NSF Proposal 2010

From InterSciWiki
Jump to: navigation, search

Award types: Sociology: Early Faculty (CAREER) Research, RAPID Response Research, and Early Concept Grants for Exploratory Research (EAGER). Anthropology: Senior proposals; Faculty Scholars Program, Grants for Rapid Response Research (RAPID) and Early-concept Grants for Exploratory Research (EAGER).

  • and other products ("deliverables") from this Data Share project in UK are here:
  • Duran Bell's Proposal 6987122 has been successfully submitted to NSF
  • Cross-cultural causality project
  • Doug White Proposal 1024515 has been successfully submitted to NSF

Discussion of the methods Wed, January 27, 2010 Hal

From: "hal white" <drhalwhite@yahoo.com> Date: Wed, January 27, 2010 9:06 am To: Douglas.White@uci.edu (more) Cc: "Judea Pearl" <judea@CS.UCLA.EDU> (more) Options: View Full Header | View Printable Version | Download this as a file | View Message details | View as HTML

Hi Doug and Judea -- Well, I must also admit that I haven't dug in to understand the full context of what Doug is proposing, but I can offer the following somewhat context-free two items of advice:

  • 1. For linear models (relevant here), 2SLS and the family of Chalak/White Extended

IV (XIV) estimators are all special cases of IV. The difference lies in precisely what instruments are selected. Therefore, it is efficient to code IV, which will handle the general case. One can then write the code to construct the 2SLS instruments and exercise the IV code with that. Other non-2SLS instrument constructions can then be added sequentially, once the IV and 2SLS code are robust.

  • 2. Precisely which instruments are selected depends on the causal structure,

relevant to Judea's point. The 2SLS instruments are relevant when Y= Xb + U (where Y is the response of interest, X is the cause of interest, and b is the effect of X), X is correlated with U, and there exists Z correlated with X but not U, e.g. X = Za + V, with Z uncorrelated with V and U. In this case, the 2SLS instruments are X^ = Za^, where a^ is estimated in the first stage of 2SLS.

Other instruments are relevant in other cases, as extensively detailed in our "An Extended Class ..." XIV paper. For example, with Judea's back-door structure, the instruments are what econometricians might call Hausman-Taylor instruments. With Judea's front-door structure, the effects of interest are computed as functions of XIV estimators.

So specifying the structure of the causal system, specifically the exclusion and conditional independence relations, matters crucially for deciding which instruments to construct.

I hope this is helpful!

Best, Hal

Causal and Cluster analysis

R cluster analysis -

Short 2SLS primer on Pearl Causal Graphs


https://www.fastlane.nsf.gov/jsp/homepage/proposals.jsp 000386024

Anthropology --- https://www.fastlane.nsf.gov/jsp/homepage/proposals.jsp
Deborah  Winslow-Program Director  	dwinslow@nsf.gov  	(703) 292-7315   	995 N  
Susan  D. Penfield-Program Director 	spenfiel@nsf.gov 	(703) 292-4535  	995 N  
Jill  E. Avalos-Program Assistant 	javalos@nsf.gov 	(703) 292-5017