You probably need to calculate time
since last test explicitly and let
that be part of your criterion for spells,
that is, observations are in the same spell
if time since last test remains under 1
year AND some other criterion is satisfied.
bysort id (dot) : gen time_since_last = cond(_n == 1, 0, dot - dot[_n-1])
Then, as earlier advised, create a pseudo-time variable
so that you can -tsset- and then -tsspell-.
. bysort id (dot) : gen t = _n
. tsset id t
etc.
Nick
[email protected]
Raphael Fraser
> For illustrative purposes, subject 31 would be classified as sustained
> due to Trace amounts being present in 1989, 1990 and 1991. Therefore
> the gaps are included when counting consecutive years. If we had
> another subject with gaps, say
>
> id dot protein
> 5 1989 T
> 5 1992 T
> 5 1993 0
> 5 1994 T
>
> The above subject could not be classified due to the gaps at
> 1990 & 1991.
>
>
> On 10/18/05, Nick Cox <[email protected]> wrote:
> > As earlier said, you need to decide how gaps are to be handled!
> >
> > Nick
> > [email protected]
> >
> > Raphael Fraser
> >
> > > Yes, gaps do exist in the data since some patients did
> not turn up for
> > > their yearly test while others may have turn up three
> times for the
> > > year.
> > >
> > > > In addition, it seems possible in principle that
> > > > an individual could be assigned to two or more
> > > > classes according to different parts of their history.
> > >
> > > This is true. Therefore if a patient is classed as
> minimal, sustained
> > > and heavy, the latter is chosen. If another is assigned
> minimal and
> > > sustained then sustained is chosen.
> > >
> > > On 10/18/05, Nick Cox <[email protected]> wrote:
> > > > Your example indicates that gaps may exist in the data.
> > > > Subject 31 was not measured in various years e.g. 1987,
> > > > 1988. Thus if you want classification according to
> > > > consecutive years you need to specify how missing
> > > > [meaning, not present in the data] values are to be
> > > > treated. Do you mean just consecutive tests?
> > > >
> > > > In addition, it seems possible in principle that
> > > > an individual could be assigned to two or more
> > > > classes according to different parts of their history.
> > > >
> > > > So, I am not clear that writing specific code is
> > > > the best answer to you until these ambiguities are
> > > > resolved. But if you go
> > > >
> > > > . bysort id (dot) : gen t = _n
> > > > . tsset id t
> > > >
> > > > you can look for spells in your data according to
> > > > your stated criteria. The user-written
> > > > program -tsspell- from SSC can be then used. It has
> > > > a fairly detailed help file.
> > > >
> > > > Nick
> > > > [email protected]
> > > >
> > > > Raphael Fraser
> > > >
> > > > > I have a longitudinal data set that contains nearly 500
> > > patients. All
> > > > > patients were tested at these times dot (date of test)
> > > for the level
> > > > > of protein in the blood; the result being 0 (no
> protein) T (trace
> > > > > amounts of protein), 1, 2, 3 or 4. I would like to
> classify these
> > > > > subjects based on the criteria below:
> > > > >
> > > > > "Minimal" if protein is T on at least 2 out of 3
> > > consecutive years.
> > > > > "Sustained" if the result is minimal and lasts 3
> years or more.
> > > > > "Heavy" if sustained with protein 2 or greater lasting 3
> > > > > years or more.
> > > > >
> > > > > id dot protein
> > > > > 31 15mar1985 T
> > > > > 31 14mar1986 0
> > > > > 31 15mar1989 T
> > > > > 31 15mar1990 T
> > > > > 31 15mar1991 T
> > > > > 31 15mar1993 0
> > > > > 31 18feb1994 T
> > > > > 31 07jun1995 0
> > > > > 31 23aug1996 1
> > > > > 31 10may1999 T
*
* 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/
