Dear Dave,
Sorry for the delay.
To clarify what is in the 'HAT' column,
run the simple job
Test hat value
TRT 2 Y
hat.asd
Y ~ mu
with data
1 12
1 13
1 32
1 42
1 38
1 35
2 29
2 25
2 19
2 15
The residual variance is 118
The .yht file contains
Record Yhat Residual Hat
1 26.000 -14.00 11.80
2 26.000 -13.00 11.80
3 26.000 6.000 11.80
4 26.000 16.00 11.80
5 26.000 12.00 11.80
6 26.000 9.000 11.80
7 26.000 3.000 11.80
8 26.000 -1.000 11.80
9 26.000 -7.000 11.80
10 26.000 -11.00 11.80
For this simple design,
X is a vector of 10 ones
R is 118 I
Calculating W Ci W' (page 40 of the release 2 User Guide)
is a matrix with every value 118/10 which is
what is reported in the Hat column.
In the simple case where R is scaled identity,
the variance can be factored out. On the basis that
W(W'W)^{-1}W' is idempotent, its diagonal elemts
should always be LE 1 which is what you expect.
In the more general case I believe that also applies.
Hence, if you have hat values greater than the residual
variance, that suggests a problem. If you can send me an
example, I can investigate it.