R help - Oct 2006 - Goodness of fit measures for model with autocorrelation?

I wonder if anyone can help me with this query?

I am undertaking textual content analyses which result in time series
data (i.e., relative frequency counts of word categories in successive
segments of a larger text or in successive individual texts in a
temporal order).  In regard to the content category that interests me,
previous research has suggested an underlying five-stage psychological
process and has proposed a fifth-degree polynomial as a general model
for this in texts:

cat_freq ~ I(time) + I(time^2) + I(time^3) + I(time^4) + I(time^5)

I want to test the adequacy of this model on new data.

If I fit the model using lm, I get two measures of goodness of fit: RSq
and the significance of the ANOVA.  However, autocorrelation is normally
present in these data, so I should instead be using gls with an
appropriate autocorrelation process such as cor(AR1).  However, gls in R
does not offer the RSq and ANOVA measures.

Is there any way of getting these measures for a gls fit, or is there
perhaps even a better way of testing the theoretical model on new data?

Many thanks,

Andrew Wilson

Email: eiaaw at exchange.lancs.ac.uk

I am trying to generate a vector of random numbers with the constraint that 
they have to sum up to one with uniform distribution.

eg. {0.1,0.7,0.2 }

any function to do this? Thanks.