TRANSITIVITY OF ODOR PREFERENCES: CONSTANT AND PARTICULARITIES IN HEDONIC PERCEPTION

Abstract

Transitivity of preferences has been investigated for a long time in decision-making. In the field of perception, the pleasantness of odors raises several questions related to individual versus cultural or universal preferences and the existence of a classification in a delimited hedonic space. The aim of this study was to test transitivity in olfactory hedonicity using a first panel of 10 mixed odors and a second panel of 10 odors from a delimited floral category. Data were collected by paired comparisons in a two-alternative forced choice. Results in both panels showed a strong transitivity for each participant leading to a linear range of 10 odors classified by preference. However, ranges varied from one participant to another and the mean preferences of the group did not allow one to infer individual's hedonic classification of odors. Moreover, the individual classification appeared stable over time and undisturbed by odorant distractors. These findings suggest that humans have considerable ability to classify odors hedonically as a model of individual preferences in a sensory space usually considered to be more involved in affective/emotional states than in cognitive performances.

Stochastic transitivity is a property of preferences among pairs of objects formed from three alternatives, a, b and c. Stochastic transitivity means that if a is preferred over b, and b is preferred over c, then a will be preferred over c (Tversky & Russo, 1969). Most studies on transitive preferences present each paired comparison once, in a two-alternative forced choice from each of N respondents, and count the number of a, b, c triples for which the respondent chose a over b, b over c and a over c (Regenwetter, Dana, & Davis-Stober, 2010; 2011). This number is used to indicate the degree of transitivity demonstrated by that respondent. In the case of multiple respondents, group mean transitivity values are typically reported.

Transitivity of preferences has been investigated in several experiments both in economics, especially in the field of consumer decision-making (Budescu & Weiss, 1987; Chen & Corter, 2006; Lee, Amir & Ariely, 2009; Mellers, Chang, Birnbaum, & Ordonez, 1992; Ranyard, 1977; Sopher & Narramore, 2000) and in psychology (Bradbury & Nelson, 1974; Zemach & Teller, 2007) including perception. Surprisingly, there is not an extensive published literature on stochastic transitivity with odor preferences. A very pleasant odor will obviously be preferred to a pleasant one and a pleasant odor will be preferred to an unpleasant one, which would imply that a very pleasant odor will be preferred to an unpleasant one. However, a number of questions related to pleasantness in olfaction remain unanswered.These questions are mainly related to stimuli and participants' characteristics listed below.

Olfactory hedonicity is influenced by several parameters such as stimulus intensity (Henion, 1971; Moskowitz, Dravnieks, & Gerbers, 1974; Moskowitz, Dravnieks, & Klarman, 1976), trigeminal activation (Brand, 2006; Jacquot, Monnin, & Brand, 2004), familiarity with the odorant (Cain & Johnson, 1978), physiological states such as hunger (Albrecht et al., 2009; Cabanac, 1979), pregnancy (Cameron, 2007), experimental task (Bensafi et al., 2002) and emotional state (Chen & Dalton, 2005). In contrast, few studies have been devoted to invariants in hedonic perception. An extended line of research has explored how physicochemical properties of molecules, for example, molecular weight and length of chains (Khan et al., 2007; Schiffman, 1974a,b) could be related to hedonic perception. However, this research line has not produced systematic findings. For instance, some authors have hypothesized that pleasantness could be related directly to the complexity of molecules but this hypothesis has never been confirmed. These perspectives imply a universal hedonic perception that is at odds with research demonstrating that odorant preferences vary by culture (Chrea, Valentin, Sulmont-Rossé, Nguyen, & Abdi, 2005), age (Distel et al., 1999) and gender (Brand & Millot, 2001). Concerning food odors, for instance, some people may like a particular odorant while others show disgust towards the same odorant. Therefore, odors could also be ranged by preferences in a strictly individual manner and the aim of this study was to test the order of preferences in olfactory hedonicity, participant by participant and comparatively to the group and comparatively to subjective numeric rating of olfactory hedonicity as a commonly used method.

