Syndicate content

Add new comment

I think the norm is still evolving here. If sample sizes are small (either in terms of units or clusters for clustered trials), then just reporting the randomization inference p-values seems a reasonable approach. But I've also seen both sets reported (because they are testing different nulls).