This is just a short rant about R's name... as you have already noticed!
Google-searching "R" and whatever else I want brings about 90% random crap results and 10% actual results.
I mean, some things it understands ("R" and "statistical package") but some things completely blow it up-- ("R" and "extract data" or "R" and "get initial parameters")
On a side note, I'm doing some curve fitting here (or trying to learn how to do it on R)-- this is my first "non-SAS curve fitting." I can tell the shape of my curve is definitely a logarithm kind, but not a ln-ln transform. In fact, the best R squared I got with a transform (I know, don't touch the R-squared in non linear fits, but it does help me get ideas informally) was with the X value cubed and the normal Y value. I did also test the "is it a straight line on log-log plot" and the answer was "my data is really fat"... I can't really tell.
It would be fun to find a power law, but I'm just hoping for one. I'll take a regular ln fit, if I need to.
In any case, I am SO GLAD to be doing statistics and not excel right now. Tomorrow and into next week I am going to be working on learning programming in Processing language. Strangely enough, I really like to learn about programming; it's very relaxing, sort of like a puzzle or something. I could make a habit of this.
(Is it sad to say that I much more enjoy doing this curve-fitting than going out with people... it's nothing against the people... I just take much joy in statistics).
Without seeing your plots myself, it's hard to say what might fit well, but a few things you could try are log-normal, stretched exponential, sums of exponentials, power law, and power law with exponential cutoff. If your curve is fat-tailed but not so fat tailed to be a power law, stretched exponential might be a good place to start -- it's a pretty flexible fitting function.
ReplyDelete