lme( . ) question

["Paul, David A" <paulda@BATTELLE.ORG>]

I am aware that least squares estimates of regression coefficients
in autoregressive models tend to be downward biased, particularly
for small samples [Hamilton, "Time Series Analysis", 1994].

I would like to include the lag(1) response in my set of predictors
within lme( ) for the simple reason that I am trying to put as much
structure on the "trend" as possible, and as little structure on the
residuals as possible.

Is this a reasonable thing for me to do?  Does including the lag(1)
response cause significant biases to appear in the estimated fixed
and random effects or otherwise affect the estimated confidence
intervals?  I know that lme( ) performs its estimation using ML or
REML, and is therefore very different from least squares estimation. 

Much thanks in advance,
  Respectfully,

  David Paul, Ph.D.
    Battelle Memorial Institute
    505 King Avenue
    Columbus, OH  43201
    614.424.3176