[rkward-devel] feedback

[Aaron Batty]

Piggybacking on Meik's feedback comments, I though I'd let you guys know
what's going on in my undergraduate stats class, where we're using RKWard.

1) Installation issues.

I was surprised to see that there is a problem in OSX Yosemite, as most of
my students use Macs (as do I), and I haven't heard any caterwauling from
those who have updated. I know that the one student who actually seems
capable of following screenshot-laden instructions for things in RKWard is
running it under Yosemite seemingly without issue...

The struggle I've run into with many Mac users, though, is just that they
have to install R and then RKWard. This is very unintuitive for people who
aren't used to installing anything more involved than the Facebook app on
their cellphones. I.e., "digital native" undergrads. It would be great if
those could be bundled together again.

With the Windows package, the issue is that it's just a .zip and you can
run it from anywhere. People really aren't used to that, and they don't
know how to add a shortcut to the Start
menu/giant-useless-screen-with-live-advertisements. Also, many people are
confused/freaked-out when a scary-looking black terminal window comes up
upon starting it. The Windows package also seems pretty fragile. Most of
the Windows students have tossed it and reinstalled it at least once. It's
intermittent, and I haven't been able to reproduce it, but it just stops
working at some point. The terminal window comes up, and that's all. I'd
give you guys more information if I had it. Just generally speaking, the
Windows version is not as reliable as the Mac, I'm finding.


2) Data import

This is just rather confusing. You go to File or the Open... menu in the
toolbar, and go to Import, and then there are a bunch of options. You have
"Import Data," then a line, then "Import format," which is a submenu about
importing data. If you use the first one, it just opens up the SPSS data
importer, and there's no way to tell it otherwise. I have started providing
my classes with .sav files for everything to just spare myself the headache
of pointing out again and again, that they actually need to go to "Import →
Import format → Import text / CSV data." It doesn't make sense because the
way that menu is organized, it looks like that is some sort of separate
function from importing data, because it's under the line separating it
from "Import Data."

Once they navigate to "Import text / CSV data," however, the problems
continue, because despite the fact that we've already ostensibly told the
software that we're working with .csv or whatever, there are a bunch of
options where we have to tell it again. The default format is "None," and
the options for the quick formats (which are likely all anyone will need in
most cases, especially in a class where a teacher is providing the data
set) are on the bottom left, looking not-very-important. This could maybe
be fixed just by moving elements around, like putting the format selection
settings in a dropdown menu at the top, labeled with "Please select the
format of your data" or something. Then rename "None" to "Custom" and put
it at the bottom of the list.

Every time I try to use a CSV in class, I import it in front of god and
everyone on the projector, then demonstrate running the analysis, and say,
"Okay, is everyone with me?" and half of the students' hands shoot up and I
go look at their screens and they have the entire table in one column and
the analysis won't run, of course.

The quickest/easiest fix right away for this would be to just kill the
"Import Data" entry, rename "Import format" to "Import data format" and
move it above the line in the menu, separating it from the script options.
Just doing that would at least avoid the problem of people finding
themselves stuck in the SPSS format importer, trying to feed it CSV data or
whatever.


3) Distributions dialogs

I only started using these this semester, so I hadn't noticed, but these
are unintuitive enough that I decided to just get the students to do them
in the R Console.

In the "r" dialogs (i.e., the ones that call rbinom and rgeom, etc.), the
values that will be fed into R are filled in by default. Why? That seems to
communicate, for example, that the "normal" thing to do when looking at a
binomial probability is to set the number of "successes" to 0.95, and only
1 trial, and with a 50% probability, when actually, these are likely to all
be different. I think these should just be blank.

"Vector of quantiles" really ought to be renamed "Number of successes."
I'll be honest here: I couldn't figure out what I was supposed to do here,
so the only way I figured out how to use these dialogs was by clicking the
"Code" button to see what commands it would be calling, and then reading
the CRAN documentation on those functions. The CRAN documentation was much
clearer.

It would be nice if there was another calculator that called the "d"
functions as well... Not that I've really ever used either of these, but
they are pretty important for teaching undergraduate stats.


