Great idea about using prediction markets. I'll think about making a few predictions PredictionBook for a few studies, but this is obviously inferior to your suggestion because the researchers in that case have incentives to be honest and encourage replication.