ecological ideology models
whoooo26505 at yahoo.com
Thu Jun 15 08:01:02 PDT 2006
As some of you know, one of the problems I have always had with ideology models in political science is that they aggregate their dependent variable prior to performing regression analysis. I repeatedly make the claim that this is improper. But in reply, some have said that aggregation is a matter of style -- that it is merely looking at the data at another level of analysis. (Some even suggest that I do not know what multi-level analysis is). Still others have simply said that I have not proven that ecological inference is troubling these models.
Well, I think I have just fixed that. I think I have just shown, conclusively, that throwing your voting data into percentages is, in effect, "cooking the books." It is not merely "analyzing from a higher vantage point." Nor is it analyzing voting "variance" instead of the votes themselves. What it is, is model misspecification that causes inaccurate results. The choice of an ecological model actually creates misinformation. Those of you who are using these models to suggest to graduate students that 60% (or better) of the voting variance in civil liberties is explained by the justices' political values really need to think about whether this claim is manufactured by (and dependent upon) a modeling defect.
I don't want to bore people who are not interested in this topic. My argument is linked below. It only addresses goodness of fit. (I'm addressing other issues later). I'd appreciate any feedback you might have (in here, in private, or on my web page -- whatever you like). Here is the analysis:
Here is a printable version: http://ludwig.squarespace.com/display/ShowJournalEntry?moduleId=595563&entryId=542739&printerFriendly=true
Dr. Sean Wilson, Esq.
Penn State University
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Conlawprof