By default HPTEST fits the model under a regularising prior. Priors are seperately specified for each parameter (i.e. each parameter is a priori assumed independent). By default the priors are as follows.

For the genetic predictor: a $\text{log-F}(2,2)$ prior is placed on the main effect, and a $\text{log-F}(4,4)$ prior on the overdominance effect in the general model.

For covariates: a diffuse $\text{log-F}$ distribution with 95% of its mass similar to that of a $N(0,40^2)$ prior is currently specified (this is $\text{log-F}(0.077,0.077)$).

You have the following options for specifying priors.

• Use the -prior option to specify a different prior for each parameter. The syntax is:

-prior '[parameter name]/[outcome]=1~logf(<a>,<b>)'

which specifies a log-F prior with the given parameters, or

-prior '[parameter name]/[outcome]=1~gaussian(<mean>,<variance>)'

which specifies a Gaussian prior with the given mean and variance. See the help for -prior for examples.

Warning: when using -prior you should check the output log file carefully to ensure that HPTEST has applied your priors properly. You should see something along the following lines in the output:

- Model 2 ("gen"): BinomialLogistic( 2000 of 2000 samples ): (outcome=1) ~ baseline/outcome=1 + add/outcome=1 + overdominance/outcome=1
  with priors:
            add/outcome=1 ~ logF( 2, 2 ).
  overdominance/outcome=1 ~ logF( 4, 4 ).

If there is no 'with priors' line, or if priors for one of the parameters is missing, then it means HPTEST has not correctly interpreted your prior specification. (This is usually because the parameter names are not specified correctly - check the log output for the correct parameter names.)

• Use the -no-covariate-priors option to turn off the priors on covariates.

• Use the -no-prior to turn off all priors - on the main genetic effect and on the covariates. (This therefore implements an unpenalised regression.)