[R] gls function, very old results

[Bill Oliver]

----- Original Message -----
From: "Raphael Gottardo" <raph at alvie-mail.lanl.gov>
To: <r-help at stat.math.ethz.ch>
Sent: Tuesday, July 10, 2001 11:50 AM
Subject: [R] gls function, very old results


> Hello R-users,
>
> I am currently trying to learn how to use the function gls of the nlme
> library. I fitted the following model:
> Generalized least squares fit by REML
> Model: response ~ array + dye + genes + variety + variety * genes +
> array * genes + dye * genes
> Data: data
>
> I have 11 arrays, 2 dyes, 2 varieties, 3200 genes, and 2 replications
> for each.
> Therefore I should have the corresponding degrees of freedom and number
> of coefficients, but instead I have the following:
> Coefficients:
>   (Intercept)         array           dye         genes       variety
>  5.955503e+00  2.695750e-02  4.120987e-01 -2.499571e-04  2.686421e-01
>   array:genes     dye:genes genes:variety
>  1.319176e-06 -7.112527e-05  2.660801e-05
>
> Degrees of freedom: 110386 total; 110378 residual
> Residual standard error: 1.030704
> > anova(fit)
> Denom. DF: 110378
>               numDF F-value p-value
> (Intercept)       1 7590769  <.0001
> array             1   21263  <.0001
> dye               1    3069  <.0001
> genes             1    4277  <.0001
> variety           1    2493  <.0001
> array:genes       1      38  <.0001
> dye:genes         1      99  <.0001
> genes:variety     1      15   1e-04
>
> So I would like to know what I am doing wrong?
> I use the following command:
>
fit_gls(response~array+dye+genes+variety+variety*genes+array*genes+dye*genes
,data=data)
>
> and my dataset looks like this:
>    array variety dye genes response flag
> 1     79       1   1     1 8.395252    0
> 2     79       1   1     1 8.583917    0
> 3     79       1   1     2 8.544225    0
> 4     79       1   1     2 8.423542    0
> 5     79       1   1     3 7.502186    0
> 6     79       1   1     3 7.524021    0
> 7     79       1   1     4 8.188411    0
> 8     79       1   1     4 8.072779    0
> 9     79       1   1     5 7.629976    0
> 10    79       1   1     5 7.524021    0
> 11    79       1   1     6 7.684784    0
> 12    79       1   1     6 7.610358    0
> 13    79       1   1     7 8.366138    0
> 14    79       1   1     7 8.369621    0
> 15    79       1   1     8 7.166266    0
> 16    79       1   1     8 7.038784    0
> 17    79       1   1     9 7.474205    0
> 18    79       1   1     9 7.805067    0
> 19    79       1   1    10 8.339501    0
> 20    79       1   1    10 8.407155    0
>
> Any suggestion would be greatly appreciated.
> Thank you,
> raphael
>

It appears that the variables "array", "dye", etc., need to be treated as
"factors". Probably the most convenient approach would be to convert them in
your data frame before carrying out the analysis. For example, the values of
dye could be converted with the following code (mutatis mutandis).

> data$dye<- factor(data$dye,labels=c("red","blue"))

Finding an appropriate model for these data is likely to be a challenging
exercise. I highly recommend the book by Pinheiro and Bates entitled "Mixed
Effects Models in S and S-Splus". These authors explain very clearly how to
carry out mixed-effects modeling.

-Bill

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._