Reflections of a CS Professor and an End-User Programmer

In my last blog post, I talked about the Parsons problems generator that I used to put scrambled code problems on my quiz, study guide, and final exam. I’ve been reflecting on the experience and what it suggests to me about end-user programming.

I’m a computing professor, and while I enjoy programming, I mostly code to build exercises and examples for my students. I almost never code research prototypes anymore. I only occasionally code scripts that help me with something, like cleaning data, analyzing data, or in this case, generating problems for my students. In this case, I’m a casual end-user programmer — I’m a non-professional programmer who is making code to help him with some aspect of his job. This is in contrast:

• To Philip Guo’s work on conversational programmers, who are people who learn programming in order to talk to programmers (see his post describing his papers on conversational programmers). I know how to talk to programmers, and I have been a professional programmer. Now, I have a different job, and sometimes programming is worthwhile in that job.
• To computational scientists and engineers, which is the audience for Software Carpentry. Computational scientists and engineers might write code occasionally to solve a problem, but more importantly, they write code as part of their research.  I might write a script to handle an odd-job, but most of my research is not conducted with code.

Why did I spend the time writing a script to generate the problems in LaTeX? I was teaching a large class, over 200 students. Mistakes on quizzes and exams at that scale are expensive in terms of emails, complaints, and regrading. Scrambled code problems are tricky. It’s easy to randomly scramble code. It’s harder to keep track of the right ordering. I needed to be able to do this many times.

Was it worthwhile? I think it was. I had a couple Parsons problems on the quiz, maybe five on the study guide, and maybe three on the final exam. (Different numbers at different stages of development.) Each one got generated at least twice as I refined, improved, or fixed the problem. (One discovery: Don’t include comments. They can legally go anywhere, so it only makes grading harder.) The original code only took me about an hour to get working. The script got refined many times as I used it, but the initial investment was well worth it for making sure that the problem was right (e.g., I didn’t miss any lines, and indentation was preserved for Python code) and the solution was correct.

Would it be worthwhile for anyone else to write this script facing the same problems? That’s a lot harder question.

I realized that I brought a lot of knowledge to bear on this problem.

• I have been a professional programmer.
• I do not use LiveCode often, but I have used HyperTalk a lot, and the environment is forgiving with lots of help for casual programmers like me. LiveCode doesn’t offer much for data abstraction — basically, everything is a string.  I have experience using the tool’s facility with items, words, lines, and fields to structure data.
• I know LaTeX and have used the exam class before. I know Python and the fact that I needed to preserve indentation.

Then I realized that it takes almost as much knowledge to use this generator. The few people who might want to use the Parsons problem generator that I posted would have to know about Parsons problems, want to use them, be using LaTeX for exams, and know how to use the output of the generator.

But I bet that all (or the majority?) of end-user programming experiences are like this. End-users are professionals in some domain. They know a lot of stuff. They’ll bring a lot of knowledge to their programming activity. The programs will require a lot of knowledge to write, to understand, and to use.

One of the potential implications is that this program (and maybe most end-user programs?) are probably not useful to many others.  Much of what we teach in CS1 for CS majors, or maybe even in Software Carpentry, is not useful to the occasional, casual end-user programmer.  Most of what we teach is for larger-scale programming.  Do we need to teach end-user programmers about software engineering practices that make code more readable by others?  Do we need to teach end-user programmers about tools for working in teams on software if they are not going to be working in teams to develop their small bits of code? Those are honest questions.  Shriram Krishnamurthi would remind me that end-user programmers, even more than any other class of programmers, are more likely to make errors and less likely to be able to debug them, so teaching end-user programmers practices and tools to catch and fix errors is particularly important for them.  That’s a strong argument. But I also know that, as an end-user programmer myself, I’m not willing to spend a lot of time that doesn’t directly contribute towards my end goal.  Balancing the real needs of end-user programmers with their occasional, casual use of programming is an interesting challenge.

The bigger question that I’m wondering about is whether someone else, facing a similar problem, could learn to code with a small enough time investment to make it worthwhile. I did a lot of programming in HyperTalk when I was a graduate student. I have that investment to build on. How much of an investment would someone else have to make to be able to write this kind of script as easily?

Why LiveCode? Why not Python? Or Smalltalk? I was originally going to write this in Python. Why not? I was teaching Python, and the problems would all be in Python. It’d good exercise for me.

I realized that I didn’t want to deal with files or a command line. I wanted a graphical user interface. I wanted to paste some code in (not put it in a file), and get some text that I could copy (not find it in one or more files). I didn’t want to have to remember what function(s) to call. I wanted a big button. I simply don’t have the time to deal with the cognitive load of file names and function names. Copy-paste the sorted code, press the button, then copy-paste the scrambled code and copy-paste the solution. I could do that. Maybe I could build a GUI in Python, but every time I have used a GUI tool in Python, it was way more work than LiveCode.

I also know Smalltalk better than most. Here’s a bit of an embarrassing confession: I’ve never really learned to build GUIs in Smalltalk. I’ve built a couple of toy examples in Morphic for class. But a real user interface with text areas that really work? That’s still hard for me. I didn’t want to deal with learning something new. LiveCode is just so easy — select the tool, drag the UI object into place.

LiveCode was the obvious answer for me, but that’s because of who I am and the background that I already have. What could we teach future professionals/end-user programmers that (a) they would find worthwhile learning (not too hard, not too time-consuming) and (b) they could use casually when they needed it, like my Parsons problem generator? That is an interesting computing education research question.

How does a student determine “worthwhile” when deciding what programming to learn for future end-user programming?  Let’s say that we decided to teach all STEM graduate students some programming so that they could use it in their future professional practice as end-user programmers.  What would you teach them?  How would they judge something “worthwhile” to learn for later?

We know some answers to this question.  We know that students judge the authenticity of the language based on what they see themselves doing in the future and what the current practice is in that field (see Betsy DiSalvo’s findings on Glitch and our results on Media Computation).

But what if that’s not a good programming language? What if there’s a better one?  What if the common practice in a field is ill-informed? I’m going to be that most people, faced with the general problem I was facing (wanting a GUI to do a text-processing task) would use JavaScript.  LiveCode is way better than JavaScript for an occasional, casual GUI task — easier to learn, more stable, more coherent implementation, and better programming support for casual users.  Yet, I predict most people would choose JavaScript because of the Principle of Social Proof.

I’ve been reading Robert Cialdini’s books on social psychology and influence, and he explains that social proof is how people make decisions when they’re uncertain (like how to choose a programming language when they don’t know much about programming) and there are others to copy.

First, we seem to assume that if a lot of people are doing the same thing, they must know something we don’t. Especially when we are uncertain, we are willing to place an enormous amount of trust in the collective knowledge of the crowd. Second, quite frequently the crowd is mistaken because they are not acting on the basis of any superior information but are reacting, themselves, to the principle of social proof.

Cialdini PhD, Robert B.. Influence (Collins Business Essentials) (Kindle Locations 2570-2573). HarperCollins. Kindle Edition.

How many people know both JavaScript and LiveCode well?  And don’t consider computer scientists. You can’t convince someone by telling them that computer scientists say “X is better than Y.”  People follow social proof from people whom they judge to be similar to them. It’s got to be someone in their field, someone who works like them.

It would be hard to teach the graduate students something other than what’s in common practice in their fields, even if it’s more inefficient to learn and harder to use than another choice.