Imagined eating – An investigation of priming and sensory-specific satiety

.


Introduction
Obesity remains a problem at the population level (The Lancet Diabetes & Endocrinology, 2020). Watching television has been implicated in the development of obesity (Boulos et al., 2012). An important factor is food advertisement (Boyland et al., 2016). Priming, i. e, exposure to stimuli to increase desire and affect behaviour, is one mechanism through which food advertisements affect viewers' eating behaviour (Harris et al., 2009). In fact, a recent large meta-analysis concluded that exposure to visual food cues can prime and affect eating behavioural outcomes to the same extent as real food (Boswell & Kober, 2016).
Meanwhile, television is being overtaken by new media forms, at least among adolescents (Twenge et al., 2019). In general, more time is nowadays spent on the internet and social media platformsalso by the older demographic (Vogels, 2019). Photography is ubiquitous on these platforms (Heyman, 2015), which is partially explained by its coevolution with the then-emerging, nowadays commonplace smartphone (Fridman, 2021;Pew Research Center, 2019). Food represents an exceptionally popular topic (Kantchev, 2014;Prince, 2014). At the time of writing, about 490 million Instagram posts were tagged with "food".
The continuous access and such unprecented exposure to food content and advertisement may, thus, constantly prime users for food intake.
How may the connection between visual food contentwhich, after all, isn't edible (Tooming, 2021) and its priming effect be explained? Grounded cognition offers a theoretical basis. According to grounded theories, perceptions are processed in modality-specific brain areas (Barsalou, 2008). In the words of Simmons et al. (2005), "brain areas representing knowledge for a particular category are those typically used to process its physical instances" (Simmons et al., 2005). This has been termed "mental simulation". Imaging studies demonstrate how viewing food images activates brain areas involved in taste perception and reward processing (Simmons et al., 2005). This neural activity enables the inference of a food's taste, energy content, and hedonic value (Toepel et al., 2009;van der Laan et al., 2011). Such sensory processing-related neural activity can further affect the internal physiological state and external behaviour (Fadiga & Craighero, 2004;Ganis et al., 2004;Jeannerod, 2001). Priming, then, is the sensory processing of relevant stimuli that activates strongly associated goal states, which promote respective goal-directed behaviour. A more concrete example would be the sensory processing of food stimuli that activates goals of experiencing pleasure, which promote food seeking and consumption.
In the physical instance of eating, at first, pleasure derived from a meal tends to increase (Crolic & Janiszewski, 2016). However, as the meal continues, the pleasure derived from it declines, until the meal is terminated (Cornil, 2017). Construing priming as an analogue of this initial increase of pleasure invites the question, if there is also an analogous case to the following satiation.
Some evidence suggests there is indeed an analogous case to satiation, with repeated exposure to visual food stimuli inducing states of satiation. Larson et al. (2014) had study participants evaluate the taste of either 20 or 60 salty or sweet food pictures, followed by tasting three peanuts. Participants that had evaluated 60 salty food pictures rated the three peanuts as less enjoyable, suggesting that they had satiated. The presumed cause for the satiation is the repeated implicit mental simulation required to infer the depicted food's taste, bringing about the subsequent physiological and behavioural response.
While a three-peanut enjoyment-rating may not appear as a particularly reliable outcome measure, similar mental imagery-based research similarly found that imagining eating 30 (vs. 3) M&M's translated into a subsequently lower actual consumption of M&M's (Morewedge et al., 2010). This finding has been replicated in much larger samples (Camerer et al., 2018), and generalized to different food (Missbach et al., 2014). Imagining the "multisensory" eating experience can also lead people to choose smaller portions at no hedonic cost (Cornil & Chandon, 2016). Findings such as these have led some authors to preliminarily suggest digital food content consumption as a potential dieting strategy Spence et al., 2016).

