This chapter delineates the methodology common to all five experiments. Experimental procedures specific to each study are described in the relevant method sections. All studies employed the same selection criteria for subjects, and similar experimental materials. All experimental designs were variations on a basic recognition memory exclusion paradigm similar to that employed in the process dissociation procedure (Jacoby, 1991). The ERP recording parameters were identical for Experiments 1 and 2. Adoption o f a new ERP recording system following the completion o f Experiment 2 m eant that electrode placement differed slightly between Experiments 1 - 2 and Experiments 3 - 5 . Experiments 3, 4 and 5 all employed the new ERP recording system. Data processing and data analysis also
remained constant across the five studies. All experiments were approved by the joint ethics committees o f the University College London and the University College London Hospitals.
Subjects
Experimental subjects were recruited from the undergraduate and postgraduate student populations o f UCL. All subjects were right-handed, native English speakers, were aged between 18 and 35, and had normal (or corrected-to-normal) vision. Subjects were screened for neurotropic medication, and were remunerated at the rate o f £5.00/hr (Experiments 1, 2, & 3) and £7.50/hour (Experiments 4 & 5). Subjects were also reim bursed for travelling expenses.
Materials
Stimulus lists were constructed from a pool o f 560 concrete nouns ranging in frequency between 1-30 per million (Kucera & Francis, 1967) and in length between 4 and 9 letters. Experimental stimuli were presented visually as upper case white letters on a black background. Stimuli were presented centrally on a computer monitor, and subtended an approximate vertical visual angle o f 0.4 degrees and a maximum horizontal visual angle of 2.0 degrees.
Experimental procedures
A basic three phase recognition memory paradigm derived from the exclusion component o f the process dissociation procedure (Jacoby 1991) was employed in all five experiments, although certain aspects o f this design varied between experiments. Exclusion instructions in recognition memory studies direct subjects to respond positively to old items from a specified source, and to select against old items from an alternate source. An exclusion task is
typically used in conjunction with an inclusion task (in which subjects respond positively to all old items) as a method o f separating estimates o f the contributions o f familiarity and recollection to overall recognition performance (Jacoby, 1991). The exclusion task was used in isolation in the following studies, and was manipulated in a num ber o f different ways in order to investigate the functional characteristics o f ERP correlates o f recognition memory.
Subjects were fitted with an ERP recording cap (described below) prior to the experiment, and were then seated in a sound-attenuated recording booth situated approximately one metre in front o f a computer monitor. It was explained that the experiment would consist o f three tasks, each o f which focused on a different aspect o f word processing. Subjects were not informed that they were participating in a memory experiment. They were instructed to relax, to keep still, and to maintain fixation at the centre o f the screen. During study phase 1, subjects were required to carry out a specified encoding task on visually presented words. Subjects were required to carry out a different encoding task on visually presented words during study phase 2. In Experiments 1-3, words presented in the two study phases were overlapping to a varying extent (i.e. 50% or 100% o f study phase 2 words had been
previously been presented in study phase 1). In Experiments 4-5, words presented in the two study phases were non-overlapping. During the test phase in all five experiments, subjects were presented with old items from both study phases along with new items. They were instructed to respond ‘old’ to recognised items from study phase 2, and to reject both new items and items from study phase 1.
ERP recording
Experiments 1 - 2
The EEG recording locations employed in Experiments 1 and 2 were based upon the international 10-20 system (Jasper, 1958). EEG was recorded from 25 tin electrodes
em b ed d ed in an elastic cap, w ith an ad ditional electrode p laced on the rig h t m asto id (see figure 4.1). All E EG ch an n els w ere referen ced on-line to an electro d e p laced on the left m asto id electro d e, and w ere su b seq u en tly re-referenced o ff-lin e to lin k ed m asto id s. EO G (electro -o cu lo g ram ) w as reco rd ed b ipolarly from one electro d e p laced on the o u ter can thus o f the left eye and a second electro d e placed above the supraorbital ridge o f the right eye. EEG reco rd in g locations con sisted o f three m idline sites [F z,P z,C z]; left and rig h t h em isp h ere sites [F P 1/F P 2, F 3 /F 4,F 7/F 8] and ad d itio n ally L F/R F (frontal, 75% o f the d istan ce betw een FZ and F7/F 8); C 3/C 4, T 3/T 4 and ad d itio n ally L T /R T (an terio r tem p o ral, 75% o f the distance betw een C Z and T 3/T 4); P3/P4, T 5/T 6 and ad d itio n ally L P /R P (p arietal, 75 % o f the distance b etw een PZ and T 5 /T 6 ) and occipital sites [ 0 1 ,0 2 ] .
