The over all grade in the course is composed of 300 pts hour exam + 300 pts final exam + 300 pts lab grade + 100 pts Concept Tests. On the basis of my experience and my understanding of the difficulty of the exams the various grade cuts are made (i.e., the A/B division, the B/C division and the C/NC line). In the past I have given the class a percentage above which an A is obtained. Since I do not have control of the Lab there is a bit of uncertainty, but the A/B line is in the 85% range and might be a bit lower. That is, if the total % for the whole course is above something in this range the letter grade is an A. To understand the next step you are going to have to read about my theory of measurement!
We give exams in an attempt to measure knowledge and the ability to think. But we also give exams as a learning technique. There is the learning due to review and studying for exams and learning in attempting to work the problems under the pressure of taking the exam and the subsequent review of the test questions. Do exams measure knowledge? I do not know how well exams measure the ability to think, or test the practical use of knowledge. I have known students who were very good at taking exams, but were not very good at applying that knowledge later, for example, in another course or in graduate school. Measuring by means of exams is like measurements in sociology, or even public opinion. It measures something at the moment and under stresses of the moment. However, exams are the only thing we have that is practical to do (at least this semester) and we must make the best of them. The Concepts Tests and especially the "reason you selected the multiple choice answer" box lets me understand something more than the stress of the moment. Concept Tests are supposed to be done on the fly and give feedback at least about a group effort if a group works on these. For someone along for the ride in a group-who knows, they cheat themselves especially if some of this conceptual material finds its way on a final!
Each hour exam is a measurement of understanding in the above sense. Each hour exam is averaged and the hour exam average is combined with the other components of the course. This a typical case of measurements of small numbers. If we deal with all this by the use of statistics, which the sociologist, etc. are fond to do, then we expect a gaussian distribution (Bell Curve) of scores having a certain width (related to the standard deviation) and a position (the mean score). Even in a larger class, say 100 people, there are strange looking distributions of scores. In fact if we divide the range from 0 to 100 into bins of a given width the distribution of scores can be made to look more or less gaussian. A gaussian distribution is the basis of the idea of "grading on a curve." That is, the distribution is divided into segments the top 1/4 being A and the bottom 1/4 assigned F (NC in Brown's lingo), etc. The use of this scheme, whose grade divisions are also arbitrary is based on the notion that the fundamentals of statistics prevail. I do not think they really do because of so many variables. Each class may be more or less prepared than the class the year before, the questions one year may be less difficult than in other years, each exam is taken by an individual and that individual may have a bad or good day, and on and on. This is what I call a lack of statistical control over the experiment and in my opinion it is the flaw in the work of sociologists, medical statisticians, epidemiologists, Deans of the Faculty, Provosts, etc. and has led to lots of problems. In physical science we deal with inanimate objects and have a greater measure of statistical control over experiments. Even then troubles can arise. This is the reason I reject "curving", or using the Bell Curve for grade determination. As far as I am concerned everyone can have an A if everyone knows the material. Of course everyone can have an NC too.
In an experiment there may be results that are off what we expect. That is, results that deviate from the mean by some amount, say several standard deviations. How are we to treat that result? Such results are called outliers. But even an outlier has some chance of occuring. Now in a measurement if the experimenter (the prof giving the exam) knows that something went wrong, like the lights going out so no one could see, or an illness, then there is reason for expunging a measurment (test), but in the absence of an iron clad reason there is no cause for exclusion. Further experimentation will bury the outlier in a mass of measurement distributed about the true mean.
This becomes impractical when experiments are difficult and when there are few measurements it is even more difficult. Do we want to give an exam every day to increase the statistics? How then do we deal with the so called outliers? Whatever we do it must be unbiased and consistant in the treatment of the data. One can say, for example, that the probability of a result that is three standard deviations or greater from the mean is so improbable that it should be discarded. That of course applies to high scores as well as low ones. A student I know scored exceptionally high on the SAT tests and was told to take them over (at ETS expense) because they felt these must be outliers. Of course, the added pressure the second time around was unmeasured by ETS as were a host of other elements over which they had no control. What did their remeasurement mean? (No pun indended.) We could apply such an arbitrary method to eliminate outliers.
In chemistry 22 there are only three hour exams. The standard deviation of three measurements is hardly amenable to statistical analysis even if the total class is. But there, in my opinion, can be outliers nevertheless. On the other hand, I do not believe in throwing out data unless there is a medical excuse, or the lights went out! What to do? What I do is not throw out scores, but I do not let one poor score determine a letter grade. Therefore, after determining the A/B, B/C, C/NC lines I look at all the people below the cut off and throw out a low grade. Actually, I program a spread sheet to do most of this. If that causes a grade change I will look at the laboratory TA comments and the rest of the scoring and make a decision whether to "toss" that person over the line-that is, give the higher letter grade. Think about it however. There are 1000 points in the course. Each hour exam is 100 points and if one got 50% points on one hour exam that is 50/1000 or a 5% point change. One has to be close to a grade line to make it over. Although it may not sound like it, this is not based on the science of statistical analysis, but on compassion.