The bootstrapping and repeated bootstrapping procedure
As was explained in the Introduction, bootstrapping P300 is a technique
for randomly sampling (with replacement) ERPs from a single subject’s
dataset. In the studies concerning detection of concealed information
with the CTP, a single dataset for one participant usually consists of
~30 probe sweeps (i.e., single trials) and
~150 irrelevant sweeps (e.g., 5 irrelevants presented 30
times each). Each resampling (bootstrap iteration) is based on the same
set of sweeps (e.g., 180 sweeps total). Because the sampling with
replacement method is used, the probability that subsequent samples of
sweeps will be identical to previous samples is reasonably low.
In each iteration, two sets of sweeps are sampled with replacement from
whole datasets, separately for probes and irrelevants (see Figure 1).
Then, an average probe ERP and an average irrelevants ERP is computed.
Next P300 amplitude values are calculated. We used the peak-to-peak
(p-p) method, which has been shown to yield the highest classification
performance in P300-based CITs (Soskins, Rosenfeld & Niendam, 2001).
The methods for the p-p analysis are described in detail in several
recent reports (Olson, Rosenfeld, Perrault, 2019; Lukács et al., 2016).
Importantly, the p-p method derives P300 amplitude using averages across
time windows, not peak amplitude values that are easily
susceptible to variation and error. When the Probe (P) and Irrelevants
(I) P300 amplitudes are determined, the P-I difference (d ) is
calculated and is marked as 1 if P-I > 0 or 0 if P-I
<= 0. The whole process is repeated b times (usually
100) and provides the researcher with the proportion of times the P-I
differences were larger than 0 (see Figure 1, steps 1-2). For example,
such an analysis might result in BSITER score = 92, which indicates that
in 92/100 of comparisons the Probe amplitude was larger than the
Irrelevants amplitude. Such a result in concealed information studies is
usually interpreted as a sign of “guilt”, or recognition of the probe
item.
In the repeated bootstrap procedure (rBS) the calculation of a BSITER
score is performed r times (see Figure 1, step 3) which provides
new information about the consistency of results and their distribution,
among other things. Figure 1 presents the whole procedure graphically.