Fpl Fp2 F8 LF F4 RF T3 T4 LT C3 C4 RT Cz T6 T5 LP P4 RP
oi
02
Figure 4.1. Selected sites from the international 10/20 system em ployed in Experim ent 1-2.
Experiments 3 - 5
EEG w as reco rd ed from 31 silv er/silv er chloride electro d es, 29 o f w h ich w ere em b ed d ed in an elastic cap, and 2 o f w hich w ere placed on the right and left m asto id s (see figure 4.2 for m ontage). T he 31 site m ontage em ployed selected sites from the m o n tag e 10 61 channel eq u id istan t m o n tag e (h ttp ://w w w .e a sy c a p .d e /ea sy c a p /e n g lish /sc h e m a e .h tm ). T w en ty -fiv e o f these sites w ere co m p arab le to those em ployed in the 10/20 m o n tag e d escrib ed for
E x p erim en ts 1 and 2. A n additional four sites w ere located o v er in ferio r lateral fro n to p o lar and occipital regions. A ll ch an n els w ere referen ced to Fz d u rin g reco rd in g , and w ere su b seq u en tly re-referen ced o ff-lin e to linked m astoids. H orizontal E O G w as reco rd ed b ip o larly from electro d es on the ou ter can th u s o f each eye, and vertical E O G w as reco rd ed b ip o larly from electro d es p laced above and below the cen tre o f the rig h t eye.
Figure 4.2. S elected sites from ttie m ontage 10 61 channel equidistant m ontage em ployed in Experim ent 3-5.
Experiments 1 - 5
The following EEG recording parameters remained constant across all five experiments. On line sampling was performed with a sampling interval o f 6msec per point for a total o f 1,536 msec. This included a 102 msec pre-stimulus baseline period, resulting in a post-stimulus recording epoch o f 1434 msec. All channels were amplified with a bandpass o f 0.03 - 30Hz (3 dB roll-off;.
Data processing
Trials on which baseline drift exceeded 54.9 microvolts at any site were rejected, as were trials containing A/D saturation. Blink artefacts were minimized by estimating and correcting the contribution o f the vertical EOG channel to the ERP waveforms via a regression
technique. Trials exhibiting horizontal movements and non-blink vertical eye movements were identified during visual inspection and rejected. Waveforms in all experiments were re referenced to linked mastoids, and were digitally smoothed with a low-pass frequency o f 17 Hz (roll o ff 3Db). Averaged ERP waveforms were formed for each o f the response
categories o f experimental interest (defined within each experimental chapter) for each subject. Only subjects contributing a minimum o f 16 artefact-free ERP trials to each o f the critical response categories were included in subsequent statistical analyses both o f ERP and o f behavioural data. This criterion was imposed in order to achieve an adequate signal-to noise ratio in the ERP data. A minimum o f 16 subjects contributed ERP data towards the analysis o f each critical response category.
Data analyses
Both behavioural and ERP data were analysed by repeated measures ANOVAs (with exception o f Experiment 3, which employed ANOVAs o f a mixed design with one between- subjects factor). The Geisser-Greenhouse correction for inhomogeneity o f covariance (Greenhouse & Geisser, 1959) was implemented in all analyses. This procedure is necessitated by the fact that all data analysed does not necessarily exhibit the sphericity assumed by the ANOVA model. Sphericity is characterised as homogeneity o f covariance between levels o f factors. Although this issue is relevant to o f all types o f data, it is particularly pertinent to the analysis o f ERP data. ERP analyses treat each electrode site location as a separate observation. However, the covariance shared by two geographically
close electrodes is likely to be greater than the covariance shared by two geographically distant electrodes. ERP data can therefore easily violate the sphericity assumption. The Geisser-Greenhouse correction procedure estimates the extent to which the sphericity assumption has been violated, and reduces the degrees o f freedom employed by the ANOVA accordingly, thus reducing the probability o f a type I error.
