-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Marcus G. Daniels wrote: > Fine, but more models won't help that problem. The data is the > data. In contrast, Phil's example would be addressed by AIC.
How so? I'll reformulate Phil's statement as: "Because understanding a referent requires multiple simplifying projections (models), the question of which particular model is the correct one is confusing." Phil, if that isn't a good paraphrase, please correct me. But, if it is a good paraphrase, selectors like the AIC that assume some ideal perfect (largest) description of the referent will NOT help. In fact, they hinder understanding because they _imply_ that there is a single, true, perfect, ideal, largest model, which is false. Operationally, they do not help because the data are at least one (probably many many more) modeling level(s) removed from the referent system. To be clear, the process works this way: 1) casual observation and psychological induction leads to a (usually mental) model 2) an experiment is designed based on that model 3) data are taken from the experiment 4) a more rigorous model is derived from the data (perhaps regulated by the prior model) 5) repeat as necessary Each data set is derived from a prior model. Hence, the best a data-driven model selector can do is find the model(s) upon which the data are based. It only targets the referent to the extent that the original (usually mental) models target the referent. And if those original models were induced by a perverse person with perverse thoughts, then the original model is probably way off and false. Hence, selectors like the AIC will only lead us to that false model, not to the truth. Worse yet, because they used a rigorous (hermeneutic) mathematical technique to find that false model, they will be strongly inclined to believe in that false model. Just like the old adage "Don't believe everything you read", we could state an analogous adage in modeling and simulation: "Don't believe everything described mathematically." Of course, those aphorisms are way too moderate. [grin] In fact, to quote Sturgeon "Sure, 90% of science fiction is crud. That's because 90% of everything is crud." So, we should change the aphorism to "Don't believe 90% of the math you read." > Phil Henshaw wrote: >> It does confuse that we seem >> to need to look at real systems with simplifying projections that >> look different from each other. The answer as to which 2D >> projection is the correct one is what seems most confusing. >> > Glen E. P. Ropella wrote: >> My most heated was with a guy who claimed that model selectors (e.g. >> AIC) that rely on a posited "perfect" or "optimal" model can actually >> help one get at the true generator of whatever data set is being >> examined. It took a lot of verbage on my part to draw a detailed enough >> picture for him to understand that the data were taken by a method that >> presumed a model (because all observation requires a model and all >> models require observations). And the best a selector can do is find >> that occult model by which the data was collected. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Shallow men believe in luck ... Strong men believe in cause and effect. - -- Ralph Waldo Emerson -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFGg+bXZeB+vOTnLkoRAn48AKCSs6TtvhgzoPivHZV9qp3QTtfTewCfdThI Jx3uOb0ycXiJ9/DpYW6dF2U= =RkyW -----END PGP SIGNATURE----- ============================================================ 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
