Experiment 1: Olfactory n-back partial replication .1 Introduction .1 Introduction

Experiment 1 is a partial replication of Jönsson et al. (2011). The purpose of this replication is threefold. First, above chance olfactory n-back performance has only been shown in two previous studies (Dade et al., 2001; Jönsson et al., 2011), so the experiment seeks to replicate this finding with a different set of odours. Second, this experiment can validate recent normative data from Chapter 2, by demonstrating the same facilitative effect for highly verbalisable odours as that reported by Jönsson et al.

(2011). Indeed, the findings in Chapter 2 indicate that verbalisability of odours is an unsuitable dimension with which to control odours, due to high levels of individual differences. However, not only has a similar measure been used successfully in other research (Jönsson et al., 2011), but the verbalisability scores in Chapter 2 correlated strongly with other dimensions deemed suitable for use in olfactory memory experiments. Although this is not tested, one might speculate that verbalisability for low and high extremes of the verbalisability dimension are less susceptible to individual differences, and consequently the use of these odours enable an effective manipulation of odour verbalisability in the present tasks. Third, an additional testing sequence is introduced to investigate discriminability changes due to perceptual and verbal learning

throughout the task (Jönsson et al., 2011, employed a single testing sequence). That is, does performance improve for low verbalisability odours as a result of repeated exposure to those odours?

The present experiments use an index of recognition ability derived from signal detection theory, and additionally reports independent analyses of hits and false alarms.

This is necessary to provide insight into specific changes in the ability to reject lures and accept targets, in addition to shifts in response bias. For example, the findings in Lyman and McDaniel (1986) have been criticised due to the facilitation from labelling on recognition performance occurring through a reduction in false alarms rather than an increase in target recognition. However, signal detection is preferred over hit rates because nameability of an odour has been proposed to affect response strategy, with an item that is not identified judged as ‘new’ more frequently (R. A. Frank et al., 2011).

This may be because the information about an odour that is not named is very limited (Jönsson & Olsson, 2003), with participants therefore reluctant to respond old when they cannot report any information about the odorant.

Three hypotheses are presented based upon previous evidence of a working memory advantage for verbalisable odours and considering the effect multiple exposures may have on n-back strategy. It is predicted that (1) odour working memory will be above chance for low verbalisability odours but (2) performance for high verbalisability odours will be significantly better (Jönsson et al., 2011). Across testing sequences, it is predicted that (3) greater improvement for low verbalisability odours from repeated presentations, through a process of perceptual learning and verbal learning (Nguyen et al., 2012; Stevenson, 2001; Stevenson & Mahmut, 2013a).

3.1.2 Method 3.1.2.1 Participants

Twenty participants (12 males and 8 females, mean age = 20.0, SD = 2.7) participated in exchange for course credit. Participants who self-reported olfactory impairments (e.g.

symptoms of cold) and smoking (Katotomichelakis et al., 2007) were excluded, as were participants aged over 40 years (Doty et al., 1984). Ethical approval was obtained via the Bournemouth University Ethics Committee.

3.1.2.2 Materials

The odours were as described for Chapter 2.

Twelve odours were randomly selected from the twenty highest and lowest verbalisability scores to form the low and high verbalisability odour sets used in the n-back task (see Appendix B). The verbalisability judgment in Chapter 2 followed closely that of Jönsson et al. (2011) such that stimuli were scored from 0-3 according to the quality of the verbal labels provided, with a lower score indicating vague or absent verbalisability and a higher score reflecting use of a specific noun. Verbalisability for the two odour sets differed significantly, t(12) = 26.38, p < .0005, d = 15.23, BF10 >

1,000 (Mhigh = 2.66, SDhigh = 0.11; Mlow = 1.12, SDlow = 0.09). An additional two odours were selected from the high and low verbalisability odorant samples. These were chosen to act as non-analysed buffer items (i.e. these are used at the start of the task and are not included in the analysis).

As discussed in Chapter 2, a normative verbalisability score may not be the most suitable dimension on which to base odour selection, due to high variability across participants. However, familiarity scores deemed more suitable for odour selection

covaried with verbalisable ratings, such that the odour sets were also significantly different across familiarity scores, t(12) = 21.62, p < .001, d = 11.57, BF10 > 100.

