Analysis of visual literacy of biomolecules

Indicators of performance: PLC and NS

PLC: path length correlation. How connected are two nodes. The closer to one more connected. Compared to expert.

NS: Neighborhood similarity. Compared to the experts.

August 2022: Comparing 3 Biochem courses at UMR, UT, and UNL clustering_indeces_v2.html
Exploring how to measure performance: PLC vs Coherence vs NS
- PLC seem to clearly distinguish experts from students (with significant overlap though)
- NS does not distinguish as much as PLC, but the highest score and median is still significantly different between experts and students
- Coherency does not seem to distinguish well and we will discard it in the subsequent analysis
Clustering the pairs PLC/NS: 3 groups better than 4
The problem with clustering is that it is an iterative method and different “initial seeds” will yield to different results. It is only reproducible when the k-means method uses “set.seed(42)”
Automatically labeling the 3 clusters as High, Intermediate, and Low performance. Using only PLC to label what is HP, IP or LP
Demographics: only Biochem at UMR vs UNL vs UT
November 2022: Comparing all Chem semesters at UMR clustering_and_heatmaps_onlyUMR.html
Some of these students are the same because the data was taken during Spring 2021 and Fall 2021. Therefore, students in CHEM2 in Spring 2021 took the survey again in OChem2 in Fall21. The same applies for students in GenChem2 in Spring2021 who are then in Biochem in Fall 2021.
Approach 1 - PLC Only: compare how PLC score is distributed among demographics and courses.
Approach 2 - NS Only: compare how PLC score is distributed among demographics and courses.
Approach 3 - Cluster NS/PLC pairs: combine students’ PLC and NS indexes as a measurement of students performance.
Cluster the PLC,NS pairs into three groups: Low performers, intermediate, and high performers.
Analyze the composition of those three groups by its demographics, year and courses. The year is not by the number of credits, rather, if they are in CHEM1 or CHEM2 they are labeled as first_year, CHEM3 and CHEM4 will be second_year, and Bioc1 and Bioc2 will be third_year.
Being the three cluster HP, IP, and LP categorical data and not numerical (like PLC or NS) we can use a Chi-Square analysis instead of student-t or anova
Are cluster groups unevenly distributed among these categories?
A chi-square analysis will give us the probability that all three cluster groups (HP,IP,LP) contain statistically similar proportions of this category (course, year, sex, race…)
March 2023: Comparing all Chem semesters at UMR (and two years of Biochem F22 and F21) clustering_onlyUMR_Feb23.html
Same as the previous point, but with an additional semester of Biochem (F21 and F22)
There is now some data that could allow for following a set of students as they progress through the curriculum. For example, students who took the survey in Spring21 in Ochem1, Fall21 in OChem2, and Fall22 in Biochem.
Some changes with data The file "umr_all_preFall22.csv" is missing 2 rows of students who didn't have PLC or NS data (not sure why) and 11 more students who I was not able to find what course they were enrolled When I merge it with "bioc3321_f22_allquestions.csv", we write the "UMR_all_for_R_with_courses.csv" Experts dataset: Experts_all_for_R.csv Other campuses dataset: "Dennison_UNL_UT_all_for_R.csv"
March 2023: Comparing only Biochem1 at UMR "Fall 2021" vs "Fall 2022" clustering only Biochem at UMR F21 vs F22
March 2023: Comparing all Biochems across campuses including UMR's BIOC3321 in F21 and F22 clustering all Biochems at UNL, UT, and UMR
March 2023: Same students. Longitudinal study as same students take OChem1, OChem2, and Biochem1 taking the same survey
Python plots with same students taking OChem1, OChem2, and Biochem1
May/August 2023: Same students. Longitudinal as same students take OChem1, OChem2, and Biochem1
Study with ANOVA and Chi-Square analysis