How to teach sensible elementary statistics to lower-division undergraduates?
statmodeling.stat.columbia.edu
statmodeling.stat.columbia.edu
As much as I love learning about the nuts and bolts of statistics, would the general populace not be better of learning the basics of why (experiment design, independence, covariance, etc) and then the basics of how augmented with tools like probability toolkits that do the math for you?
Because, honestly, the math in statistics can be boring and error prone. I'd much rather a population capable of setting up experiments and using the right abstractions and then using a calculator for it.
And as much as I want to strive for greater mathematical literacy, perhaps that should be sought independently elsewhere. Diving too deep into the math might impede understanding of statistics.
Of course, I don't know where or how you draw the line on who should be taught what...
I'd also tend towards Bayesian approaches too (p-hacking - or equivalents-;is of course still possible, but there are presentations of results (i.e., probability distributions of a result value) which make hacking to an arbitrary value less tempting) which does tend to have a bit more complicated math past the first introduction.
This is how I learned statistics the first time. We started with experimental design and sampling and built up from that. By the time we'd actually started getting to the actual math and statistics we already had a pretty firm grasp of the principles behind what we were doing and why. The math itself by the time we started learning it really wasn't too bad, then we moved onto using R for stastics.
I ended up having to take a second stastics course afterwards, the teacher taught it differently. There were people in my class that hadn't been through the other course. They struggled horribly with it. The material wasn't any more difficult, it was mostly the same material we'd been taught before(my degree was odd, I ended up having to take a few redundant, repeat classes to satisfy credit requirements). The teaching method made a huge difference, I even found some of the second course confusing despite it covering almost the same stuff.
I think if one has a statistics class that's aimed at giving people better statistics literacy it would be a remarkably good fit for a gamified interactive curriculum using an in class currency to make bets/predictions!
You could earn currency the same way you do in games, completing mundane tasks you really don't want to do, like having every homework problem you successfully complete give you a few gold. This incentivizes studying and the effort adds some real emotional value to the currency you wager later.
You could have a whole final fantasy style rpg underlayer where currency is used to buy armor/weapons to make you stronger and help you fight bigger monsters for more currency and advancing in the "game" with the ultimate goal of amassing enough currency through your activities to "buy" your A in the class or whatever.
Then you could have the meat of the class be in lecture scenarios where everyone is presented with a situation that's implicitly meant to test your knowledge of a common gotcha in statistical literacy like base rate neglect in a live statistical simulation. You get to watch as your characters are subjected to "rolls" of the dice that determine their fate based on if you chose A or B. Do you believe the wizard with the diagnosing spell with a x rate of false positive/negative who says you have disease such and such with a y incidence rate?
It's just the basic base rate neglect fallacy scenario, but putting people in a situation where they're incentivized to care about it and rewarded for getting it right.
And by doing these simulations live in class it makes for a lot of spectacle and fun with everyone getting to gamble on these scenarios and then get to see the results play out in real time with all the suspense that gambling normally entails.
You could even have group scenarios where the entire class has to pick option a or b as a whole with everyone's money collectively on the line, and have a heated discussion period where different parties are trying to explain why the class should choose a or b. Imagine being the sole voice of reason trying to valiantly explain to your class how base rate neglect works with everything on the line!
This just seems like such a natural fit to me, and I'm really excited thinking about it.
That seems like entirely too much overhead for a topic that already doesn't have enough time to teach the fundamentals properly. Not to mention that, at some point, you hit a wall of severe diminishing returns on promoting interest through gamification without enormous payoffs. People will either be engaged with the material or they won't. Your class/their grade is simply not high enough stakes for someone not interested in the material to slog through with that extra effort.
What would happen is a couple folks who would have already been motivated will suggest their answer and the rest of the room will follow one of them.
I like the 'pick and choose'-additive model you mentioned. It builds in extra credit along the way, too. 12 weekly assignments worth 5 pts apiece (say 1 question, 1 pt, they can be longer questions), 3 exams worth 20 points each.
Of course, that is merely making explicit a fairly normal system where you just don't know the assignment counts or weights beforehand.
But you could adjust it so that, instead of 12 homeworks, you can do a handful of projects that require deeper understanding. For people uninterested, just getting their credit, they can slog through the simple homework. For those motivated to learn the material more deeply, they can do the harder, deeper work (that hopefully takes less overall time - no need to punish them).
I had some lectures of the rules-of-thumb type and nothing stuck for me. It was all jumping across different approximations, implicit assumptions, "use this formula if 5<n<30 and this other one if n>=30", the whole thing felt very ad hoc.
Bayesian formulations and the general didactic style of machine learning texts did the trick for me. By understanding the Bayesian approach, I can see through the frequentist style much better, things don't seem nearly as scary as before.