Studies concerning preferences such as those on consumer choices frequently consider the preferences of the group, while the preference profiles of individual members of the group can differ between participants, and between one participant and the group. To our knowledge, the stochastic transitivity approach has not been used to explore classifications of odor preferences. Logically, preferences could depend on perceptive characteristics of odors included in a specific panel (i.e., the array of odorants used), especially in relation to classical categories defined in olfaction, for example, floral, ethereal, burned (Amoore, 1965). Using the transitivity model, the present experiment used a first panel of 10 mixed odorant categories and a second panel of 10 stimuli from a delimited floral odorant category. Because olfaction is usually considered as a sensory system poorly related to cognitive processes (Rouby, Schaal, Dubois, Gervais, & Holley, 2002) due in great part to neuroanatomical specificities (Hawkes & Doty, 2009), we expected that there would be a low level of transitivity in odor preferences at the individual and group level. If transitivity is demonstrated, comparisons between panels will allow us to determine how the level of transitivity depends on hedonic proximity, anticipated to be greater in the second panel (floral odorants) than in the first (mixed odorants). To that end, transitivity was tested in the same group of participants with a third panel composed of five odorants from the two previous panels. This part of the experiment tested the role of interference—known to be weak in olfaction (Issanchou, Valentin, Sulmont, Degel, & Köster, 2002) —in the hedonic classification of odors.

METHOD

Participants

Twenty volunteers participated in this experiment (10 males and 10 females). Their age ranged from 19 to 27 years (mean 24 years 4 months). Participants reported normal smell sensitivity and none of them had a history of nasal/sinus disease. The study was conducted in accordance with the Declaration of Helsinki – Hong Kong.

Odorants

Before the present experiment, a pretest was conducted to identify odorants with similar ratings of intensity (i.e., those with nonsignificant differences in intensity ratings). Twenty odorants identified in the pretest were used in the present study. Stimuli were presented in jars (with small holes in their covers) containing polycrylamide (nontoxic and with no detectable smell) crystals together with the odorant (Sentosphere, France). A first panel (panel 1) of 10 odorants was constituted from mixed categories as follows: A1 = Apricot; B1 = Mushroom; C1 = Caramel; D1 = Coffee; E1 = Orange; F1 = Pepper; G1 = Grapefruit; H1 = Walnut; I1 = Milk; J1 = Vanilla. A second panel (panel 2) of 10 odorants consisted of a floral group: A2 = Honeysuckle; B2 = Mint; C2 = Rose; D2 = Lilac; E2 = Lily of the valley; F2 = Grass; G2 = Lavender; H2 = Eucalyptus; I2 = Violet; J2 = Pine tree. A third panel (panel 3) of 10 odorants was composed of 5 odorants from each of the previous panels. The third panel was constructed based on each participant's ranking of odorants within panels 1 and 2 (see below for details). The odorants from these panels that were ranked 1, 3, 5, 7, and 9 were combined into a third panel of 10 odorants.

Procedure

Each panel of 10 odorants was tested in separate sessions (3 to 4 weeks between sessions). Participants were first asked to rate the perceived pleasantness of each odor on an anchored scale of 0 (very unpleasant) to 10 (very pleasant). Next, preferences were tested by presenting all possible combinations of odor pairs:

with each odorant presented nine times. The order in which odors were presented within a pair was randomized. Overall, each odor was presented equally often first and second. A rest period of 30 sec was allowed between pairs. The 45 pairs were presented in random order but a given odor could not be presented in two successive pairs. After sampling both odors in a pair, participants were asked to choose their favorite (indifference and equality were not accepted). Sessions lasted about 35 min.

