Experiment 6

3.5.3.1. Method Participants

Twenty University of Lyon students (14 females, 6 males; mean age = 24.45 ± 7.3 years) took part in this experiment. Before starting the experiment, participants completed a consent form.

They were individually tested and paid sixteen Euros for their participation. Exclusion criteria included olfactory and/or neurological diseases. All participants reported a normal sense of smell. The study was conducted in accordance with the Declaration of Helsinki and experimental procedures were approved by the local ethical committee. Due to recording issues (for two participants) and to failure to respect sniff instructions (for one participant), three participants had to be excluded from the sniff analyses.

Stimuli

Forty-eight monomolecular odorants, routinely used in Lyon Neuroscience Research Center, were selected. Odorant concentrations were equalized by dilution in mineral oil. Odorants were presented in 15 ml vials (opening diameter: 1.7 cm; height: 5.8 cm ; filled with 5 ml solution) and absorbed on scentless polypropylene fabric (3x7 cm; 3 M, Valley, NE, USA) to optimize evaporation and air/oil partitioning. A number coded each odorant.

These forty-eight odorants were split into three lists, created based on a previous study’s ratings (Kermen et al., 2011). According to these ratings, the three lists did not significantly differ in terms of pleasantness, familiarity, intensity, edibility, molecular weight or complexity [Fs(2,45) = 0.36, 0.05, 0.02, 0.68, 0.54, 0.04 ps = .703, .955, .980, .512, .587, .960 respectively]. The three lists did not differ significantly in terms of pleasantness also according to our participants’ own ratings [F(2,45) = 0.03, p = .975].

Odor List Mean

pleasantness during Rating 1

Mean pleasantness during Rating 2

Mean pleasantness during Rating 3 (+)-fenchone 1 5.21 (± 1.29) 5.03 (± 1.14) 5.33 (± 1.33) Allyl Caproate 1 5.25 (± 1.45) 5.52 (± 1.56) 5.37 (± 1.53)

alpha ionone 1 5.45 (± 2.29) 5.70 (± 1.98) 5.28 (± 2.17)

butanol-1 1 3.07 (± 1.67) 3.33 (± 1.38) 3.33 (± 1.60)

Carvone-l 1 5.25 (± 1.84) 5.08 (± 1.58) 5.35 (± 1.73)

citral 1 6.96 (± 1.33) 7.05 (± 3.35) 6.65 (± 1.47)

Citronellal 1 5.65 (± 1.55) 6.25 (± 1.39) 5.75 (± 1.76) Ethyl

phenylacetate 1 2.95 (± 1.76) 3.05 (± 1.86) 3.03 (± 2.00)

Guaiacol 1 3.19 (± 1.66) 3.10 (± 2.05) 3.15 (± 1.95)

Heptanol1 1 4.80 (± 1.60) 5.28 (± 1.16) 4.90 (± 1.52)

Hexanoic acid 1 2.31 (± 1.25) 1.98 (± 1.32) 2.25 (± 1.30)

Hexanol3 1 4.60 (± 1.18) 4.47 (± 1.12) 4.46 (± 1.17)

Isoamyl acetate 1 6.37 (± 1.68) 6.21 (± 1.54) 6.54 (± 1.52) isobutyric acid 1 2.23 (± 1.28) 1.93 (± 1.03) 2.09 (± 1.24)

R-(+)-limonene 1 5.53 (± 1.56) 5.28 (±

1.48)

5.17 (± 1.46)

S-(-)-limonene 1 4.89 (± 0.74) 4.70 (± 1.34) 4.62 (± 1.17)

1-Decanol 2 3.62 (± 2.11) 3.28 (± 1.98) 3.55 (± 2.03)

alpha Pinène 2 4.92 (± 1.70) 5.46 (± 1.29) 5.39 (± 1.39) Amyl

PhenylAcetate 2 1.82 (± 0.82) 2.18 (± 1.25) 2.00 (± 1.01) Benzaldehyde 2 6.10 (± 2.01) 6.59 (± 1.96) 6.55 (± 2.30) Benzyl Acetate 2 6.75 (± 1.67) 6.99 (± 1.26) 6.42 (± 1.93)

Myrcène 2 3.64 (± 1.33) 4.41 (± 1.68) 4.24 (± 1.24)

d-Carvone 2 5.58 (± 1.91) 5.68 (± 1.92) 5.88 (± 1.65)

Dodecanal 2 2.75 (± 1.53) 2.96 (± 1.74) 2.63 (± 1.59)