People treat statistics as some dark art, when its principles are actually quite simple and the applied techniques can be derived nicely by making your assumptions explicit and knowing exactly where and by how much you're approximating and why. You need to go through some simple proofs of how distributions, e.g. Poisson and binomial become normal in the limit.
Maybe different people are different, but for me learning things at a shallow level is quite difficult. It rather works like I either understand it or I'm just memorizing for a test with a very confused and scrambled mental model. The point of "getting it" or grokking it is often quite sudden, not a linear progression through understanding it. If I stop in the middle, I may be able to reproduce the content of a course convincingly, by knowing what I'm supposed to say and regurgitating facts, but not really being convinced of it deep inside.
1) Mathematics majors should be taught how to prove. 2) Engineering/Applied Math majors should be taught to use.
Fortunately or unfortunately they take the same classes. Personally, I think the obvious solution would have been to just take whichever classes you want to and to get "a degree" once you fill some sort of criterion. But the CAs and actuarial scientists, uhm basically everyone, don't want that human resources headache to actually read through someone's CV patiently and ask a few patient questions.
So you might drop out of either of the two by year two and so you don't have the full four years. [1]
[1] In South Africe, a bachelor's degree is three years. I don't know whether this is a good or bad idea, but that is how it works. Your fourth year is called "honours" and is a separate degree.
Most people just want to calculate a P value or a 95% confidence interval for the mean of whatever they’re researching. They’re not interested in how it all works.
Another child poster pointed out that in special cases (like the limiting case of the binomial distribution) you can get away with other arguments. And you can certainly 'prove' it via simulation. So maybe you don't need the full generality of Fourier analysis in the end.
I would call that a colloquial definition of understanding. Very different from mathematical understanding.
If you ask a mathematician what you’d need to know to understand Fermat’s last theorem, he won’t say “high school pre-algebra.” That’s only enough for you to understand the basic statement of the theorem. It doesn’t get you to the why. To understand the problem involves a deep dive into both algebraic number theory and analytic number theory.
For example, most mathematicians would say they understand large numbers of theorems that they cannot prove off the top of their heads. On the other hand they probably have what Borovik in "Mathematics under the microscope" called a "recovery procedure." That is, a set of constraints or path that reproduces the result. They know they can reconstruct the proof if they need to from the various gambits and skills they keep polished.
Also, the proof via the normal distribution being an attractive fixpoint of convolution is fine, but it only works on a particular subset of functions. We know the theorem applies beyond that subset, and there's a cottage industry of extending it in bits and pieces and calculating better convergence bounds. There is no proof available today that says central limit theorem applies iff conditions x, y, and z. So in this case the proof really can't be said to be understanding.
Now, that's well and good for probability alone, which is a field of mathematics. Statistics isn't a subfield of math, or is a subfield of math the way physics is. For a statistician, understanding the central limit theorem is much more about knowing what kind of observations it is reasonable to expect it to approximately apply to, what kind of tests rely on it and which don't, how to check if it applies in a rigorous way, what kind of visualizations and exploratory data analysis is enabled if it does, how the normal distribution and convergence to it fits into a whole family of distributions and features thereof...
There are a huge number of hot button debates where it is a pointless and uphill battle talking to anyone who isn't familiar with Simpson's Paradox. Pushing ideas like that out into the broader Arts, etc, communities would do a lot of good. A course leading up to that insight would probably make a good elementary stats course.
My degree is in Math but I've been impressed that Stats professional societies take education very seriously. There is widespread, evidence-based, discussion of how to present the materials.
I want to say one more thing, based on prior experience dscussing this kind of thing here: we should acknowledge that some students find this hard. I've been teaching the subject for twenty five years and I really do have some idea of what I am doing. Many students are taking this for their Liberal Studies course and I don't teach at Harvard so those may be factors, but another factor is that different people think differently. For some folks a five-step argument- this is so, and from this we conclude, etc.- is just not how they usually work. I think this course is a help to them, both in developing intellectually and in worthwhile life knowledge, but a person also needs to respect that they are doing something hard.
This is a real book! http://www.lock5stat.com/
As far as societal bang for your buck, I'd give people bullshit detecting classes. I'd focus on this goal specifically, not incidentally.
In my college econometrics courses we used Stata for this, but I'd probably recommend R if you have a choice. The book "R for Data Science"[1] is really good for teaching the basics of data manipulation, graphing, and running regressions. However, it's not a statistics book - you'd need to consider it a "supplement" to teach applied skills. You'd also want to skip the chapters that focus on cleaning data, programming, etc.
> Lower division courses are any course taken at a junior college or community college or courses offered at the freshman and sophomore level at a four-year college or university regardless of the title or content of the course
Titanic: Machine Learning from Disaster
https://www.jwilber.me/permutationtest/
If nothing else, it’s amusing.