A Review of the Effects of Valenced Odors on Face Perception and Evaluation

Z-Curve

We used the z-curve package in R to conduct z-curve analyses (Bartoš & Schimmack, 2021). The z-curve method is designed to estimate publication bias by using observed p-value or z-scores in published studies. The z-curve method attempts to estimate the conditional mean power by using all the significant results in the published studies. Using this estimate of mean power, it is possible to calculate the ERR, that is, the success rate if these studies would be exactly replicated. The z-curve 2.0 method includes a feature to calculate the unconditional mean power, that is, an estimate of the power in studies that never were published because of statistically nonsignificant findings. Publication bias can be calculated from the unconditional mean power, that is, if the EDR (an estimate of the proportion of statistically significant results in all conducted studies, both published and unpublished) is lower than the observed discovery rate (ODR; i.e., statistically significant results in published studies). Using the EDR, we can also estimate the proportion of false positive findings in the literature. Detailed descriptions of these methods are published elsewhere (see; Bartoš & Schimmack, 2021; Brunner & Schimmack, 2020).

Of the 25 published studies, 24 contained test-statistics or exact p-values that could be analyzed with the z-curve method. From these studies, we manually extracted test statistics that were relevant for odor effects on face perception. Where applicable, we extracted statistics for the full analysis of variance and not follow-up contrast analyses. We also computed EDR and ERR on a random sample of p-values from each study because the z-curve has only been validated on test statistics in independent studies (i.e., one focal test per study). Specifically, we ran the z-curve analysis for 500 iterations and randomly sampled one p-value from each independent experiment (n = 27) in the studies with enough information to perform the z-curve analysis (n = 24).

Odor Effects on the Processing of Face Stimuli

In this review, nine studies used response times as a measure to investigate the effects of valenced odors on emotion recognition. Leppanen and Hietanen (2003) were the first to show that pleasant odors facilitated the classification accuracy of happy faces. In contrast, the speed of classifying disgusted faces was not affected by either pleasant or unpleasant odors. In the same study, a second experiment confirmed the effect of pleasant odors on happy faces. Although several studies have used similar methods as Leppanen and Hietanen (2003), their odor-face congruency effects have to our knowledge not yet been replicated. Instead, in a similar experimental design, D. Li et al. (2020) found that facial expressions (happy, neutral, and fearful) were overall recognized faster in an unpleasant food odor context. This effect was most pronounced for fearful faces. The effect was not present for happy faces in the pleasant odor condition. In two separate studies by Seubert et al., the results showed only an overall effect of odor presence on improving face classification (Seubert, Kellermann, et al., 2010; Seubert, Loughead, et al., 2010). In one of these studies, participants were more accurate and faster in recognizing disgusted expressions in both the pleasant and unpleasant odor conditions and slower at recognizing happy faces than in the neutral condition (Seubert, Kellermann, et al., 2010). In the second study, the results were similar, but this time only present for disgusted faces (Seubert, Loughead, et al., 2010). Hence, these results are difficult to consolidate. One reason for the heterogeneity of results might be that performance is already close to the ceiling level in a simple task such as emotion classification. If odor effects are generally relatively small, there is not much room for improvement. The results of Leleu et al. (2015) hint at this possibility. In this study, emotional expressions were morphed with neutral faces to show various levels of emotional intensity. The results demonstrated that facial expressions morphed with neutral faces were recognized at lower levels in congruent odor conditions (e.g., happy faces in the pleasant odor condition), for happy, disgusted, and angry facial expressions (Leleu et al., 2015). However, the results in a recent study (Syrjänen et al., 2017) showed that the recognition of dynamically morphing happy or disgusted faces (gradually emerging from neutral baseline face stimuli) was not affected by congruent odors. In the latter study, exploratory results indicated that both happy and disgusted expressions were recognized faster during exposure to an unpleasant odor (Syrjänen et al., 2017). Some studies have also presented stimuli in the left and right visual fields. In Stankovic et al. (2020), participants performed an emotion classification task on neutral or happy, surprised, fearful, angry, disgusted, and sad facial expressions. Pictures of faces were presented in the left or right visual fields while participants were exposed to an unpleasant odor in the left or right nostril. Participants were faster at recognizing faces overall in the left visual field (right hemisphere) during exposure to the odor in the right nostril (right hemisphere), indicating a cross-modal confluence in the right hemisphere resulted in a processing speed advantage; however, the study had no neutral control condition, so it is possible that the arousing effect of the odor, irrespective of its valence, improved processing of faces in the stimulated hemisphere (Stankovic et al., 2020). The right-hemisphere effect is consistent with results from our team that the face-sensitive N170 ERP component is more prominent in the right hemisphere; furthermore, our study showed that the responsivity of the N170 to disgusted faces is enhanced by an unpleasant odor (Syrjänen et al., 2018). Another recent study investigated whether odors might direct attention toward odor-congruent happy and disgusted facial expressions in a dot-probe task (Syrjänen et al., 2019). However, this study’s results gave no evidence that attention was automatically directed toward odor-congruent facial expressions. However, the results instead indicated that probes in the unpleasant odor condition were uniquely detected more rapidly over repeated trials within a block, raising the possibility that visual processing benefits of malodor exposure might accumulate over time. Similar time-on-task effects were found by Damjanovic et al. (2018) in a visual search task. The authors found that happy faces were recognized faster initially in a pleasant and an unpleasant odor context; however, this effect was reversed toward the end of the task.

