3.3.1 Regression
To further consider the validity of the STP approach, linear regression
analyses were completed to determine the extent to which the component
measurements obtained from this approach accounted for the
between-participant variance of components obtained the from averaged
ERPs. A separate regression analysis was conducted for each component
(i.e., P1, N1, P2, N2 and P3) using the same general model of four input
variables entered in three steps. Step 1 included the participants’ mean
amplitude of the STP (baseline-to-peak) component; in Step 2,
participants’ the standard deviations of both the latency and amplitude
of the STP component were added; in Step 3, participants’ mean noise
value was added. For all components, the regression models showed a
significant amount of variance accounted for in each successive step.
Therefore, R 2, R 2Change, and the standardized coefficients only after the third step are
reported in Table 10.
In the models of each component, the results after the third step
indicate that of the four input variables, the mean amplitude variable
from the STP approach had the highest beta weight and accounted for a
significant amount of variability in its corresponding component
measurement from the averaged ERP. Noise accounted for a significant
amount of variability in the early components (P1 and N1); however, the
beta weight values decreased suggesting it accounted for a smaller
portion of the variability in the later components. Although the beta
weights of the standard deviation of latencies were significant
indicating that the latencies obtained in the STP approach accounted for
some variability, this was much less so than the mean amplitude or
noise. Across all components, the beta weights of the standard deviation
of the amplitude were not significant. Interestingly, the results of the
second step, the step prior to the addition of the noise factor, the
beta weights of the standard deviation of the amplitude were significant
in each of the regression analyses. The fact the amplitude variability
is no longer a significant predictor of the averaged ERP component when
noise is introduced to the prediction model suggest that these two
variables are colinear, i.e., they share a common portion of the
variance. Taken together, these results of these regression analyses
indicate that the single trial peak measures are strong predictors of
the stimulus-locked averaged ERPs, and thus, may be considered valid
measures of brain activity.