Stat applied courses: what's good

| 5 Comments


We would love to hear what parts of the current MS and PhD in Stat curriculum are things that you enjoyed, or found useful or both. Concentrate on the applied Statistics courses, primarily on Statistics 8051-54 if you have taken them. We would love to hear about other courses as well, of course.

Here is what I think what's good about our 8051 and 8052 classes.


  1. Linear regression is supremely important, we do this well.

  2. We do classical design of experiments thoroughly.

Add in your comments below.

Also, just as importantly (perhaps more so), we are eager to hear your criticism. There's a separate blog entry for that, and another blog entry for the kind of material you'd like to see as part of the applied core program, and we look forward to your comments.

5 Comments

Pluses for 8051:

Extensively used R for homework, good textbook, lots of case studies

Pluses for 8052:

Multiple comparison, false discovery rate are really things people care about in industry. Also linear mixed models are useful in real world.

For 8052, linear mixed models, longitudinal analysis, probably can add some survival analysis. I would also be happy to have topics like mixture regression, EM algorithms etc.

Student consultants in the clinic invariably pull out and review their 8052 notes upon starting, specifically the parts on random effects/multilevel models/repeated measures/split plots, etc. This part is definitely working.

Sandy's "grab bag" of 8053 topics: I wish I'd had more of this when I was a student, and several times have reviewed his notes to prepare for clients.

We are doing "classical experimental design" thoroughly? No, we are not; the 8052 course is a race against time to cover the most basic concepts that people need when they get jobs in industry. We are not covering at all notions of A-optimal or D-optimal designs, and the mention of incomplete block designs, mixture designs and response surfaces is at a bare minimum.

In real life, 90% of the problems can be solved really well with classical linear models, think normal errors regression/ANOVA, GLM, mixed effects models. Yes, we want to have a modern curriculum -- but let's not forget to equip our students with a solid knowledge of "classical methods". Using bootstrap to get the p-values just right is important knowledge, but it's not helping much when a split plot is being analyzed as a factorial CR design.

We are doing "classical experimental design" thoroughly? No, we are not; the 8052 course is a race against time to cover the most basic concepts that people need when they get jobs in industry. We are not covering at all notions of A-optimal or D-optimal designs, and the mention of incomplete block designs, mixture designs and response surfaces is at a bare minimum.

In real life, 90% of the problems can be solved really well with classical linear models, think normal errors regression/ANOVA, GLM, mixed effects models. Yes, we want to have a modern curriculum -- but let's not forget to equip our students with a solid knowledge of "classical methods". Using bootstrap to get the p-values just right is important knowledge, but it's not helping much when a split plot is being analyzed as a factorial CR design.

Leave a comment

About this Entry

This page contains a single entry by published on September 30, 2013 1:27 PM.

Applied Statistics topics: A wishlist was the previous entry in this blog.

What's not working in Stat 8051-52 is the next entry in this blog.

Find recent content on the main index or look in the archives to find all content.