Sensory-specific satiety
Classic research has shown that satiation is sensory-specific (B. Rolls, Rolls, et al., 1981). Sating on one food type (e.g., sausages) predominantly decreases liking and subsequent ad-libitum consumption of that very food, while affecting responses to other food (e.g., cheese on crackers) much less. In fact, after having satiated on, e.g., sausagesa savoury foodliking for not eaten sweet food may increase (B. Rolls, Rolls, et al., 1981). Importantly, these effects are apparent already 2 min after consumptionand also remain so after 20 mini.e., before substantial nutrient absorption could take place. Thus, these effects are not directly related to physiological homeostatic needs but rather related to sensory stimulation and cognition (B. Rolls, Rolls, et al., 1981;B. J. Rolls, 1986).
If satiation is sensory-specific, then eating a varied meal might increase overall intake. This has also been established (McCrory et al., 2012;B. J. Rolls, Rolls, et al., 1981). For example, during a meal of four ad-libitum courses, within-subjects energy intake was 44% higher with different food served at each course compared to when serving the same food (B. J. Rolls et al., 1984). The higher intake came mostly from the latter two courses, which corresponded to a lack of a decline in food liking. Moreover, even after the varied four courses, participants had still not reached general satiation, as small taste samples of uneaten food remained palatable.
The perception of food variety seems to be multisensory and additive, and isolated changes in food properties may exert only minor influences on consumption (B. J. Rolls, 1986;B. J. Rolls et al., 1984). In one study, serving participants multiple yoghurt flavours did not increase food intake over serving only a single yoghurt flavour (B. Rolls et al., 1982). In another study, pleasantness changes were specific to the consumed food's colour (Smarties), but total intake remained unaffected (B. Rolls et al., 1982). Hence, affecting food intake appears to require changes also in basic taste, such as sweet vs. savoury, other sensory properties, or a combination (B. J. Rolls, Rolls, et al., 1981).
However, a more recent laboratory study with potato chips mainly differing in flavour found that participants having sated on a large preload of one flavour tended to subsequently switch to the most dissimilar flavour (Maier et al., 2007). This finding was confirmed in a free-living consumers' diary study. It is also known that colourful salads are more attractive than pale ones (Mielby et al., 2012;Paakki et al., 2019), which seems to imply that, after all, visual aspects have appetite-stimulating effects.
Overall, the current literature contains several gaps. Studies on mentally induced satiation did not compare outcomes to participants' baseline state. It is therefore unclear whether the experimental outcomes were comparatively related to priming or satiation. These studies did not consider priming at all. One might further argue that actual satiation requires depression of wanting or liking below baseline levels, just as a satiating meal would leave one less hungry than before. Studies have also given limited consideration to sensory-specific satiety. In particular, the authors have not distinguished between general and specific eating desires. Lastly, the role of food flavour and colour on the satiation response remains unclear, and this has certainly not yet been investigated in the context of mentally simulated eating.
It is worth disambiguating the terminology around satiation and satiety, and our use thereof. In general, the internal appetite drive can be conceptualised in terms of meal initiation and meal termination (de Graaf et al., 2004). The inter-meal appetite state is referred to as satiety, while the within-meal appetite state is referred to as satiation. Satiation can further be divided into a physical factor, the feeling of fullness, e.g., due to stomach distension, and a cognitive or neurological factor, the food's pleasantness or reward value (de Graaf et al., 2004;Kringelbach, 2003;B. J. Rolls, 1986). The cognitive factor is also the subject of sensory-specific satiety, as described above. In this work we are using the terms satiation and sensory-specific satiety interchangably, in the specific sense of the latter.

Aims
The present paper poses the following two questions: 1) How does repeated mentally simulated consumption affect sensoryspecific satiety? 2) How does stimulus variety modify the aforementioned satiation response?
Based on these general research questions, the specific aims are to: 1) Replicate previous findings of repeated mental simulation leading to satiation 2) Extend previous literature by a. Comparing experimental effects to baseline scores b. Disentangling general and sensory-specific eating desires 3) Systematically explore the effect of food variety, specifically a. Visual variety b. Flavour variety Three studies were carried out to address these aims. The first study investigates the effect of imagining eating 3 or 30 M&M's of the same colour on sensory-specific eating desires and measures of food-specific liking and wanting. The second study adds visual variation by using M&M's of different colours. The third study uses differently coloured Skittles to introduce both colour and flavour variation. Fig. 1 visualizes the studies' contribution toward the overall aims.

