DNA Prediction Tool Evaluation | DNA Science

Relationship predictions optimized for the DNA Prediction Tool Evaluation

Please enter your shared cMs and/or # of segments below. Or anything, really.

# of Segs.

Sexes of the testers

Check to enter HIR cMs for 23andMe (not recommended)


Group Probability

Any Relationship


If you think this relationship predictor is absurd, you're right. Only an absurd comparison would allow this predictor to score higher for both of two metrics than any other relationship prediction tool. That betrays a complete failure to properly set up a comparison. All of this should go without saying.

This predictor will always assign 100% probability to the correct relationship type and will always list the correct relationship within the top three listed probabilities. Those are the two metrics in the comparison above. They simply don't work if comparing individual relationships of one tool to relationship groups of another tool.

It should now be apparent that there is value to breaking up the relationships into smaller groups and even individual relationship types. Who gets to decide how far to break up the groups? You do. And for a month people were choosing SegcM because of how large the probabilities were assigned to individual relationship types.

It's already very clear that predictions that include the number of segments and total cMs assign a much higher probability to the correct relationship type than with cMs alone.

This is the predictor that Blaine Bettinger and Leah Larkin would like to see. Perhaps if a survey is created by a scientist or data scientist who doesn't have their name on one of the tools, then this tool can be taken offline and SegcM can come back. I do not consent to SegcM being used in a comparison where the Ancestry probabilities are guaranteed to look better than any other tool except the page you're on now.

The SegcM tool is currently password protected. The comparison study is set up with a bias that guarantees that AncestryDNA simulated probabilities will outperform any other predictors that are compared. Except for the page you're on now.

SegcM will perform exceptionally well in any study that compares the probabilities assigned to the correct relationship if individual relationship types are compared to each other rather than comparing individual probabilities to those of groups comprised of six or more relationship types. Of course probabilities of large groups will always be larger than probabilities of the individual relationships that make up those groups.

A double cousin predictor is also available.