Wayne's Github Page

A place to learn about statistics

The notes are organized by semester for now.

The Ed teaching platform
- Can host lecture materials
- Notebooks/Rstudio/terminal
- Auto-grading
- Enabled coding in exams with the help of Secure Exam Browser
Teaching without a textbook
Teaching regression from D. Freedman + Simulation standpoint
Final exam was all multiple choies
Exams/questions were made by blending 2+ concepts into one problem to avoid pattern matching
In-class quizzes/surveys were useful for attendance
Completely changed Applied Stat Methods given the background of the students
Tried an oral exam for the final for Applied methods

Students were very active in office hours (they made friends!)
Students who stayed with the course found their lessons very useful in the next class
Some students sent very nice notes
Students could see how the problems were relevant in the future
I enjoyed teaching about wrong models
I think the oral exam final after a final project is a good system!
No grade complaints

Evaluations are hard, they pressure instructors to lower the standard and make courses less relevant to the real world.
Students’ study patterns:
- (some) study first
- (most) hear it in class
- (all) do it
- (more than I thought) read it up again
Students want to have a textbook to follow.
Grading is still painful
Need to standardize the concentration for Stat Department
Students code a lot slower than I thought (5x)
Using Ed for lectures might be too much
TAs were helpful but can be more so…

Applied Stat Computing (90)
Applied Linear Regression (9)
Advised one QMSS student
- are women included on online conversations?
Advised one undergrad researcher
- ranking wrestlers with lots of missing values
Helped out with MA reading group

Wanted to transition from Excel to R using Fisher’s dataset
Wrote a learning R through examples for future references with hopes to strike a balance between examples and CS rigor. Intentionally avoided tidyverse
Lots of datasets (Fisher’s experiment, planting date from text, airport connections, NYC salary, historical corn yields per state, COVID death + confirmed cases, course descriptions from Columbia, NYC 311, job descriptions, twitter, grade distributions from Madison)
Tried Github pages + Ed (exam was tough with Rstudio on Ed)
Advising undergrad on collecting data
Made videos for lecture

First breakout room was a fail (no one spoke!)
Gave up on quizzes midway due to COVID, should have continued
Final projects from regression was below expectation, need to add more real examples
Ed avoids some issues for students (it’s too good for an intro class).

We should restrict the type of R functions allowed on some exams, students have a lot of copy/paste from online which prevents them from understanding the underlying data. A lot of effort is spent in the wrong dimensions.
Maybe it’s better to ask the researchers to mentor the next generation researchers (so 2 semester cycles) as part of the agreement
Making exams take a lot of time
Academic honesty issues were not noticeable, maybe because P/F only.
Some students asked more questions online, some just stopped showing up to class.
Responding to multi-day exams is tiring (Victor recommended letting people ask questions up front)
Might try to teach students to reproduce each other’s reports
Undergraduates need time to ramp up expectations for research (…not sure how yet)
MA students really want to try a lot of different models in research

Teach logistic regression in Applied Stat Methods
- cutoffs
- different measures of accuracy
- cross validation (leave last few games out vs leaving some samples)
Go through a paper slowly… let students attempt the same problem
T-test vs regression (with and w/o intercept)
dogma that more data is better, more granular is better