review predicted probability section

[norcalbiostat]

choosing a cutoff value with "max" accuracy is picking a value that is actually the max predicted probability. Not the value of the max accuracy

[norcalbiostat]

no, its working fine. the example sucks.

[norcalbiostat]

Because you have so much overlap between the distributions of predicted probabilities the accuracy value (True Pos + True Neg) / N is really at it's max if everything is classified as "eating 1-2 F/V".
Since there are fewer data points overall in the 3+ group, it doesn't matter if you mis-classify ALL of them as eating 1-2FV, you'll still have a higher % accuracy than if you used any lower cut point.

Explain this example better and then find another example where moving the cutpoint actually helps.