If the idea is that the error in estimation is random around 0, you would expect about half to be overestimating. I think 5 out of 8 is not a very strong diffe

If the idea is that the error in estimation is random around 0, you would expect about half to be overestimating. I think 5 out of 8 is not a very strong difference from 4 out of 8, which is the comparison that makes sense in this context.