A meaningful interpretation of the solution from the logit model can be
obtained by computing EXP(solution or contrast). This gives Relative Risks,
also called Odds Ratio's.
I agree with Arthur that your solutions from the probit model are way off
scale: they should be roughly from -3 to +3 for probit (underlying standard
normal distribution) or about -5 to +5 for logistic (underlying logistic
distribution). These very high values for solutions for Fixed Effects
indicate usually an unidentifiability of some effect or level.
> P.O. Box 65
> 8200 AB Lelystad, The Netherlands
> -----Original Message-----
> From: Kim Bunter [SMTP:firstname.lastname@example.org]
> Sent: Monday, September 24, 2001 2:41 AM
> To: email@example.com
> Subject: Solutions for FE under link functions
> Dear Arthur,
> when 0/1 data is analysed with the probit (or logit) link function, do the
> solutions for fixed effects have any meaning? For example, the following
> solutions are obtained for management groups under either animal (A) or
> sire (S) models, when assuming normality (N) or using the probit link
> function (P).
> AN SN SP
> 0.33 0.33 71.61
> 0.38 0.38 71.80
> 0.43 0.43 71.86
> 0.58 0.59 72.72
> What is the interpretation of these results, and why are SP values of such
> a different magnitude? The differences between solutions for each level
> also vary. Are computed F_values informative for modelling purposes when
> using the link functions?
> Any comments much appreciated.
> Kim Bunter
> Research Scientist
> Animal Genetics and Breeding Unit
> University of New England
> Armidale, NSW, 2351
> Ph (ISD): -61-2-67733788
> Fax (ISD): -61-2-67733266
> email: firstname.lastname@example.org
> Asreml mailinglist archive:
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml