Estimating the Correlation between Variables under Destructive Testing, or How to Break the Same Board Twice

A procedure is described for estimating the correlation between two variables that, individually, can only be observed through destructive testing. The procedure involves proof loading a unit in one failure mode and then failing survivors in the other mode. Under the assumption that the variables have a bivariate normal distribution, several properties of the estimation procedure are evaluated using simulation. Primary results provide guidance concerning the experimental design aspect of selecting a proof-load level.