Bootstrapping in Psychophysiology
Bootstrapping is a data-based simulation technique for making
statistical inferences (Efron & Tibshirani, 1994). It is especially
useful for estimating the repeatability and reliability of a statistical
result. Over 30 years ago, Wasserman and Bockenholt (1989) introduced
bootstrapping to the field of psychophysiology. It has been shown to be
useful in a variety of applications, including research on event-related
potentials (ERP) such as the P300 waveform (e.g., Fabbiani et al.,
1998). More recently, it has been used to interrogate
psychophysiological methods, such as computing standard measurement
errors to obtain a universal metric of ERP data quality (Luck et. al.
2021).
Although bootstrapping is used in a variety of fields, it is commonly
utilized in the detection of concealed information. For instance, the
P300-based Concealed Information Test (P300-CIT), has been studied for
over 30 years and reliably uses the bootstrapped P300 to detect
concealed knowledge of privileged, typically crime-related, information
(Rosenfeld, 2020). In the P300-CIT, a series of crime relevant(Probe) and crime irrelevant (Irrelevant) stimuli are presented
to a participant, and based on their responses to these stimuli one can
classify them as being knowledgeable or unknowledgeable of crime
relevant information. Another stimulus called a Target is used to
maintain the attention of the participants by asking them to produce a
unique response (e.g., on a keyboard) everytime they see it (Rosenfeld
et al., 2006; Gamer & Berti, 2012). Guilty/knowledgeable individuals
will have a differential psychophysiological response to crime relevant
information compared to crime irrelevant information. For instance, the
probe, which is crime relevant, would be salient only to someone who is
“guilty” resulting in a larger P300 for probes in comparison to
irrelevants. For this reason, the probe-irrelevant difference (i.e.,
“CIT effect”) is used to determine if the individual is “Guilty”
(i.e., knowledgeable) or “Innocent” (i.e., unknowledgeable)
(Rosenfeld, 2020).