modeling time series data using Generalized Linear Model in
in [SAS]
|
Prev: modeling time series data using Generalized Linear Model in SAS
Next: Probit dose-response model with second, random main effect
From: Katy Seib on 8 Mar 2010 17:15 Ming, I don't know very much about R, but I have done glm for mv longitudinal data. This is a good book for theory: http://www.amazon.com/Applied-Regression-Analysis-Multivariable-Methods/dp/0495384968 I think you can see a good bit of it on google books. There are also a couple of good relevant chapters at the very end of Kleinbaum's other book "Logistic Regression" - you may be able to find pdfs of them online. If you need more thorough learning tools I recommend a subscription to Safari Books - http://my.safaribooksonline.com/ They have a ton of textbooks and programming books online. I have the really paired down subscription (plus they do a free trial period) and have yet to use up all my "bookshelf" space. I think it's about 24USD/month. UCLA has some great stuff - annotated code and output for stats - (SPSS, SAS and Stata) - that might be a good start. They also have some other online resources that might be helpful. http://www.ats.ucla.edu/stat/ Hope some of this helps. If you have more specific questions, I can try to help. Katy On Mon, Mar 8, 2010 at 4:45 PM, Ming Chen <chenming(a)gmail.com> wrote: > Hi All, > > Has anyone have used generalized linear models to model time series > data especially multivariate time series? I came upon a book on this > topic. > the book name is "Regression Models for Time Series Analysis" and you > can find it sold on amazon. > > There is no way that my company would buy SAS/ETS. Now I am on the > steep learning curve of modeling multivariate time series data using > R. The time series stuff itself is new for me and the lack of > documentation and different and confusing R library makes matter even > worse. Actually I focus more on model reference than the forecasting > accuracy. Maybe the regress analysis is a better options. Is there any > papers or examples I can read and learn? > > Thanks. > > Ming Chen >
From: Sigurd Hermansen on 8 Mar 2010 17:03 Ming Chen: I'd say that PROC MIXED PROC NLMIXED could be good procedures for use in analyzing "repeated measure" data; for example, observations of more than one event or response per subject. The fuzzy dividing line between repeated measures and time series would be related to the strength of influence and complexity of prior observations on future ones. Secular time series often have lags in responses due to trends, cycles, and seasonality. SAS/ETS programs decompose time series into components and help model complex lag structures. I'd also say that few on the 'L will respond unless you specify in more detail what you are trying to do and what data you have. S -----Original Message----- From: SAS(r) Discussion [mailto:SAS-L(a)LISTSERV.UGA.EDU] On Behalf Of Ming Chen Sent: Monday, March 08, 2010 4:46 PM To: SAS-L(a)LISTSERV.UGA.EDU Subject: modeling time series data using Generalized Linear Model in SAS Hi All, Has anyone have used generalized linear models to model time series data especially multivariate time series? I came upon a book on this topic. the book name is "Regression Models for Time Series Analysis" and you can find it sold on amazon. There is no way that my company would buy SAS/ETS. Now I am on the steep learning curve of modeling multivariate time series data using R. The time series stuff itself is new for me and the lack of documentation and different and confusing R library makes matter even worse. Actually I focus more on model reference than the forecasting accuracy. Maybe the regress analysis is a better options. Is there any papers or examples I can read and learn? Thanks. Ming Chen
From: Ming Chen on 9 Mar 2010 10:41
Thanks for the replies. I didn't make my self clear from the first post. According to the book "Regression Models for Time Series Analysis", it is possible to model time series using PROC NLIMIXED. For example, ARMA(1,1) model can be analyzed by PROC NLIMIXED. The book link at amazon is http://www.amazon.com/Regression-Models-Analysis-Probability-Statistics/dp/0471363553/ref=sr_1_1?ie=UTF8&s=books&qid=1268149029&sr=8-1 I just read the preface of that book from amazon and it seems generalized linear model can deal with time series data very well. I searched online and there are not much reference and examples about that. I am just wondering anyone on this list has read that book and tried real projects. Ming On Mon, Mar 8, 2010 at 5:03 PM, Sigurd Hermansen <HERMANS1(a)westat.com> wrote: > Ming Chen: > I'd say that PROC MIXED PROC NLMIXED could be good procedures for use in analyzing "repeated measure" data; for example, observations of more than one event or response per subject. The fuzzy dividing line between repeated measures and time series would be related to the strength of influence and complexity of prior observations on future ones. Secular time series often have lags in responses due to trends, cycles, and seasonality. SAS/ETS programs decompose time series into components and help model complex lag structures. I'd also say that few on the 'L will respond unless you specify in more detail what you are trying to do and what data you have. > S > > -----Original Message----- > From: SAS(r) Discussion [mailto:SAS-L(a)LISTSERV.UGA.EDU] On Behalf Of Ming Chen > Sent: Monday, March 08, 2010 4:46 PM > To: SAS-L(a)LISTSERV.UGA.EDU > Subject: modeling time series data using Generalized Linear Model in SAS > > Hi All, > > Has anyone have used generalized linear models to model time series > data especially multivariate time series? I came upon a book on this > topic. > the book name is "Regression Models for Time Series Analysis" and you > can find it sold on amazon. > > There is no way that my company would buy SAS/ETS. Now I am on the > steep learning curve of modeling multivariate time series data using > R. The time series stuff itself is new for me and the lack of > documentation and different and confusing R library makes matter even > worse. Actually I focus more on model reference than the forecasting > accuracy. Maybe the regress analysis is a better options. Is there any > papers or examples I can read and learn? > > Thanks. > > Ming Chen > |