Suggested Ordered Subset Analysis Reporting
Results of an Ordered Subset Analysis can be inadvertantly
misinterpreted or misrepresented. We propose the following
Return to FLOSS home page
- Report the reason for using each covariate. The covariates
must be identified without reference to the
genetic data. Data-snooping can result in a meaningless and
misleading permutation p-value.
- Report relevant features of the covariate data. For example,
consider reporting correlation coefficients for pairs
of covariates that are highly correlated, and consider reporting
covariates with high levels of missing data.
When two or more covariates yield significant p-values in the same
region, report the number of families shared in common for the optimal
subsets for each covariate.
- Never report an increase in LOD score or z-score without reporting
the corresponding permutation p-value. Since the Lod score is
maximized overall all ordered subsets, the
significance of a LOD score increase can not be determined without a
- Report the total number of regions tested and the covariates used
in each region. The significance of a p-value can not be interepreted
without knowing the number of tests performed.