Analyses

To determine if preferences made in pairs demonstrated transitivity, for analysis purposes we combined odorants into every possible triple:

For each participant, we counted the number of these120 triples in which every paired choice was internally consistent, thereby demonstrating transitivity (for example if A>B, B>C, and A>C then A>B>C and the triple ABC was transitive). The number of transitive triples was converted to a percentage for each participant and for the group. Results were analyzed comparatively to transivity relative to chance. With the triples method, the transitivity score by chance is always 75% due to six possibilities (A>B, B>A, B>C, C>B, A>C, C>A) arranged in eight possible triples of which six are transitive. When the transitivity score is 100%, it is described as strict transitivity.

Comparative analyses were performed using t-tests (paired and independent) and Spearman correlation coefficients. P-values below 0.05 were considered significant. Nonsignificant analyses are noted as ‘ns'.

RESULTS

Panel 1

The total number of times an odorant was preferred in a pairing is shown for the group in Table 1 (20 participants x 9 presentations = a maximum of 180 selections). The mean transitivity score for the group was 97.5% (corresponding to three intransitive triples). Individual data (Table 2) showed that some participants had strong or strict transitivity, for example, 6 out of 20 participants were strictly transitive and all other participants scored at above chance levels. Thus, it was possible to classify all odors in a strict linear order (> = preferred to). For example:

Table 1.

Total number of preferences (odorants presented by couples) in both Panels 1 and 2 for a group of 20 participants.

Odors – Panel 1. Mixed categories In preference order	Number of choices For the group (N=20) (max: 9x20=180)	Odors – Panel 2. Floral category In preference order	Number of choices For the group (N=20) (max: 9x20=180)
A1= Apricot	143	A2= Honeysuckle	122
G1= Grapefruit	140	E2= Lily of the valley	108
C1= Caramel	121	D2= Lilac	104
E1= Orange	115	F2= Grass	101
J1= Vanilla	94	I2= Violet	97
B1= Mushroom	70	B2= Mint	87
F1= Pepper	69	C2= Rose	83
I1= Milk	55	G2= Lavender	77
H1= Walnut	49	H2= Eucalyptus	62
D1= Coffee	44	J2= Pinetree	59

Table 2.

Individual transitivity scores for both Panels 1 and 2 expressed in number of transitive triples and percentage (number of transitive triples/120 × 100).

Participants	Sex	Panel 1. Mixed odorant categories	Panel 2. Delimited odorant category
1	M	120 (100%)	113 (94.17%)
2	F	120 (100%)	116 (96.67%)
3	M	112 (93.33%)	116 (96.67%)
4	M	118 (98.33%)	115 (95.83%)
5	M	119 (99.17%)	117 (97.5%)
6	F	119 (99.17%)	117 (97.5%)
7	F	115 (95.83%)	120 (100%)
8	M	120 (100%)	115 (95.83%)
9	M	110 (91.67%)	120 (100%)
10	F	114 (95%)	120 (100%)
11	F	120 (100%)	120 (100%)
12	F	120 (100%)	120 (100%)
13	M	117 (97.5%)	118 (98.33%)
14	F	119 (99.17%)	112 (93.33%)
15	M	119 (99.17%)	120 (100%)
16	F	115 (95.83%)	119 (99.17%)
17	F	112 (93.33%)	118 (98.33%)
18	M	113 (94.17%)	114 (95%)
19	F	118 (98.33%)	115 (95.83%)
20	M	120 (100%)	120 (100%)

• Participant 1: G1>E1>A1>C1>J1>B1>I1 >F1>H1>D1

• Participant 2: A1>J1>I1>C1>E1>F1>G1 >B1>D1>H1

• Participant 3: I1>G1>A1>J1>E1>C1>D1 >B1>F1>H1

Classification appeared to differ strongly from one participant to another and group classification, that is, average rankings did not allow prediction of individual preferences.

