I recently obtained my exam results at university and while looking at the score distribution I noticed a peculiar remark.
They state that the distribution is obtained using data from the last 5 years, but where the past years are weighted degressively by a factor 0,8.
After looking on the net (and finding for example What is the most scientific way to assign weights to historical data?) I don't really understand why one would apply exponential smoothing in this case. Is there some obvious reason to put more emphasis on the current year that I am missing?
I can think of a scenario where this would make sense to me: If the exams in different years tend to be similar but the subjects of the course slightly change over time (when the subjects change there might always be a small change in difficulty) then it might be reasonable to choose the score distribution using exponential smoothing. At least I would think that this would be reasonable.
In general this is not something that I would use when grading exams though (but this might depend a lot on the structure of the university/course/content etc.). In my experience exams tend to be very different when the courses were held by different lecturers. But for one fixed lecturer usually exams are very similar in structure and difficulty. At my university lecturers for different courses change regularly and so this technique would not make a lot of sense.
Edit: If the lecturer makes some adjustments to his course/exam every year (for example due to feedback) then it makes sense to suspect that the exam from last year is more relevant than the exam from $5$ years ago.