R vs Gretl

nodoodahs · Post by **nodoodahs** » Fri Sep 05, 2008 3:08 am

Have any of you used both R and Gretl?

I'm interested in impressions – preferably from people who have used BOTH – on comparing the two software packages: learning curve, user interface, quality of output, calculation speed, etc.

Assume that the usage will be along the lines of data mining, linear and logarithmic regression, and generalized linear modeling, and that a maximal data set would not exceed 2 million records at 100 fields per record.

tristanreid · Post by **tristanreid** » Fri Sep 05, 2008 4:42 pm

I use gretl more than R. It's very easy to use, you don't have to remember anything, it's all menu driven. You should download and try it, the learning curve is zilch.

That said, I pretty much ignore the command-line interface. If you wanted to do something more powerful or complicated, my impression is that R would be your bet.

Sorry that there's not much specific substance to my thoughts about this, just impression. The usage you describe is pretty much exactly what I do with gretl.

-t.

nodoodahs · Post by **nodoodahs** » Fri Sep 05, 2008 5:59 pm

Excellent, that's actually along the lines of the response I was looking for.

Does Gretl support using distributions such as the Tweedie in GLM? I'm specifically looking to model pure premiums (loss cost/exposure) directly, as opposed to modeling a combination of frequency and severity.

Thanks!

Maggette · Post by **Maggette** » Fri Sep 05, 2008 9:00 pm

Hey nodoodahs,

sorry, I don't realy know what you mean by "support", but if I got you right, you want to use a general linear model do to some loss modelling using the normal and some other distributions? Then I have to say, that I would recommend R to you. I have to admitt that I didn't use Gretl extensively. I liked it, when I used it to do basic stuff...but for something a little bit "out of the box and away from the standard econometrics textbook" like your problem I would prefer R. It is just more powerful and better to "customize" ...

Cheers

nodoodahs · Post by **nodoodahs** » Fri Sep 05, 2008 10:24 pm

I guess "perform GLM natively with choice of distribution = Tweedie" would have been a better choice of words than "support" - my bad, but we're past it now.

Thanks!