I must agree with the suggestion that any Master's program in Stats include significant exposure to various areas of application, as opposed to concentrations on theory.
I was lucky (?) enough to to attend a Master's program that was just starting at the University of Arkansas. Because of this there was a shortage of courses in the Math department, so I attended Design of Experiments in the Agronomy department, Simulations, Sampling and Advanced Distributions in the Industrial Engineering Department. Because of this, I had significant exposure to a variety of applications. There were also courses in the Psych Department, Education, and Business. Today, there are additional courses in BioStats and other fields. Each of these have some of the same basic statistics with their own little tweaks for the specific applications.
When I counsel new analysts one of the first things I tell them (with a smile) is to forget what they learned in school, especially regarding constraints on how when and how to use various analytical tools. The key is to try to understand what works and you can only find that out by trying various solutions and having the appropriate samples to test the results.
One of the advantages of having an advanced stats background is understanding when and how to use the tools and how to interpret the results. Certainly, understanding the limitations is important; but, one of the best tools a stats person has is common sense, and that common sense comes with experience. Which is why students should be exposed to a lot of data from a lot of different applications AND those examples should include data that has flaws to give the students experience in data cleansing.
As an independent reviewer in the latter stages of my career, I saw too many analysts who didn't know how to look at the initial data exploration to see if the basic stats "made sense." Part of "making sense" is to understand the area of application, which requires becoming familiar with the area of application itself. Do the variable distributions make sense for the are of application? If this is a model, do the bi-variate relationships with the target make sense? In other applications do the preliminary results make sense?
No matter how good your automated analytical tools, you can't teach it common sense. There are no shortcuts.
Building models or completing sophisticated analysis based on flawed data either results in bad analysis or having to do everything from scratch when the final results are garbage (GIGO).
------------------------------
Michael Mout
MIKS
Original Message:
Sent: 01-05-2016 03:16
From: Andrew Ekstrom
Subject: Routes leading to Statistics/Biostatistics
Hey Mary,
I've always enjoyed talking with biostatisticians that came from a science background. Far too many statisticians come from a pure math/stats background for my taste. Most are great at analyzing messy data, yet fall short in understanding where the data comes from and what can be done with it. I've had several discussions with PhD statisticians about elementary ideas from chemistry/science.
Simple ideas for a chemist. Completely foreign ideas for most statisticians. Usually, these conversations end with, "Oh really? Why didn't professor X say you can do that?" The sad answer is, most scientists have less than an AP high school level of training in stats. Most think multiple regression techniques don't exist because, "You can't change more than one thing at a time!"
At the MS/MPH level, most programs require students to all start at nearly the same place anyway. So why not allow biology, biochem, chem students start by taking an intro to biostats course or 2.
It might help too, if your department allows students to work on dual degrees and let's them take cognate courses in their home departments. (U of Mich does this. Don't know about Northwestern.) I might ask, if not demand some of the biostats profs reach out to science departments on your campus. It wouldnt surprise me to find out a chemist that takes 6-8 stats/biostats classes will make an excellent chemometricist. (They'll at least know what they are doing and what others get wrong.)
With the right electives, and an allowance for genetics/molecular biology students to substitute a Bioinformatics course or two for a "regression analysis 5 & 6" class can help attract bio students. (Most of the biostatistics programs by me require 5-6 regression modeling classes. The last 2 tend to cover the same thing as other 4, just in less detail and have the same pre-reqs and only add to student loan debts, not gained knowledge.)
If you can cross list some courses, or count them as equivelent, you might be able to get some econ, psychology, industrial engineering students to do dual degrees too... or at least take some extra biostats classes. (Oddly, industrial engineering students cover many of the same topics as biostats students. They just do it faster with more applications.)
If there was an "applied biostats" track for MS/MPH students, (where they skip a lot of the heavy theory based courses and take relevant other courses) I would suspect that will attract interested science students and help keep them there.
------------------------------
Andrew Ekstrom