Study 1
Study 1 had three purposes. The first is the initial validation of the experimental paradigm. Therefore the study's design was heavily influenced by prior research. It adopted the scheme from Morewedge et al. (2010), comparing imagining eating 3 vs 30 M&M's. Following Larson et al. (2014), participants also viewed images of the food, in this case, M&M's. The following hypotheses operationalize this purpose: To extend the existing literature, this study's second purpose was to measure outcomes both before and after the experimental manipulation. By establishing a baseline it is possible to distinguish between three different outcome patterns: 1) priming, i.e., elevation above baseline; 2) partial satiation, i.e., decrease from a primed state back towards baseline; and 3) full satiation, i.e., a decrease to or even below baseline (cf. Groves & Thompson, 1970). Prior research did not distinguish between these outcomes.
The third purpose of this study was to better understand the role of sensory specificity in repetition-induced satiation effects. Three additional hypotheses operationalize study purposes two and three: H1.3. Participants prime to the exposed tastes after 3 trials H1.4. Participants satiate to the exposed tastes after 30 trials H1.5. Unexposed tastes remain unaffected by the experimental manipulation 2.1. Methods

Participants
Approximately 400 subjects -200 per condition and balanced by sex -were recruited off the research platform Prolific.co. In comparison, a large-scale replication of Morewedge et al.'s (2010) study sampled at five times the original size for approximate 80 participants per group (Camerer et al., 2018). Additionally, purely imagined eating may produce smaller effects than actual eating, warranting a higher sample size. Ultimately, our sample size is in line with common online research practice, with group sizes of a hundred participants and upwards (e.g., n = 200 Prolific participants in Hagen, 2020;Krishna & Hagen, 2019). Participants had to be at least 18 years old, located in the United Kingdom, and not following any diet, incl. Vegetarian or vegan. All participants gave their informed consent before taking part in the study.
Due to the large active participant pool, data collection was rapid and therefore time of day was relatively controlled. All responses were collected between 14:30 and 17:30 local time.
To conservatively account for extremely low participation effort, unusually fast or slow submissions were removed according to the Interquartile Range (IQR) heuristic. 1 Participants with only a few missing data were not removed 2 ; instead, these participants were removed only for the corresponding analyses.
The final sample consisted of 352 participants (see Table 1 for demographic details). This number was lower than the target. Of the approximately 50 missing participants, about 40% were due to study dropout or low-quality responses, and 60% were removed due to very fast or slow study completion.
In the experimental manipulation, participants were explicitly cued to imagine eating either 3 or 30 M&M's ("Please vividly imagine eating the pictured M&M"). After an obligatory 2 s delay to prevent participants from casually skipping through the stimuli, participants could advance to the next image at their own pace.

Measures
Before the experimental manipulation, participants evaluated their hunger level, general liking for the stimulus food, current desire to eat, as well as desire for something sweet, salty, and fatty on visual analogue scales (VAS; anchors: 0 = "not at all", 100 = "very much" or "extremely"; Duerlund et al., 2019). Eating desires have been shown to correlate with implicit food preferences and subsequent ad-libitum intake (Griffioen-Roose et al., 2010).
After the experimental manipulation, participants reassessed their eating desires, as well as their expected enjoyment ("How much would you enjoy eating this M&M right now?"; cf. Larson et al., 2014) and desired number ("How many M&M's would you like to eat right now", to be selected from a picture menu, options 1-10 pieces; cf. Morewedge et al., 2010;cf. also Cornil and Chandon, 2016) of the same M&M that they had been exposed to before. Fig. 3 illustrates the overall experimental flow. Finally, participants could give feedback on the overall study experience in an optional freetext field.

