R help - Apr 2004 - regression and dw

Dear R People:

Suppose we have a regression model that we will call
y.lm

We run the Durbin Watson test for autocorrelation
and we find that there is positive autocorrelation,
and phi = 0.72, say.

What is our next step, please?  

Do we calculate the following
yprime_t = y_t - 0.72y_t-1,
x1prime_t = x1_t - 0.72x1_t-1,

and so on, and re-fit the linear mode?

I haven't done this in a while.  Just wanted to double check.

Thanks so much!
Erin Hodgess
Associate Professor
Department of Computer and Mathematical Sciences
University of Houston - Downtown
mailto: hodgess at gator.uhd.edu

On Fri, 16 Apr 2004, Erin Hodgess wrote:
> Dear R People:
> 
> Suppose we have a regression model that we will call
> y.lm
> 
> We run the Durbin Watson test for autocorrelation
> and we find that there is positive autocorrelation,
> and phi = 0.72, say.
> 
> What is our next step, please?  
Look at the residuals more closely, e.g. look at the acf.
> Do we calculate the following
> yprime_t = y_t - 0.72y_t-1,
> x1prime_t = x1_t - 0.72x1_t-1,
> 
> and so on, and re-fit the linear mode?
Better to use arima with AR residuals and an xreg matrix.  

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

> > Dear R People:
> > 
> > Suppose we have a regression model that we will call
> > y.lm
> > 
> > We run the Durbin Watson test for autocorrelation
> > and we find that there is positive autocorrelation,
> > and phi = 0.72, say.
> > 
> > What is our next step, please?  
> 
> Look at the residuals more closely, e.g. look at the acf.
> 
> > Do we calculate the following
> > yprime_t = y_t - 0.72y_t-1,
> > x1prime_t = x1_t - 0.72x1_t-1,
> > 
> > and so on, and re-fit the linear mode?
Hello Erin,

beside the points mentioned by Prof. Ripley, you might also want to consider
test for order of integration of y and x and if cointegration exists between
these variables. A high R2 and a low DW is often a hindsight for a spurious
relationship, which needs to be investigated further. There is the
contributed package 'urca' available. Incidentally, a package update
will be
put on CRAN shortly. If you want to receive it now, pls. contact me offline
and name your OS (i.e. zip or tar.gz).

HTH,
Bernhard  


> 
> Better to use arima with AR residuals and an xreg matrix.  
> 
> -- 
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
http://www.R-project.org/posting-guide.html


--------------------------------------------------------------------------------
The information contained herein is confidential and is inte...{{dropped}}