That the marks (or CGPA or as we call it at IIT Kanpur, CPI) are only an approximate indicator of a very narrow aspect of education is known to all educators. Narrow, because we are only looking at academic performance and that too in an exam setting. It is not a measure of broader definition of education, which should include co-curricular and extra-curricular activities. It is not a measure of important skills like leadership, communication, and so on, which are important for future success. And approximate indicator because those marks and grades depend on many factors - did the student chose only those electives where the grades are normally easy or were the courses chosen based on interest, did the student do 6 courses a semester and tried to finish the BTech program in 7 semesters, or did the student do 5 courses a semester and took 8 semesters to complete. Also, luck factors in exam. While the luck of having those questions asked which you just studied before the exam will get even out over a large number of courses, but the bad luck of being ill just before the exam or something causing stress can cause lower marks/grades.
Because of all this, I used to believe that even in the narrow interpretation of academics, two students with close enough CPI should not be necessarily distinguished. And my own belief (not based on any scientific study) was that all those factors can cause a difference of about 0.5 in CPI. So, if I see a student with a CPI of 8.0 (out of 10.0), I would like to really believe that his true academic merit (in a narrow sense) would be somewhere between 7.5 and 8.5. Not any more.
Over the last many years, I have consistently found any where between 5 and 10 percent students copying in the exam in my course. Of course, I am quite a tough guy on ethics, and I have an elaborate procedure to catch copying. I asked my TA to note down the seat number of every student in the exam. I then arrange the exam answer-books in the order of seats, so that two students who sat next to each other in the exam, their answers will be checked together. I then check one question for all students, and then the next and so on. And sure enough I catch many students with identical uncommon errors. This number has been between 5 and 10 percent in each course. (There are some more whom I would suspect, but if I don't think I can convince a third party with those answers, I end up taking no action on that.)
Last semester, I decided to figure out how many people use unfair means in my exams. I had completed grading of end-semester exam, and there were 3 pairs (6 students) whose copying was obvious, and there were 3 pairs in which case, I was suspicious, but the proof did not seem strong enough. I sent an email to all students of my course giving them an option to either admit that they have copied in the final exam, in which case I would give them 0 marks in the questions that they have copied and give them one grade less after that (so potentially, it could be two grades less). On the other hand, if they did not admit, and I had them on my list, they will get an "F" grade.
5 pairs admitted, two pairs from the first group, two pairs from the second group, and one pair who was not even in the suspicion list. I called the pair from the first list who had not admitted, and showed them their answerbooks, and they finally agreed to have copied (but both of them were given F grades). So we had 6 pairs involved in copying in my course. (Actually, in one case, one student convinced me that he copied from the other without his knowledge. So 11 students were involved in using unfair means.)
11 out of 75. That is 15 percent. Do I believe that everyone admitted. Of course, not. Notice two students even in the first category did not admit (but they are included in 11). And I am not counting in 11, two students whom I strongly suspect, but still gave them benefit of doubt, since they did not admit. Do I believe that no one else copied in the mid-semester exam. Of course, not. So my guess is that about 20 percent students used unfair means in my course in the two exams alone. There is no way we can guess the number in assignments and project.
Now, I have a reputation of being extremely tough with unethical behavior. The "sources" in hostels tell me that many students are afraid of copying in my course. I also try to be very active during the invigilation. I will keep taking round and ask any student to move to a different seat if I have slightest suspicion of their collaboration. All this make me believe that the number of students who use unfair means in other courses is much higher than 20 percent. It is estimated that the number of students who copy in programming lab course, for example, is around 40 percent.
Let us assume that on an average 25-30 percent students use unfair means in some aspect of the course or the other. It is, I believe, fair to assume that these students are willing to copy since they think they are on the borderline of the two grades, and copying just might push them to the higher grade. If even 10 percent of them got pushed to the next higher grade (and an equivalent number of honest students remained behind due to relative grading), that would have a huge impact on CPI/CGPA of the two groups as well. If we take this into account then CPI of an honest student versus a dishonest student with similar academic preparation could be substantially different, of the order of 1.0.
So, now, when I look at a student with a CPI of 8.0 and not knowing whether s/he is honest or not, I would have to assume that his/her academic preparation could be similar to one with a CPI of 6.5 on one hand, and 9.5 on the other. With this kind of variance, the CPI has lost whatever little value it had in at least that narrow aspect of judging academic performance.
With this kind of variance, we are only discouraging the honest students (who still appear to be in majority despite the huge temptation and the unfairness of the system to honest students) and encouraging the cheats. Notice that higher CPI is a ticket for higher education, better placement, many scholarships, many awards, and so on. If CPI has no meaning, we must find out alternative ways of selecting students for admission to higher degrees, scholarships, and awards.
If this is the state in IIT Kanpur, one of the premier institutes, where we still take teaching rather seriously, I don't even want to imagine the situation in colleges where it is in the short-term interest of the college to have better marks/grades of their students.
The status of our university/institutional performance metrics is the main reason why we need even more exams later on. It is not just that we can't compare academic performance of two universities and hence we need GATE as an equalizer, but we can't even trust the reported performance of the students and hence we need GATE. This is what is making us a society where we have exams for everything, but that is making things only worse, since learning how to cheat is beneficial in the society with too many exams.
Our universities must tackle this problem urgently or risk losing all relevance in the society.