Re: Evaluating PSPP

[Andy Choens]

Several people have responded to my original request for information. 
Thank you. I will try to respond to what people have written below:

John Darrington wrote:
> If you say what your company's needs are, then it'd be easier to
> answer that question.  Other uses have identified repeated measures
> anova (aka GLM), factor analysis and reliability analysis as things
> they need, so these are probably going to be implemented in future
> releases.  I personally have an interest in non-parametric statistics,
> so you might see a few more of those too.
>
>   Also, a better output system is on the cards
> ( http://jstover.motd.org/cgi-bin/oddmuse.cgi/NewOutputGoals )
>   

Like most companies (I assume) we have different "levels" of users. Most 
of our SPSS users need to be able to open .sav files and generate 
tables, cross-tabs, etc. for use in their reports. For these users, I 
think the limited graphical output system is the biggest problem I can 
see with us using PSPP. Being able to output the results to CSV (for 
further analysis) and HTML (for cut/paste into a report) are both really 
really important.

Other users, such as myself and a couple of others, write syntax and do 
most of the actual analytical work. My existing syntax does use MANOVA 
in a few places, which I can see is one function that PSPP does not 
currently support. Factor analysis and reliability are also used quite 
heavily. Based on your comments, I think PSPP is headed in all of the 
right directions.

I would also love to see syntax highlighting / line numbers in the 
syntax editor. Since SPSS syntax is not something that has traditionally 
been useful on a linux box, none of my favorite editors support syntax 
highlighting and I have tried to use/maintain a vi syntax file, but it 
would be nice to have something more complete/integrated.

I also do some GIS work, but I've been doing more and more of this in 
R-CRAN and that will probably not change in the foreseeable future.

> A python interface is something that I think should definitely be
> done.
>
>   
I agree. As SPSS users familiarize themselves with Python I fully expect 
SPSS syntax to increasingly rely on Python, which will make for some 
interesting looking syntax. Any mixing of PSPP syntax and python will 
make syntax highlighting increasingly important.


> With regard to R integration,  most people seem to suggest that
> spss/pspp could be a front-end to R's engine.   That might be useful,
> but I'd be much more interested in the opposite; would it be possible
> to use R's syntax with PSPP's back-end?   A few simple tests should
> convince you that PSPP is magnetudes faster than R, so I'm not keen on
> the former option.
>
> J'
>
>   
When I mentioned R-CRAN on this thread I was a little worried that I 
might get flamed (vi v. emacs anyone?). Fortunately, I can see that this 
isn't going to be an issue. I myself have not done any speed comparisons 
between R and PSPP, but speed isn't why I mentioned R-CRAN. For what it 
is worth, R-CRAN is an important tool in academic statistics. R already 
has far more features than SPSS-base or PSPP. SPSS appears to have 
developed an R plug-in for newer versions of SPSS to let advanced SPSS 
users take advantage of these tools.. I don't think they have any 
intention of ever replacing their engine with R's. Nor do I think PSPP 
should replace it's engine with R's. The PSPP FAQ does a good job of 
explaining why this wouldn't be a good idea.

That being said, I think a plug-in could be useful for two reasons.
1) If SPSS does it, users like me will eventually get some syntax from 
someone that relies on the R plug-in. I'd like to be able to run this 
syntax in PSPP.

2) R-CRAN has a wealth of useful tools that PSPP does not have. There 
are some domains, such as GIS, that PSPP is not really well suited for. 
If I can use R via PSPP, then I can save myself having to output my data 
structure from PSPP and import it into R, and let the plug-in do all 
that for me. This is time effective AND less error prone. I think there 
are some good arguments for giving access to R's functions via the 
syntax/command-line but not via the menu system.



I do agree, it would be really neat if we could use R syntax in PSPP.


--andy

--
Andrew Choens, MSW
Research Policy Analyst
Hornby Zeller Associates, Inc.
(518) 273 - 1614