Re: Re: st: Suggestions for Second Edition of A Visual Guide to StataGraphics

[n j cox <n.j.cox@durham.ac.uk>]

The common structure to this and Fred's report seems to
be

1. A desire to plot results, not raw data. This distinction
is as much the researcher's as Stata's.

2. A structure to the results that does not easily match
the structure of the dataset.

It is #2 that can make graphs difficult. #1 is by comparison
a piece of cake. The assumption that Stata should be smart
enough to understand almost any desire is admirable, but
difficult to achieve until nanotechnology and programming
advance to allow each Stata executable to encapsulate a 
statistically-trained Stata programming-capable cyborg.

I think Fred and Ronan have mostly answered their own questions, but in 
broad terms there are various strategies, and without more
specificity it is difficult to say more than that they could
all be useful.

1. If you are doing the same irregular thing repeatedly,
it can be automated via a do-file or program.

2. Stata gives (limited) support to "immediate" graphs.

3. Add new observations and/or new variables, or make
a new dataset. I've often gazed at one data structure,
some tables of results, and wondered how to get from
the tables to a new structure -- and on more than one occasion
decided the easiest way to do it was to input the results
as new data. Sometimes, although you wouldn't find me
admitting this in public, I just go into the Editor
and start on the right and the top and type in stuff
ad hoc. After doing one kind of thing about three times
I tried to generalise and the result is -textbarplot-
on SSC. (I see from the Ack'ts that Ronan was in at the
creation.) The perhaps odd-sounding name is meant to convey the
structure

              text  baaaaaaaaaaaaaaaaaar
         more text  baaaaaaaaaaar
		
     yet more text  baaaaaaaaaaaaaaaaaaaar
 you get the idea!  bar

(including gaps), so that a hybrid graph-table can
be produced. The bars are not compulsory. The help
is quite detailed, so I won't say more here.

4. Your problem is really a meta-analysis. You just
have get your data in the shape of your favourite meta-program.

In essence, the dilemma is this. You can complain all you
like that sometimes things are too complicated (for you)
to explain to Stata, but that will get little beyond sympathetic
consent. If you can explain any generic problem concretely enough
for others to understand, that either that's soluble
or it's an item for the StataCorp tothinkaboutlist.

Nick
n.j.cox@durham.ac.uk

Ronan

On 25 Jul 2007, at 18:34, Nick Cox wrote:

> I am not clear what the difficulty here is at all. Please give
> a specific example or more detail.

I have encountered the same problem, trying to make a graph for a
researcher. She wanted to plot something for patients with antibody
A, antibody B, either antibody, both antibodies and without either
antibody.

Clearly the groups are not mutually exclusive, and a nuisance to
plot. While Stata is good at groups which are mutually exclusive, it
does not seem to foresee overlapping groups on the same plot -
probably because you can't see them ending up in the same model.

I ended up calculating relative risks and confidence intervals for
each of the groups above, then making a little dataset of them and
using -metan- to display the results.
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/