st: RE: RE: RE: RE: RE: xtabond and cluster

["Nick Cox" <n.j.cox@durham.ac.uk>]

The way this would work is that 

company_number distid, gen(id_) 

is not recognisable by Stata as a command, but using -cap- as prefix
just eats the corresponding error. But equally no work gets done and the
-id_*- are not created. So -xtabond- complains. 

The code would be better as 

tab company_number  distid, gen(id_)
xtabond Y l.Y X1 X2 id_*, robust

However, Rodolphe made what looks like a telling point about whether
this makes econometric sense, which is a different issue. 

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

Ma, Yu-Luen

Thanks for responding. Below is my program: 

cap company_number  distid, gen(id_)
xtabond Y l.Y X1 X2 id_*, robust

Basically I am just trying to figure out how to create a series of dummy
variables to be included in the model. Thanks.

Nick Cox

I commented on the one command line I saw, which just wouldn't work. 

You may well be right. Perhaps if Yu-Luen were to follow advice and show
exactly what was typed when calling -xtabond- then better advice would
be forthcoming. 

Martin Weiss

"3. The argument to -generate- must be a stubname. If you're trying to
include -id_*- in the argument to -generate()-, that's wrong. The test
of a stubname is that you can append to it and produce legal new
variable names, but * can't appear in variable names.
If a varlist had been legal for that option, the syntax would say so."

My take on Yu-Luen`s question is that this point was quite clear, and
that
the problem arose later when "id_*" was fed to -xtabond-. Indeed, the
code
supplied shows that the -generate- option was used with the perfectly
legal
"id_". The "variable list" mentioned at the end of the post was probably
the
one meant for use in -xtabond-...

Nick Cox

I focus on the dummy question. 

Your syntax looks wrong on various counts. 

1. I assume what's missing here is -tabulate-. The -capture- is
irrelevant to the rest of the question. 

2. -tabulate- can only be used to generate dummy variables when called
with one variable. If you're trying to include -id_*- in the variable
list, that won't work. (But you wouldn't get this error message, so you
can't mean what you say.) 

3. The argument to -generate- must be a stubname. If you're trying to
include -id_*- in the argument to -generate()-, that's wrong. The test
of a stubname is that you can append to it and produce legal new
variable names, but * can't appear in variable names.
If a varlist had been legal for that option, the syntax would say so. 

P.S. I'd advise against the expression "a data" to mean "a dataset".
It's bad enough when people treat "data" as singular, even though many
regard that as defensible. I'd regard "a data" as unacceptable English
in any draft handled by me as adviser, assessor or editor. 

Ma, Yu-Luen

I have a panel data and I use xtabond because my model includes the
lagged dependent variable. My question is: am I able to cluster my
standard errors by subject using xtabond? If not, what is my
alternative? 
 
I also have a question about creating dummy variables. I used the
following command for creating subject dummy variables:

cap company_number  distid, gen(id_)

But when I include id_* in the variable list, I received an error
message suggesting "variable id_* not found". Can anyone assist?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/