An unbiased errors-in-variables approach to detecting unconscious cognition

Greenwald, Klinger and Schuh (1995) have proposed a regression approach for detecting unconscious cognition. An errors-in-variables approach is presented that corrects for measurement error in the predictor and takes into account that the latent predictor variable is assumed to be non-negative. The new approach requires the same input as the uncorrected regression analysis and provides consistent estimates of regression weights as well as valid statistical tests of their significance. In particular, the method yields a consistent estimate of the regression intercept that provides critical evidence for unconscious cognition. A simulation study illustrates these aspects of the new technique. Several data sets are then reanalysed by means of the new method.