4) R Console

It would be great if there were a way to permanently set text size there,
or at least have a keyboard shortcut to change it. It's tiny by default,
and the only way to get it bigger is to keep going up to "View" and
clicking "Enlarge font" over and over. Then upon restart of RKWard, it's
back to tiny again. This is another teaching issue more than anything. If I
need the students to do something in R, it's very hard to demonstrate
without bumping that way up, but it takes forever as I keep going up to
that menu over and over until I see the students stop squinting at the
projector screen!


5) Plotting issues of various kinds

I can't seem to figure out how to make a grouped barplot out of the Plots →
Barplot dialog. I've been telling my students to do it via the "N to 1
Crosstabulation" dialog, where it works fine.

A Q-Q/Normality plot in the Scatterplot dialog would be awesome. The only
place I can find that in RKWard seems to be one of the options in the
Scatterplot Matrix dialog.

It would be very helpful if the Scatterplot pulled axis labels from the
Label header, or, if empty, the variable name itself, for the variable by
default. Right now you have to enter labels manually. This results in me
getting a lot of test answers labeled "Xrange" and "Yrange," which I still
give full marks for if they are obviously the right variables.


6) Outputs

Exporting outputs is the area of the most confusion for my students. The
only option for getting things out of the output is to export the entire
giant HTML file, which still just pulls graphics from the invisible .rkward
directory. To get a histogram out to submit for my test (I give tests with
Moodle), the process is:

   1. Export the output to HTML.
   2. Open the output in your browser.
   3. Locate the desired graphic on the giant page of everything you've
   ever done with the software ever, unless you've flushed the output, but you
   were afraid to do that so it's still there.
   4. Right-click the graphic and save it to a file.
   5. Upload it to Moodle.

A big part of this, of course, is that RKWard is running KDE, not the host
OS, really. But this is a lot of steps for an undergrad or
not-terribly-savvy user to complete correctly. Every time I've taught with
RKWard, whether it be to undergrads or colleagues, everyone gets lost here.
They can see the graphic right there in front of them. They don't
understand why they can't just export it directly from the output (they'd
love to just drag it out, but I point out that even SPSS doesn't do this,
and JMP is a pain, too).

Even I, as a pretty confident user of RKWard, find this to be a massive
PITA, and what I do is just navigate to the .rkward directory and keep it
open as I work, sorted by date, and just copy stuff out of there as needed.
But even when I show people this workaround, they invariably MOVE things
out of that directory and then cry, "Where did it gooooooo???" the next
time they open RKWard.

The whole system is extremely unwieldy and unintuitive, and what it means
is that I get a lot of screenshots of RKWard on the tests. Even of the
tabular outputs!

This, I suspect, is a much bigger fix. What I would love to see, though, is
that output HTML file and the associated images be in a folder with the
workspace if saved, similar to how it works in SPSS, but less locked down
and headache-inducing. So perhaps you open RKWard up to just play with some
ideas, as I often do, and then opt to discard the workspace on close. While
you're working on an unsaved workspace, the behavior is the same as now,
but when you close it, the output files are also discarded. If you save the
workspace, however, it creates a folder containing the R and RKWard
workspace files, the same as now, but also a folder containing the outputs.
Loading that workspace later will use that output folder instead. This
solves the problem of an endless scroll of everything you've ever done with
the software because if you even need one thing in there, you can't flush
it, and makes it much easier to find things later. The current HTML-based
output system is wonderful in that it can be opened in your browser and all
the graphics files are already "exported" to PNG, but since they are locked
away in an invisible directory, and necessarily chained to everything else
you've ever done, it makes it very hard to get information out of it.

Last comment about the outputs is one that has already come up here before:
The tables are, by and large, very ugly. There needs to be more of a margin
inside cells, and it would be nice if there were borders by default. Also,
many of them have far, far too many decimal places. The contingency
table/crosstabs are the worst for this. If any of the values need decimal
places, they all get them. So even though most of the cells are actually
just counts/sums, they all have approximately one million zeros behind
them. It's just very hard to read.


I hope this doesn't come across as caviling. Regardless of the various
spots and bumps and blemishes, I've become a pretty noisy evangelist for
this project. It makes R accessible for more people, and even for me, if
there's something that RKWard has a GUI for, that's what I use. It gets me
most of the way to where I want to go, and then the R Console can take me
the rest of the way. It's awesome and I love it. But I'd like it to be even
easier to get others to love it, too.


Aaron
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/rkward-devel/attachments/20141123/d29a1c9c/attachment.html>