The assumption in cognitive neuroscience is that electrophysiological activity maps directly (or indirectly) onto psychological phenomena. It is important to keep in mind, however, that the observed ERPs merely correlate with the cognitive processes under investigation and cannot be assumed to be straightforward manifestations of those processes. Regardless, when the ERPs have been extracted from the ongoing EEG recording, an attempt must be made to somehow interpret them with regard to their cognitive meaning. The first step in this process is to appropriately identify and select the ERP components to be examined.
2.4.1. Component Selection
In the early days of ERP research, components were defined in terms of their polarity, latency and distribution on the scalp (Luck, 2005), however these qualities of are not very informative as a way of identifying the cognitive processes that the ERPs correspond to. Many researchers (e. g. Donchin, Callaway, Cooper, Goff, Hillyard & Sutton, 1977) have therefore adopted a “functional approach”, focussing on an ERP component’s relationship with experimental variables rather than its peaks and troughs. To follow the functional approach it is necessary to design tasks that have the potential to isolate and contrast specific cognitive processes, allowing ERPs elicited in two different experimental conditions to be subtracted from one another (see Rugg & Coles, 1995). The resulting component reflects the difference in activity that distinguishes the experimental variables.
The functional approach includes two underlying assumptions: the latency of the ERPs to be subtracted must be equal and the experimental conditions that produce them must differ with regard to only the cognitive process of interest. If the first assumption is not met, the subtraction will produce separate peaks in the waveform and thereby mistakenly give impression that the two processes differ qualitatively.
The second assumption, also known as the pure insertion principle (Donders, 1868), presupposes that cognitive functions are additive and do not interact with each other. In most cases, however, this assumption is unlikely to be valid; two conditions will consist of a number of shared cognitive components, each of which will be influenced by the introduction of additional components. Consequently, the subtraction will reflect a combination of the added and the shared (but adapted) components. It is worth noting, however, that violation of the pure insertion principle is not unique to ERP research but applies to all experiments that involve comparisons by subtracting data (including behavioural experiments and other experiments using other neuroimaging methods).
2.4.2. Making Inferences from ERPs
Identifying that experimental manipulations give rise to different patterns of brain activity does not in itself inform the specific nature of these differences. Interpreting ERPs is a notoriously difficult process, complicated by many of the issues covered in above sections. Nonetheless, the consistency of findings across numerous studies provides confidence in its value as a tool for investigating human cognition. ERPs can be interpreted in terms of their temporal, size and distributional characteristics, each of which will be discussed in turn below.
The major advantage of using the ERP technique over haemodynamic imaging methods (such as fMRI and PET) is their high temporal resolution; latency differences can help establish the time it takes the brain to differentiate between two experimental conditions. Importantly, however, the ERPs can only provide an upper-bound estimate of timing differences, because earlier differences could occur which are not detectable on the scalp. Amplitude differences, on the other hand, are believed to correspond to the strength or degree of processing. Higher amplitudes elicited by one condition over another suggest that the same process is occurring in both cases but is differentially engaged across the conditions (although it is also possible that differences in amplitudes are caused by an ERP effect being present on a different proportion of trials, rather than being smaller in magnitude per se). Also, as noted earlier, differences in latencies across individual trials can result in erroneous amplitude differences in the averaged waveform; hence the interpretation of quantitative differences must always be made with caution.
When one experimental condition gives rise to an ERP with a particular amplitude and latency at one location of the scalp, and another condition gives rise to an identical ERP but at a different location, it is reasonable to assume that the two conditions engage neurally and functionally distinct processes which happen to overlap in time (Rugg & Coles, 1995) 2. Although ERPs cannot provide accurate information about the specific anatomical structures involved, the differential distribution of effects is informative in itself. Unfortunately, in practice there are
2 The polarity of ERP effects are also of interest in this regard: when two effects differ in polarity it does not mean that different neural structures are giving rise to the two effects, but it does necessitate that different cognitive functions are operating (that might or might not be supported by the same underlying structures). Note, however, that polarity of an ERP effect does not carry any additional
serious challenges associated with statistically verifying that such qualitative differences actually exist. The repeated measures ANOVA (also used for analyses of quantitative differences) is based upon an additive model, whereas differences in dipole strength are multiplicative rather than additive. This mismatch has the potential consequence of producing the appearance of differences between conditions at some locations compared to others, which are not caused by differential activation in different underlying sources. During the analysis of effects, a simple main effect of condition could be wrongly interpreted as an interaction between condition and location. As a possible solution to this problem, McCarthy &
Wood (1984) recommend that ERP data are rescaled prior to the analysis of topographic distribution, as this would minimise the unwanted multiplicative effects. The most commonly used scaling strategy is the minimum-maximum method which involves normalisation of the data. The use of rescaling is vigorously debated (see Haig, Gordon & Hook, 1997; Ruchkin, Johnson & Friedman, 1999;
Urbach & Kutas, 2002; Wilding, 2006), but is still preferred by many researchers due to the reduced likelihood of type 1 errors.
2.5. Summary
Event-related potentials reflect activity (predominantly caused by postsynaptic potentials) originating mainly in the cortex which is consistently associated with the processing of a stimulus event. ERPs are extracted from the ongoing EEG, which is recorded by using electrodes situated on the surface of the scalp. The EEG needs to be amplified, digitised and filtered before multiple trials can be averaged together and the ERPs revealed. ERPs can be characterized in terms of their latency,
amplitude and distribution on the scalp – all of which provide information regarding the processes believed to be producing the signal.
ERPs are considered to be an important and useful tool with which to examine functional models of cognition, allowing cognitive processes to be defined according to their neurophysiological correlates. Although the spatial resolution offered by the ERP technique is rather poor, it provides excellent temporal resolution and is therefore an optimal choice for investigating timing aspects of mental operations.