[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: ci question
"Nick Cox" <email@example.com>
RE: st: ci question
Thu, 10 Apr 2008 23:20:19 +0100
When I want to do something similar, I use tactics like these:
1. Reduce (e.g. -collapse-) the dataset to what you want to show.
2. Alternatively, typing the results into new variables, or even a new
dataset, may be quicker and
easier than working out how to automate that, especially if your graph
design is a little unusual.
3. Although there are exceptions, using -twoway- is more flexible than
-graph bar-, -graph hbar-, or -graph dot-, especially for combinations
elements (most obviously point symbols plus bars or other intervals).
4. Overlaying is your friend.
5. An axis for categories can be constructed by working out a simple
A group of three, another group of four, then a group of two might be
plotted against 1 2 3; 5 6 7 8; 10 11.
If you want finer control of spacing, use larger integers, not
The -egen- function -axis()- from -egenmore- on SSC is one tool in this
area. A better tool called
-seqvar- is in preparation.
6. Once the spacing is good, attach text to the integers as value
labels. -labmask- from -labutil-
from SSC is one tool here.
7. Despite tribal conventions in many fields, text plotted horizontally
is easier for the programmer to manipulate
and the reader to follow.
Something in this territory should appear in Speaking Stata in Stata
Journal 8(2), or so I gather.
I'd like to be able to have multiple levels by which the plots are
separated/grouped along the x-axis (like the "over" command in "graph
box" allows). For example, using eclplots, I'd like to be able to
include two "parmid_varname" variables....or using ciplots, I'd like to
include two variables in the "by" option.
Now, I've also plotted this data by overlaying rcap plots with scatter
plots of the means. Using this plotting method, is there a way to
horizontally offset the rcap plots within a single x-axis value so that
the rcaps do not overlap? This might be an acceptable presentation for
me as well, although it would be better if I could do this through
ciplot, eclplot, or serrbar.
* For searches and help try: