|Date: ||Fri, 26 Aug 2011 13:38:51 -0400|
|Reply-To: ||Sophia Tong <sophiDT@HOTMAIL.COM>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Sophia Tong <sophiDT@HOTMAIL.COM>|
|Subject: ||time series modeling|
I am new to time series modeling, got a time series project and found the
challenge and hope to get help from the list.
I have a 60 months time series data, with dependent variable as monthly
prescription drug use rate, and independent variable is month. I am trying
to model if certain drug policy change alters the drug prescription rate.
Here is the model I am trying to use based on self study:
PROC AUTOREG DATA=Policy;
MODEL DRUGRATE=MONTH PA PST_PA/METHOD=ML NLAG=3;
TEST MONTH + PST_PA=0;
Here PA is the policy indicator: 0= before policy period, 1=after policy
PST_PA is the months after policy implemented coded as 1,2,3...
The challenge to me is how to choose appropriate nlag= and method=. It
changes estimates greatly.
according to SAS doc, it seemed nlag=13 and method=yw (defualt) is
appropriate, but the publications I can find use nlag=1 and method=ml.
Is there a rule of thumb? What are the important numbers (ie. MSE, AIC,
total-R square, or something else?) to look as to make judgement of best
Any input will be greatly appreciated.