Open AccessPremorbid IQ estimates from a multiple aptitude test battery: regression vs. equating
Author(s) -
Daniel R. Orme,
Malcolm James Ree,
Paul Rioux
Publication year - 2001
Publication title -
archives of clinical neuropsychology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.909
H-Index - 98
eISSN - 1873-5843
pISSN - 0887-6177
DOI - 10.1016/s0887-6177(00)00091-3
Subject(s) - equating , statistics , linear regression , regression , psychology , regression analysis , equivalence (formal languages) , econometrics , bayesian multivariate linear regression , proper linear model , artifact (error) , regression toward the mean , mathematics , discrete mathematics , neuroscience , rasch model
Estimation of premorbid abilities remains an integral part of neuropsychological evaluations. Several methods of indirect estimation have been suggested in the literature. Many of these methods are based in prediction via linear regression. Unfortunately, linear regression has the well-reported tendency to underpredict high IQ scores and overpredict low IQ scores. This can be shown to be an unavoidable statistical artifact of linear regression. We demonstrate a procedure to estimate premorbid IQ without the regression artifact. The procedure has two steps: confirmation of construct equivalence and psychometric equating. An example using real data is presented which shows the regression to the mean problem with prediction and compares it to the results from equating.