April 2009
Sun Mon Tue Wed Thu Fri Sat
      1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30    

Authors' Committee

Chair:

Matt Blackwell (Gov)

Members:

Martin Andersen (HealthPol)
Kevin Bartz (Stats)
Deirdre Bloome (Social Policy)
John Graves (HealthPol)
Rich Nielsen (Gov)
Maya Sen (Gov)
Gary King (Gov)

Weekly Research Workshop Sponsors

Alberto Abadie, Lee Fleming, Adam Glynn, Guido Imbens, Gary King, Arthur Spirling, Jamie Robins, Don Rubin, Chris Winship

Weekly Workshop Schedule

Recent Comments

Recent Entries

Categories

Blogroll

SMR Blog
Brad DeLong
Cognitive Daily
Complexity & Social Networks
Developing Intelligence
EconLog
The Education Wonks
Empirical Legal Studies
Free Exchange
Freakonomics
Health Care Economist
Junk Charts
Language Log
Law & Econ Prof Blog
Machine Learning (Theory)
Marginal Revolution
Mixing Memory
Mystery Pollster
New Economist
Political Arithmetik
Political Science Methods
Pure Pedantry
Science & Law Blog
Simon Jackman
Social Science++
Statistical modeling, causal inference, and social science

Archives

Notification

Powered by
Movable Type 4.24-en


« April 2, 2009 | Main | April 7, 2009 »

4 April 2009

Can Nonrandomized Experiments Yield Accurate Answers?

Here is some latest progress (at least to me) on causal inference. William R. Shadish, M. H. Clark, and Peter M. Steiner published a paper on JASA (December 1, 2008, 103(484): 1334-1344.) based on "a randomized experiment comparing random and nonrandom assignments". Basically "In the randomized experiment, participants were randomly assigned to mathematics or vocabulary training; in the nonrandomized experiment, participants chose their training." As the authors acknowledged, unsurprisingly, the randomized and nonrandomized experiments provided different estimates of the training effects, very likely through the selection bias caused by math phobia. The key finding is that statistical adjustment including propensity score stratification, weighting, and covariance adjustment can reduce estimation bias by about 58-96%.

Here is a link to the PPT of the paper. The comments on the paper are also very insightful.

Posted by Weihua An at 10:31 PM