Ethyl butyrate 2 4.60 (± 1.91) 4.99 (± 2.08) 5.66 (± 2.20)

Geraniol 2 6.07 (± 1.51) 6.02 (± 1.65) 6.05 (± 1.73)

MethylAnthranilate 2 5.80 (± 2.09) 5.89 (± 1.70) 6.09 (± 1.53) propionic acid 2 3.23 (± 1.40) 3.15 (± 1.30) 2.95 (± 1.11) Terpinen-4-ol 2 4.65 (± 1.78) 4.64 (± 2.15) 4.63 (± 2.06)

trans-2-hexenylacetate 2 4.98 (± 1.15) 5.10 (± 1.01) 5.14 (± 1.12) trans-anethole 2 5.99 (± 1.99) 6.15 (± 1.78) 6.40 (± 1.77) Valeric Acid (ISO) 2 1.48 (± 0.66) 1.59 (± 0.66) 1.80 (± 1.02) (-)-fenchone 3 5.05 (± 1.60) 4.95 (± 1.10) 5.03 (± 1.27)

1-8 Cineol 3 4.30 (± 1.98) 3.65 (± 2.07) 3.80 (± 2.02) Acetophenone 3 5.50 (± 1.76) 5.20 (± 1.86) 5.75 (± 1.93) Butane dione 2-3 3 3.93 (± 2.22) 4.43 (± 2.22) 4.03 (± 2.50)

cis-3-hexenylacetate

3 4.45 (± 2.17) 4.38 (± 1.75) 4.15 (± 1.99) Citronellol 3 5.84 (± 2.08) 5.95 (± 1.82) 6.10 (± 1.62) Ethyl salicylate 3 4.34 (± 1.72) 4.83 (± 1.52) 4.78 (± 1.63)

Eugenol 3 3.63 (± 1.65) 4.25 (± 2.23) 4.18 (± 1.83)

Heptanal 3 3.60 (± 1.90) 2.99 (± 1.63) 3.21 (± 1.97)

Ionone-beta 3 4.68 (± 2.49) 4.30 (± 2.31) 4.38 (± 2.19)

Linalool 3 6.48 (± 1.88) 5.73 (± 1.61) 6.38 (± 0.89)

p-cresol 3 2.90 (± 1.92) 3.10 (± 2.01) 3.83 (± 2.03)

pentanol 3 2.45 (± 1.56) 2.90 (± 1.55) 2.38 (± 1.11)

Phenyl ethanol 3 5.33 (± 2.16) 5.48 (± 1.99) 4.95 (± 2.06)

propanol 3 4.90 (± 1.64) 4.98 (± 1.62) 5.20 (± 1.16)

Alpha terpinène 3 5.34 (± 1.77) 5.36 (± 1.87) 5.15 (± 1.39) Table 5. Name of the 48 odors used in Experiment 6 and their mean ratings. Standard deviations are presented in parentheses.

Sniffing measurement

Sniffing data was measured following the exact same methodology described in the previous experiment.

Procedure

Testing was performed in an experimental room designed specifically for olfactory experiments, and proceeded in three blocks, corresponding to the three sequences of measurements, in a within-subjects design, with three different lists of odors. Within each of these sequences, four steps were present: Rating 1, Rating 2, Rating 3 and Choice. The order of the four steps varyied by sequence (RCRR, RRCR, RRRC). Each of the 48 odors was consequently presented four times to each participant, which means that in total, participants were exposed to 192 odor perception trials across the entire experimental session. The experimental duration was about two hours. In order to prevent habituation effects, during the rating phases, the interstimulus interval was fixed at 20 seconds. During the choice phases, 8

seconds separated the odors presented in a pair. Moreover, a break of five minutes was systematically inserted between the three blocks (so every 64 presentations of odors).

The following three parameters were counterbalanced across participants: (i) the order of the sequence of measurements, (ii) the list of odors associated with a particular sequence of measurement, and (iii) the presentation order of the odors.

In every rating phase, we assessed individuals’ pleasantness ratings of the smells.

After each odorant presentation, participants were asked to verbally indicate a continuous value from 0 to 10 to rate the smell’s pleasantness, from “very unpleasant” to “very pleasant”.

Participants were informed that they could use decimal numbers.

