During the reign of Stata 7, the most common complaint was about the
absence of publication quality graphics. With Stata 8, there are often
complaints about publication quality tables and regression outputs, with
some people suggesting SPSS as a Stata alternative for tables.
I want to suggest to the Stata community extensions to the way Stata
handles variable labels, as it I think such additions can lead to better
looking tables and other outputs.
I want to suggest 3 new labels as options and additions to Stata's
-variable label- Many of these comments may reflect my personal usage,
but I suspect that there is a generalizability here that may be useful
to all.
-label- may have many uses, only one of which may be the production of a
publication quality labels. In the 2 examples below, the variables names
are continued for historical compatibility reasons and the labels
provide different types of information,
variable name type format variable label
-------------------------------------------------
fatigue_ float %9.0g Sx-fatigue
haq_disa float %9.0g Disability Index
Label 1 tells me that the fatigue variable came from the symptoms
section and label 2 that the variable is one of many disability indexes.
I cant use any of these label for publication, however. So, if I copy a
table using either varname or label, I have to reformat in my word
processor.
What I think is needed is a publication or "table label." This can be
done, for example, by using variable characteristics:
. char list haq[tlabel]
haq_disa[tlabel]: HAQ (0-3)
. char list fatigue[tlabel]
fatigue_[tlabel]: Fatigue (0-10)
I used this in a program (on SSC) called -fsum-:
Variable | N Mean
SD
----------------+---------------------------
Fatigue (0-10) | 6309 4.33 2.88
HAQ (0-3) | 6270 1.08 0.72
The output is publication ready.
These table labels, however, are not of much use for column labels in
tables, as they are much too long. In Stata, the -list- command contains
an option to display as a column label the text in char
varname[varname]. With the help of Nick Cox, I wrote a program called
-corrtab- that will be placed on the SSC Archives on Kit Baum's return
next week. This program, that display correlations, is an example of the
use of tlabels (table labels) and clabels (column labels) together. Here
are some examples.
clabels in columns and rows
. corrtab haq pain glb age sleep,v(3) all
Pearson correlations
+--------------------------------------+
| Variable HAQ Pain Global |
|--------------------------------------|
| HAQ 1.000 0.609 0.591 |
| Pain 0.609 1.000 0.664 |
| Global 0.591 0.664 1.000 |
| Age 0.123 -0.042 0.028 |
| Sleep 0.411 0.505 0.507 |
+--------------------------------------+
tlabels in rows, clabels in columns
. corrtab haq pain glb age sleep,v(3) t c
Pearson correlations
+------------------------------------------------------+
| Variable HAQ Pain Global |
|------------------------------------------------------|
| HAQ (0-3) 1.000 0.609 0.591 |
| Pain (0-10) 0.609 1.000 0.664 |
| Global severity (0-10) 0.591 0.664 1.000 |
| Age (years) 0.123 -0.042 0.028 |
| Sleep disturbance (0-10) 0.411 0.505 0.507 |
+------------------------------------------------------+
Notice that the clabel is short and serves as an identifier rather than
being very informative.
A third type of label is a graphics label. It usually differs from other
labels for a variety of reasons.
So, I suggest the label extensions that are carried out by variable
characteristics:
tlabel
clabel
glabel
The labels suggested are useful for people who repeatedly work with the
same set of variables. In addition, they give control over the output.
They would not replace Stata variable labels, but would be extensions.
If Stata were to adopt extensions like these, it might be an additional
step toward better output throughout its many programs. I could see them
being used in various regression commands.
I don't know if Statalisters think this is a good idea, but if they do
it might be useful to develop a consensus regarding what kind of
extensions there should be and into which char they should be placed.
Perhaps these comments might stimulate discussion on the issues of
publication quality output and how it might be accomplished.
Fred Wolfe
Fred Wolfe
National Data Bank for Rheumatic Diseases
Wichita, Kansas
Tel (316) 263-2125 Fax (316) 263-0761
[email protected]
