From Jean_Michel Vedrine (jean-michel.vedrine at wanadoo.fr) Thursday, 28 July 2005, 12:51 AM:
Well this is in fact not a bug in the code, but this help file doesn't speak about a very special case.
excerpt from the help file:
The discrimination coefficient is a correlation coefficient between scores at the item and at the whole quiz. Here it is calculated as:
DC = Sum(xy)/ (N * sx * sy)
where Sun(xy) is the sum of the products of deviations for item scores and overall quiz scores,
N is the number of responses given to this question
sx is the standard deviation of fractional scores for this question and,
sy is the standard deviation of scores at the quiz as a whole.
Now imagine that all student gave the same response to this question, then sx = 0 and we have a divide by zero operation and DC is infinity. In that case by design -999 is displayed.
So to close that bug, all we need is to rewrite the help file
From Jean_Michel Vedrine (jean-michel.vedrine at wanadoo.fr) Thursday, 28 July 2005, 12:51 AM:
Well this is in fact not a bug in the code, but this help file doesn't speak about a very special case.
excerpt from the help file:
The discrimination coefficient is a correlation coefficient between scores at the item and at the whole quiz. Here it is calculated as:
DC = Sum(xy)/ (N * sx * sy)
where Sun(xy) is the sum of the products of deviations for item scores and overall quiz scores,
N is the number of responses given to this question
sx is the standard deviation of fractional scores for this question and,
sy is the standard deviation of scores at the quiz as a whole.
Now imagine that all student gave the same response to this question, then sx = 0 and we have a divide by zero operation and DC is infinity. In that case by design -999 is displayed.
So to close that bug, all we need is to rewrite the help file