Summary: choose method for initial estimates of parameters No Arguments
This command determines how the variance components command makes its initial parameter estimates. To use, enter "set starting values" and choose the desired option:
npl:1> set starting values Genehunter currently uses a constant fraction of total phenotypic variance. Possible starting values: 1. ML estimate from adjacent position 2. Constant fraction of total phenotypic variance Enter the index of the start values you want to use [2]:The first method simply divides the total trait variance evenly among the variance components of the model. Thus it uses the same initial values for each position. The second method does this for the first position, but thereafter uses the maximum likelihood values for the last position. The second method is generally faster, because adjacent positions usually have similar maximum likelihood estimates, hence the algorithm requires fewer iterations to converge when it starts near its destination. However, the method can sometimes prevent convergence on the true maximum likelihood estimate, instead settling on a local maximum near the maximum likelihood values of the adjacent position. The first method, slower but more reliable, is the default.