Bias against negative results are a problem. Frequentist "p values" are another, especially when samples are small and biased ("20 undergraduate students") and

Bias against negative results are a problem. Frequentist "p values" are another, especially when samples are small and biased ("20 undergraduate students") and 0.05 is considered "significant." Now, the system is still designed to correct itself for errors over time, and reinforcing this design is the most important meta-thing we can do. So it's not all doom and gloom. It's just that "over time" is a longer and messier horizon than we would like!