WHAT IS A COMPUTER SCIENTIST DOING ON A STATISTICS WEB
PAGE?
Good question, and I am glad you asked it before dwelling into the linked
publications because, contrary to possible expectations,
they do not contain information on computer programs or
statistical packages.
They are about studies that involve a rich body of substantive information,
thus requiring mathematical machinery for
expressing and manipulating such information,
combining it with data, and drawing new conclusions
from the combination.
Don't we have such mathematics already?
Not well developed, unfortunately. For example, try to express
in standard statistical language the understanding
that the rooster's crow does not cause the sun to
rise, or, more generally, see if you can
define (mathematically) when it is appropriate,
in studying the effect of one factor (X) on another (Y),
to adjust for some other factor (Z), sometimes called "covariate",
"concomittant" or "confounder".
Can structural equations modeling provide answers to
such question?
It sure can, to the surprise of its current practitioners
and its relentless critics. In fact,
what you will find in the linked
papers
(especially
R-207 and
R-218-B)
is a principled, nonparametric resurrection of the path-analytic program
initiated
by Sewal Wright in 1919, whose basic ingredients are:
- Courage to communicate causal assumptions forthrightly and mathematically
- Use of graphs as communication language and inferential tool
Have a safe journey, and let me know of any problems or comments.
