for "<mean agef
> value>"?  Is "<mean agef value>" the
> mean of age as a categorical or continuous variable for each
> year?
>
>     estimate "Year 2005 exponentiated mean"
>     intercept  1
>     year       1 0 0
>     agef       <mean agef value>
>     expf       0.25 0.25 0.25 / exp;

Jason,

Oops, my mistake.  I didn't notice agef on the list of class
variables.  Let's start over.  Please describe your data
including how many levels there are for agef and expf.  Also,
in your first e-mail, you indicated that you wanted predicted
values for each "level of year".  When I saw that, I thought
that you would have year on your class statement, but you do
not.  That means that you believe that the log mean injury
count is linear in year.  Is that really reasonable?  Why
do you not include year in your class statement?

I would be interested, too, in how many levels of year you
have in your data.  I am guessing that you have only three
years worth of data.  If that is true, then you really
don't have enough years of data to examine whether the
relationship between log mean injury count is or is not
linear in year.  You really should place year on your CLASS
statement, I believe.

One more question.  Do you believe that it is reasonable to
assume a balanced distribution of agef and expf.  If you
do assume a balanced distribution for those variables and
the reality is that the data are far from balanced, then you
could produce "mean" values far from the observed mean value.

Dale

---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph:  (206) 667-2926
Fax: (206) 667-5977
---------------------------------------