Testing Models

There are two ways to test a pair of Heritability Models: the first is based on likelihood; the second is by creating a hybrid model and allowing the data to choose how much weight to assign to each.

For both approaches, the first step is to create the Tagging Files corresponding to each model. It is important that both tagging files contain exactly the same predictors. Therefore, if you are using predictor weightings, you should add --reduce NO when computing the tagging file to ensure that LDAK retains predictors with zero weightings.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

To compare models based on likelihood, first regress the test statistics onto each file using --sum-hers (this command is explained in SNP Heritability). When performing this regression, the only difference between the two commands should be the choice of tagging file. The values to compare are the alternative (log) likelihoods reported in the files ending .extra.  When examing these files, you should check that the null likelihoods match (this confirms that the same predictors and settings were used for each regression).

To create a hybrid model, then having computed the two tagging files, merge these using the command --merge-tagging, specifying the names of the two tagging files via the option --taglist. Next you regress the summary statistics onto this merged tagging file. The weighting assigned to each heritability model corresponds to the shares reported in the file with ending .share.
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Example to come here.