Should Auxiliary Variables Be Included Under Not Missing at Random?

Time and Location
Poster Session III - Board: III-003
Friday May 23, 2008, 1:30 PM - 2:30 PM
Exhibit Hall

Chung-Ping Cheng
National Chengchi University, Taipei, Taiwan

Estimates of full information maximum likelihood method (FIML) are biased under missing not at random. FIML may benefit from adding auxiliary variables which are not concerned in the research. By simulation, we found that when the auxiliary variable is uncorrelated with the cause of missingness, inclusion of it may bias the estimates.

Go back

 The Role of Functional Neuroimaging