Likelihood function surface

[td2mafea@uco.es]

[Originally Posted: 12/13/99-- Transferred by ReliaSoft Moderator]

1- How to interpret the "likelihood function surface" plot? Witch is the best adjustment: the one with a very litle surface and very tiny, like "Eifell tower", or the one with a large surface?

2- How can I perform a test to decide witch is the best mobel (weibull versus log-normal)? Do you known any bibliografy to perform this test?

My best regards

[David J. Groebel]

[Originally Posted: 12/14/99-- Transferred by ReliaSoft Moderator]

First, it is more desirable to have a thin, Eiffel Tower like likelihood function surface. This implies that there is less uncertainty in the parameter values at the maximum value. A larger surface implies that there is large uncertainty in the parameter values. You can usually decrease your uncertainty by increasing your sample size. For your second question, there a multiple ways you can go. You can use Weibull++ Distribution Wizard to get a suggestion. You could also conduct a Kolmogorov Smirnov Test or a Chi-Square test. You can view these results in Weibull++ from the Tools menu then select Show Other then select Goodness-of-Fit Test Results. You can only access these results from a data sheet. If you are using maximum likelihood theory, you can consider the distribution which gives you the largest likelihood value. If you are using rank regression, you can consider the distribution which gives you the largest correlation coefficient. Our website contains a listing of references. You can access the references at the following address, http://www.reliasoft.com/Books/index.html.