st: RE: Panel Data, timevar with gaps, Stationary series

 From "Nick Cox" To Subject st: RE: Panel Data, timevar with gaps, Stationary series Date Fri, 14 Jul 2006 17:17:25 +0100

```What is it about analysis that gives rise to these
questions?

Quite a substantial fraction of questions on Statalist
-- and possibly an increasing one -- is loosely of this
form:

People want to use some short-cut, approximation, or
fudge and also to be told that their procedure is "correct".
(Often, bizarrely, "econometrically correct", which
apparently means something different, even if no economics
is anywhere in sight. Do people say
"biostatistically correct" or "psychometrically correct",
depending on tribal affiliations?)

Or they would rather ignore some error, warning, wart,
quagmire, pathology or perversity and also be told that
what they are doing is "correct", or that something else
they are doing remains "valid" nevertheless.

Let's have a new maxim on Statalist.

Analysis of data is an adult activity, with all that
that implies.

What is meant by this is

(a) There is no parent looking over your shoulder saying
"correct" or "incorrect". We are not in a position to judge.
Far from having an Olympian perspective, the other
people on Statalist just can't see your project, let
alone your data, from where they sit.

(b) Fantasy can be fun, but adults are defined by
knowing the difference between reality and fantasy.
Pretend that something doesn't exist or isn't important
if it suits you, but don't kid yourself that makes it
disappear.

(c) What happened to quantitative thinking? Try
parallel analyses with and without to see how
much difference it makes. "Correct" or "incorrect"
is not an absolute. An old joke problem in
Newtonian mechanics starts "An elephant, whose
mass can be neglected, ...". Is it correct to neglect
the mass of the elephant? Possibly, if the other body
in the problem is the Earth. Probably not, if it is
a mouse.

That's quite enough sermon, no doubt.

In your case, I am not sure that I follow all of
this, but the key principle is that you are in charge.

If you are convinced that it makes sense to use time order
defined by something other than calendar time, then that
what is what you should use in -tsset-. -tsset- won't ignore
gaps unless it is led to believe that they don't exist.
So, I too on occasion have used a pseudo-time variable
to get the benefits of panel analysis.

Alternatively, if it makes sense to treat a spell
as a kind of "panel", feel free.

Equally, you should take responsibility for interpretation
and for whatever bending of the truth is implied by your
convenient fiction.

The same applies, but multiplied by some large factor,
to your proposed dodge for stationarity tests. Who
can tell from what you say? Maybe the gaps are not
that big a deal. Maybe different ways of dealing
with the gaps would put you on different sides of some
magic significance level, and then where would you be?

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

Vincenzo Lombardo

> I am just joining the list..for a couple of questions:
> 1) I have a panel dataset; my timevar has gaps (1977...1984,
> 1986 each two years), but i want to tell stata that the
> series have no
> gaps, because i want to deal with spells. How to do? i have
> just created a
> variable "spell" (without using tsspell) and i gave it a sequence of
> numbers (from 1 up to 18), just to set (tsset) my timevar the spell
> variable; could this procedure be correct? I do not know how,
> otherwise,
> tell stata that my series has no gaps..How to do?
>
> 2)then, i have to perform stationary test; one feasible for
> me is the Im,
> Pesaran and Shin (ipshin), but this test only works
> throughout all the
> sample (i.e. using all the time variables) if there are no
> gaps; could I
> use as timevar my "spell" var, above described? Could there be some
> problems, related to the fact that my original series has gaps and i
> replaced it with one without? or better, for instance, i have
> yearly data
>  from 1977 to 1984, then from 1986 to 2004 I have data each
> two years (i.e.
> 1987, 1989, 1991, 1993, and so on), but i set continously my spell
> variable (1...18). could I perform my test?what could be the problems?
>
> thanks a lot for helping me.
>
> --
> Vincenzo Lombardo
> University of Sussex, Dept. of Economics
> University of Napoli Parthenope, DSE
> BN2 4DN Brighton - UK
> tel. (UK):++44(0)7765991711
> tel. (ITALY):++393284164302
> *
> *   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/
>

*
*   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/
```