Stabilize simple imputation for use in primary experiments

[drisso]

@mbcole, what is this issue? Was it addressed when you changed the imputation methods?