In every choice phase, eight pairs of smells were created on the basis of the initial pleasantness ratings of each individual participant. These pairs were divided into two groups of four: 1) The first group consisted of four pairs of odors that the participant had rated as similarly pleasant (i.e., difficult choices; mean rating differences = 0.17 ± 0.20 on the 11-point subjective scale); 2) The second group contained four pairs of odors previously rated differently for pleasantness (i.e., easy choices; mean rating differences = 3.44 ± 0.69 on the 11-point subjective scale). Within each pair, participants were required to choose which odor they preferred.

For each odor presentation, participants were instructed to smell the odor as naturally as possible during one inspiration.

Data analysis

Data were analysed in the same way as in Experiment 5.

3.5.3.2. Results Chosen stimuli

We will begin with the analyses conducted on pleasantness ratings.

First, we checked whether an increase in self-reported preference was observed between Ratings 1 and 2 in the classical sequence of measurement (RCRR), but also in the two control sequences (RRCR and RRRC). To do so, we performed repeated measures ANOVAs on the pleasantness ratings with the Rating (Rating 1, Rating 2) as within-subject factors for each Sequence of Measurement (RCRR, RRCR and RRRC). Pleasantess ratings were significantly increased in the second rating session in comparison to the first one, in all

sequences of measurement [Fs(1,19) = 12.61, 5.93, 9.46, ps = .002, .025, .006, η²s= .40, .24, .33, respectively RCRR, RRCR and RRRC] (see Figure 8).

Second, in line with Chen and Risen’s (2010) criticisms, we checked whether the preference modulation observed between the two rating sessions varied as a function of the sequences of measurement. To do so, we conducted a repeated measures ANOVA on the pleasantness signed difference scores between Rating 1 and Rating 2 with the sequence of Measurement (RCRR, RRCR and RRRC) as a within-subject factor, which revealed no significant difference [F(2,38) = 1.94, p = .158]. A planned comparison contrasting the sequence RCRR to the sequences RRCR and RRRC revealed that preference modulation in the RCRR tend to be higher than in the RRCR and RRRC sequences [F(1,19) = 3.56, p = .075]. Although this effect was only marginally significant, this tends to suggest that beyond Chen and Risen’s (2010) criticisms, the act of choosing a smell could impact its subsequent pleasantness ratings.

Pursuing the same goal, we also tested whether the difference between Ratings 1 and 3 was different according to the sequence of measurement. To do so, we performed a repeated measures ANOVA on the pleasantness signed difference scores between Rating 1 and Rating 3 with the sequence of measurement (RCRR, RRCR and RRRC) as a within-subject factor.

This did not reveal any significant effect [F(1,19) = 1.46, p = .245]. A planned comparison contrasting the sequences RCRR and RRCR (where a choice has been made at the stage of Rating 3) to the sequence RRRC (where it was not the case) did not reveal any significant difference [F(1,19)= 1.19 p = .289]. Using this second analysis strategy to control for Chen and Risen’s (2010) criticisms, and contrary to the results found with the previous strategy (described in the previous paragraph), results do not reveal that choice does impact subsequent pleasantness ratings for chosen smells.

Third, we explored whether this statistical trend could be explained by mere exposure effects, i.e. whether the preference modulation between the first and the second rating for chosen stimuli in RCRR could tend to be higher than in RRCR or RRRC. This is what one would expect if a mere exposure was at play, because during Rating 2, stimuli have been presented one time more (during the choice phase) in RCRR. To do so, we conducted a repeated measures ANOVA on the pleasantness signed difference scores between Rating 1 and Rating 2 in RCRR and on the pleasantness signed difference scores between Rating 1 and Rating 3 in RRRC. This analysis did not reveal a significant effect [F(1,19) = 0.98 , p = .336].

We could not consequently reject the possibility that the statistical trend for preference modulation between the first and the second rating for chosen stimuli in RCRR tending to be higher in comparison to RRCR and RRRC is driven by mere exposure effects.

Let us now move on to the analyses conducted on sniff parameters.

We conducted repeated measures ANOVAs on sniff parameters with the Rating (Rating 1, Rating 2) as within-subject factors for each Sequence of Measurement (RCRR, RRCR and RRRC). We could not find any systematic effect between Ratings 1 and 2 in any of the three sequences of measurement for sniff duration, total volume and upward phase of the inspiratory volume. More precisely, for sniff duration [Fs (1,16) = 0.02, 1.89, 8.36, ps = .893, .188, .011 for RCRR, RRCR and RRRC respectively], the only significant effect in the RRRC condition revealed that durations were longer during Rating 1 than during Rating 2. For sniff total volume [Fs (1,16) = 0.59, 8.02, 4.38, ps = .453, .012, .053 for RCRR, RRCR and RRRC respectively], the volume was larger during Rating 2 in the RRCR condition and tended to be smaller during Rating 2 in the RRRC condition than during Rating 1. The upward phase of the inspiratory volume, [Fs (1,16) = 0.24, 0.53, 3.48, ps = .629, .477, .080 for RCRR, RRCR and RRRC respectively] tended to be larger during Rating 1 in the RRRC condition.