F ratios are reported with the corrected degrees o f freedom where appropriate. Analyses were performed on accuracy and reaction time (RT) data from all subjects contributing ERP data. Observed interactions in the behavioural analyses were decomposed by means o f targeted Bonferroni corrected t-tests. Averaged ERP data associated with the critical response categories were subjected to two types o f analyses.
Magnitude analyses
Averaged ERP waveforms associated with each o f the response categories o f experimental interest were contrasted in order to determine the extent to which they differed in amplitude. Magnitude analyses were also used to investigate the scalp location/s over which ERP effects were maximal in amplitude. ERP data were quantified by measuring the mean amplitudes associated with the various response categories o f interest within specific latency regions, relative to the mean o f the pre-stimulus baseline.
Two sets o f magnitude analyses were carried out for each study. Both sets o f analyses altered slightly between Experiments 1-2 and 3-5 due to the adoption o f the new electrode placement system. One set o f analyses employed a grid o f distributed sites selected to permit an
assessment o f which scalp locations were sensitive to the experimental manipulations. In Experiments 1-2 these global analyses employed the factors o f response category (defined within each experiment) and three site factors (frontal/temporal/parietal location, left/right hemisphere and inferior/mid-lateral/superior site). Global analyses in Experiments 1-2 included lateral frontal sites (F7, LF, F3, F4, RF, F8), lateral temporal sites (T3, LT, C3, C4, RT, T4) and lateral posterior sites (T5, LP, P3, P4, RP, T6). In Experiments 3 -5, global analyses employed the factors o f response category (defined within each experiment) and three site factors (frontopolar/frontal/parietal/occipital location, left/right hemisphere and inferior/superior site), and included lateral frontopolar (49, 50, 37, 36), lateral frontal (33, 19, 9, 22), lateral parietal (30, 29, 25, 26) and lateral occipital sites (45, 44, 41, 42) sites. A second set o f analyses were guided a priori by the ERP literature. Two principle ERP
old/new effects have been associated with dual-process models o f recognition memory in the ERP literature; the early frontal old/new effect and the later parietal old/new effect. This literature is reviewed in full in Chapter 3. However, these effects are briefly summarised below in order to explain the rationale underlying the ERP analysis strategies adopted in the following studies. As the aim was to fractionate established ERP correlates o f recognition memory and to investigate their functional characteristics, it was desirable to maximise sensitivity to the presence or absence o f these effects under various experimental
manipulations. Focused analyses on the sites o f interest permitted a sensitivity which is often lost in more general global ERP analyses.
Data from mid-frontal sites were analysed during the 300-500 msec latency region. These analyses were guided by previous findings indicating an early frontal positivity associated with old items and familiar lures relative to new items within this latency region (Curran,
1999; Curran, 2000; Rugg et al., 1998). This effect has been linked to familiarity-based recognition memory, as it is associated with a positive recognition response rather than veridical recognition. Magnitude analyses within the 300-500 msec latency region were restricted to three mid-frontal sites in order to explore this effect. The mid-frontal sites employed in Experiments 1 and 2 were F3, Fz and F4, whereas the sites employed in Experiments 3-4 were sites 19, 8 and 9. Mid frontal analyses in all five experiments
employed the factors o f response category and site. Many previous studies have also reported parietal positivity associated with old relative to new items within the 500-800 msec latency region, an effect that is often left-lateralised and has been linked to recollection (Wilding & Rugg, 1996; Smith, 1993; Wilding, 2000). This ERP effect is often referred to as the ‘parietal old/new effect’. Consequently, magnitude analyses carried out within the 500-800 msec latency region focused upon lateral parietal sites. In the 10/20 system employed in Experiments 1 and 2 these sites comprised T5, LP, P3, P4, RP and T6. In the montage 10 61 channel equidistant montage employed in Experiments 3-5 these sites consisted o f sites 46, 30, 29, 26, 25 and 40. A priori analyses at parietal sites in all five experiments employed the factors o f response category, hemisphere and site.
