I'm currently analysing a two site chick pea trial with primary focus
being fungicide application time in relation to yield.
The ranking of the five chick-pea varieties is also of importance.
The raw means of the five varieties are unusual as one variety yields
quite poorly at the first site and exceptionally well in another, while
all other varieties yeild relatively consistently.
The model Yield ~ mu c(site) !r site.VARIETY
produces effects in the solution file:
site.VARIETY 1.BohatyrNZ -0.3062 0.1959
site.VARIETY 1.Dundale 0.7861E-01 0.1959
site.VARIETY 1.Bluey 0.6107E-01 0.1959
site.VARIETY 1.Jupiter 0.2128E-01 0.1959
site.VARIETY 1.PSE23 0.1452 0.1959
site.VARIETY 2.BohatyrNZ 0.3062 0.1959
site.VARIETY 2.Dundale -0.7861E-01 0.1959
site.VARIETY 2.Bluey -0.6107E-01 0.1959
site.VARIETY 2.Jupiter -0.2128E-01 0.1959
site.VARIETY 2.PSE23 -0.1452 0.1959
This is clearly incorrect as each variety has an effect of the same
magnitute, negated, at both sites. I'm wondering why these effects have
been produced in this manner.
No error variance structure has been applied to this interaction term.
Fitting the model again with !r site.VARIETY replaced by
!r site|1/VARIETY site|2/VARIETY
produces the desired effects.