Dear Rune,
am I right in thinking of Model3b as a zero-correlation-parameter
model (m_zcp) but with the variances of the operators-related effects
constrained to equality?
Specifically, is the difference between Model3b and m_zcp that m_zcp
estimates variance components for each level of the factor 'machines'
and Model3b assumes equal variances across the levels of machines and
estimates only one variance for all levels?
Model3b <- lmer(Y ~ machines + (1 | operators) + (1 | machines:operators))
m_zcp <- lmer(Y ~ machines + (machines || operators))
Cheers,
Maarten