While there is some evidence to suggest valenced odors might improve visual processing of emotionally congruent facial expressions, especially the combination of unpleasant odor and disgusted or fearful faces, the evidence is mixed. Increasing task difficulty, employing brief face duration times, subtle or morphed expressions, and so on, might be needed to achieve consistent effects of background odors. There are also some indications that odor effects develop as a function of exposure time during the experiment. Not accounting for such effects might to some extent explain why the findings are mixed.

Odor Effects on Evaluations of Faces

Ten studies assessed how valenced odors affect the subjective properties of faces. Two of these studies showed that neutral faces were rated as more unpleasant in the context of an unpleasant odor (Cook et al., 2015, 2018) and more pleasant during a pleasant odor context (Cook et al., 2015). However, one study did not find any effect of pleasant odors on classifying neutral faces as either pleasant or unpleasant, which might indicate that binary classifications of faces are less sensitive than ratings to the presumably subtle influence of odors (Bensafi, Pierson, et al., 2002). Several studies have shown that faces, regardless of expression, are rated as more arousing and that the face valence is affected in the direction of the odor valence (Syrjänen et al., 2018, 2019). However, results are not entirely consistent. Using Bayesian statistics, Syrjänen et al. (2017) found evidence against odor effects in face evaluations.

Ratings are also used to study valence congruency effects. One study showed congruency effects between facial expressions and odor valence; faces displaying happiness were rated as more pleasant during exposure to a pleasant odor, and disgusted faces were rated as more unpleasant during exposure to unpleasant odor (Cook et al., 2017). Odors may also affect facial attractiveness, a construct closely related to pleasantness; in one study, faces were rated as younger and more attractive-looking in a pleasant odor context (Seubert et al., 2014). In a similar vein, Dematte et al. (2007) found that faces were rated as less attractive in the presence of an unpleasant odor. However, Cann and Ross (1989) found no effects of pleasant and unpleasant odors on attractiveness evaluations in a blocked design. Finally, one study indicated that such facial evaluations were affected by odor exposure at a subthreshold level. Here, W. Li et al. (2007) found that only participants unaware of the presence of unpleasant odor were influenced to rate neutral faces as less likable. However, the evidence reviewed earlier suggests that also readily perceived odors may have similar influences on face evaluations.

