Analogies and heuristics are useful tools when teaching new or difficult concepts. As a part of your course design, you will write a list of analogies for concepts, heuristics for best practices, and any other non-technical explanations of things that may be helpful to students. Read on for tips, tricks, and common questions.

What is the difference between an analogy and a heuristic?

Analogies are comparisons between similar things. In this case, you want an analogy that compares the technical concept in your course, with a concept from every-day life.

Heuristics are sometimes called "rules of thumb". They are practical rules that work a lot of the time.

This is hard. Can I skip it?

You are not alone. Many instructors struggle with this. If you are stuck, come back to this question after you've written the rest of the README. You will need to write something here eventually, since it's very important to try to find non-technical explanations for the concepts in your course.

Write that down!

Many heuristics are the things that you do in your day job, but don't write down because you just know them. Many DataCamp students don't know these things yet, so now is the time to write them down!

Be opinionated

It's okay to have opinions, and to tell the students what they are (and why). Why do you use one package over another? Do you have an aversion to support vector machines? Explain why! Does p-hacking irritate you? Let it off your chest.

I can't think of anything

Examples

From a course on single cell RNA-seq workflows. This analogy is intuitive even without a biology background, and the origin of the idea is cited.

From a course on forecasting product demand. Signal and noise - It's like trying to hear someone across a crowded room. Remove the noise and you can understand easily what they are telling you.

Here's a funny image analogy used to explain Type I and II statistical errors.

An analogy from Albert Kim. R packages are like apps, and CRAN is like the App Store.

IBM heuristic on machine learning model choice. Support Vector Machines work well with wide datasets, such as those with a very large number of input fields.

Computer science is full of heuristics on good coding practice:

  • From The Elements of Programming Style: Debugging is twice as hard as writing a program in the first place
  • From Testing R Code: Most R functions should be 7 lines of code or less.

Siddiqui 2013 provides some heuristics on minimum sample sizes for various statistical techniques. For multiple regression analyses the desired level is between 15 to 20 observations for each predictor variable.

Did this answer your question?