Stats Analysis Software Preferences

Although I'd prefer not to trigger a "Holy War" (see, e.g.,  "Word v. WordPerfect" or, perhaps, "Tastes Great v. Less Filling") I invite comments, suggestions, observations, etc. relating to the various stats software packages on the market.  In particular, two specific discussion threads strike me as possibly useful.  First, which packages work the best for empirical legal scholarship and why.  Second, what would you advise a colleague (or student) making their initial stats software selection.

To be sure, products like Stats Transfer make stats software selections far less critical.  That said, I analogize the stats software selection to a law prof's casebook selection.  Due to a necessary front-end investment of time and energy path dependency tends to explain a lot of post-initial decision behavior.  Also, for many (most?) initial software selections pivot on whatever their professors used/required for statistical analysis courses in college or graduate (and, increasingly, law)  school.

Two final points.  First, a fairly thorough and objective comparison of the Stata, SPSS, and SAS packages is found here.  Second, in the interest of full disclosure, insofar as I use both Stata and SPSS on a daily basis I have no "dog in this fight."

Comments

[Andreas Broscheid]

One argument in favor of learning some R is the fact that one can easily transfer data from one format to another. Working at an institution with restricted computing resources, I use an old version of Stata that does not open, for example, the Spaeth Supreme Court database, which is available in a newer Stata format. Since I do not have dbms/copy, I use R to format the database "down" to my Stata version.

[Joe Doherty]

These are all great comments (I'm glad to hear that Datadesk is still around). I agree with Christopher Zorn that Stata>S+/R>SPSS>SAS for the applied user, and I would add the Excel plug-in "Analyze It" to the right of SAS.
Strengths: Relatively inexpensive, easy to use, no need to import data from Excel into another package, generates graphs along with coefficients, free 30-day trial, good for teaching.
Weaknesses: The set of statistical tools is limited (e.g., no logit).
http://www.analyse-it.com/

[frank cross]

I started with SPSS, which I still very much like.

However, I fell in with a co-author who was a proselytizing disciple of Stata and have been converting. Now that Stata has pull down menus, it's nearly as accessible as SPSS, though it retains some annoying features. I think a new chooser should go with Stata, though I'm not sure I would have changed my choice absent heavy lobbying by a co-author

[Tracy Lightcap]

I use STATA for most of the heavy lifting in my empirical projects, but only in the final stages. For dataset formation, exploratory analysis, and initial model testing I use DATADESK. Paul Velleman, who designed this app, is one of the leading lights in exploratory data analysis. Many of the techniques in ILGI (linked interactive graphic interface) programs were originated by Velleman and, usually, are still best implemented in DATADESK. The program is easy to use - far easier than the Big Three - and is the only pure window-based program I've seen that self-documents effectively. Add in an extensive set of macro commands that can be used for repitious analysis and you have an excellent preliminary data analysis tool. There's a student version as well.

And, no, I don't get a commission for this!

[Paul Collins]

For those of you who are new to Stata, or thinking about taking the plunge into Stata, I have posted some lecture notes on data management issues in Stata on my webpage, along with the related dataset and do file. Working with Harold Spaeth’s Original United States Supreme Court Judicial Database, I cover a number of topics, ranging from the relatively simple (e.g., calling data into Stata) to the somewhat advanced (e.g., transforming the unit of analysis in a dataset). Note that I do not discuss statistical modeling, rather, the discussion focuses exclusively on issues related to managing data in Stata. These notes can be found here: http://www.polsci.uh.edu/collins/data.htm.

[Michael Heise]

Christopher makes a helpful point: This discussion is really about assessing comparative strengths and weaknesses. I take his good point as a friendly amendment to my post.

[Christopher Zorn]

It's really not about "better" or "worse," but rather strengths vs. weaknesses. My sense of the four most widely-used packages is:

SPSS:
- Strengths: Widely available, simple to use.
- Weaknesses: Not as powerful as some, costly (if not site-licensed), modular, weak on graphics.

SAS:
- Strengths: Powerful, huge user base (it's long been the de facto standard in applied biostatistics and epidemiology).
- Weaknesses: Hard to learn, cumbersome.

Stata:
- Strengths: Powerful, strong and growing user base (though not as large as SAS), wide usage in the social sciences.
- Weaknesses: Can be costly, not as easy to learn as SPSS (but easier than SAS or S+/R).

S-Plus/R:
- Strengths: Powerful, large user base, strong graphics, free (R, anyway).
- Weaknesses: Hard(er) to learn.

My own view is that, if you're a techie, S+/R>Stata>SAS>SPSS. For a more applied user, Stata>S+/R>SPSS>SAS. Of course, all of them are fine for 95% of what folks here are likely to do, so if you've already invested the time and energy in learning one, it's probably best just to stick with it.

[Jeff Yates]

In the past I preferred using SPSS, largely due to it's intuitive feel and windows commands (I guess I'm dating myself by saying this). However, while it works pretty well in OLS and certain MLE applications, I found that it wasn't as suitable for time series or pooled cross-sectional time series work.

So, I tried out SAS and STATA at the same time. I found STATA to be much easier to use and it had about everything that I wanted. Of course, both packages now come in windows, although I find myself just using STATA commands anyway. STATA also has a listserve (although being on it can yield entirely too many emails) and a pretty helpful staff. On the other hand, I still dont really like it's graphics, at least for certain kinds, and sometimes use SPSS for such things.

Jeff Yates - J.D., Ph.D.
Associate Professor
Department of Political Science
University of Georgia
Web: http://www.uga.edu/pol-sci/yates.htm
SSRN page: http://ssrn.com/author=454290