Identity by descent (IBD) regression
------------------------------------

^ibdreg^ [varlist] [if] [in] [, ^po^sterior^(^varlist^)^ ^pr^ior^(^numlist|varlist^)^
   ^nocons^tant ^off^set^(^varname^)^  ^r^obust ^cl^uster^(^varname^)^ ^nolog^]




Description
-----------

Using data comparing genetic and phenotypic similarity betweens pairs of,
relative this program carries out a logistic regression of the probability 
that the pair share a gene identically by descent (IBD). Each data record
describes a related pair of individuals. The prior probabilities that the 
members of the pair share zero, one or two genes IBD are determined by the 
relationship between members, and the "posterior" probabilities will have 
been calculated using a program such as ^genehunter^ or ^merlin^. For most 
relationships, the pair can only share zero or one genes IBD and the 
probability of 1-sharing, p say, is modelled by a logistic regression equation.
For sibpairs, however, the pair may share zero, one or two genes IBD, and the 
probabilities of these outcomes are assumed to be, respectively, (1-p)^2, 
2p(1-p), and p^2, where p is again given by a logistic regression equation. 
This represents an independence assumption for IBD status for maternal and 
paternal copies and this in turn implies that the effects of the two copies 
of a disease susceptibility  carried at the locus by a subject are 
multiplicative upon risk of disease. The dependence of the method on this
assumption can be relaxed by use of the ^robust^ option.If several relative 
pairs within the same pedigree may be considered, the ^robust^ option should
be used together with the ^cluster^ option to indicate that records are 
clustered by pedigree.  

A known "offset" may be incorporated in the regression model. It will often be 
appropriate to set the offset to the "prior" log odds of IBD sharing:

                            Pr(1-IBD)
                        log ---------
                            Pr(0-IBD)

for each relative pair (for sib pairs zero no offset is required).


Options
-------

^posterior^	Variables holding posterior IBD probabilities in the order
		zero, one and (possibly) two. If not specified these are
		assumed to be those used in the last call to ^ibdreg^

^prior^		The prior IBD probabilities. If the relationship between
		members of pairs is constant, this is simply a list of 
		values. Otherwise it is a list of variables containing 
		this information. Default is 0.25, 0.5 and 0.25 --- the 
		correct value for sibling pairs

^noconstant^	Omit the constant term (intercept) from the regression model

^offset^	Include a known offset in the regression equation

^robust^	Use "robust" Huber-White method to calculate Wald tests and
		standard errors of regression coefficients

^cluster^	Records are grouped in potentially correlated "clusters", 
		for example describing multiple pair-waise comparsions 
		within th esame pedigree

^nolog^		Suppress iteration progress log



Example
-------

^ibdreg mean_age, post(post0 post1 post2) pre(0.25 0.5 0.25)^ 

^ibdreg mean_age, robust cluster(pedigree)^


Reference
---------

Holmans, P.(2002) "Detecting Gene-Gene Interactions Using Affected Sib-Pair 
Analysis with Covariates", Human Heredity, In press.

Author
------

David Clayton

Diabetes and Inflammation Laboratory		Tel: 44 (0)1223 762669 
Cambridge Institute for Medical Research	Fax: 44 (0)1223 762102
Wellcome Trust/MRC Building			david.clayton@cimr.cam.ac.uk
Addenbrooke's Hospital, Cambridge, CB2 2XY	www-gene.cimr.cam.ac.uk/clayton