Why publishing everything is more effective than selective publishing of statistically significant results

Why publishing everything is more effective than selective publishing of statistically significant results