Leigh Tesfatsion wrote:

> The labor market experiment I reported in an earlier email is a case
> in point.  As discussed in more detail in that earlier email, a key
> **observation** in empirical labor studies is "excess heterogeneity"
> -- substantial unexplained variation in wage earnings that persists
> even after attempts are made to control for all relevant structural
> factors (gender, industry type, schooling, etc.).    Recently
> (Econometrica 1999), using very long panel data (observations!), John
> Abowd et al. uncovered a surprising effect:  Simply adding workers'
> names to the list of regression variables led to substantial reduction
> in the unexplained variation in wage earnings across workers who
> appeared to be in structurally similar circumstances.  These
> *empirical observations* then led me to wonder whether personal
> non-price interaction effects on the work-site could -- even in
> principle -- sustain persistent variance in wage earnings among
> workers whose observed structural characteristics were absolutely
> identical.   I set up a simple experiment to test this conjecture, and
> received a strongly affirmative answer.   More to the point, I was
> able to see two distinct sources for the persistent wage variation
> (behavioral effects and network effects).   I was also able to see
> that outcome distributions for any given treatment did not take a
> "normal" central tendency form (a common social science a priori
> assumption) but instead were spectral (multiple peaked) in nature.
> These findings, while understandable after the fact, were certainly
> not anticipated by me in anywhere near this specificity.
> As I now continue on with my agent-based computational modeling work
> focusing on unemployment benefit programs (e.g., in Iowa), and on a
> reliability study of New England's restructured electricity market as
> a consultant for the Los Alamos National Lab,  I take with me from
> these simple labor market experiments the cautionary warning that
> non-price behavioral and network interaction effects can be very
> strong indeed, leading to spectral rather than central tendency
> distributions of outcomes even for similarly structured entities, so I
> should be extremely careful not to engage in inappropriate pooling
> purely on the basis of a priori structural categorizations (e.g., a
> priori lumping together data for all generators of a certain size and
> fuel type).
> In short, observation led to theorizing which in turn is changing the
> way I am organizing observed data in subsequent empirical studies, and
> so it goes in an endless feedback process.

Leigh uses a prisoners' dilemma game theoretic formulation.  Either this
was an arbitrary design choice based only on its use in other similarly
arbitrary model designs or she validated the design against some
evidence about the behaviour of workers.  She does not say that she has
validated the spectral distribution of outcomes.  In the worst case,
therefore, Leigh has a wholly unvalidated model inspired by an empirical
observation.  However inspirational she finds the results, I do not
understand how she could use such a model with confidence to formulate
social policy.

> By the way, I believe your original question asked for "excellent
> examples of simulation providing explanation in any field."  Somehow,
> in all the email that has ensued from your original email, I have seen
> general dismissive remarks and abstract discussion but not a response
> to your request per se -- which I interpret as a request for *specific
> constructive examples*.
> How about it, other readers?  How about engendering more constructive
> discussion on the basis of *specific* examples?

Off the top of my head, there is the VDT model produced by Ray Levitt
and colleagues at Stanford, the models of the Anisazi by George Gumerman
and colleagues, models of domestic water demand produced for the UK and
Catalonia as part of the European FIRMA project, I'll chuck in one of my
papers on critical incident management in JASSS a few years ago, a model
of electricity usage by Jan Treur's team in Amsterdam (reported I think
in ICMAS-98), Kathleen Carley's recent work on biological weapons use.
What these all have in common is that they are descriptive first.  It is
also the case that some generalisable techniques have been developed in
the course of mplementing these models.