A “nice example”:http://www.crescatsententia.org/archives/week_2004_03_21.html#003440 via Crescat Sententia of an issue I’ve “mentioned before”:https://www.crookedtimber.org/archives/001259.html, namely, a case where the stylized facts lend themselves to an elegant bit of modeling that seems to analyze things very neatly, but the empirical details turn out to be much messier or a different kind of process altogether. Here it’s the debate about the Hijab in French schools. This is why fieldwork is important. The identification of mechanisms like sub-optimal conventions, failed co-ordination, tipping phenomena, self-fulfilling prophecies or auto-equilibrating systems are amongst the most useful and powerful tools in social science, but the number of phenomena they _appear_ to explain is much larger than those they _in fact_ explain. This can lead to odd consequences. For example, John Sutton’s little book “Marshall’s Tendencies”:http://www.amazon.com/exec/obidos/ASIN/0262692791/kieranhealysw-20/ref=nosim/ (which I didn’t read carefully enough when I picked it up) makes the point that we can be led to misapply standard models not just when the reality is much more complicated or otherwise difficult, but even when there’s a perfectly good alternative model available, just not the obvious one.
Reaction to my last few posts make me want to add disclaimers like “Look, this doesn’t mean formal modeling is unimportant or bad,” “Yes, yes, of course there are lots of very smart game theorists,” and “No, Libertarians, I am not talking about you, so please relax.”
And sorry to anyone who was expecting this post to be about the attractions of the other kind of model.
{ 14 comments }
arthur 03.27.04 at 3:38 am
your last sentence was the most insincere sentiment ever expressed in a blog.
Kieran Healy 03.27.04 at 3:52 am
_the most insincere sentiment ever expressed in a blog._
I dunno, that seems like a pretty high bar.
msg 03.27.04 at 4:26 am
I vote for Laetitia Casta.
Phill 03.27.04 at 5:14 am
As a network security architect I design security protocols for the Web and Internet. I currently serve as editor for several active proposals.
The connection to the modelling question is that computer architecture frequently involves building models of the real world.
The big difference is that in computer science it is very hard to escape the brutal facts indefinitely. If your model is wrong you will usually end up being forced to admit it. Sometimes you can pretend you were right because no matter how bad something is the costs of change are far too great. But the pain is still there. The informed CompSci world knows that the ‘object persistence’ model developed ten to fifteen years ago is vastly superior to SQL but with close to a trillion dollars of sunk costs change is hard.
Ten years ago Bruce Schneier and myself both got seduced by the fact that cryptography could create unbreakable computer security protocols. We both eventually saw the light, Bruce wrote his appology in his book ‘Secrets and lies’. Being a protocol designer rather than a protocol analyst (in the security world you need both but analysts who break stuff tend to get more press), I was forced to face the fact a little earlier. When Bruce changed his mind he went way beyond my critique. Last month I was at a roundtable crisis meeting looking at a current security problem and suddenly everyone in the room realized that the model we had all being following was completely wrong, not just a bit flawed.
This could be crammed into the ‘Paradigm shift’ model, but thats not really what had happened. We all understood each other fine. We even showed our students slides with fine slogans like ‘Security is risk control (not risk elimination)’. But then we had gone off and built systems whose principal stated goals were to meet the needs of the military and principal unstated goal was to prevent Louis Freeh’s wiretap ambitions.
Because we thought we could build perfect security systems we tried. Predictably we ended up failling to build even good schemes half the time. One of the dirty secrets of the Internet security story is that the most widely used Internet protocol SSL was not very good at all when first proposed. In fact a catastrophic flaw was discovered fifteen minutes into the first public presentation, not a subtle error, a rookie mistake.
But in the end SSL did more than the perfection model ever did.
yabonn 03.27.04 at 5:30 am
In the same post, the guys at crescat :
– discovers that this law, albeit french – like, y’know, statism, taxes and other assorted fascisms – is there because it helps people,
– and mentions the words “overreaction”, and “hysteria”.
… cute :-)
bad Jim 03.27.04 at 6:46 am
I hate to follow phill’s thoughtful, informative, evocative post. Many of us have tenderly husbanded elegant solutions that withered in the glare of hostile reality.