Taken together, we do not find evidence that the sniff-related measures are significantly affected by the Rating (Rating 1, Rating 2) in the case of chosen smells.

Figure 8. Signed pleasantness difference scores for chosen smells between Rating 1 and Rating 2 and between Rating 1 and Rating 3 from Experiment 6 according to the sequence of measurements, Rating1-Choice-Rating2-Rating3 (RCRR), Rating1-Rating2-Choice-Rating3

(RRCR) and Rating1-Rating2-Rating3-Choice (RRRC). Error bars represent the standard error of the mean.

Rejected stimuli

Pleasantness ratings were not significantly decreased between Ratings 1 and 2 for any of the three sequences of measurement [Fs(1,19) = 0.22, 1.36, 1.18, ps = .644, .258, .291, respectively RCRR, RRCR and RRRC].

Regarding sniff parameters, results again lack systematic coherence. We conducted repeated measures ANOVAs on sniff parameters with the Rating (Rating 1, Rating 2) as a within-subject factor for each Sequence of Measurement (RCRR, RRCR and RRRC). More precisely, for sniff duration [Fs (1,16) = 0.06, 7.34, 2.83, ps = .807, .015, .112 for RCRR, RRCR and RRRC respectively], the only significant effect in the RRCR condition revealed that duration was longer during Rating 1 than during Rating 2. For sniff total volume [Fs

(1,16) = 0.51, 8.14, 7.79, ps = .484, .012, .013 for RCRR, RRCR and RRRC respectively], the volume was decreased during Rating 2 in comparison to Rating 1 in the RRCR and RRRC conditions. For the upward phase of the inspiratory volume, there was no significant effect [Fs

(1,16) = 0.79, 2.78, 1.62, ps = .387, .115, .222 for RCRR, RRCR and RRRC respectively].

Taken together, we do not find evidence that the sniff-related measures are significantly affected by the Rating (Rating 1, Rating 2) in the case of rejected smells.

Influence of pleasantness on choices

We checked whether the pleasantness of the odor before the choice varied as a function of the participant’s choice. The repeated measures ANOVA with the factor choice (chosen, rejected) performed on pleasantness scores during Rating 1 did not reveal any significant effect [F(1,19) = 0.09, p = .763].

3.5.3.3. Discussion

Experiment 6 aimed at extending Experiment 5, which lead to promising but inconclusive results regarding the possibility of sniff as an implicit physiological correlate of preference modulation in the free-choice paradigm. Experiment 6 enabled us to control for Chen and Risen’s (2010) criticisms regarding the methodological validity of the free-choice paradigm by adding a RRCR sequence of measurement to the classical RCR sequence of the

free-choice paradigm, where a third rating phase was added. Experiment 6 also permitted us to control for potential mere exposure effects (Zajonc, 1968; see section 3.4), via a third sequence of measurement, RRRC.

We found that pleasantness ratings between Rating 1 and Rating 2 were increased for chosen odors in our three sequences of measurement, RCRR, RRCR and RRRC. This suggests, as in section 3.4, that it is particularly important to control Chen and Risen (2010)’s criticism that pleasantness ratings modulation can be observed in the free-choice paradigm without any genuine preference change, when investigating preference modulation induced by choice. However, this increase tend to be higher in the RCRR sequence (where a choice phase has taken place in between) in comparison to RRCR and RRRC (where no choice phase had occured before Rating 3). This tends to suggest that choices do impact preferences for chosen smells. To the contrary, in applying the same line of though to the difference between Rating 1 and Rating 3, results revealed that it was not significantly higher in the RCRR and RRCR conditions in comparison to the RRRC condition. The pleasantness difference scores between Rating 1 and Rating 2 in RCRR was not significantly different from the pleasantness signed difference scores between Rating 1 and Rating 3 in RRRC. We cannot consequently exclude that the increase in pleasantness ratings following choice for chosen smells in RCRR condition was not driven by mere exposure effects, rather than a preference change induced by choice. Taken together, and despite the results in sections 3.4 and 3.6, these results are inconclusive regarding whether choice has a true impact on preferences. Of course, there is the possibility that this is due to a lack of statistical power, as the number of participants in this experiment (20 participants) was much less than in experiments 3 (100 participants) or 4 (60 participants).