Eight line drawings, printed on individual A5 sheets of paper, were taken from Snodgrass and Vanderwart (1980) and used as a 2-back practice task at the start of the experiment.

3.1.2.3 Design

A continuous yes/no recognition task was employed on two testing sequences of 52 odour trials, where each trial necessitated a judgment as to whether the present odour was the same or different to the odour presented two items previously (i.e. the 2-back task). The experiment employed a within-participants multifactorial (2x2) design. The first within-participants factor concerned whether the block of odour trials contained odours categorised as high or low on verbalisability. This was operationalised as a block of 26-trials employing high verbalisable odours and a block of 26-trials employing low verbalisable odours. There was no interval between blocks (i.e. it was presented as a continuous 52-trial sequence). The second within-participants factor concerned testing sequence. Participants undertook two 52-trial testing sequences, with each testing sequence containing a block of high and low verbalisable odour trials. These odours were the same items used in the first sequence. The presentation order of trials was predetermined before testing, and the order of blocks was counterbalanced via a Latin square design.

Within each (high or low verbalisability) block, the six (high or low verbalisable) odorants appeared as a ‘target’ once (25% of trials), and three times as a ‘lure’ (75% of trials). Targets were odorants that had been presented two trials previously, and thus required a ‘yes’ response. Lures were odorants not matching the odour presented two

trials previously, and therefore required a ‘no’ response. Thus, a block comprised 24 critical trials (and 2 buffer trials) with each odorant presented four times. The first two trials in a sequence would always be lures, so preceding each block were two additional buffer trials. For the high verbalisability block, ‘Pear’ was presented for these two trials, and for the low verbalisable block ‘Nag Champa’ was presented. These buffer odours were not repeated elsewhere in the sequence, and responses for the buffer trials were not entered into the analysis. Recent lures at positions n+1 and n-1 were allowed to occur, and randomly appeared in sequences. Differences in the target to recent lure ratio across participant trials were equated across verbalisability conditions using the counterbalancing methods described below.

When determining the order of trials within blocks, the nature of the 2-back task required that six lures were tethered two positions before the six matching targets. To be clear, for that odour to be a target, it must first be employed as a lure two trials previous.

The remaining 12 lures in each block were placed pseudo-randomly, with the caveat that their position did not result in itself or a previously positioned lure becoming an unintended target, nor result in a target becoming a lure. The predetermined trial orders were counterbalanced, such that a sequence of lures and targets was re-used for another participant with the alternative set of 6 odours.

The number of correct target identifications (Hits), and incorrect identifications of a lure as a target (False Alarms, FA) were recorded and used to compute the proportion of Hits to FA via A’. The mismatched number of recent lures across participants made analysis of only these lure types for incidences of false alarms unsuitable. Instead, false alarms were calculated from all lure probes, at the cost of having slightly inflated correct rejection proportions and A’ scores. This measure of signal detection theory was selected due to the unequal trial numbers for lure and targets, and because it allows FA

rates to exceed Hits. A’ was calculated as 0.5 + ((Hits – FA) x (1 + Hits – FA)) / ((4 x Hits) x (1 – FA)) when Hits exceeded FA, and as 0.5 – ((FA – Hits) x (1 + FA – Hits)) / ((4 x FA) x (1 – Hits)) when FA exceeded Hits (Stanislaw & Todorov, 1999). Unlike d’

where Hit rates of one or FA rates of zero result in an indefinite value, use of A’ allows these results to remain unadjusted.

3.1.2.4 Procedure

The experiment was conducted in a quiet, well-ventilated room with a fan to circulate fresh air. Participants sat opposite the experimenter, separated by a wooden screen with a central fixation cross to prevent visual inspection of the odorants. Prior to the olfactory task, participants performed an 8-item visual version of the 2-back task in order to familiarise themselves with the procedure.

Figure 2. Schematic diagram of the 2-back task. Two buffer items precede the 24 test trials.

The 2-back task (Figure 2) presents participants with a sequence of stimuli, where each