After further experimenting, it is not a small number problem. I took the random numbers and multiplied them times 1000 to make them larger and got negative SS's on my third recalc. Then I tried times 1,000,000 and it happened again.
Ronny Richardson At 10:45 AM 12/21/2002 -0700, Paul Bernhardt wrote: >Ronny Richardson said on 12/21/02 9:53 AM: > >>I've repeated this experiment several times and sometimes the sums of >>squares are positive and sometimes one or more of them are negative. Is >>this a bug in Excel? I don't see how a sum of *square* can be negative even >>though I am using random numbers so all of the SS should be in error. > >You would be right to consider this wierd behavior. What version of Excel >are you using? A couple of versions back, Excel would have trouble with >data composed of small numbers. Something about the determinant of the >matrix getting close to zero causing problems. > >Paul >. >. >================================================================= >Instructions for joining and leaving this list, remarks about the >problem of INAPPROPRIATE MESSAGES, and archives are available at: >. http://jse.stat.ncsu.edu/ . >================================================================= > . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
