General Method
5.4. Data processing and analysis
5.4.1. Measuring discrimination accuracy
Discrimination accuracy is reported for all experimental tasks that required discrimination between previously studied and unstudied stimuli. Estimates of discrimination accuracy were calculated using the two-high threshold model (Snodgrass
& Corwin, 1988). The discrimination index (Pr = Hits - False Alarms) was used to correct discrimination scores for lucky guesses. Although the discrimination index proposed by Snodgrass and Corwin, (1988) is applied routinely in the recognition literature, the index breaks down under certain circumstances. For instance, where a participant makes no errors, the two-high threshold model becomes undefined for hit rates of 1.0 or false alarm rates of 0 because the corresponding z scores are infinite (Snodgrass & Corwin, 1988). As a correction for values of 1 and 0, Snodgrass and Corwin, (1988) proposed that both the hit rate (number of hits + .5/number of old items
139 +1) and false alarm rate (number of false alarms +.5/number of new items +1) are routinely adjusted. This transformation has been applied with hit rates and false alarm values of 1 and 0 throughout the current thesis.
Importantly, studies of discrimination accuracy typically employ a measure of response bias (see Section 1.2.1. for more detail) to determine if participants are making conservative or liberal response judgements. In the current thesis however, the Unitization tasks employed required participants to make a three-way discrimination response between Intact, Recombined and New word pairs, a procedure identical to other ERP associative recognition studies (i.e., Bader et al., 2010; Rhodes &
Donaldson., 2007, 2008). The inclusion of a Recombined response is important to ensure that participants do not identify an Intact pair based on the recognition of a single item – instead, the presence of Recombined items forces the participant to retrieve the association between pairs to make a successful judgement. The inclusion of a three-way response, however, makes the task inherently ambiguous with regards to response bias (which only accounts for bias between old and new responses). To date, the problem of accounting for response bias during a three-way decision task has not been resolved and it is therefore not included in the thesis. Regardless, the exclusion of a response bias measure does not have any direct bearing on the principle concern of the current thesis – to demonstrate difference in discrimination accuracy between Intact and New pairs between Unitized and Non Unitized tasks.
The three-way associative discrimination task employed in the current thesis is also distinct from previous behavioural associative recognition tasks that require participants
140 to discriminate Intact from Recombined pairs. The inclusion of a third response will therefore make discrimination on any one trial more uncertain. Consequently, the results from the current experiments may not be directly comparable to behavioural studies employing a more simplified binary judgment. Employing ‘New’ pairs does, however, facilitate comparison with old/new effects strongly associated with recollection and familiarity (as outlined in Chapter 4). To ensure that discrimination accuracy accurately reflected the ERP Hit-Correct Rejection comparison, we treated Recombined responses as Hits; resulting in two types of false alarm (i.e., to Intact and Recombined pairs: identical to the procedure employed by Rhodes & Donaldson, 2008). Discrimination accuracy was therefore calculated separately for Intact and Recombined pairs. Although we accept that Recombined pairs can be treated as Correct Rejections, treating them as Hits allowed us separately analyse Intact/Correct Rejection discrimination that was of primary interest in the current thesis.
5.4.2. ERP data acquisition
EEG was measured at the scalp using 62 silver/silver chloride electrodes embedded in an elasticated cap (Neuromedical supplies: www.neuro.com) in accordance with an extended version of Jasper’s (1958) International 10/20 system: (FP1, FPZ, FP2, AF3, AF4, F7, F5, F3, F1, FZ, F2, F4, F6, F8, FT7, FC5, FC3, FC1, FCZ, FC2, FC4, FC6, FT8, T7, C5, C3, C1, CZ, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPZ, CP2, CP4, CP6, TP8, P7, P5, P3, P1, PZ, P2, P4, P6, P8, PO7, PO5, PO3, POZ, PO4, PO6, PO8, CB1, O1, OZ, O2, CB2). The ground electrode GND was positioned midway between AF3 and AF4. During recording, each electrode was referenced to an additional electrode midway between CZ and CPZ. All channels were re-referenced offline to a virtual
141 mastoid that was calculated by averaging the signal from electrodes located on the left and right mastoids. Vertical and horizontal EOG was recorded from bipolar pairs of electrodes placed above and below the left eye, and on the outer canthi. Electrode impedance was kept below 2kΩ. EEG and EOG data were amplified with a band pass filter of 0.1 – 40Hz, digitized by a 16 bit analogue to digital converter at a sampling rate of 250Hz and recorded on a desktop computer using Neuroscan Aquire software (Version 4.3). EEG data were processed using Neuroscan Edit (version 4.3).
The raw EEG was inspected and segments of data including high levels of noise (i.e.
artefacts including excessive muscle movements) were removed. An ocular artefact reduction procedure was applied to reduce the effects of eye blinks: 32 optimal blinks from each participant were selected to estimate the individuals blink pattern and remove the contribution of the average blink from all channels. The EEG data was then epoched and time-locked to stimulus presentation at test using a 2040ms time window (starting with a 104ms pre-stimulus baseline). Epochs were rejected if they had a drift from baseline exceeding ±75 µV, or where the signal change exceeded ±100 µV. Averaged ERPs were then formed from correct responses and the data was baseline corrected and smoothed with a 5-point kernel. To ensure a good signal-to-noise ratio a minimum of 16 artefact-free trials was required from each participant, in each of the critical response conditions. The mean numbers of trials contributing to the grand average ERPs are described in the relevant experimental sections.
142
5.4.3. ERP analysis
In the current thesis, ERP ‘effects’ correlated with successful memory retrieval were of primary interest. For Chapters 6, 8, 9 and 10, memory retrieval was analyzed by comparing the difference in magnitude between ERP waveforms elicited by Hits (correctly identified ‘old’ items) and Correct Rejections (correctly identified ‘new’
items). Initial analysis focused on the 300-500ms and 500-800ms time windows;
typically found to capture the neural correlates of familiarity and recollection respectively. To characterize the neural correlates of successful memory retrieval, mean amplitudes were calculated over the duration of each time-window for different electrodes and submitted to repeated measures ANOVA. In order to reliably capture the topography of old/new effects, selection of the number and location of electrodes are described in each experimental chapter. Only significant effects involving retrieval are reported and the Greenhouse Geisser correction for non-sphericity was applied where appropriate. Topographical analyses were employed on re-scaled data using the Min-Max method described by Mccarthy & Wood, (1985).
143