A couple of weeks ago, I taught a one-hour workshop on the basics of writing functions in R. Overall, it went as well as I would expect an hour introduction workshop to go, which is to say OK but not great. Why not great? It turns out teaching skills in workshops is really difficult. But why?

I think it boils down to a couple of aspects of workshops:

Workshops:

1. are too short
2. teach to the lowest skill level in the room
3. don’t reflect how you actually learn these skill

1. Workshops are too short

Ask around, and you will find that almost everyone loves the short format of workshops. Life is busy and time is precious. Asking for a four week commitment is a lot, but ask for one hour out of someone’s afternoon easy.

However, what can you cover in an hour?

Turns out, not much. When I teach, I typically assume that you can drive home one minor idea or concept per hour of instruction. As for major themes or ideas, I figure you can convey 4-5 in a typical 3 credit hour course. After talking to many other instructors and professors, I don’t think I am alone in this assessment.

So, if we assume we can get a single minor concept across in an hour workshop, what is the issue? We know that we are not going to teach all of R, but giving an introduction to writing a function should not be an issue, right?

Well, the trouble comes in with #2… `

2. Workshops teach to the lowest skill level in the room

Let’s say we are teaching the single skill of “writing R functions” during a workshop. It is a simple task, well within the criteria of “one minor idea per hour”.

But as we give the talk, it turns out there are a range of experiences in the room. It turns out some people have not really been introduced to functions at all, so we cannot teach how to write functions without giving a brief overview of what functions are. Some people have never been introduced to R’s method for using and finding variables, so we need to also introduce that. Related to that, most people don’t know about scoping, and get confused about how variables could be local to a function, or why global variables are a bad idea. Lazy evaluation? Debugging? The list goes on and on…

When I teach a class, I can start at the fundamentals. Once we get to writing functions, the students should be comfortable with the basics of the R language. The foundation has already been built, so adding to it is an easy task.

With workshops, all bets are off.

If you look at most workshop series (e.g. Software Carpentry workshops), they just don’t bother to teach “fundamental” skills. (Note, do not confuse “fundamental” with “basic” - fundamental skills are skill which form the foundation for deep understanding.) But, the reason this becomes an issue is …

3. Workshops do not reflect how you actually learn a skill

Find an expert in some data/software skill, and ask them how they became an expert. I would bet good money that none of them became an expert by attending workshops or reading a book. Why? Because to transition from novice to expert, you need to apply the skills to a non-trivial problem. You need to hammer away at something that is difficult, work through issues with the tool and your understanding of how it functions, and come out the other side with hands-on experience to be an expert. It is the reason why most companies give problem sets to new data science hires. Many people attend workshops, few know how to use the skills in practice.

But you cannot solve a difficult problem in a workshop. Doing so would require time that you just don’t have. Instead, in a workshop, you are presented with a skill and you might get to see a toy problem or two, but you never actually apply the skill. It is extremely common for workshop participants to leave feeling good about “learning” a skill, only to be completely unable to apply it in practice.

So, what’s wrong with that? Well, when you tackle a difficult issue, you need the fundamental understanding. You need to be able to reason about what the language is doing and why. Otherwise, you end up in the extremely common workflow of “I have this script, and I will modify it with half-educated guesses until it appears to work”. And yes, I have seen many many workshop attendants in this mode of working. It doesn’t really work.

How do I know this? Because the only reason I see these people is that eventually these workshop “orphans” reach out to me for help, having exhausted the resources provided by the workshop.

So, why are workshops so popular?

I have heard many reasons, including time constraints and ease of scheduling. But I think there might be a deeper issue at play.

Workshops have been created to optimize the wrong thing. Since it is probably impossible for a workshop to teach fundamental or deeper understanding, most workshops aim to produce two things: 1) capable beginners and 2) a sense of accomplishment. Most participants of workshops are happy at the end. They feel empowered. And there is nothing wrong with this, per se. But we also should not delude ourselves into thinking workshops will produce a cohort of skilled data scientists.