1 / 21

Comments on Diamond and Sekhon

Comments on Diamond and Sekhon. Philip Schrodt University of Kansas 21st Summer Meeting of the Society for Political Methodology. or APSA Organized Section on Political Methodology. or “those !*$%#$^s who reviewed my article”. Disclaimer.

mairi
Télécharger la présentation

Comments on Diamond and Sekhon

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Comments on Diamond and Sekhon Philip SchrodtUniversity of Kansas 21st Summer Meeting of the Society for Political Methodology orAPSA Organized Section on Political Methodology or“those !*$%#$^s who reviewed my article”

  2. Disclaimer The Board of Education of the State of Kansas is likely to determine that the theory of evolutionis of questionable scientific merit and consequentlyshould be viewed with skepticism. See Genesis 1:1-31 Or Genesis 2:4-9 for a different version whatever…

  3. Disclaimer The Board of Education of the State of Kansas is likely to determine that heliocentric astronomyis of questionable scientific merit and consequentlyshould be viewed with skepticism. See Joshua 10:13

  4. Disclaimer The Board of Education of the State of Kansas is likely to determine that round-earth geographyis of questionable scientific merit and consequentlyshould be viewed with skepticism. See Daniel 4:10-11, Matthew 4:8

  5. Disclaimer The Board of Education of the State of Kansas is likely to determine that the assertion π = 3.14159… is of questionable scientific merit and consequentlyshould be viewed with skepticism. See 1 Kings 7:23

  6. Creeping Artificial Intelligence:Primary Tools of AI from 1980s • Expert Systems/ID3 • Neural Networks • Genetic Algorithms • Nearest neighbor clustering • Diamond-Sekhon: Use a GA to optimize a nearest neighbor metric

  7. Why is this important? • SWWC “So what? Who cares?

  8. Henry Brady on statistical pedagogy Theological seminaries distinguish between theology, or the systematic study of religious beliefs, and homiletics, the art of preaching the gospel convincingly. Theologians ask hard questions, and often espouse opinions that would shock and horrify the practicing members of the religion’s congregations. Homiletics is about homilies: sermons that are practical, down to earth, simple and reliable interpretations of the faith. The social sciences have a great deal of theology, but very little homiletics.Brady and Collier, eds. Rethinking Social Inquiry 2004) pg. 53 [WRONG] [well, maybe] You decide…

  9. Gary King on statistical pedagogy “I have found lying to be one of the most effective techniques when teaching statistics”Comment from floor at some previous PolMeth Summer Meeting, probably ca. late Pleistocene

  10. “Why they hate us” Students, colleagues, random social encounters, etc. PoliticalMethodologist

  11. “Why they hate us” • What they want: “If X is applied to case J in situation Z, what difference will it make in Y?” • What we give them: “The estimate bi of the population coefficient ßi is significantly different from zero at the p = 0.043591 level” [unless Xi is co-linear with some other independent variables—and in a sufficiently large sample, it almost certainly is—in which case the sign of bi may be reversed, and if bi isn’t significant, don’t pay attention to that either] [or maybe Xi has no causal effect whatsoever but happens to correlate with something not in the model that does]

  12. Phascinating Phacts: State Nicknames • Illinois Land of Lincoln • Alaska Land of the Midnight Sun • Minnesota Land of 10,000 Lakes • Massachusetts Land of Propensity Functions

  13. Matching Cases on Covariates Variable Mass. Calif. Kansas Traffic problems   Only for opossums Not with current Board of Education High tech economy  Marriage Equality for a while Passed anti-gayamendment GOP governor   in liberal state Democratic governorin conservative state Outrageous   housing costs Assistant professorscan afford houses Treatment: Sekhon From here… To here…

  14. Genetic Algorithms • Originally proposed in John Holland's Adaptation in Natural and Artificial Systems(1975). The problems of interest to Holland are characterized by: • The impossibility of enumerating all possible devices for solving the problem. • The performance of a device had a large number of local minima: i.e. the initial introduction of a component might initially degrade performance even if it ultimately improved performance. • The structure of the problem-solving device was sufficiently complicated that it was not always obvious which components were responsible for improvements in performance

  15. Optimization surfaces assumed in OLS and theoretical econometrics Faculty meetings as described during job interview {

  16. Optimization surfaces in the real world Your office Actual faculty meeting Your Dean APSA JobPlacement Service

  17. Fundamental Principle of Genetic Algorithms + =

  18. “Intelligent Design” • Choice of the fitness/loss function • Method of encoding the solution • Probabilities of selection, mutation, recombination and transposition

  19. Queries: Pragmatic • How long does this take to run? • Does running it in parallel provide a significant increase in speed? • GA’s are inherently parallel at the point where fitness/loss functions are evaluated, but the communications costs are only justified if these functions are computationally intensive • How consistently does the method converge to a particular optimum matching? Is it sensitive to the initial values of the simulation

  20. Queries: Comparative Advantage • How does the GA compare to a conventional numerical optimization algorithm? • Hill-climbing • Simulated annealing • How does the parameterized formulation compare with simply selecting matching cases directly? • Note that this scales with N2 • Directly selecting the subset would be completely nonparametric and well-suited to the strengths of genetic algorithms • “We are all agreed that your theory is crazy. The question which divides us is whether it is crazy enough to have a chance of being correct. My own feeling is that it is not crazy enough”—Niels Bohr

More Related