January 2006
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 31        

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


« January 27, 2006 | Main | January 31, 2006 »

30 January 2006

Applied Statistics - Jim Greiner

This week, the Applied Statistics Workshop resumes for the spring term with a talk by Jim Greiner, a Ph.D. candidate in the Statistics Department. The talk is entitled "Ecological Inference in Larger Tables: Bounds, Correlations, Individual-Level Stories, and a More Flexible Model," and is based on joint work with Kevin Quinn from the Government Department. Jim graduated with a B.A. in Government from the University of Virginia in 1991 and then received a J.D. from the University of Michigan Law School in 1995. He clerked for Judge Patrick Higginbotham on the U.S. Court of Appeals for the Fifth Circuit and was a practicing lawyer in the Justice Department and private practice before joining the Statistics Department here at Harvard. As chair of the author's committee, he is a familiar figure to readers of this blog.

As a reminder, the Applied Statistics Workshop meets in Room N354 in the CGIS Knafel Building (next to the Design School) at 12:00 on Wednesdays during the academic term. Everyone is welcome, and lunch is provided. We hope to see you there!

Posted by Mike Kellermann at 12:30 PM

Ecological Inference in the Law, Part III

Jim Greiner

In two previous posts here and here, I discussed the ecological inference problem as it relates to the legal question of racially polarized voting in litigation under Section 2 of the Voting Rights Act. In the latter of these two posts, I suggested that this field needed greater research into the case of R x C, as opposed to 2 x 2, tables.

Here's another suggestion from the courtroom: we need an individual level story.

The fundamental problem of ecological inference is that we do not observe data at the individual level; instead, we observe row and column totals for a set of aggregate units (precincts, in the voting context). This fact has led to some debate about whether a model or a story or an explanation about individual level behavior is necessary to make ecological inferences reliable, or at least as reliable as they can be. On the one hand, Achen & Shively, in their book Cross-Level Inference, have argued that an individual level story is always necessary to assure the coherence of the aggregate model and to assess its implications. On the other hand, Gary King, in his book A Solution to the Ecological Inference Problem, has argued that because we never observe the process by which ecological data are aggregated from individual to group counts, we need not consider individual level processes, so long as the row counts (or percentages) are uncorrelated with model parameters.

From a social science point of view, this question is debatable. From a legal point of view, we need an individual level story, regardless of whether such a story produces better statistical results. When judges and litigators encounter statistical methods in a litigation setting, they need to understand (or, at least, to feel that they understand) something about those methods. They know they will not comprehend everything, or perhaps even most things, and they have no interest in the gritty details. But they will not credit an expert witness who says, in effect, "I ran some numbers. Trust me." What can quantitative expert witnesses offer in an ecological inference setting? The easiest and best thing is some kind of individual level story that leads to the ecological model being used.

Posted by James Greiner at 6:01 AM