15:00 - 16:30
Talk Session VII
+
15:00 - 16:30
Wed-HS1-Talk VII-
Wed-Talk VII-
Room: HS1
Chair/s:
Simone Malejka
Can we detect regularities in our environment and adapt behavior accordingly in the absence of awareness? The demonstration of unconscious (implicit) cognition hinges on participants’ unawareness of stimuli, processes, or products involved in a task. The gold standard is to establish an indirect-without-direct effect, that is, an uninstructed effect of a stimulus on behavior under conditions that preclude any effect of the stimulus on a response according to explicit instructions. This symposium will bring together researchers working on new methods tailored to investigate the possibility of indirect-without-direct effects. The first two talks will present novel indirect and direct
measures for well-known experimental paradigms. Sascha Meyen will demonstrate a new test of reaction-time differences, which offers an improved indirect measure and provides evidence against unconscious processing in contextual cueing. In the area of priming, Thomas Schmidt will talk about a new theory of visibility focusing on the critical stimulus feature that generates the indirect effect and must be assessed in the direct measure. The final three talks will present new analyses for data that presumably show an indirect-without-direct pattern. These data often suffer from regression to the mean (RttM), defined as the statistical phenomenon that makes natural variation in repeated data look like real change. When direct measures are contaminated with measurement error, low awareness scores will tend to be followed by awareness scores closer to the mean. Itay Yaron will outline a solution to the RttM problem that uses a widely applicable bootstrapping algorithm based only on a small set of assumptions. Simone Malejka will present a method of true-score estimation based on the Bayesian principle of shrinkage, which corrects noisy data and can solve RttM and related measurement biases. Lastly, Zoltan Dienes will demonstrate how Bayes factors can provide evidence for (or against) one’s theory in the presence of measurement error by testing an interval null hypothesis of zero awareness in post-hoc trial selection.
Back
A simple formula for Bayesian shrinkage to correct regression to the mean
Wed-HS1-Talk VII-04
Presented by: Simone Malejka
Simone Malejka 1, Miguel A. Vadillo 2, David R. Shanks 3, Zoltan Dienes 4
1 University of Cologne, 2 Universidad Autónoma de Madrid, 3 University College London, 4 University of Sussex
As a method to investigate the scope of unconscious mental processes, researchers frequently collect a behavioral measure (e.g., some assessment of learning) and a measure of awareness (e.g., recognition judgments) of the critical cue or contingency. Evidence that behavioral change was indeed unconscious may require that participants were unaware of the critical regularity or that behavior and awareness are independent—both of which are commonly demonstrated using standard statistical tests (e.g., t-tests, correlation/regression analysis). We highlight a critical limitation in these approaches: Systematic bias caused by ignored measurement error (e.g., regression to the mean, regression attenuation) can lead to false-positive rates up to 100%. As a solution, we propose a correction formula for observed data based on the ideas of true-score estimation in educational testing and shrinking estimates towards a grand mean in Bayesian modeling. Because error is defined as imprecise measurement of an individual’s true score, information about the individual (e.g., their group membership) and the measure (e.g., its reliability) can be used to move observed scores closer to their true scores. Our formula provides corrected estimates as a weighted combination of the observed score and the group mean. We discuss different weighting methods and compare their performances in simulation studies. Our work shows that true-score estimation provides the means to correct data for measurement error: The corrected data offer a more representative sample to test the research hypothesis and consequently better inferences in scientific decision-making. We conclude by offering best practices for correcting measurement error in psychological research.
Keywords: unconscious processes, implicit cognition, measurement error, reliability, regression to the mean, Bayesian methods
Back