MRC INTEGRATIVE EPIDEMIOLOGY UNIT (IEU)

SEMINAR SERIES

 Thursday, 5^th February, 2015

 16.00 – 17.00

Room OS6 – Oakfield House

 Eleanor Sanderson

Research Assistant in Econometrics Applied to Epidemiology Studies

MRC Integrative Epidemiology Unit

 

“Testing for Weak Instruments"

Abstract

Mendelian Randomisation relies on the assumption that the SNPs used as instruments for the endogenous exposure variables have a strong association with those exposure variables. If the association between the instruments and the exposures is weak then this can lead to substantial bias in the estimation. In models with one endogenous exposure variable a rule of thumb test that the instruments strongly predict the exposure is that the first stage F statistic is greater than 10. Stock and Yogo (2005) extend this to models with multiple endogenous variables and develop a test to test when the rank of the matrix of reduced form parameters is close to zero. Here I consider testing for weak instruments in models with multiple endogenous variables where there is a near rank reduction of one in the matrix of reduced form parameters. I propose a conditional F-statistic, similar to that proposed by Angrist and Pischke (2009), which can be used to identify weak instruments of this type. When there are more than two endogenous exposure variables in the model this new test provides additional information about the nature of the weak instrument problem than existing tests.

Biography

Eleanor recently started working at the IEU in the Mendelian Randomisation programme. Her background is in Econometrics and before starting working at the IEU she completed a PhD on 'Using first stage F-statistics to test for weak instruments in Two-Stage Least Squares estimation of linear Instrumental Variables models' in the Department of Economics at Bristol University.

 

ALL WELCOME