Olfactory Conditioning: Learned Odor-Face Associations

In conditioning paradigms, associations between stimuli are created by simultaneous presentations. Here, we focus on olfactory conditioning studies that use face and odor stimuli. In studies on this topic, an unpleasant odor (unconditioned stimulus [UCS]) is usually associated with a face conditioned stimulus (CS+) by presenting these concurrently; faces that are not associated with the UCS (CS−) are used as a control. Some studies have used odors as a contextual cue and instead used startle sounds (sudden loud noises) as UCS (Hermann et al., 2000). These designs allow studying how learned cross-modal associations shape the processing of face stimuli; faces associated with unpleasant odors as CS+ may be processed differently from the CS− associated faces. Two studies have shown response-time effects in a gender classification task, but results are inconsistent; CS+ faces associated with an unpleasant olfactory UCS were initially faster at the beginning of the task (Gottfried & Dolan, 2004). However, in another similar study (Gottfried et al., 2002), results showed slowed responses to the CS+ faces conditioned with either pleasant or unpleasant UCS odors. More research with high-powered samples is needed to establish how valenced odor conditioning might modulate face classifications.

Several conditioning studies have shown similar results as when odors and faces are presented concurrently, faces previously conditioned by an unpleasant odor are evaluated more negatively (Hermann et al., 2000; Steinberg et al., 2012, 2013). Steinberg et al. (2013) found that faces conditioned with pleasant odors were preferred over faces paired with clean air and an unpleasant odor. In four experiments, Todrank et al. (1995) found that neutral faces associated with pleasant and unpleasant odors were liked and disliked, respectively. However, this effect was only consistent with odors that were people-related (e.g., soap or sweat like odors) but not for object-related odors such as woody odors. One study showed that faces were rated as more arousing if they were paired with either pleasant or unpleasant odors (Hermann et al., 2000). Thus, similarly to contextual odors, odor conditioning seems to have robust effects on facial evaluation, with odors biasing the evaluation of faces in the odor valence direction. This capacity is highlighted in a study by Steinberg et al. (2012). The authors showed that in a stimulus set of 208 faces, only two CS+ pairings (neutral faces) with the unpleasant UCS odor were enough to affect facial evaluations negatively. We now know that animals can associate aversive odors with visual objects in a fast and flexible manner (Handler et al., 2019). These findings show a similar capacity in humans to rapidly associate faces with aversive odors.

Z-Curve Results

Using z-curve, we found that 75% of the focal tests that were reported were statistically significant. However, if we assume that publication hinges on statistically significant results, 75% might be an overestimate of the true hypothesis confirmation rate. Because many of the studies reported several statistical tests, resulting in a higher chance of obtaining a statistically significant result. To estimate an unbiased ODR, we sampled one test from each study. The results were similar compared with the full sample of p-values, with a mean of 77% significant results. The results of the z-curve analysis indicated clear evidence of publication bias in the studies in this review. Given the estimated power in the included studies, only 10% of the results were expected to result in statistically significant findings. This discrepancy between the ODR and the EDR resulted in an expected replicability rate of only 47%. That is, if these studies would be exactly replicated, only 47% of the studies would yield similar findings. In the following, we highlight outstanding questions for future research studies and suggest measures that could improve their reliability.

Future Directions

In the review mentioned earlier, we have highlighted key results emerging from the literature on odor effects on face perception. Several avenues remain to be explored in further research. For example, only a few studies investigated how interindividual variation in odor sensitivity affected odor effects in facial perception. Individuals vary considerably in their odor preferences, which might influence how different odors affect facial perception. For example, people vary in disgust sensitivity (Tybur et al., 2018), which might be especially relevant for how unpleasant odors might affect perception of facial expressions. Here, a newly developed Body Odor Disgust Scale might be a suitable instrument for investigating individual differences in olfactory related disgust (Liuzza, Lindholm, et al., 2017). Although the Body Odor Disgust Scale was developed for measuring body odor-related disgust, it has excellent construct validity in relation to existing disgust scales and has been shown to predict disgust responses to human sweat bio-samples (Liuzza, Olofsson, et al., 2017).