Panel 2

Total preference for each odor in Panel 2 is also reported in Table 1. The mean transitivity score for the group was 97.7% (corresponding to 3 intransitive triples). As with Panel 1, individual data (Table 2) showed that some participants had strong or strict transitivity, e.g. 7 out of 20 participants were strictly transitive. Thus, it was possible to classify all odors in a strict linear order (> = preferred to). For example:

• Participant 1 : D2>E2>G2>A2>C2>F2>B2 >J2>I2>H2

• Participant 2 : I2>G2>B2>F2>C2>D2>A2 >E2>H2>J2

• Participant 3 : A2>D2>C2>J2>H2>G2>B2 >F2>E2>I2

Comparison between Panel 1 and Panel 2

Transitivity scores were very high regardless of whether mixed odorants (Panel 1) or delimited odorants (Panel 2) were arranged. Moreover, Panel 1 had a wider range (143 – 44) than Panel 2 (122 – 59) suggesting that the transivity scores were independent of hedonic proximity of odorants. A paired t-test indicated that the number of transitive triples did not differ between panels [t₍₁₉₎ = 0.245; ns]. The correlation between transitivity scores for Panel 1 and Panel 2 was not significant (rho = 0.183, ns) suggesting that a participant with a high transitivity score in Panel 1 did not necessarily have a high transitivity score in Panel 2.

Blended Panels

For each participant five odorants from both panels were chosen as follows:

• Participant 1: G1-A1-J1-I1-H1 and D2-G2-C2-B2-I2

• Participant 2: A1-I1-E1-G1-D1 and I2-B2-C2-A2-H2

• Participant 3: I1-A1-E1-D1-F1 and A2-C2-H2-B2-E2

Similarly to Panels 1 & 2, the 10 blended odorants were tested in 45 pair presentations. Results showed that the 5 odorants of a panel were strictly ordered as they were the first time experienced by each participant, suggesting a total absence of interference. This result indicated that olfactory hedonic transitivity was stable over time.

• Participant 1: G1>D2>G2>A1>C2>J1>I1 >B2>H1>I2

• Participant 2: I2>B2>A1>I1>C2>E1>A2 >H2>G1>D1

• Participant 3: A2>I1>C2>A1>H2>B2>E1 >E2>D1>F1

Relation between Pleasantness Ratings and Odorant Transitivity

Correlations between pleasantness ratings (0 to 10 scale) and the rank of odorants by the transitivity method are reported in Table 3. Between participants, the correlation coefficient varied from moderate to strong. Thus, participants classified the odorants by pleasantness in the same manner as when they chose between odorant pairs in Panel 1 as well as in Panel 2. The order in which odorants were presented within a pair had no significant effect on preference.

Table 3.

Spearman correlation coefficients between self-rating of pleasantness (0 – 10) and range of odorants by transitivity.

Participants	Sex	Panel 1. Mixed odorant categories	Panel 2. Delimited odorant category
1	M	0.833	0.682
2	F	0.673	0.639
3	M	0.621	0.555
4	M	0.764	0.652
5	M	0.788	0.715
6	F	0.969	0.870
7	F	0.812	0.806
8	M	0.582	0.688
9	M	0.724	0.809
10	F	0.818	0.745
11	F	0.876	0.891
12	F	0.939	0.952
13	M	0.515	0.685
14	F	0.688	0.552
15	M	0.698	0.988
16	F	0.679	0.591
17	F	0.655	0.933
18	M	0.609	0.606
19	F	0.642	0.642
20	M	0.915	0.924

DISCUSSION

Transitivity of preferences in odor hedonicity was very strong or strict for each participant irrespective of the panel of odors, mixed categories or delimited category. Thus, for each participant, the 10 odors in a panel can be ordered linearly from most pleasant to least pleasant although several odors are almost alike from a hedonic appreciation standpoint. Moreover, the hedonic classification is specific for each participant, and the group classification does not allow one to infer any individual classification. Hedonic autoestimation is related to transitivity but the two-alternative forced choice method provides a more robust classification insofar as results of retests were quite similar to those of tests, implying a possible prediction of choices for each participant.

