Retiring statistical significance would give bias a free pass