This modulation in pleasantness ratings for chosen smells did not correspond to systematic statistical modulations in sniff parameters. Note that our sample size for sniff analyses was small (17 participants), and could consequently have impaired statistical reliability in answering our question of interest.

We did not find any significant statistical modulation for rejected odors in either pleasantness ratings or sniff parameters. Regarding pleasantness ratings, as we already mentioned in Experiment 5’s discussion, the scope for post-choice preference modulation depends on the pleasantness spectrum of the stimuli used. A choice between a-priori pleasant stimuli is assumed to lead to a large decrease in hedonic evaluation of the rejected stimulus,

while a choice between a-priori unpleasant stimuli is assumed to lead to a large increase in hedonic evaluation of the chosen stimuli (Shultz et al., 1999). This might explain why the evaluation of pleasantness ratings was not significantly decreased for rejected smells in this experiment. Regarding the absence of modulation in sniff parameters for rejected odors, both this effect and a low statistical power could be involved.

3.5.4. General discussion

The aim of the two experiments reported in the section was to investigate sniff as a potential psychophysiological correlate of preference modulation induced by choice. The modulation of sniff is largely involuntary and hence unlikely to be due to experimenter demand effect, which makes it particularly interesting to investigate implicit processes engaged in post-choice preference modulation. It makes all the more sense that decision-making processes and motor system responses could be related as early stages (see e.g., Selen, Shadlen, & Wolpert, 2012).

After making a choice between two similarly rated smells, participants rated the chosen one as more pleasant (Experiments 5 and 6), and the rejected one as less pleasant (Experiment 5). In Experiment 5, this preference modulation goes along with a statistical trend for a higher-magnitude decrease in sniff duration for previously rejected smells in comparison to chosen ones. However, in this experiment, results were inconclusive regarding a true modulation of preference by choice, as no control of the true impact of choice on preferences (see Chen & Risen, 2010 for further details) or of mere exposure effects (see section 3.4) was run in this experiment.

Experiment 6 aimed at replicating those effects while controlling for this potential flaw of the free-choice paradigm and for mere exposure effects. We obtained mixed evidence regarding the true impact of choice on preferences. Thus, the increase in preference between Rating 1 and Rating 2 in the RCRR sequence (where a choice phase has taken place in between) tended to be higher than what was observed in RRCR and RRRC (where no choice had occured in between), which is consistent with (though not sufficient for showing) choice impacting preferences. However, the difference between Ratings 1 and 3 was not significantly higher in the RCRR and RRCR conditions (where a choice phase had taken place in between) in comparison to the RRRC condition (where no choice had occured in between).

Furthermore, we could not exclude that mere exposure effects were at play. Finally, we did not find any systematic modulation linked to choice on sniff parameters.

One can speculate about the reason(s) why we failed to observe the hypothesized impact of preference modulation following choice on sniff characteristics. One potential reason is that the number of participants in those two studies was quite restricted (respectively 22 and 20, with 17 participants viable for sniff analyses), which goes along with low statistical power. Both the decreased duration of sniff for rejected odors in Experiment 5 and the preference increase in the RCRR sequence (where a choice phase has taken place in between) in comparison to RRCR and RRRC (Experiment 6) were merely tendencial effects.

This suggests that the hypotheses might not be entirely off the mark. Testing them empirically might benefit from larger samples of participants.

Note that we could also investigate the links between sniff, preference modulation and decision-making processes in taking the question the other way around. In vision, it has been shown that gaze duration can shape which face out of two is more attractive (Shimojo, Simion, Shimojo, & Scheier, 2003). One might therefore want to ask whether given patterns of sniff create given preferences. For instance, can training participants to sniff according to a pattern typical for a pleasant odor make it more pleasant? And the other way around, can imposing a sniff pattern typical for unpleasant odors make a given smell less pleasant?

Can the duration of the sniff be sufficient to induce preference modulation? Such an experiment would however require making sure that training allows the conscious modulation of sniffs, which untrained participants do not appear to be able to do, when spontaneously asked to do so (see Bensafi, Sobel, & Khan, 2007).

3.6. Experiments 7-9 - When flexibility is stable: Implicit long-term shaping