Robert,
we seem to be within reach of one another on most points, so I will reply only
to this one.
"Ah, there's your problem - you can probably only use this model in a limited
range of %-similar-wanted. I suspect that at high values of %-similar-wanted
there simply aren't enough turtles of the other color to give the target
%-similar-wanted. Is that phase change at 75% an artefact of the model? To me,
the fact that 75% = (an integer)/(the number of neighbors) -- specifically 6/8
-- suggests that it is an artefact. I'd be REALLY surprised if this phase
change corresponded to something in the real world. "
Well, perhaps you should prepare to be surprised. On your account, one would
expect the model to be exquisitely sensitive to the number of turtles. If one
increases the number of turtles to 2500, the tipping point doesn't seem to
change much. I would further increase the number of turtles, but model limits
me to 2500.
Oh, I see! It is the number of neighbors you see as being too coarse-grained.
Well, since there are 8 neighbors (in the procedures) one would expect that if
this were the problem that .875, not .75 would be the tipping point. But even
if .875 WERE the tipping point, why is that an artifact? Assuming that a
neighbor is somebody whose bedroom window can be seen from your bedroom window,
and given that there is no such thing as half a neighbor, I think
"neighbor-number" in the real world is probably pretty coarse grained. One of
the things I learned sitting in at the SFI summer school is that discreteness
of input is one of the important sources of complexity.
In any case, having implied that I didn't study the model, don't you need, in
order to keep the record straight, to amend your original claim that
"As you increase the "%-similar-wanted" slider, the end-point of the
"percent-similar" plot get closer to 100%. It does NOT suddenly start dropping.
"
Otherwise, you might be accused of Gerry-mandering the data.
Speaking of cups of coffee.
Nick
Nicholas S. Thompson
Emeritus Professor of Psychology and Ethology,
Clark University ([email protected])
P.S. Oh: Whether or NOT the drop at 75 percent is an artifact or not, the
model still points to at least ONE fundamental flaw in our thinking ... the
belief that reasonable flexibility in choosing one's neighbors will lead
automatically to reasonably integrated neighborhoods. I really do believe we
agree much more than you say.
----- Original Message -----
From: Robert Holmes
To: [email protected];The Friday Morning Applied Complexity Coffee
Group
Sent: 2/1/2009 8:20:08 PM
Subject: Re: [FRIAM] Contra Thompson: problems with the explanation
ofexplanations
Miscellaneous responses below... -- Robert
On Sun, Feb 1, 2009 at 3:38 PM, Nicholas Thompson <[email protected]>
wrote:
Robert,
<snip>
I DID play with the model! Although you are the last person I would want to go
up against on such a geek matter, I think it performs pretty much as the paper
describes. In fact, I have it running at this very moment. Percent similar
sought is set to 85, average percent similar achieved is running around 50
percent and everybody is unhappy. There is, I think, a dramatic phase change
between 70 percent sought and 80 percent sought, in fact, now that I explore
it, between 75 and 76%.
<snip>
Ah, there's your problem - you can probably only use this model in a limited
range of %-similar-wanted. I suspect that at high values of %-similar-wanted
there simply aren't enough turtles of the other color to give the target
%-similar-wanted. Is that phase change at 75% an artefact of the model? To me,
the fact that 75% = (an integer)/(the number of neighbors) -- specifically 6/8
-- suggests that it is an artefact. I'd be REALLY surprised if this phase
change corresponded to something in the real world.
===>Hempels' symmetry of explanation and prediction has been dead and buried
for years so really can't be used to support any argument; <===
I think the paper puts it around the other way: that scientific practice
supports Hempel, not other way around. "Precisely the same point holds for the
other examples—which, collectively, serve to confirm, not undermine, Carl
Hempel's sixty-year old "Symmetry Thesis" concerning explanatory and predictive
power (Hempel 1948)."Can you think of any examples of good scientific theories
that do not provide good, clear, expectations of observation?
No I can't, but that's not actually the symmetry issue with Hempel. I'll stick
that in another email once I've had time to think it out properly.
===>hypothesizing micro-rules in models is actually a perfectly reasonable
thing to do;<===
I think I agree; where do Derr and I contradict this assertion?
Para 1.5: "Models are chosen or designed to be in accord with what is already
known about the phenomenon we are trying to explain."
But consider flocking models: we ignore what is already known (or don't attempt
to find out what is known) and instead hypothesize that we can get flock like
patterns using the three rules (i) avoid collision (ii) align direction (iii)
align speed. In actual fact, from the papers Steve posted earlier this week it
actually looks like the rules going through the creatures' brains are (i)
pursue thing in front and (ii) escape from the thing behind.
Actually, perhaps your statement is the more accurate: a good model is one that
IS based on known micro-rules; making up the micro-rules (which is what we
usually do in an ABM) is actually cheating...
===>the burden-of-proof should rest with the modeler, not with anyone who dares
to disagree with her.<===
Para 1.7: "How can this counter intuitive result come about? Is it just an
artifact of all the artificialities in the contrived model? Or does the model
point to some fundamental flaw in our thinking about segregation? Modelers
would rightfully claim the latter." (emphasis added).
That last sentence is an awfully big claim and is why I'd say that the
burden-of-proof rests with the modeler. e.g. I should not be expected to
consider your 75% phase change above as a real-world effect - the burden would
be very much on you to show that it was not a model artefact.
Again, not clear what you have in mind, here, or how it stands in contradiction
to Derr and Thompson.
Thanks for the comments,
Nick
Nicholas S. Thompson
Emeritus Professor of Psychology and Ethology,
Clark University ([email protected])
----- Original Message -----
From: Robert Holmes
To: The Friday Morning Applied Complexity Coffee Group
Sent: 2/1/2009 1:20:35 PM
Subject: [FRIAM] Contra Thompson: problems with the explanation ofexplanations
Although I agree with the overall tenor of Nick's "Contra Epstein" piece
(http://jasss.soc.surrey.ac.uk/12/1/9.html), there's one glaring error:
Schelling's segregation model is completely misrepresented. The notion that
segregation decreases as the individuals' desire to be segregated increases is
wrong. Nick - have a play with the Netlogo model! As you increase the
"%-similar-wanted" slider, the end-point of the "percent-similar" plot get
closer to 100%. It does NOT suddenly start dropping. The interesting point that
the model illuminates is that you need surprisingly low values of
"%-similar-wanted" to generate high "percent similar" environments.
Robert
P.S. There's some other parts of the paper I'd argue with, viz:
Hempels' symmetry of explanation and prediction has been dead and buried for
years so really can't be used to support any argument;
hypothesizing micro-rules in models is actually a perfectly reasonable thing to
do;
the burden-of-proof should rest with the modeler, not with anyone who dares to
disagree with her.
...but having already demolished 40% of Nick's paper, I thought I'd better give
it a rest :-) Nick - buy me a coffee and I'll give you details!
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org
