Researchers from Harvard, Stanford and Microsoft extensively analyzed the "Potential Duplicate Registrant" data supplied to Iowa by Crosscheck in 2012 and 2014 and published their results in a paper entitled "One Person, One Vote:  Estimating the Prevalence of Double Voting in U.S. Presidential Elections"

Their analysis of "Potential Double Registrants" for 2012 revealed that what is presented by Crosscheck as 64,092 probable doubly registered voters is most likely 6,454 double registered voters plus 115,276 validly registered American voters included due to poor data matching. 

Quality of Data Match for 64,092 "Pairs" from 2012 Iowa Crosscheck

90% of "pairs" are really two unique American voters matched using inadequate data