A coauthor and I just recently submitted a revision of our manuscript to a journal. If we’d known it was going to be so much work, we probably never would’ve written the paper in the first place. . . . It’s a surprising amount of work between idea and execution (even forgetting about issues such as writing the letter in response to the referee reports). And, actually, this particular review process was very easy, as such things go. Still a lot of effort, though. It reminds me that being able to something once is a lot less than describing a method clearly and in appropriate generality.
The gap that you talk about – from doing something once to being able to describe it clearly and in appropriate generality – is what ultimately separates those that influence from the masses.
Many of us have the ideas and general thoughts on how things can be done better. But it requires equal doses of patience, drive and clarity of thought to put that down into a format that can be shared with multiple people and used to change their opinions about things. You have that rare gift, Prof. Gelman. So don't kill what you are good at and what the world needs to hear from you, even if it kills you!
being able to something once is a lot less than
Not quite sure what you mean; do something? describe something? … and a lot less what?
@anon, I think we can be pretty sure that the sentence is supposed to read "…being able to do something once is a lot less effort than describing a method clearly and in appropriate generality." But I think you knew that.
Andrew, dude, you've been sitting on this great Philip K Dick reference for five years, and you waste it on this post for which it is barely appropriate? Aw, man, what a waste!
no, not quite.
Could be "do something once", or "implement something once", perhaps "derive something once"… the subsequent "appropriate generality" tag made me wonder.
For the other one, "less effort", "less time-consuming", "less difficult"? They carry different nuances.
The exact same thing goes for programming. It's relatively easy to code something once for a particular data set shape than it is to write something that's generally reusable.
I agree with Krish that it's really important to generalize if you want your papers or software to have broader impact.
The problem's often what seems to be a tension between generality and simplicity (and efficiency [fast] and scalability [large data sets] in programming). With enough "drive, patience and clarity" as Krish put it, you can often achieve both.
what a can of worms you're opening up! This is a major, major issue in data mining. No one has shown that a given predictive algorithm is superior over general classes of data sets. you have a proliferation of many algorithms all claiming to have achieved small incremental (often practically meaningless) predictive improvement on specific data sets. Those who try to claim general superiority typically resort to the dozens of toy data sets publicly available (boston housing data, etc.) but neither are those data sets representative of the big picture. Are there general algorithms that are dominant over others? Are we chasing after something nonexistent?