As for the blended panel, the initial rankings were strictly maintained which suggests that odorant distractors do not disturb hedonic classification. Moreover, these findings suggest a stability of olfactory hedonic transitivity insofar as the experiment was conducted several weeks after the initial classification.

These findings raise a number of questions. First, what are the mechanisms underlying hedonic olfactory transitivity? In a recently published study investigating postchoice preference changes (Coppin, Delplanque, Cayeux, Porcherot, & Sander, 2010), the transitivity of odor preferences was related to implicit memory mechanisms (Issanchou et al., 2002). Second, concerning individual preferences, no gender comparisons were conducted in this experiment. Insofar as gender differences are frequent in olfaction research (Brand & Millot, 2001), further study with larger samples should be conducted to determine if the greater olfactory abilities of women would also be shown in the hedonic classification of odors.

For methodological reasons, transitivity was considered in triples in this study which allowed better comprehension and reliability. However, transitivity can also be analyzed with quartets (A, B, C, D with 24 transitive quartets among 64 possible quartets), quintets (A, B, C, D, E with 120 transitive quintets among 1024 possible quintets) and so on up to decets. The probability of classifying by chance the items in a strict transitive order among the total number of possibilities increases logarithmically with the number of items. As previously noted, the probability of classifying by chance 10 odorants in a strict transitive order by preference is about 1/10⁷ (i.e 0.00001%). Therefore, the transitivity demonstrated by each participant in the present study was extremely strong (recall that, for each panel of odors, about a third of the participants showed strict transitivity) and different from results obtained by chance. However, other factors not considered in the present study could have an influence on transitivity of preferences. Among them, the familiarity of the odorants could be the most prominent, insofar as many exprimental studies have shown a positive correlation between familiarity and pleasantness (e.g., Distel et al., 1999) although this conclusion is currently disputed (Delplanque et al., 2008). Furthermore, the present study shows great stability of transitivity over a short re-test period. Future research could test this stability over longer durations to determine if a possible flexibility of perception exists, for instance, in relation to familiarity or experience towards odorants.

Because the ranking activity is an important cognitive process, the present findings also suggest a cognitive influence on the emotional space of odors classically considered as less sensitive to cognitive impact than other sensory modalities such as vision or hearing (Rouby et al., 2002, Hawkes & Doty, 2009). This notion contributes to the understanding of the interactions between emotion and cognition, usually investigated as the influence of emotion on cognitive processes (i.e., learning, attention, memory, judgement, etc.), and seldom the other way around. Interestingly, the present experiment was conducted with a homogeneous population of healthy young adults. Further research is needed to investigate cases in which transitivity of odor hedonicity can be disturbed and thus reflect cognitive or emotional impairment, for example in senescence and neuropathological diseases (e.g., Alzheimer's disease or Parkinson's disease). Similarly, developmental research could explore olfactory transitivity in infants according to age and perhaps analyze the relation between transitivity and familiarity of odors. Furthermore, our findings raise questions about recent research on transcultural approaches to olfactory preferences (Chrea et al., 2005; Chrea, Valentin, & Abdi, 2009), showing specific odor categories in relation to cultural membership. An individual classification of odor preferences may in fact be superimposed on cultural preferences as observed in food preferences (Rozin, Fischler, Shields & Masson, 2006). Finally, recent research has investigated the relation between hedonicity and the structure of odorant molecules. For example, it is well known that the presence of an atom of sulfur induces an unpleasant perception. By contrast, the possibility that pleasantness might be related to the complexity of molecules (Khan et al., 2007; Poncelet et al., 2010), including the prediction of odor pleasantness by an electronic nose (Haddad, Medhanie, Roth, Harel, & Sobel, 2010) could be relevant for a subpopulation but does not, in the light of the present findings, appear to be informative with respect to individual rankings.