There has actually been some research on this. I didn't save the link, but the paper I found showed that time spent reading code is a much stronger predictor of skill development than time spent writing code.
I'm guessing in practice it's probably comparable to music and language learning, which seem to work similarly in this respect. You do need to spend time practicing by doing, but having that be most of your skill development time is sub-optimal for people who aren't already at an advanced level. Most your time should actually be spent carefully observing what the experts are doing. Because that's what builds the intuition you need to critique your own performance. Which, in turn, acts as an enormous force multiplier for your active practice.
That's arguably why contributing to open source projects is such a great method. Because, unlike drilling yourself on practice problems, the vast majority of time spent contributing to an existing project will be spent reading and understanding the code that's already there.
I'm guessing in practice it's probably comparable to music and language learning, which seem to work similarly in this respect. You do need to spend time practicing by doing, but having that be most of your skill development time is sub-optimal for people who aren't already at an advanced level. Most your time should actually be spent carefully observing what the experts are doing. Because that's what builds the intuition you need to critique your own performance. Which, in turn, acts as an enormous force multiplier for your active practice.
That's arguably why contributing to open source projects is such a great method. Because, unlike drilling yourself on practice problems, the vast majority of time spent contributing to an existing project will be spent reading and understanding the code that's already there.