On Jul 31, 10:30pm, "Andrzej Pownuk" wrote:
} Subject: RE: [UAI] random degrees
} Dear Judy,
} 
} I general I agree that there are some random components in the process
} of giving degree.
} 
} However if we consider a very simply tests then 
} I don't think that the influence of randomness is significant.

I argue here that there is both uncertainty and randomness
present in all of your examples.

} 
} Let us consider the test which contains two questions:
} 
} 1) When did Columbus Discover America?
} 2) When did the First World War begin?
} 
} Probability of guessing the answers is very low.
} What is random in the fact that somebody answers 
} 0%,50%,100% of these question?
} 

It's arguable whether Columbus "discovered" America.
That's true in the sense that he didn't know whether it was
there, or that it was there.  However, many people already
knew this fact.

Many countries were involved in World War I.  Began for
which countries?  What is the definition of the beginning
of a war?  (Or in the case of the US aggression in Iraq,
what defines the end of a war?)

These questions, and many exam questions, are open to
discussion.

} If somebody learned then he know 
} if somebody didn't learn then he doesn't know. 
} This is quite deterministic relation.

Given the way my memory access works, I am not sure what
it means to "know" something.  I may be able to address you
by name in our conversation, and a few minutes later be
unable to introduce you to someone else.  

My students generally have better access to their memories,
but still exhibit glitches, especially under pressure.

There are many factors that affect student performance on
exams.  Some are personal:  has the student slept, eaten,
etc. appropriately before the exam?  Have there been any
external distractors?

Can the student access the information?  Can the student
frame the relevant information appropriately for the
question?   Some students prefer to memorize, some to
derive or reason.  Some prefer short answers, some open-ended
questions.

One might be able to quantize these as parameters/factors
in a multidimensional space.  I prefer to model the outcome
probabilistically, where the probabilities take into account
both the probabilities of external affectors and the 
probabilities of internal glitches and moments of clarity.


} 
} Of course in reality there are some random "noises" 
} for example sclerosis, influence of stress etc.
} 
} However I still don't understand 
} what is random in the fact that somebody remember some data?
} (if we neglect sclerosis and other problems)
} 
} 
} 
} Well this is really not a very good example
} because randomness play an important role in this case.

Reply via email to