Foundations of response time measurement
20.10.2017 (Fr) 15:00 Uhr, VMP 11, Hörsaal
Abstract
Response time (RT) is one of the most important variables in psychology.
In RT research and diagnostics, participants repeatedly perform a task,
and the average duration between presentation of the stimulus and
execution of the correct response is estimated. If incorrect responses
occur, this estimation is difficult because the time between
presentation of the stimulus and the execution of the correct response
is not known in all task repetitions. The ad hoc solution of this
problem is to determine RT solely on the basis of the correct responses,
which yields an overly optimistic, and incomplete, picture of
performance. In this presentation I investigate imputation methods for
incorrect responses based on the two canonical response time models for
two-choice decisions. In counter models with independent accumulators
for the different response options, the incorrect response is masking
any further accumulation of evidence for correct alternative. In
contrast, random walk models (e.g., the two-barrier diffusion model)
assume a single continuum of evidence that is mapped to an ordinal
performance scale (fast correct < slow correct < slow error < fast
error). The two model classes suggest different summary statistics for
response time data (independent accumulators: Kaplan-Meier-estimator,
random walk: rank-based statistics). Using simulations and validation
data from a Neuropsychological test battery I investigate reliability
and validity of these different performance measures and make
recommendations on the treatment of contaminated data in RT research and
diagnostics. The improved estimates allow for flexible experimental designs with variable response deadlines.