st: re: programming with Stata

[Kit Baum <baum@bc.edu>]

Tunga said

Thanks for your answer Maarten.
There are really not many people who can compare ox and mata languages.
Therefore I would like to further the question:

Is mata a clean language, as ox? What possibly would I be missing by not
going for Mata?
What would make me choose mata instead of ox? Or, so many people use  
matlab.
Why should I go for mata instead of matlab?

My question may not be easy to answer because I am simply asking an
objective comparison of programming languages.
There are articles which compare ox, gauss, s plus etc...
But stata is a different issue here. As one uses Stata's higher level
language capabilities,
it makes little sense to switch to another program for lower level
languages, if mata is as good as gauss or ox...

Also I don't really get objective advises. People advise one over other,
without giving clear reasons.
Perhaps this is because they are not able to compare, but they don't  
mention
this and advise the one they know.
This is not an advise I am seeking.
I will invest time into this. And I want to know for what I am  
investing my
time.
I want to know what I would be missing by leaving other options away.
Or would the differences be less than I am thinking of?


I have no knowledge of GAUSS (other than that the functions are  
exceedingly unpredictable in their calling sequences)-it has not  
been, generally,  platform-independent. I have programmed in Ox for a  
number of projects, and done some MATLAB and Mathematica programming  
as well as extensive work in Fortran. Mata is IMHO an excellent  
solution for anyone needing the lower-level facilities of a matrix  
programming language without the hassle of dealing with all the fussy  
stuff of handling output, data manipulation, etc. Furthermore it is p- 
code, compiled on the fly (or it can be stored in compiled format)  
which makes for a **significant** speed improvement for most tasks  
over interpreted code. MATLAB is slow at many things unless it is  
compiled and apart from Jim LeSage's Econometrics Toolbox has  
virtually no built-in support for econometrics---you have to write it  
all yourself. Ox is very fast and C-like, but a bit fussy when it  
comes to things like data import/export. Mata is C-like to a large  
extent, with many of the features you would find in a C-like  
language: structures, pointers, header files, function passing, and  
the like. Furthermore it has the advantage of being very close,  
syntactically, to something like MATLAB, so that it is generally very  
easy to translate existing code in MATLAB (or Ox, or GAUSS, or even  
Fortran--see Bill Gould, Stata Journal article on that) into Mata.   
And if you want to share your code with others, no one makes that  
easier than Stata with -net from- and -ssc-.

I agree wholeheartedly with Maarten in that the combination of high  
level features in Stata's language (e.g., -syntax-, -simulate-, - 
bootstrap-, missing data handling) and low-level, efficient  
programming features in Mata are unparalleled. Those who suggest it  
is not a good platform for econometric work don't know enough about  
what they're saying to make an informed statement. The recent  
addition of -optimize- to Mata has given it hugely valuable  
capabilities in that regard: Mark Schaffer and I put together a GMM- 
CUE routine (see our paper with Stillman, SJ 7:4) that is a fraction  
of the ado-file code in -ivreg2- and runs from five to twelve times  
faster. The construction of that routine is illustrated in the last  
section of my forthcoming "Introduction to Stata Programming", soon  
to be available from Stata Press.

Kit

Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


*
*   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/