Description Usage Arguments Details Value Warning Author(s) See Also Examples
When a mixed model is run in ASReml-R, this function can estimate the profile likelihood of variance components from the output.
1 |
full.model |
An |
component |
A character vector indicating for which variance component the profile likelihood will be constructed. Must be an object in |
negative |
Logical indicating whether or not the |
nsample.units |
Number of sample units to be used in constructing the profile likelihood |
nse |
Number of standard errors on either side of the estimate, over which the profile likelihood should be constructed |
alpha |
The critical value for determining the Confidence Interval |
threshold |
Acceptable distance, between actual sample values and interpolated values, for determining the upper and lower limits of the Confidence Interval. Actual samples will be no more than this distance from the true value of the estimate |
For the negative
argument, this should be used if the profile likelihood of a covariance component is to be constructed.
The function uses a grid search to obtain the approximate confidence limits. Therefore, nsample.units
should be carefully thought about beforehand when running the function. Increasing this value will ensure a smoother surface, but at cost to time. The function weights this quantity to ensure more samples for regions of the profile likelihood greater than the estimate. The lower bound of the sampling interval to the estimate is made up of 2 times nsample.units
, whereas the sampling interval from the estimate to the upper bound is 3 times nsample.units
.
Similarly for nse
, the function will include twice as many standard errors from the estimate to the upper bound than it does from the estimate to the lower bound. If negative
is FALSE, and the lower bound of the sampling interval extends beyond zero, this will instead be set to 0.
lambdas |
negative log Likelihood ratio test statistic. Estimated from the log Likelihood of the |
var.estimates |
value along the sampling interval for which the |
UCL |
approximate Upper Confidence Limit |
LCL |
approximate Lower Confidence Limit |
component |
the component for which the profile likelihood surface has been constructed |
May be unfeasible to estimate profile likelihoods for complex models with many variance components
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 | ## Not run:
library(asreml)
ginvA <- asreml.Ainverse(warcolak[, c(1,3,2)])$ginv
ginvD <- makeD(warcolak[,1:3])$listDinv
warcolak$IDD <- warcolak$ID
warcolak.mod <- asreml(trait1 ~ 1, random = ~ped(ID) + giv(IDD), ginverse = list(ID = ginvA, IDD = ginvD), data = warcolak)
summary(warcolak.mod)$varcomp
profileA <- proLik(full.model = warcolak.mod, component = "ped(ID)!ped", negative = FALSE, nsample.units = 3, nse = 3)
profileA
profileD <- proLik(full.model = warcolak.mod, component = "giv(IDD).giv", negative = FALSE, nsample.units = 3, nse = 3)
x11(w = 6, h = 8)
par(mfrow = c(2,1))
plot.proLik(profileA)
plot.proLik(profileD)
## End(Not run)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.