Join 3,497 readers in helping fund MetaFilter (Hide)


StatFilter: the time vs. money test
October 11, 2010 10:14 AM   Subscribe

I am a grad student in a social sciences field who needs to do some "lightweight" statistics work now and then. I used SPSS for a course last year, and it was dandy, but now I've lost my school access to the software. Should I suck it up and buy my own copy of SPSS ($200) or try and learn R?

I'm doing very basic stuff -- a lot of descriptive statistics, some chi-squares, t-tests, one-way ANOVAs. A lot that can be done with Excel, but I find it a little clunky. I am trying to weigh the time it will take me to learn (free) R vs. spending $200 for SPSS which is a lot more tool than I probably need. But time is such short supply - I could hit the ground running with SPSS.

Any suggestions on R vs SPSS for simple stuff?
Any thoughts on SPSS 17.2 (which I can get for $99) vs. SPSS 18?

Thanks!
posted by pantarei70 to Education (12 answers total) 8 users marked this as a favorite
 
Another alternative: look for an equivalent of R or SPSS in a programming library. I can't help you there though, sorry.
posted by devnull at 10:23 AM on October 11, 2010


Do you know any students who aren't going to use SPSS and leverage their student discount?
posted by calistasm at 10:29 AM on October 11, 2010


I use R and I am definitely not a maths sort of person. For the kind of stuff you're describing, I don't think R would be too hard to learn unless you're seriously pressed for time, and in the long run I think it's probably a more valuable skill. R is great because you have a lot more control over how you set up the statistical tests (though obviously this can be a double-edged sword), and also it gives you much more specific output, rather than the pages and pages of stuff SPSS spits at you.

The documentation for R can be a pain to figure out if you're not a programmer, but there are textbooks available (some even online for free, e.g. Baayen's Analyzing Linguistic Data) that describe the whole process you'd need to take data through. (Also I have some tutorial stuff that I wrote up when I was first learning it, memail me and I can send it to you if you want.)
posted by SymphonyNumberNine at 10:31 AM on October 11, 2010


You may want to at least check out this informative list of free stats software and maybe take one or two for a quick test drive before deciding whether to spend $200.

PSPP is said to be very similar to SPSS, so you may be able to leverage your familiarity with that.
posted by philipy at 10:40 AM on October 11, 2010


There's also pspp, which claims to be a FOSS workalike to SPSS.

A lot of this stuff depends on what field you're going to be doing light statistics in. In my own field, I'd advise against SPSS as there's been a long-term but very firm shift away from SPSS towards Stata and R, but in other fields SPSS is dominant.

The $99 or $200 isn't significant. What's significant is what people in your field are using; you want to be able to walk down the hall and ask someone "How do I do WHATEVER in PACKAGE?" and not have the answer be "Jeez, I dunno. Nobody uses that." In that sense, even if you're going to be using it for light statistics, you'd probably find it useful to use whatever the quantoids in your field tend to use.
posted by ROU_Xenophobe at 10:41 AM on October 11, 2010 [2 favorites]


I suggest R with giant helpings of Google. As SymphonyNumberNine mentioned above, the docs for R are almost uniformly atrocious, where they're even present. But for any simple thing you want to do, a thousand people have already asked for and received help on how to do it online.
posted by gurple at 11:59 AM on October 11, 2010


What ROU_Xenophobe said. And I'll suggest another factor. If you can pay $100-200 to shave a week or more off of completing your thesis, it's money well-spent. Although doing some digging around, I'm a bit disturbed that I can't get better than quote for a six-month lease for IBM SPSS without providing contact information.

If you do go into R, take a good look at one of the GUI interfaces that simplify data entry and running basic tests. R Commander and JGR are under active development and seem a bit more mature than Deducer. It might be worth an hour to kick the tires in order to see what the learning curve might be like. I believe the Mac port comes with GUI system as well.
posted by KirkJobSluder at 12:00 PM on October 11, 2010


ROU is right; what the people around you use is what matters. If time matters, use what you already know. Personally I think that R is a better bet in the long run, but the long run can start after whatever is providing your time pressure.
posted by a robot made out of meat at 12:59 PM on October 11, 2010


Thanks for all the good advice.

The "what people around me" is good advice. I'm in information studies (the discipline formerly known as library science) and we kind of tend to be all over the map. The education-related people stick with SPSS, a lot of sociologist types like SAS, and the quantoids in my school (who also tend to have CS degrees) love R.

So, I'm giving R at least a little shot. I'm a year out from dissertation-writing, so there is some time to try it out on this project and go spend $200 later if it comes to that.
posted by pantarei70 at 3:28 PM on October 11, 2010


You may want to grab some books or develop a nice set of web-resources on R.
For books there's R for spss and sas users, which started as a free 80 page pdf. There's also Data Management using R and SAS which has a great blog as well.

One more link.
posted by stratastar at 10:18 PM on October 11, 2010


Why not just use excel. For most basic statistics excel is good enough.
Link will tell you how to do anova/ t test etc using .. excel

Otherwise there are number of addins for "relatively" cheap that can enhance on the basics.
Here is one called Sigmazone $250 license (http://www.sigmazone.com/spcxl.htm)
I use it at work and it can handle histograms,control charts and various tests even obscure distribution like weibull etc.
posted by radsqd at 12:49 PM on October 12, 2010


Go with SPSS.

Older versions are fine (I'm still partial to v12) but somewhere around v17 they switched to their new Java-based platform and that completely screwed things. PASW18 is much more stable.

As KirkJobSluder says, this is one of those few times when you can purchase some sanity for yourself. You could try figuring R (which I've used for clustering) but in the time it takes for you to do that, you could make some good progress with SPSS. The grass might look greener on the R side of the fence, but I would pay the gardener and buy SPSS instead.

Sorry, that metaphor doesn't really work but the rest of the advice is sound.
posted by Sutekh at 6:46 AM on October 13, 2010


« Older I'm asking this question on be...   |  What is the most relevatory, s... Newer »
This thread is closed to new comments.