Badly formulated questions

Some students complained about the unclear formulation of question 20 and question 25. Here is the percentage of wrong answers as function of the question number:
.
The question number 20 got 52% wrong answers while question number 25 got 36% wrong answers. I guess these numbers are consistent with a random hypothesis (binomial distribution) within one sigma of 1/sqrt(19)~20%. So I agree to remove the questions number 20 and 25 from the analysis of the examination results. Thus only 44 of the original 46 questions will by included.

Passed/not-passed

Under (a slightly violated) assumption that the questions are uncorrelated, a hypothesis that a student gave random answers and obtained k wrong answers out of n questions corresponds to a binomial distribution Bin(n,k)/2^n. Here are the experimental results together with the theoretical hypothesis:
.
A reasonable threshold would be 4% probability that a student answered randomly. This threshold corresponds to 17 wrong answers. Apparently all students have answered not randomly (with a probability larger than 96%) with a clear bias toward correct answers. Therefore since all students have less than 17 wrong answers and since everybody has delivered the obligatory exercises, all students have passed the course.

Self-evaluation and the evaluation formula

Here plotted is the self-evaluation character a student has claimed as function of the number of wrong answers they got, together with a linear fit to the data, and a line, 12-0.5*k, from 12 at zero wrong answers to 4 at the maximum allowed wrong answers (4 been chosen as lowest mark for those who submitted the (allegedly a bit too big)) homework:
.
Apparently the students with the least wrong answers (the 9 left points, some of them are multiple) are actually well correlated with the linear estimate. The last 5 points seem to be an overestimation. So, the simplest linear formula 12-0.5*k seems to correlate well with the expections of the students with the least wrong answers and will then be used to mark the exam.