Nevertheless, individual variation is not only relevant for unpleasant odors. For example, we have observed that odors typically regarded as pleasant (e.g., lilac) might be perceived as unpleasant at higher concentrations by some participants (Syrjänen, 2020; Syrjänen et al., 2018). Conversely, an unpleasant odor such as valeric acid is by some participants rated as positively valenced (see Figure 2 for a summary of valence ratings for these odors at three concentrations; data from Syrjänen et al., 2018). Although olfactory valence is sometimes thought of as a hard-wired response, perhaps universal (Yeshurun & Sobel, 2010), our observations suggest this assumption may lead to an underestimation of the individual variation in olfactory valence evaluations. Future work could instead use different odors for each participant to maximize valence effects.

Among the reviewed studies, many have used unpleasant, presumably disgusting odors, and facial expressions such as angry or fearful that do not have a natural connection to these odors. This usage might reflect the fact that anger and fear are well investigated in affect perception research but that these emotions are not easily mapped onto the affective space of odors. Instead, we argue that the basic emotion most well-matched by unpleasant odors is disgust, and for pleasant odors, happiness (Bensafi, Rouby, et al., 2002; Croy et al., 2011; Glass et al., 2014). Future research should consider odors congruent with distinct emotional expressions, and this olfactory-visual congruency should be empirically validated. For example, odors such as smoke may elicit fear in specific contexts (Stevenson, 2010). Thus, the odor might modulate the processing of fearful expressions in a context-dependent way. Future studies might also explore the connection between odors and mood and how these relate to the perception of emotional expressions. Also, interindividual factors such as gender, age, and culture should be investigated in future research. These factors typically affect the perception of emotional information in other sensory modalities such as vision and hearing (Cortes et al., 2017, 2019) and therefore probably also in the context of different odors.

It is important to note that the studies reviewed here are typically not characterized by open research practices such as transparent and preregistered study protocols and open data. In recent years, the low reproducibility of research findings has led to a discussion about methodological improvement in psychological science (Open Science Collaboration, 2015). Transparent and open study protocols might reduce practices that result in nonreplicable findings, such as hypothesizing after results are known (HARKing; Kerr, 1998). Considering that at least half of the studies utilized small sample sizes and that odor effects typically may be assumed to be subtle, we can expect that some of the findings in this area are false positives or negatives (Button et al., 2013). Almost all studies found some effects of odor on facial perception. If we assume that many studies in the field are underpowered, there might be a file drawer problem of conducted but not published studies due to nonsignificant findings. These concerns were confirmed by the z-curve analysis. Given the power in the included studies, the EDR was only 10%. Furthermore, the expected replicability rate indicated that 47% of the results would replicate in exact replication studies. The research in this area could also benefit from standardized measures, testing procedures, and stimuli. Practices that have been highly successful in emotion research include using standardized stimuli such as the international affective picture set (Lang et al., 2008) and assessments such as the self-assessment manikin (Bradley & Lang, 1994). Standardization would benefit the research community by making it easier to compare results across studies and sustain a cumulative progression of research.

Because of the relatively small number of published studies to date, open data would be beneficial for the whole community of researchers. Open data practices will enable reinterpretation and creation of novel hypotheses using existing data, and such practices will reduce the need for duplications. On a positive note, several large multinational collaborations are formed to better assess the reliability and generalizability of research findings. In the olfactory research community, the Global Consortium for Chemosensory Research (www.gcchemosensr.org) studies olfactory-related changes in COVID-19 infection (Parma et al., 2020). The first publication from the Psychological Science Accelerator, another large-scale research collaboration, assessed how faces are evaluated in different world regions (Jones et al., 2021). Such efforts might be used as templates for conducting large-scale replication efforts.