It should be noted that post-hoc analyses o f ERP data are rarely corrected for multiple comparisons, and this is true o f the analyses presented within this thesis. There are two principle reasons for this. Firstly, ERP analyses contain a number o f electrode site location factors in addition to factors relevant to specific experimental manipulations. Post-hoc
analysis o f ERP data do not therefore merely focus on interactions between factors pertaining to experimental manipulations, but also on interactions between these and electrode site location factors in an attempt to determine the scalp distribution o f each ERP effect o f interest. As a relatively large number o f post-hoc analyses are required to meet both o f these aims, correcting for multiple comparisons would result in a greater number o f Type 2 errors and might therefore be overly conservative. Secondly, ERP data is particularly susceptible to violations o f sphericity as is discussed above. Therefore, any post-hoc analysis technique must be able to correct for these violations o f sphericity. No satisfactory technique has yet been developed which corrects for violations o f sphericity and which is suitable for the post- hoc analysis o f ERP data. Although it is acknowledged that post-hoc analyses that do not correct for multiple comparisons are unsatisfactory, they are currently necessitated by inadequate statistical techniques. In order to protect against Type 1 errors, the majority o f ERP research relies instead on the replicability o f results, and draws on previous findings in the ERP literature both to minimise the number o f post-hoc comparisons required and to interpret the results o f these comparisons. This approach is adopted throughout this thesis.
Topographic analyses
W here appropriate, topographic analyses were conducted in order to determine the scalp distribution o f ERP effects identified in the magnitude analyses. Topographic analyses are employed only when one wishes to demonstrate a qualitative dissociation between an ERP effect at a particular latency region either with a second ERP effect, or with the same ERP effect at a different latency region. These analyses are carried out on ‘difference’ waveforms (i.e. differences between the ERPs to the two response categories forming the contrast o f interest). A potentially problematic characteristic o f ERP data is that any change in the activity o f an underlying generator has multiplicative effects on the amplitude o f activity detectable at the scalp. This characteristic is at odds with the ANOVA model, which assumes that the same change in generator activity will produce additive effects. If an experimental manipulation elicits a change in dipole strength, this will not produce a constant change across electrode sites (which would result in a main effect o f condition), but will produce different changes at different electrode sites, thus resulting in a condition/site interaction. Consequently, magnitude analyses do not dissociate qualitative differences from quantitative differences. Topographic analyses detect qualitative differences between response contrasts
as they are performed on data rescaled to satisfy the additivity assumptions o f the ANOVA (McCarthy & Wood, 1985).
The rescaling method employed in the following studies was formulated by M cCarthy & Wood (1985), and calculates the amplitude o f the ERP effect o f interest at each electrode site relative to all other sites. The rationale underlying this form o f rescaling can be understood by representing ERP data as points in multidimensional space (McCarthy & W ood, 1985). The scalp distribution o f an effect is represented as a vector in N-space, with each axis reflecting the voltage at each electrode site. The shape o f the distribution is a function o f the vector’s orientation. The length o f the vector represents the amplitude o f the distribution, and is derived by calculating the square root o f the sum o f squared voltages over all electrode sites. The aim o f rescaling is to maintain differences in shape while eliminating differences in amplitude. This is achieved by scaling the voltage at each electrode site by the vector length associated with the distribution. The scalp distribution o f the effect is thus maintained while removing differences in amplitude (McCarthy & Wood, 1985). Any resulting
interactions between response category and electrode site can then be cited to support the assertion that the response categories o f interest reflect the activity o f at least partially non overlapping neural populations. However, the drawback o f topographic analyses is that they cannot detect main effects o f condition. It is therefore desirable to carry out magnitude analyses in order to identify main effects and differences in amplitude prior to conducting