Soon, I'll likely teach a stats methods course at the masters level as part of my policy PHD program. If it were up to me, the class would only have one or two assignments determining your grade (putting attendance and other mandatory things that I can't change aside). This class would have lectures/readings, but 0 homework, no quizzes, no discussion posts, no nonsense.
The only assignments that determine your grade in this case are the final paper, and the first draft of that same paper. The paper would be a real paper, where students must collect, clean, and analyze a dataset, applying one of the methods we'd discuss that semester. I view there being several advantages to this.
The first advantage is pedagogical: to me, the only way for me to REALLY know if you understand something is by you writing it out to me and proving to me that you do. People can be nervous test takers. In-class tests have arbitrary times as determined by your class. The professor can sometimes choose questions that're worded unclearly or unfairly. A written paper? That you have many months to prepare about a topic that you choose? No, there's less of an excuse here. Why?
Well, you'll have books and freeware to explain the statistical material. You'll also have me for this purpose. You can also call me, come to office hours, email me for feedback on topics, data sources, whatever. You also have no quizzes or homework to worry about, so the burden my class has can't be used as an excuse for why the draft and final aren't done. In other words, you'll have all the time in the world and all the resources you could ever dream of to ensure the assignment goes well.
I believe it also prepares masters/PHD students for their careers one day where, if they go into academia, they'll be evaluated by their publications. A class that demands a quality paper they conceptualize and write is more representative of what they'll actually be doing in the real world as researchers, generally speaking. I'm not a professor yet, but I've always figured that this would be a good model for instruction (at least for stats methods in the social sciences). I was wondering if this model made sense or not from a pedagogical perspective.
EDIT: I should've specified more the first time, but yes, the degree to which I'd pursue this specific model will depend a lot on if it's a PHD class and if it's elective. If it's elective and PHD, naturally I'll presume everyone has certain working background knowledge (the masters level stats courses including causal inference) and wants to dive further. Now that I think about it more, this model would only be tenable for PHD students interested in methods. For a masters course, the way I've outlined it would be much too cruel of an expectation.
If I did anything like this at the masters level, I would likely only ask them to design a study, likely using synthetic data (that's what we did in my masters course), and implementation would be completely optional. I would also likely have additional assignments since I would presume most people have little to no exposure to the material and would need that to learn better