[Rkward-devel] Comments on design issues

[thomas]

Hi!

>   Though I'm very competent in R, I'm not a C++ programmer, and am not
> volunteering on that end.  I would be interested in discussing design
> issues.  You mention following the SPSS paradigm.  I must say that I find
> that very limiting.  The thing I love about R is that I can have multiple
> objects/datasets in the workspace, including analyses and functions. 

Yes, I mentioned that, but what I was really trying to say is, that the GUI 
should be intuitively useable for people migrating from SPSS. In the 
beginning I also thought, that support for multiple objects might be hard to 
implement, but I have grown more and more confident, that this is not 
actually a problem. So the first release of RKWard (expected soon) will 
indeed only have a single table like SPSS, but this is not meant to stay that 
way. And definitely, when talking about the ease of use of SPSS, I'm neither 
talking about it's horribly limited syntax (as a matter of fact, I came 
across the R-project, when one day I got so frustrated with trying to 
automate a really silly task in SPSS, that I concluded there must be 
something better), nor am I talking about tiny dialogs that can't be 
resized...
One of the things I have grown rather fond of, however, is the ability to 
specifiy meta-information such as long descriptive variable labels, value 
labels, scale level (and here - in contrast to SPSS - I'd be planning to 
actually _use_ that information to assist those users that would like such 
assistance, in checking whether the requirements for a certain test are met), 
and potentially some other fields.
Admittedly, though, another reason, why I mentioned SPSS, is that I have 
hardly any experience with other statistics packages.

> If I
> were designing a GUI, I would make it like that of
>   DataDesk. The basic workspace is more like a desktop with objects.
>   It has the nice feature that results or plots can be clicked in a
>            corner to obtain a list of further suggested steps.  For
>            instance with a scatterplot, you would have the choice of
>            adding a regression line.  WIth  regression output, a choice is
>            to plot residuals.
>         http://datadesk.com/
>
> Also interesting:
>   ViSta   -- charts the analyses which have been run, and makes it easy to
>              go back and modify analyses.  Again it has a sort of desktop
>             of results.
>         http://forrest.psych.unc.edu/research/index.html

These two do indeed look quite interesting, and it might indeed be a good idea 
to integrate some of those ideas. I'll have to have a closer look one day.

>   A few programming questions to illustrate my ignorance:
> Would it not help to use Gtk and the library at Omeghat with connections
> to R?  http//www.omegahat.com

I'm currently discussing a possible fusion with the obveRsive-project 
(http://obversive.sf.net) and there might be quite a few more opportunities 
for cooperation with different projects. When I started this project, I was 
largely unaware, of the other _active_ efforts for an R-GUI. Now one of my 
major goals was to come up with some results quickly and hopefully attract 
some attention from potential developers, when the release is announced.

>   If the goal is to pass data &objects to/from R and an office suite,
> does Koffice have something equivalent to Gnome-bonobo to assist in that
> chore?  I do use Linux, but not Koffice, so I'm curious about its
> advantages.

This advantage might partially be, that I'm more aware of the capabilities of 
KOffice than of e.g. Gnome-bonobo. Most KDE-applications, including all of 
KOffice are available as so-called "parts", which can be easily embedded in 
another application and to a good degree steered from outside. This is what I 
had in mind, but admittedly, this part of the project is still in the realms 
of vague ideas.

>   Is the difficulty with plots the problem of making them editable?  Does
> the xfig device in R create a plot which has editable components?

I had not even though about editable components, though that certainly is 
worth consideration. Rather, I was concerned with a consistent look. 
Sometimes when writing a report you want your figures to look "right" with 
your text. You might potentially want to use figures generated from different 
application, only some of those from R, and still you want the figures to 
look "the same". This can be a very tedious buisiness. At least with SPSS, it 
is just about impossible to get the graphics to look the way you want them.
This is the reason, why I'd rather rely on the charting-capabilities of the 
office-suite itself as much as possible.
But again, this is only a vague idea, and not necessariyl something to worry 
about right now. At first, definitely, "plain" R-graphics will be used.

>   Most of my students use R in Windows, and the copy plots to the
> clipboard and paste them into Word.  You want to be able to update plots
> as the data is modified?

That would be another neat feature (in most, but not all cases), but I had not 
even considered that.

> Congratulations on the t-test example.  I'll look for code releases soon.

Thanks! Release date will probably be Tuesday.

Thomas