Re: [Mev-tm4-devel] Option windows for statistics algorithms

[John Q. <jo...@ji...>]

Joe is right about the ANOVA - we should allow it. The reason we didn't
originally was that *everyone* knew an ANOVA with two groups was a
t-test. But we should build it in since it would be an easy fix.

Braisted, John C. wrote:
> Hi Joe,
>
> I think that several of the dialogs could be reworked similar to what
> you propose.
> Even with relatively high resolution the OK button on some of the
> modules
> is still hidden off screen.  The scroll pane is an option if everything
> cooperates and holds its size correctly.  For most dialogs the
> last part of the constructor includes the 'pack()' method.  I'm not sure
> how
> pack() is going to work if it thinks it can compress.  It might just
> take
> some work to get it going.  Probably setting a preferred size for the
> top
> level (enclosing object) object could help.  Also, the 'showModal'
> method
> that launches the dialogs currently just centers the dialog.  Perhaps if
> you
> put the dialog into a scroll pane this method could assess the screen
> size
> and modify accordingly. The JSplitPane seems like a very good idea
> between the
> group assignment area and other controls so that the group selection
> area can be expanded.
>
> *Maybe on each of the stat dialogs we could have a button that launches
> a
> widget for group assignments.  This would be in place of an embedded
> group
> selection panel.  Sort of like the SVM classification dialog.
> This could be launched in its own window.  It would also be nice if the
> grouping
> could be stored in the data structure and new runs of the dialog could
> refer
> to the defined grouping scheme.
>
> Just curious, does Madam make an accommodations for monitors set for low
> resolution that we
> could consider?
>
>
> I favor smaller dialogs rather than scrolling but I'm open to both.
>
> Here are a couple other ideas:
>
> a.) Tabbed Panes to group related controls/selections
>
> The separate JPanels within the dialogs, including the stat modules like
> ANOVA,
> are supposed to contain sets of controls that are related in some way.
> What if these big guys had their dialogs either split onto tabbed panes
> or
> if the options came up on successive dialogs?  The tabbed pane idea
> would be
> similar to EASE where options are split onto separate panes.
>
> b.) Successive Dialogs ... wizard style...  probably not such a good
> choice.
>
> The successive dialogs option is another option where one could present
> the options in a logical order.  Sample grouping (which could be
> reworked to
> produce widget for these dialogs), p-value corrections, ... could each
> be displayed successively. **Has the draw back of not being able to go
> back to check
> or change selections without some strange loop back control.
>
>
> =====================
>
> Suggestion 2 about running allowing 2 group ANOVA.  I'm not sure why
> it's not
> allowed.  Perhaps Nirmal can answer that part.  I suppose if someone
> knows
> about ANOVA they would probably know about t test.
>
> John
>
> -----Original Message-----
> From: Joseph White [mailto:jw...@ji...] 
> Sent: Monday, September 04, 2006 10:02 PM
> To: mev
> Subject: Option windows for statistics algorithms
>
> Hi Guys,
>
> I'm writing this message for 2 reasons: 1) to suggest a change to the
> GUI for the options panels for the statistics and other algorithm
> modules, and 2) to point out what I consider a statistics error on the
> ANOVA pages. 
>
> 1) Being visually impaired, I usually use lower screen resolution to
> make font and images larger on screen.  This means that any panel that
> is not scrollable can, and often does, over-fill the screen.  Without
> scrollbars, it is difficult to get to some of the data fields and
> buttons.  I can re-size the window, but this leads to the problem that
> group  selection for t-test and ANOVA, are not possible because the
> display area becomes too narrow to be useful. 
>
> My suggestion is this: add scrollbars to the options panels.  Also, add
> re-size bars between elements of the same panel so that parts of the
> panel may be re-sized without changing the position of the window. 
>
> 2) On the ANOVA options panel, the user must select at least 3 groups in
> order to do ANOVA. If a user wishes to choose 2 groups, the user must
> use t-test because the ANOVA page will not accept anything less than 3
> groups.  Although, a 2 class ANOVA is exactly the same as a t-test, it
> is still a valid ANOVA.  This should be allowed.  Those that are
> statistically challenged will simply wonder why the MeV authors made
> such a simple mistake and move on to some other program, unless there
> are some directions to use t-test instead--which there aren't. 
>
> I hope these remarks are useful, and look forward to your response.
>
> Joe White
> DFCI
>
>