Maybe for typewriters, no way that math works out for fast typists though. This exercise indicates my time-cost-per-error is roughly equivalent to typing a word; the cognitive cost of leaving the site of the error can't possibly be lower than that.
It works for most of the traditional forms of typing (audio typing, typing from shorthand, etc) where you have a body of stuff that needs to be typed and not so well for stuff that programmers type where you're creating as you type.
You must check the text for errors when you've finished typing it so it makes sense to do all the error checking then. Some people find the cognitive focus switching of correcting errors as you type tricky.
Very good point. I think my argument still holds for brainstem-layer errors, where I know as I'm pressing the key that it's not what I meant-- it's hard for me to not correct those, anyway.
For the kind of typos that you would need to be watching the output to correct, it totally makes sense to batch those up sometimes.