However, I wanted to mention the many plastic models I assembled as a teen: USS Constitution, HMS Victory, assorted knights, aircraft and antique pistols, all of which were eventually blown up with firecrackers in celebration of the Fourth of July. One moves at times from one seduction to the next.
zaoem 03.27.04 at 3:07 pm
Actually, it seems more like a game that involves both a coordination and collaboration dilemma. Such games usually have many uncooperative equilibria where intimidation and exploitation are common tactics in attempts to move to a more cooperative equilibrium in a decentralized manner. What is perhaps the more accurate characterization then is that it is an attempt to move from an uncooperative equilibrium to a cooperative one. The normative objection is that there is no obvious virtue to the cooperation (everyone opting to wear/not wear a headscarf).
humeidayer 03.27.04 at 3:11 pm
And sorry to anyone who was expecting this post to be about the attractions of the other kind of model.
“All models are wrong. Some models are Yasmine.” :)
mondo dentro 03.27.04 at 3:43 pm
My name is Mondo, and I am a modeler.
The issues that Kieran raises are really at the heart of the craft of modeling, viewed as an exercise in “natural philosophy”. I do not wish to minimize their importance, but I think, in the end, the observations are not very deep, and can be summarized as follows: “reductionist models can be very useful, but they are often wrong.”
It is a pity that fundamentalists on both sides of the modeling question, in a wide variety of fields, fail to grasp the importance of this vital, yet rather banal, truth–thus making repeated statements like Kieran’s necessary. But aren’t we all already supposed to understand the unavoidable and never-ending necessity of the modeling-empiricism dialectic? I mean, aren’t all scientists and natural philosophers weaned on the works of those two great modelers, Ptolemy and Copernicus?
I do a fair amount of interdisciplinary work, meaning that I enjoy talking to people in fields that I don’t really know a lot about and working with them to develop explanatory/predictive models of phenomena that seem important. The entire process is fraught with perile at every step, including at the very beginning, because one can not even collect data without having at least an inchoate model guiding the experimental design. A skilled modeler is thus important even before the “real modeling” has begun.
Indeed, it is not even possible to make the distinction between the epistemological categories the Kieran takes as axiomatic, Things That a Model Appears to Explain and Things That a Model In Fact Explains, without a firm grasp on various alternative models.
Finally, let me say that there is nothing more common to a would-be modeler that to hear the statement that “your model is too simple” or “the actual situation is far more complicated”. It is so common, in fact, that applied mathematicians will quickly exchange knowing chuckles upon the mere mention of “the spherical horse model”. Of course, a modeler always needs to listen to this criticism, even though it is often made by a person who does not understand the modeling process, because it just might be correct. The skill of the modeler comes in filtering such criticisms to extract problems that are legitimate from those that merely appear to be legitimate!
nnyhav 03.27.04 at 4:30 pm
Relevant perturbation analysis (via B&W).
Bob McGrew 03.27.04 at 10:10 pm
I love game theory and I think it’s quite often applicable to the real world with only minimal changes – for instance, in the auction setting, where it’s pretty clear what people’s utilities are and it’s pretty clear what people’s strategies are.
But when you try using game theory to model other things – you really have to make sure you are modelling the right game. (And that’s before issues like bounded rationality, unclear preferences, and the like come into play.)
Luc 03.28.04 at 8:51 pm
This is why fieldwork is important.
In this case that fieldwork would require understanding French. Something which the english speaking (and blogging) academia seems clearly incapable of.
harry 03.29.04 at 5:18 am
I had to read the last sentence three times before I figured out what other sort of model there was.
Luc 03.29.04 at 8:13 am
I wish there would be a formal model for understandable english.
I was a bit disappointed by the ugly trick Peter Northrup uses to get his point across. No fieldwork or modeling would correct this.
Look at the development of the following phrases:
coerced by their fellow students
insults and even violence–mostly from their male classmates
the intimidation from the boys
intimidation, backed up by credible threats of violence
a boy who [] was preparing to engage in (presumably illegal) violence
a group of criminals
This transformation is used in this conclusion:
Expecting a law condemned by much of the international press and plausibly seen as targeting sacred observances to command respect among a group of criminals is a bit much.
[todo]
Insert remark about French models here.
[/todo]
Comments on this entry are closed.