This objection is not contrived at all.
The problem exists and it bites. Over
several years I have seen it often, heard
about it often, on Statalist and elsewhere, written a FAQ on it and
written
one program jointly (-longplot-, with Zhiqiang Wang) and one
individually (-linkplot-) aimed at least in part at helping
users overcome this problem.
Note that a prior sort on time and -connect(L)- are _not_
sufficient to solve the problem. That's the whole point.
Even in the particular example of growth curves,
with human data at least large datasets will not be experimental,
but will have the imperfections of panel data with dropouts.
Allan's point about -xtline-'s legend is well taken and underlay
my reference to -linkplot-. However, -xtline, legend(off)- is
easy enough whenever the legend is pointless.
Nick
[email protected]
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of
> Allan Reese
> Sent: 19 January 2004 12:55
> To: Stata distribution list
> Subject: st: Plotting multiple (was Growth) Curves
>
>
> On Mon, 19 Jan 2004, Nick Cox wrote:
> > That said, users re-creating graphs for [time series]
> > data from first -graph- principles are likely to be bit
> > by a problem of spurious connections joining together
> > curves from two or more individuals whenever
> > the last data point for one person is no
> > later than the first data point for the next person.
> > This may be rare, but it does occur. The graphical problem
> > is soluble, but fiddly.
>
> That's an interesting contrived objection, and raises the point that
> authors of graphics be sensitive to problems and should
> proofread graphs
> just as carefully (or more so!) as someone writing text
> should check for
> typos, malapropisms and other unintended errors of communication.
>
> The distinction c(l) versus c(L) that enables Stata graphs to draw
> multiple lines within one variable has always struck me as
> a nice, elegant
> feature that gives power when articulating the drafting
> process. The
> suggested problem will not arise in a conventional
> repeated-measurements
> experiment but will occur in a longitudinal study where
> panel members drop
> out and are replaced. Having just tried xtline, one
> difference between
> the two suggested methods is that xtline does not allow
> point labelling
> on the lines, but appears to demand a legend. That does
> not seem feasible
> with 604 lines.
*
* 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/
