There's more too. There's a boatload of research in CHI and Human Factors about input mechanisms. Much of it is crappy. Some of the best was done in the last

There's more too. There's a boatload of research in CHI and Human Factors about input mechanisms. Much of it is crappy. Some of the best was done in the last 60's and revolved around the "McRuer Crossover Model". There's a neat bit of work looking at what kind of models a user has to create in order to control a noisy system efficiently. It turns out that you can cleanly calculate when the task gets too hard. This has since been corroborated with the flight testing of (among other things) fighter aircraft. The bottom line is that it says that the less (mental) work you have to do to, the better you'll do. Each integration is considered work, and by this: touchscreen < [mouse|trackpad] < [joystick|nubbin] With a mouse, you have to think "I have to move it for this amount of time to get to where I want". You integrate its motion to get the final position. Joysticks & pointing nubbins are one step worse: "I have to hold it in this deflection to cause it to move the cursor for this amount of time." Touch typing on a keyboard fall outside this space because there is no pointing/tracking task at all. And, for reference, reversing a two-stage tractor trailer is 5 levels of integration. Getting an oil tanker into harbor is 6!