Divorcing SAS

[ted00]

I'm pretty dang good in SAS, and I always have been (if I may say so myself ). And let me say this: SAS is a great product; it's become so successful for good reason.

But lately I've consciously been distancing myself from it. I've been doing things in R, mysql, and even C that I would've only tried in SAS in the past. Why? Main reason for me is, I can take free software with me anywhere -- what if one day my employer/client doesn't provide SAS? And I want to be as competent in a SAS-free world as I can.

So, in the spirit of this post , I'm considering a full-fledged divorce from SAS. Anyone got any reasons why I should try to reconcile things with my 'ole sweetheart? Other than that I should use SAS at least sparingly in order to maintain a high level of SAS competency in the event that a future employer actually puts a high value on it, I don't see a lot of good reasons.

I'm very curious to know thoughts of other folks here.

 

[TheEcologist]

I started open-source, and was only forced to use SPSS, SAS and MatLab in undergrad courses. However, R isn't quite as efficient as the ol' mainframe era SAS, thus keep her (SAS) on.. when you hit a R related bottleneck (e.g. Big data), call SAS from R, and make your life a little easier (SAS is easier to develop than C )

 

[noetsi]

Personally I don't really support one product over another. I prefer SAS, or SPSS, because I am familiar with those, had I learned another instead I am sure I would prefer those. As someone who hates to code, although I do, the SAS GUI is a very nice feature as is the code window which captures mistakes quite well. People better at coding likely prefer other products. Another nice feature is PROC SQL which does a complete SQL - not sure if R for example does this, but I have never seen this in SPSS.

Some work places might expect you to work in SAS and, as in my workplace initially not allow R on their system. The advantage to an organization of a product like SAS is that it is a known product. They may not trust an organization which is not corporate in nature - but ultimately I think you should use what you are comfortable with.

I always find amusing the extreme commitment of some to one software other another.

 

[hlsmith]

I suggest polygamy. That way you can have the best of everything.

Happy V-day!

 

[vinux]

SAS is rich and R is intelligent. Choose wisely

Happy V-day!!!!!!!!!!

PS: I agree with hlsmith

 

[noetsi]

I agree with hlsmith and vinux-I am always amused by the fact that posters think the software they use the most is "better" (anything you use a lot is likely better to you).

 

[ted00]

polygamy :yup:

has anyone actually seen how expensive SAS is?? I know someone who bought a personal license for consulting ... I'll always use free software for consulting

I'm trying to pull colleagues off SAS
there's such a hubub about budget
... as much as I love SAS, we frankly don't need it!

so, for me it's not about any kind of fanboyism, it's good 'ole pragmatism

I <3 u SAS, but you're high maintanen$e

 

[ted00]

the SAS GUI is a very nice feature as is the code window which captures mistakes quite well

lots of translated code has a nice free IDE available: R--rstudio; MikTex--TexWorks; javascript--dev tools in IE/FF; all else ... notepad++ (my fav! ... okay, so it's not an IDE, but it's a very great editor under Windows)

Another nice feature is PROC SQL which does a complete SQL - not sure if R for example does this, but I have never seen this in SPSS.

I'm quite swift with PROC SQL ... it's great ... not sure it's "complete SQL", I'm pretty sure no implementation of SQL is "complete" by the various standards ... yes, R+mysql exists

Some work places might expect you to work in SAS and, as in my workplace initially not allow R on their system. The advantage to an organization of a product like SAS is that it is a known product. They may not trust an organization which is not corporate in nature - but ultimately I think you should use what you are comfortable with.

I remember a quote from a SAS exec, something about jet engines being designed (in part) using SAS, and she wouldn't want to fly on a jet whose engineers used open-source software ..... hahahahahaha, pretty sad. fact is, no software is error-free, and with FOSS at least you can open it up if you have suspicions ... god bless dicky stallman

 

[noetsi]

has anyone actually seen how expensive SAS is??

I do, I priced it last year against other products. Beyond being very expensive SAS does not sell permament licenses as say SPSS does. So you have to renew it every year a major additional cost. The only time you will use SAS is in a corporation that buys it (as I have for years).

