3.1 The Feasibility of Single Trial Measures
Signal to Noise Ratios: SNR values for each component in a
participant’s averaged ERP were calculated in a manner recommended by
Spencer (2005), Debener et al. (2006) and Hu et al. (2010) where the
“signal” was the peak amplitude of the component, and the “noise”
was the standard deviation of the amplitude of all data points in the
baseline period (i.e., -200 ms to 0 ms). For both the STP and STW
approaches, SNR values were first calculated at the trail level using
the same definitions of “signal” and “noise”. Additionally, for
direct comparison purposes to the SNR derived each individual
participant’s averaged ERPs, the SNRs at the trial level were then
averaged for each participant. For a brief summary, means across
participants and their corresponding standard deviations for each
component are shown in Table 1.
As expected, and in keeping with the justification for averaging ERPs
often found in the literature, the mean SNR for all components was the
largest when the components were measured from averaged ERPs. However,
the general assumption that the “signal” of interest is too small to
be measured in single trials may not be true for at least the paradigm
used in this study as the mean SNRs for the STP approach, while lower
than SNRs of the averaged ERPs, did
range from 1.62 to 2.9 and were
larger than the SNRs of the STW approach. More importantly, the
within-group variability (i.e., SD s) of the SNRs for the STP
approach is considerably smaller than the variability of the SNR of the
averaged ERPs but larger than those for the STW approach. A more direct
comparison of the three approaches can be made by computing their
coefficient of variation
(CV ). The lower the coefficient of variation value the better, as
it denotes that the spread of data values are low relative to the mean
of the values. When the differences in the variability of the approaches
are viewed in relation to their mean amplitudes as shown in Table 1,
other than for P1, the STP and STW approaches are nearly the same
regarding finding a “signal” within “noise”, and both have much
smaller CV s than the averaged ERPs.
Comparison of Measurement Approaches: To allow for direct
comparison between the STP (baseline-to-peak), STW, and averaged ERP
approaches, the ST data for each participant were first averaged across
trials then these averages were summarized across participants to
produce descriptive statistics; the means, standard deviations, minimum
and maximum values. These descriptive statistics for the amplitude
measures derived from the averaged ERP, STW, and STP approaches are
provided in Table 2. Descriptive data for the latencies for the averaged
ERP and STP approach are provided in Table 3. In general, the STP
approach produced the largest within-group mean amplitudes along with
having the largest standard deviation across all of the components
measured. With the exception of P1, the averaged ERP approach produced
considerably larger with-group means but only slightly larger standard
deviations than the STW approach. However, when these approaches are
viewed in relation to their coefficient of variations (Table 4), the STP
approach had considerably smaller CV values than the other two
approaches for all components. The STW approach had larger absoluteCV values than the averaged ERP approach for all components
except for the P1 component.