Allan Cousins (commenter AllanC on the blog) writes:
Just wanted to let you know that at least 3 other of my fellow students at Sheffield are taking the graduate certificate because they noticed Shravan’s comments about the program on the blog. It’s most impressive because out of a class of 20(ish) you both are directly responsible for 20% of the enrollment! You should probably contact them to negotiate some sort of royalty fee.
Also, I wanted to say thanks to you both as I wouldnt have enrolled without the blog / Shravan’s comments and thus far I am quite pleased with the class. It has been great to this point and the students seem unusually capable (we have some data scientists, former physicists, engineers, etc). It already has led to some stimulating discussions and I certainly hope for more.
Cool! And, no, we won’t be asking for a royalty fee.
P.S. Shravan adds:
The main reason I started doing the graduate certificate was that after reading and re-reading Gelman and Hill 2007 I started to feel that I didn’t really understand even the supposedly basic things (like the one sample t-test).
I also encountered really interesting people in my class too; a transplant surgeon (!, that was a surprise, given that I have been waiting for a kidney transplant since 2011), a pediatrician from Japan, many others. I lost touch with everyone over time, though.
What I really like about Sheffield’s stats offerings is the systematic way in which the grad certificate prepares you for the MSc degree program, and the online MSc program itself. I learnt a lot through their systematic exploration of the topics, and the exams (I had to fly in to Sheffield to do them most of the times) were fun too. Sheffield is an important “Bayesian center” in the UK, so to speak, probably because Tony O’Hagan was there. In retrospect, I wish they had focused more on Bayes and less on NHST, especially in the medical statistics course. There is so much interesting work from Spiegelhalter’s group on clinical trials that could be part of a med stats curriculum. In the grad certificate, I encountered the greatest teacher I ever met: Fionntan Roukema. He teaches the math module. He had a big impact on me.
It appears that Shravan has made several comments about the Certificate on this blog! IIRC the one that got me interested was from an older post in 2017 on recommendations for learning stats later in life: https://statmodeling.stat.columbia.edu/2017/02/23/wants-know-book-read-learn-statistics/#comment-429142
So you went off to Sheffield? (Perhaps share when, why and how often and by email if you prefer.)
I don’t think I would classify it as shipped off seeing as I haven’t left Canada since before Covid (last place I went was India WOW that seems so long ago now)! The certificate as well as the master’s programs are through distance learning. This was the setup even before Covid. The only time one needs to physically be at the university is for the exam period, which is once a year.
I enrolled in the certificate this year with plans to progress to the masters. The reason mostly is that while I feel I have the proper philosophical foundation (in large part thanks to this Blog / P.E. Meehl) I lack the education to be able to evaluate more technical elements of statistical modeling. It’s a deficiency that I have wanted to correct for some time.
Actually, it has been encouraging for me to read accounts like those of Bob Carpenter / Shravan who even post-PhD had trouble understanding certain chapters in BDA. I used to feel as if I had a defect in that regard, either that or I was too dumb for the subject matter. Having incredibly bright individuals share that they understood the words but still felt as if they were reading a foreign language (when picking up BDA) has made me think perhaps I would be capable of more with the appropriate formal training.
When I saw Shravan mention the Sheffield certificate / masters combo, which he specifically said allowed him to understand technical introductions such as BDA, that sold me right then and there!
That is great. Right now I’m studying the GRE quantitative reasoning section. Man, I was like, I need to review my math, some of which I had forgotten. I will do well on the verbal reasoning based on taking a couple of practice tests.
i realized that I don’t have a great grasp on basic statistics and was relying largely on my logical abilities, knowledge of literary criticism, & readings in various fields toward in my understanding of statistics. That’s how I seemed to get Sander’s perspective at least
So after I finish my GRE study, I might get the Certificate. Thanks for sharing.
Well, I guess that means you have to fly to Sheffield. Hopefully there is a meritorious statistics course in US.
Sameera, no, one doesn’t have to fly to Sheffield. One does it online. One does have to go there for the exams, but one can treat it like a holiday with some unpleasant bits (exams) in between.
Shravan, Thanks for the clarification. Honestly, I must have skimmed through that portion of the info.
Some academics characterize me as a generalist or a fox in the sense that Philip Tetlock means it; fox being in contrast to a hedgehog. It got me wondering about what competencies I have mastered. So exploring that for some reason more seriously.
Thanks again.
I temember a post by Andrew long ago on foxes vs hedgehogs.
In high school I used to feel that statistics was so boring. Now I think it should be taught more intensively to everyone, but using simulation. Studying statistics fundamentally changes how I approach my research. It also gave a more fatalistic outlook than was transmitted to me in grad school in the US.
Sameera, BTW I should warn all Americans planning to take the Sheffield classes that the grading is harsh. If you get 60% that is cause for celebration. Once I proved that the Cauchy had no mean, and wrote that it had no mean (or something along those lines), and I got a 0 because I should have said “does not exist” or something like that.
Also, I made extensive notes on the MSc material. It’s all here: https://github.com/vasishth/MScStatisticsNotes
These will be useful to anyone doing their courses.
> should have said “does not exist”
Hmm, sounds more like a lazy tutor who marks anything not matching the solutions wrong.
On the other hand, been wondering why we say that with a continuous distribution the probability of a point is zero rather than saying it’s undefined. If probability is being defined for intervals, a point ain’t an interval…
I think if you define probability measure theoretically through lesbesgue integration, then you use the integral against the lesbesgue measure over a singleton set which is zero.
Thanks, but I do remember studying measure theory.
The point I was trying to make centers on the fact that in applying statistics you never will observe anything that is not actually an interval nor will you ever be able to pin down an unknown parameter value to anything other than some interval. So you will never _need_ anything more than interval probabilities.
One way to view the use of measure theory for statistics then is not much more than a complicated means to define the probability of a point as zero (as well as other weird sets). Then students, tutors and some professors get all hung up on what to make of the zero probability…
I agree with this. I think the primitive concept in probability should be sets, and a probability densities should be considered devices for computing probability over sets. I think the problem arises from trying to transition from discrete probabilities with the classic examples like dice and coin tosses to continuous ones by way of analogy. Usually, densities are treated as drop-in replacements for mass functions, which creates confusion
Regarding Tony O’Hagan: Does anyone else remember 1stBayes? This was probably the first ever package for teaching Bayesian statistics (nowadays we would call it an app). Written in APL and very prone to crashing, I used to use it at the beginning of my graduate Bayesian statistics course. Despite its clunkiness it was a wonderful interactive tool for enabling students to explore and understand how the Bayesian “engine” works. In particular it had a triplot window which showed the prior, likelihood and posterior on a common scale – students could then see the effect of changing prior parameters on the posterior. And this was in the 90s when graphical interactive computing was in its infancy.
A belated thankyou to Tony and others at Sheffield for their pioneering work.
It’s still available – go to http://tonyohagan.co.uk/1b/