Measurement error in subliminal perception experiments: simulation analyses of two regression methods

A. G. Greenwald, M. R. Klinger, and E. S. Schuh (1995) investigated subliminal perception using a regression-based test for a dissociation between direct and indirect measures of perceptual ability. Direct and indirect measures were obtained for each observer, and a regression analysis was used to predict the amount of indirect perception at the point where the direct measure showed zero sensitivity. A significant positive intercept, obtained with both standard regression and a modified regression developed by K. C. Klauer, S. C. Draine, and A. G. Greenwald (1998), was used to argue for the key dissociation. When the assumptions of these methods are not met, however, simulations indicate that significant positive intercepts can often be obtained even if there is no dissociation. Moreover, the assumptions are theoretically implausible and inconsistent with some aspects of the results. Thus, the significant positive intercept is not strong evidence of the key dissociation.