|
Activity Number:
|
199
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #300841 |
|
Title:
|
Causal Inference in Spatial Hierarchical Setting
|
|
Author(s):
|
Natalya Verbitsky*+ and Stephen W. Raudenbush
|
|
Companies:
|
University of Michigan and University of Michigan
|
|
Address:
|
439 West Hall, Dept. of Statistics, Ann Arbor, MI, 48109,
|
|
Keywords:
|
causal inference ; potential outcomes ; spatial statistics ; hierarchical models ; violation of SUTVA ; social networks
|
|
Abstract:
|
Does drug A or drug B alleviate headache faster? Does community policing decrease the annual crime rate of a city? Causal questions arise in many facets of daily life. Rubin's causal framework is based on estimating the differences between individual's potential outcomes to treatment A and treatment B and on Cox's "no interference between units" assumption (Stable Unit Treatment Value, SUTVA). However, this assumption is no longer valid when causal questions arise in social networks or geographical settings, where the treatment received by one's neighbor may affect one's potential outcome. We propose an extension of Rubin's framework where the potential outcomes of any unit in the population depend of the treatment assignment of all units in the population. We define causal effects and discuss assumptions that make this framework tractable. Since the data are hierarchical in nature (e.g., individuals nested in neighborhoods) we propose a two-level hierarchical linear model with spatial dependence at level two.
|
