Log in

Sign up for our weekly newsletter!

When & Where
Date: 
Mon, June 8, 2015 - 9:00 AM to Fri, June 12, 2015 - 4:00 PM
Location: 
D-lab Convening Room, 356 Barrows Hall
Description
Type: 

Berkeley Methods Workshop, 2015  

Instrumental variables (IV) analysis is the most popular statistical technique for solving the endogeneity problem in the social and biomedical sciences. Conventional regression models are biased when treatment is confounded, selected, measured with error, or otherwise endogenous. By contrast, IV analysis is consistent where regression fails if certain assumptions are met.

This course teaches a rigorous applied survey of modern IV analysis and related techniques, such as regression discontinuity design. We present classical and very recent results from econometrics, statistics, biostatistics, and computer science with a focus on what is useful in practice.   

Instrument variables analysis stands and falls with the validity of the assumptions. Therefore, we emphasize an applied understanding and discuss numerous real examples from the social and health sciences to develop dependable intuition for what these assumptions mean in practice.

Topics include:

  • The problem: treatment endogeneity 
  • Just-identified and over-identified IV models
  • Wald estimation
  • Two-stage least squares (2SLS)
  • Generalized method of moments (GMM) for non-iid errors
  • Limited information maximum likelihood (LIML) for many weak IVs
  • Graphical IV-set identification
  • Multiple indicator IV models for measurement error
  • Understanding assumptions
  • Types of exclusion violations
  • Assumption testing in theory and practice
  • Non-parametric bounds
  • LATE: IV with heterogeneous treatment effects
  • Regression discontinuity design (RDD)
  • Real examples from the social and health sciences.

 The course mingles theoretical exposition with extensive hands-on training to develop transferable skills. We analyze many real applications and discuss under which conditions IV analysis and allied techniques are promising or not.  The goal is to empower social scientists and health researchers to apply these techniques appropriately in practice.

 This course is independent of Dr. Elwert’s course on Causal Inference From Observational Data. The two courses can be taken separately or together.

Fee: $1500 (USD)    Register at RegOnline

UC Berkeley Graduate Students:  A limited number of fellowships will be available to cover up to $1400 of workshop fees. Please complete the Fellowship Application to be considered.

Details
Training Host: 
D-lab Facilitator: 
Jon Stiles
Participant Technology Requirement: 
Participants are expected to bring laptops for hands-on training. A trial installation of Stata 13 will be provided for participants who who not have that installed already. A small number of "loaner laptops" will be available.