Regression software question?

hi roasthawg,

gretl is the GNU Regression, Econometrics, and Time-series library. You can download the source code for free at:

http://sourceforge.net/search/?type_...rds=regression

Interesting and thanks for the reply! Before I download this and mess with it myself maybe you could answer a few questions for me.

1. Is this easy to use? Since all I want to do is run some simple mult. linear regressions I don't want to have to work to hard to learn the software.

2. Can you do forward and backward stepwise automatically?

3. Can it handle a lot of independent variables?

Thanks a lot!

**Dave Head** · 11-17-09, 04:39 PM

hi roasthawg,

I've never used it myself. My experience with downloading free source code has been very good, though. Many of these packages come with executable binaries that don't need to be compiled.

1. Easy to use? Probably not in this case. When I download free source code, I almost always have to re-write it and then compile it. Not that it has bugs, but it's not exactly what I want.

2. Forward and backward stepwise automatically? I have no idea what you mean, and yes, I'm familiar with multiple regression.

3. Can it handle a lot of independent variables? If not, you can just re-write it. You want 72 independent variables? no problem.

By the way, the list of links that I gave earlier was in no way complete. You can do a Google or Yahoo search on "regression multiple download source" or something similar and get a lot more links.

**roasthawg** · 11-17-09, 04:46 PM

Thanks for your help Dave. The gretl seems a little out of my league as I don't know how to rewrite or compile source code. I've done a good deal of google searching without much luck so far... it's hard because I want a developed software package or excel add-on but I don't want to spend much money!

**maxdalury** · 11-18-09, 01:04 AM

Use R its free can handle any amount of variables, but it won't be able to the stepwise automatically (not that I know of), unless you right your own function

**roasthawg** · 11-18-09, 01:57 AM

Originally posted by maxdalury

Use R its free can handle any amount of variables, but it won't be able to the stepwise automatically (not that I know of), unless you right your own function

Yeah, R would take a lot of effort for me to learn... not really worth it considering all I'm doing is linear regressions. I looked into R commander to see if it would simplify things for me but it was all looking too confusing to be worthwhile.

**maxdalury** · 11-18-09, 10:38 AM

R commander is pretty easy to use. once its installed it really is easier than excel, although it doesn't have the full functionality of R but should suffice.

**dcbt** · 11-19-09, 08:39 AM

XLSTAT is an Excel add-in with a free 30 day trial. Not sure on the number of variables or your stepwise functionality, though. But I've used it for logit regression, and it's crazy easy to use, not to mention very fast. The hedge fund I used to work for used this software as well, so it must be decently powerful...

**Peregrine Stoop** · 11-19-09, 04:17 PM

Originally posted by roasthawg

Thanks but it'll only take 20 independant variables... I need more.

cannot imagine a sportsbetting situation w/ >20 independent variables

**Dave Head** · 11-20-09, 01:38 PM

hi Peregrin Stoop,

Originally posted by Peregrine Stoop

cannot imagine a sportsbetting situation w/ >20 independent variables

A typical box score for a football game has at least 50 independent variables (not including individual stats).

**Peregrine Stoop** · 11-20-09, 05:10 PM

I think you need to examine what "independent" means

**Dave Head** · 11-21-09, 12:14 PM

Hi Peregrine Stoop

Originally posted by Peregrine Stoop

I think you need to examine what "independent" means

For example, given:

total return yards = KO return yards + punt return yards + interception return yards

total return yards is not independent
KO return yards is independent
punt return yards is independent
interception return yards is independent

Since there are 2 teams, that's 6 independent variables right there.

Technically, stats regarding field goals are individual stats, but I classify them as team stats. So, I would include stats such as field goal yardage to the list.

**marcoforte** · 11-21-09, 07:09 PM

I use a medical research package called medcalc, $400. You can get a free trial. It does stepwise, backward and forward. Easy to use.

**Peregrine Stoop** · 11-23-09, 03:48 PM

Originally posted by Dave Head

Hi Peregrine Stoop

For example, given:

total return yards = KO return yards + punt return yards + interception return yards

total return yards is not independent
KO return yards is independent
punt return yards is independent
interception return yards is independent

Since there are 2 teams, that's 6 independent variables right there.

Technically, stats regarding field goals are individual stats, but I classify them as team stats. So, I would include stats such as field goal yardage to the list.

KO return yards is not independent from punt return yards
LOL @ using interception return yards for any sort of predictive modeling

I think you need to suss out your model some, or everything you obtain from it will be garbage

**Dave Head** · 11-25-09, 10:45 AM

Hi Peregrin Stoop,

Originally posted by Peregrine Stoop

KO return yards is not independent from punt return yards
LOL @ using interception return yards for any sort of predictive modeling

I think you need to suss out your model some, or everything you obtain from it will be garbage

We're not talking about my model. We're talking about roasthawg's model. I agree that he probably won't want to use all of those variables, but that's his decision.

Please explain the dependence between KO return yards and punt return yards, because I just don't see it. (Correlation or inverse correlation is not a dependence.)

**strixee** · 01-11-12, 01:43 PM

What software would you recommend for 10-15 variables regression? My requirement is that it's for free (14-day trial is OK) and it can spit out something else than just a linear combination. A polynomial function of deg 4 is enough.

**mathdotcom** · 01-11-12, 04:34 PM

R

**bztips** · 01-11-12, 05:31 PM

Stepwise regression is nothing more than a data-mining technique for fools

**strixee** · 01-11-12, 06:03 PM

Originally posted by mathdotcom

R

I came across this software yesterday and it grabbed my attention, so I'll try it.

**uva3021** · 01-11-12, 11:28 PM

stata or R, no reason for anything else

or use some sort of OOP