But standardizing weights seems to only be necessary if the weights are parameterized, and this isn't the case here because they are being provided as prior weights. In Theory.pdf on CRAN, there is a suggestion to standardize by setting the first element in the weight matrix to one.
#Asreml r weights how to
This may have to do with assumptions on how to scale the weights, as Ben mentions above. Now both versions gets virtually identical fixed effects, but the standard errors are different (as well as the random effect variances). Before my latest commit (updating Xwts in LMMs), the two versions gave completely different answers. Simon's concern was that CRAN lme4 and new lme4 were giving different answers. > # tease apart lev1 and lev2 variances using a fixed "units" level knownĪsreml(fixed = Ym ~ Xm, random = ~Xm.f + RE.Xfs, family = asreml.gaussian Gamma component std.error z.ratio constraint > # calculate means and fit sample size weighted LMM with random lack of > Nsamp REs #generate some within residual errors (i.e. > #generate some lognormal random effects (i.e. #generate some lognormal random effects (i.e.
0 kommentar(er)
