February 5, 2004

  NOTE 1:  LODs for negative DHATs indicate deficit sharing. 

           For all versions of ALLEGRO subsequent to v1.1e, ALLEGRO 
           maximizes the likelikhood over positive and negative values 
           of the sharing parameter (the estimate is listed in the 
           output under DHAT).  Often one is principally interested in 
           the one-sided test of whether the sharing parameter is bigger 
           than zero.  When the estimated value (DHAT) is bigger than 
           zero, then the LOD has the traditional interpretation of 
           evidence in favor of excess sharing at that location.  However, 
           there are situations in which the sharing parameter can be 
           estimated to be negative; this roughly corresponds to situations 
           in which the NPL score is negative), which indicates that there 
           is evidence of a deficit in sharing.  Because this information 
           may be useful to some users, rather than simply report a LOD of 
           zero for this situation, the program reports the LOD based on 
           the maximizing (negative) sharing parameter, and reports the 
           negative DHAT.

           Thus, when reporting LODs users should be careful to examine
           the signs of the DHATs (estimated sharing parameter).


  NOTE 2:  The exponential sharing model is recommended.

           If an analysis will utilize an allele-sharing model, the 
           authors recommend that the exponential model be used.  One of 
           the main advantages of the linear model is its ease of 
           computation -- it is very straightforward to compute the LOD 
           based on the family NPL scores.  A drawback is that there is a 
           bound on the range of the sharing parameter.  When the 
           likelihood maximizes near the bound, then traditional asymptotic 
           results for the likelihood can break down, and the resulting
           LODs can be misleading.


  NOTE 3:  LODEXACTP and NPLEXACTP options are only available for the 
           exponential allele-sharing model.

           The LODEXACTP is not defined for the linear model.  Prior to
           version 1.2, the program did report a LODEXACTP for the linear
           model, however, that reported value was actually the one 
           based on the exponential model.  In versions 1.2 and later, the
           LODEXACTP option is only available for the exponential 
           allele-sharing models.
           
           The NPLEXACTP does not depend on the allele-sharing model.  It 
           is computed based on numbers that are available and stored during
           the analysis of an exponential sharing model.  These numbers are
           not routinely available in the calculation of the linear model, 
           and, especially in the light of NOTE 2 above, the authors have
           only implemented the NPLEXACTP for the exponential allele-sharing
           models.

        




