Spaced repetition

One of the oldest, most robust findings on learning and memory, one that’s been found everywhere from purely verbal skills like language-learning to fine motor skills like surgery, is that if you want to remember something, you should use spaced repetition.

It gets a mention in the related article on interleaved practice, but I’ll go over it again here. Spaced repetition is, basically, the opposite of what we do when we cram for a test. The research simply tells us that we’re better off spreading out our practice sessions over days, weeks, and months. It sounds like more work, but in fact we can get better recall for the same total amount of practice time.

How big is the effect? The best meta-analysis I know of is Donovan & Radosevich, “A meta-analytic review of the distribution of practice effect” (1999), which looks at 116 studies since 1927 and synthesizes their results. There’s also a comprehensive survey of the field by Gwern Branwen, who writes of this study:

According to Donovan and Radosevich’s meta-analysis of spacing studies, the effect size for the spacing effect is d=.42. This means that the average person getting distributed training remembers better than about 67% of the people getting massed training. This effect size is nothing to sneeze at—in education research, effect sizes as low as d=.25 are considered “practically significant”, while effect sizes above d=1 are rare.36

The effect is more subtle as skills get more physically and mentally demanding, and for complex psychomotor skills like piano performance it’s not particularly well-studied. I would guess that this happens because learning music requires some combination of “massed” and spaced practice: you need to do quite a few repetitions in a row to make any progress, but to retain and build on that progress you’d do well to space out your reviews. Because of the limited scope of any research study, you tend to find comparisons that look like “practiced seven times in a row, exactly once” versus “practiced exactly once a day for a week.” Neither of those methods is going to get you to master that Mozart sonata. But the finding itself is so robust that it would be surprising if there weren’t a spacing effect for piano practice done right as well.

So why does hardly anyone do this? Why are we all up late the night before finals, if it’s in fact less work to use spaced repetition for better results? The answer probably has to do with how difficult it is for us to think long-term: why should I study now, when I won’t be tested for two months? It takes discipline to implement a spaced repetition system on your own.

That’s why there’s a colorful offering of software spaced repetition systems out there, designed to offload the discipline and scheduling onto your computer or phone. This is one of many useful ways to “scaffold” your study habits so you can focus on what really matters. My personal favorite program is Anki, an intelligent flashcard system which tests you on flashcards just before it thinks you’re going to forget them. It learns your “forgetting curve” from how well you tell it you remembered the card every time it shows up.

For a while, I used Anki to keep track of my piano repertoire. Because I didn’t want pieces to become so rusty that it would take a week of focused practice to get back up to speed, I let Anki remind me to revisit them at shorter intervals: one “flashcard” for each piece. It worked, but I found it deeply unsatisfactory, because that kind of spaced repetition system was not built for musicians.

Spaced repetition for musicians

Why didn’t the “one flashcard per piece” work for me? Well, there were many reasons. The most problematic was that a piece isn’t a single thing. It’s made up of many smaller parts. So why not make a flashcard for each smaller part? Well, because it’s just as important to think about how these parts combine into the whole.

Even worse, Anki’s spaced algorithm, while it worked for studying for tests in college, never quite felt right to me when I was practicing these pieces. I’d review them too often, or too infrequently. It’s true that the research suggests the exact spacing algorithm is barely important compared to switching from massed to spaced practice in the first place, but the system was just too inflexible. I needed more freedom to choose when and what to practice, but also still to keep track of my decaying memory of my entire repertoire so that I didn’t let things get too rusty.

Most of all, I wanted something where I could combine “learning” with “memorization.” Spaced repetition systems are built with the latter in mind, not the former. But learning and memorization are closely related, especially in music performance. There are layers upon layers, where you need to make one aspect automatic so you can focus on the next. And even though the spaced repetition literature mostly talks about recall, interleaved practice is closely related: it’s basically spaced repetition on the time-scale of a single practice session. Why can’t we combine the two?

In summary: I needed something where I could independently track progress on the subsections of a piece, as well as the piece itself. I wanted more transparency and control over the spacing algorithm. I wanted a system that would work well with good learning and deliberate practice habits, scheduled day-by-day using spaced repetition but guided minute-by-minute with interleaved practice.

A better approach to effective practice

So I came up with Piano Practice Assistant. (This website, by the way, is technically about the Android app, but I consider it much more important to communicate useful ideas about how to practice effectively. That’s why the app gets at most a small mention at the bottom of each article.) It does everything I want, and it’s flexible enough to let me enjoy practicing while encouraging me to practice more effectively. When I add a piece, I tell PPA how many large sections it has, and how many small subsections to each section. Then, when I practice, it chooses a large section and has me practice its subsections at random, switching every three minutes. I then revisit those subsections later in my practice session. That’s all described in the article on interleaved practice.

The spaced repetition wrinkle is this: when I’m done practicing a subsection, I use a slider to guess at how well it’s coming along (my “mastery” in the language of PPA). I then say how easy it was to remember what I improved last time. Over time, my “mastery” decays automatically according to how many times I’ve practiced and how easy it is to remember, although I can edit all of that manually. I can then tell PPA that I only want to practice things I’ve forgotten. As silly as it is to rate my performance with one “mastery” score, I find this leads to dramatically more effective use of practice time, and ultimately to more fun and better music.

Piano Practice Assistant spaced repetition screenshot