INVESTIGATION USING BRAIN ELECTRICAL ACTIVITY MAPPING
6.5. The lunch
Lunch for Group 1 comprised a starter of vegetable soup (Heinz, 2l()g) followed hy baked potato (4(X)g) with baked beans (Tesco, 22()g). A cup of carbonated mineral water (140 ml) was supplied as liquid refreshment. All cooked meals were prepared as described in experiment one.
6.6. EEG apparatus and recording protocol.
EEG activity was recorded in real-time from 28 electrode sites according to the 10/20 international placement system using a Neuroscience Brain Imager Series 3 model and referenced to linked ears. Each of the 5 classical EEG frequency bands- della (0-311/), theta (4-7Hz), alpha (8-12Hz), beta 1 (1.3-22H/.) and beta I I (23-3()Hz) was monitored and recorded. Recording was undertaken at 256 microvolt sensitivity. The incoming EEG waveform is represented by 4096 values. With the 256 microvolt dynamic range, an accuracy of 256/4096 or 1/16 uvis obtained. Fillers were set at default settings (0.3 Hz & 40 Hz). A Fast-Fourier Transform algorithm decomposed the raw EEG into the classical bandwidths at the rale of a 1(X) samples per second. This operation enables the display and storage of the relative amplitude for each of the frequencies (see A ppendix C2).
The EEG is recorded as 2.56 seconds of averaged EEG activity (epochs). Power (uV2) in each frequency band is calculated for each epoch. Imager software calculates the amplitude for each electrode by taking the square root of the power. Data from each electrode site are taken to determine the voltages between electrode points (via a 4-point interpolation routine). Activity within this averaged epoch is consequently represented in visual form as the topographic map. A numerical value (in microvolts) is also obtained for each electrode site in each frequency band. The sampling rale is normally
twice the maximum frequency studied [the Nyquist frequency, Johnson (1980)].
Details of the brain imaging system are given in Appendix C l.
15 frames of no odour were recorded followed by 15 frames when the odour was present. A further fifteen frames were recorded post-odour where no smell was present. During the blank condition, 45 frames of no-odour were recorded. The aim of recording the first set of frames was to accustom the subject to the experimental setting and to enhance the feeling of relaxation. The final set of fifteen frames was used in order to allow the odour lime to dissipate (the blank condition employed this condition in order to keep all trials time-constant) and to allow the subject’s nose to recover from possible olfactory fatigue. Records were visually inspected for artifact employing a technique recommended by Neuroscience. Records showing significant amounts of symmetrical theta or delta activity in frontal leads were removed from the subject's overall EEG record. Those frames remaining were used in subsequent analysis. All artifact detection was made blindly, i.e., without knowledge of which record represented which odour condition.
RESULTS
6.7. EEG data reduction.
EEG data from nineteen electrodes were used for subsequent analysis. These electrodes were selected on the basis of previous studies which have shown odour- and affect-specific responses at these sites (see Tables 4a & 4b in C hapter 4). The electrode locations represented broad delineations of lobe area: frontal (F.3, F4, F7, FX, FZ), temporal (T3, T4, T5, T6), parietal (P3, P4, PZ), occipital (01, 02) and central (C3, C4, CZ) [see Fig. 6], In addition to the theoretical parsimony of reducing the
Electrodearray usedfor analysisinexperiment two.
number of electrodes and selecting them judiciously, there is also the statistical advantage of avoiding p-inflation (see Section 4.5., C hapter 4) since large quantities of EEG factors (such as electrode site) would certainly weaken the power of an analysis of variance.
Data from each electrode in each waveband were downloaded from the Brain Imager to a Macintosh hard disk using the Autoedit 4.1.2 downloading programme (Milligan, 1992).
One method of data reduction in experiments in which variables are numerous is to create a simple macro programme which can be tailor-made to perform some specific reductive function such as averaging or summing. This task is effectively undertaken in SYSTAT (SYSTAT, Inc., 1992) by typing in the macro in the command editor window. This manual procedure, however, is time-consuming and arduous since it permits the submission of only one file at a time: alterations to the macro may have to be made for each successive file.
To combat this wasteful expenditure of time and energy a programme compatible with the SYSTAT statistical package was designed using SYSTAT software and HyperCard 2.0v2. "Systat Batch Commander" allows multiple files to be processed sequentially and automatically (hence "batch") using a single command file (Milligan & Martin.
1994; see also Appendix D). This programme rapidly converts the EEG data into a form suitable for analysis in SYSTAT. Imported and edited files were log transformed in order to avoid skewedness and to remove nonstationary variability such as increasing variance across lime. The transformed data then underwent a repeated measures analysis of variance. Each electrode was analysed separately using an odour x group x time-of-day mixed design in the first instance.
6.8.i. Odour identification.
The procedure for the correct classification of odour was as described for experiment one. A post-hoc sign test revealed significant differences in the identifiability of some odours. Spearmint was identified by more people than was cumin (p=0.039), strawberry (0.022), garlic and onion (p=0.013) and almond (0.039). Furthermore, more people identified the vegetable correctly than they did garlic and onion (p=0.021).
The odours, ranked for identifiability and familiarity, are presented in T a b l e 6a.
Table 6a. Identification of odours (from most to least well identified) and familiarity of odours (from most familiar to least). Means lor familiarity are given with standard deviations in brackets.