I doubt any SQL is complete or entirely consistant with ANSI standards. But given that ACCESS was until fairly recently the major SQL (which is absurdly far from ANSI) PROC SQL does fine. MS or T SQL is likely better than PROC SQL, the advantage of PROC SQL is that you can do both it and stats in the same package.

In the end reliability, you know what you are getting including their track record, and ease of coding (you don't have to write as much code) is why SAS is preferred by many large organizations.

 

[ted00]

and don't forget about all the university and government researchers -- there are many -- and they use SAS, too

honestly, I think the big corps et al. keep buying it *because* it's so expensive
someone already made a decision to buy, to back down would indicate that's a bad decision

I guess one thing that irks me the most about SAS is not the product itself, but how so many "analysts" types I know are so completely reliant on it
I know someone (high paid) who couldn't even figure out why proc logistic p-values didn't agree with the confidence intervals for the coefficients ... I'm just sayin', sometimes the easy things make us dumb in the end

 

[noetsi]

They buy it because it is known (and thus seen as reliable) and because it is generate by a large organization with a track record they can check. This is not true of R which is produced by individuals they do not know.

The problem you mention applies to all software not just SAS. Although, in the specific example you mention, I am trying to imagine how the CI could differ from the results of the p values signficantly. Unless they are using the rule that the test is not signficant if the CI contains 0 (the linear regression rule) not that it is not signficant if it contains 1. That happens quite often I imagine.

 

[ted00]

...I am trying to imagine how the CI could differ from the results of the p values signficantly. Unless they are using the rule that the test is not signficant if the CI contains 0 (the linear regression rule) not that it is not signficant if it contains 1. That happens quite often I imagine.

LRT p-vals
Wald CI's
:yup:

There were some coefficients where p<0.05 but CI contained 0, and vice versa.

Very true, this isn't about the software, it's really a different issue.

As for the cost, my friend reminded me about a software package that got sued (unsuccessfully) by SAS for mimicking SAS too closely, and (I forgot the name of it) that software is apparently much cheaper

 

[ted00]

...SAS is easier to develop than C ...

I completely agree with the sentiment, and mostly that's true. Running any regression in PROC ABCD is easier than coding a regression algorithm from scratch. But there are exceptions.

I coded a novel algorithm for some small, esoteric task in C ... there's no function for it in any software package. Choosing between R and SAS, I think R's easier for things like this by a long shot. I used R to prototype this new algorithm, then wrote in C, and now use R as the base system that calls the compiled code. I'm personally convinced that a SAS-only approach would've actually been more difficult than the R+C approach in this case.

But I'm not nearly as hard core as this guy: http://ben.klemens.org/pdfs/klemens-stats_with_the_GSL.pdf. I only scanned through it out of interest after googling GSL, but it seems he's advocating doing all stat analysis from your own compiled code. He makes some interesting points, though, there's a lot of source code available for basically anything we'd ever need in any analysis, it's mostly a matter of putting together the pieces.

 

[noetsi]

Generally if you are doing cutting edge/unusual functions something like R is better than SAS. But few do so which is why SAS remains highly profitable. For what most do, software like SAS is fine.

My guess is that SAS earns more these days from its business information software (which appears to be mainly operations research) than it does its statistical software.

 

[ted00]

I was thinking ...

... there's kind of a "culture" with SAS, isn't there? I thought this while reading an old post from Andrew Gelman's blog after a google search just now

This is obviously just my perspective, your mileage may vary ...
I remember when all I knew in terms of coding was SAS. I felt a little like I was in a box, like if I needed any computation of any sort done, I had to do it in SAS. When I learned R, maybe it's just me, but I began opening up after some time to all sorts of platforms for programming, analysis, simulation, tool development, etc.

Maybe my thought still isn't well-formed, kind of vauge to me even, but it's kind of like if the main tool I use is a "black box", then I can only "think inside the box". I was even generating html files with SAS for goodness sakes :shakehead ... I can only think of about a gazillion better ways to do that now

do you think linux people are a little smarter than windows-only people? It's kind of like that. Maybe they are, maybe not. But the "opening the box" they do with linux, I think, fosters a sort of creativity and productivity that is probably more common in linux rather than windows-only people

So, I gues the freedom-to-move-about reason I gave above is still true (i.e. can use free software no matter where I go or who I work for), but the more I think about it, maybe it goes a little deeper, mabye a little braveheart yearning