Professional Documents
Culture Documents
In this study, we have used a grounded theory approach, where there is no prior
hypothesis and we build a theory based on the data. We conducted a quantitative
survey of undergradutate students to understand about the courses they can take and
to understand where there priorities lie.
Purpose
Procedure
We have used a grounded theory approach, where we started the survey with an
aim to understand students’ course picking behavior without any prior hypothesis. We
chose a survey based approach where the participants where presented with a set of 6
questions (apart from basic information questions). The survey is divided into two
parts. Part 1 deals with attributing courses so that the suggestion engine has enough
data to go with. Part 2 deals with taking the students’ personal preferences so that
further analysis can be done. A set of 13 courses (out of the total 55 offered in one
semester) ranging across all fields Engineering, Humanities, Science, Math, CS were
taken into consideration.
Part I
Part 1 involves getting data about the courses from the UG4/5 students who have
already taken that course. They were presented with a list of the 13 courses mentioned
above and were asked to rate these courses based on two attributes, Attendance
Strictness and Workload on a scale of 1 to 5.
Part II
Part 2 involves getting data about the students’ preferences. They were presented
with four questions that they had to answer on a scale of 1 to 5.
Questions
1. How would like your course’s attendance policy to be? (1 being very lenient
and 5 being very strict)
2. How would like your course’s workload to be? (1 being very less and 5 being
a lot)
Results
Course Attribution
Out of the N=25 participants, a majority of them were from UG3, so they
couldn’t give information about the courses. The rest of the students (14) gave
information about an average of 5 courses each. This gave us a sparsely populated
13x14 matrix. Running an inter-rater reliability test on this gave a Krippendorff’s
Alpha[1] of 0.779. Usually, a Krippendorff’s Alpha of 0.8 is considered to be reliable.
However, in some cases a Krippendorff’s Alpha of 0.66 is also considered as the
benchmark for reliability. In this case, since the matrix is sparsely populated, we don’t
have sufficient data to test the reliability. Thus a Krippendorff’s Alpha of 0.779 shows
that the course attributes are in fact reliable.
Student Preferences
Several tests were done with this data. For each student, we had 6 pieces of
information. The batch, gender, and their 4 preferences. These 4 set of preferences shall
be referred to as AP (Attendance Policy), WA (Workload Policy), AA (Attendance
Affinity) and WA (Workload Affinity) from now on.
T-Tests. First, we ran T-Tests[2] on all 4 parameters across the 2 genders. The
results are as follows:
ANALYZING STUDENT PREFERENCES IN COURSE SELECTION 5
AP (M) AP (F)
df 23
T-Statistic -1.22
P-Value 0.235
WP (M) WP (F)
df 23
T-Statistic -0.68
P-Value 0.5014
AA (M) AA (F)
df 23
T-Statistic 2.033
P-Value 0.05
WA (M) WA (F)
df 23
T-Statistic -0.553
P-Value 0.585
As we can see, except for AA, the rest of the three parameters don’t have any
significant P-Values.
Correlation. We also calculated the pearson correlation[3] between CG and the
4 parameters of the individual.
AP 0.3739 0.0656
WP 0.3837 0.058
AA -0.854 5.316e-08
WA -0.014 0.995
As we can see, AA has very high negative correlation with the CG of a student.
Basic Statistical Tests. We measured the mean of each of the four
parameters. We also meaured the frequency of elements in each of these parameters.
AA 2.88 [3, 6, 9, 5, 2] 3
Clustering. We divided each parameter into two sections, Low ( < 3) and High
( >= 3 ). With such a division across all parameters, we got 16 clusters. The division of
students into their respective clusters is as follows:
LHHL [7.33]
HHHH [8.21]
HLHH [8.56]
HLLH [8.72]
HHLL [9.23]
Course Prediction
The participant was presented with a suitable course by using the surveyed course
attributes and personal preferences. We used weighted mean square error between each
of the course attributes and participant preferences with the respective affinities as
weights to predict the most suitable course. The percentage of satisfied participants was
used as the accuracy of our course prediction method. We find out that 72% (18/25) of
the participants were satisfied with the predicted course.
Qualitative Analysis
Case Study 1
A UG4 CSD student working in the Computer Vision Lab was interviewed and we
received the following responses:
• Q1. What are the main factors you consider to select an elective
course?
I think the main factor behind selecting a course is interest. But there are mostly
very few courses I have a deep interest in. Then, I have to consider other factors
which are either the importance of them in the future or, if not importance,
factors like workload or grading which can help me get good grades.
• Q2. What factors do you consider if you don’t have a deep interest in
the course type?
I consider various factors which help me to have either a deeper knowledge in the
subjects I have to follow after graduation or have good grades which help me in
the same. For example, I took Linear Algebra as an elective because It will help
me a lot in my research work in the future. And also, I took Databases course
becuase various resources like sample question papers were available with the
seniors and helped me a lot to get better grades. Then there are other factors like
the workload, which help me focus more on all the subjects.
• Q4. Which courses have been the easiest for you and why?
I think Databases and Computer Programming have been the easiest courses for
me because I was already ahead of the Computer Programming class as I had
ANALYZING STUDENT PREFERENCES IN COURSE SELECTION 9
studied most of the concepts in school already. Otherwise, Databases was one of
the easiest courses, because I was able to score good marks without much effort
with the last year resources available with the seniors.
The study revealed that interest in course type is a significant factor behind
course selection, but that is true for selective course types. For other electives, Students
consider factors to help them get better grades or have a deeper knowledge in their
specific stream of study. The presence of other unconventional factors like past year
resource availability were revealed which further affect the student’s decision. After the
discussions, we realized that the complex phenomenon of decision making behind course
selection can not be completely reduced to some quantifiable factors, and is also
dependent on other latent variables which are unique for all students.
Conclusions
As we can see from the T-Tests conducted on the parameters, there is a significant
difference between the way females prioritize attendance and the way males prioritize
attendance.
Analysis of the Pearson Correlation coefficients tells us that there is a significant
negative correlation between CGPA and Attendance Affinity of a student. A reason for
this could be that attendance is not a factor that they consider while choosing courses.
This may be because they perform well in a course irrespective of the attendance policy.
As usual, there are certain outliers in our dataset, but the strong correlation shows that
this is a general trend.
By looking at the most frequent elements in each of the parameters, we notice
that the most frequent in AP is 2 whereas the most frequent in WP is 4, and both of
them, by a large margin. This goes to show that students prefer a heavier, more work
intensive course over a course which has strict attendance policy. Another interesting
pattern to be noted is the absence of any obvious high frequency element in AA. This is
so because of the significant correlation between CG and AA discussed earlier. Given
this correlation, we can say that since the CG of the students in this dataset doesn’t
ANALYZING STUDENT PREFERENCES IN COURSE SELECTION 10
Limitations
There are certain limitations in this research method that we have employed. The
first one being that the process of course selection is a huge task and takes into account
a plethora of factors. The second limitation is that acquiring data about courses and
about students gets increasingly difficult as we try to increase the number of parameters.
References