August 19, 2010

Overcoming the Binary (Next Week at the Statistics Seminar)

Every human relationship is a unique and precious snowflake, but do we treat them that way when we model them mathematically? No. No we do not. Join us next week to hear not just why this is wrong, but what to do instead. As always, the seminar is free and open to the public.

Joe Blitzstein, "Strengths of Ties in Network Modeling and Network Sampling"
Abstract: Measuring and modeling the strengths of ties in a social network has a long history, and an even longer history of being ignored. How much does it matter for inference if the strengths are discarded? Dichotomizing a network may seem to be an appealing simplification, but we show that it comes at a heavy cost, through quantifying the information loss. Closely related issues arise in respondent-driven sampling, a popular method for surveying ``hidden'' populations. We suggest ways to incorporate strength of tie information in this setting, comparing design-based and model-based estimation approaches in the context of an AIDS study.
Based on joint works with Sergiy Nesterko and Andrew Thomas.
Time and place: Monday, Aug. 23, 2010, 4:00--5:00 PM in Doherty Hall A310

Let me add that Prof. Blitzstein will be visiting us from the Bible college of a prophecy-obsessed, theocratic Puritan cult clinging to the rudiments of civilization in a plague-blasted post-apocalyptic wasteland*, so I expect a good turn out to show him how we do these things around here.

*: No, really.

Manual trackback (!): The Inverse Square

Enigmas of Chance; Networks

Posted at August 19, 2010 16:30 | permanent link

Three-Toed Sloth