Data analysis
The data were analyzed in R v4.1.0 (R Core Team). Specifically, desire scores were modelled using linear mixed models (LMM). The mixed models were fitted with random intercepts for participants to account for repeated measures (pre-post). This type of analysis is statistically equivalent to a repeated measures analysis of variance; however, mixed models are a more general and flexible tools, and were chosen for convenience. Expected M&M's enjoyment was fitted to a  linear model (LM) and the desired number of M&M's to a generalized linear model (GLM) with the Poisson link function. All statistical models controlled for initial hunger, general liking for the stimulus food (i.e., M&M's), gender, and age. BMI was not included as manual inspection suggested the self-reported data to be unreliable and insufficiently precise. Likelihood-ratio tests determined the influence of experimental factors on the outcome of interest, starting with the most complex model (e.g., the interaction between the number of M&M's and time). The significance level was α = 0.05. Four contrasts were tested for betweengroup differences (time point group differences) and within-group changes (desire score pre-post differences). Bonferroni correction accounted for multiple comparisons. Cohen's d served as an effect size estimate.
The desire for something sweet was significantly affected by the interaction of time and the number of trials (χ 2 (1) = 22.234, p < .001). Post hoc contrasts (corrected for four tests) indicated no significant group differences in pre-scores (p = 1), but post-scores were significantly higher after 3 compared to 30 trials (CE = − 10.34 ± 2.74, d = 0.62, t (357) = − 3.779, p < .001). There was also an increase after 3 trials (CE = 9.76 ± 1.79, d = 0.59, t (354) = 5.459, p < .001), but no significant change after 30 trials (p = .83).  . The experimental flow of study 1. Participants first evaluated their eating desires (general, sweet, salty, fatty). Then, they saw either 3 or 30 M&M's pictures, one by one, prompted to imagine eating them. Afterwards, participants again answered questions about their eating desires, as well as their expected enjoyment and desired number of M&M's. The order of post-eating desire assessment and M&M's-related outcomes was randomized, and the order of expected enjoyment and desired number of M&M's was also randomized.
Number of trials significantly affected desired number of M&M's Fig. 4. Study 1 results. Panels A, B and D show the participants' eating desires (general, sweet, fatty) before (pre) and after (post) the experimental manipulation (3 or 30 imagined eating trials). Panel C shows participants' desire for something salty before and after the manipulations, with the two experimental groups pooled due to no significant differences. Panels E and F show the post-experimental expected enjoyment and the desired number of M&M's between the two groups.
(Deviance (1) = 4.82, p = .03). Participants in the 30-trials group desired less M&M's than those in the 3-trials group (contrast ratio [CR] = 0.91 ± 0.04, d = 0.07, z-ratio = − 2.194, p = .03). Fig. 4 visualises the results. Through the optional free text feedback, several participants commented on the repetitiveness of the orange M&M's (emphasis added): "I don't like the colour orange, so that put me off straight away. I would never eat 30 sweets in one go. After perhaps 6 I would have had enough. The longer the survey went on the more nauseous I felt. A variety of colours would have helped me enjoy a few more." "If you want people to imagine eating M&M's … at least make them different colours!" "It may have been helpful to have different colour M&M's, as looking at the same colour for all got repetitive and isn't realistic of actually eating M&M's."

Discussion
This study validates the experimental paradigm by successfully reproducing some prior results, thereby fulfilling its first purpose. There was an experimental group difference between participants that imagined eating 3 or 30 M&M's. As reported by Morewedge et al. (2010), participants that had imagined eating 30 M&M's desired fewer M&M's than participants that only imagined eating 3, supporting H1.1.
A similar effect was seen for expected M&M's enjoyment, supporting H1.2, and congruent with the results by Larson et al. (2014). The effect is much larger for expected enjoyment than for the desired number of M&M's but smaller than the effect size reported by Larson et al. (2014). This might be due to methodological differences, first and foremost the absence of actual food consumption. The present study also used fewer food images (30 vs. 60, cf. Larson et al., 2014). However, other studies have used a comparable number of trials (Missbach et al., 2014;Morewedge et al., 2010).
Towards the second purpose of comparing experimental effects to baseline values, our pre-post design demonstrated that fewer imagined eating trials resulted in priming, while more trials resulted in satiation (cf. Crolic & Janiszewski, 2016;Groves & Thompson, 1970). The results are similar to those reported by Lambert et al. (1991), who found an increased desire for chocolate after (vs. before) 90s exposure to a single M&M picture. While their participants merely looked at the M&M's, this study's participants actively imagined eating them. Total exposure time in Lambert et al.'s study likely fell in between the current study's two conditions (exposure times were not recorded in study 1, but cf. studies 2 and 3). Furthermore, it might be worth pointing out that Lambert et al.'s (1991) participants looked at only a single picture, while this study's participants were exposed to multiple pictures. Haasova et al. (2016) speculated that their participants' lack of satiation when imagining eating spoons of yoghurt might have been due to the continuous nature of the eating episode, i.e., eating one yoghurt as compared to eating several discrete food items (e.g., M&M's).
Regarding the third purpose of investigating the sensory specificity of imagined eating, the different eating desires did, in fact, not all respond the same. The desire for something sweet was above baseline after 3 trials, i.e., indicating a primed state, and (back) at baseline level after 30 trials, i.e., indicating a fully satiated state. General desire to eat was also increased after 3 trials, and significantly decreased even below baseline after 30 trials. Hence, both stimuli-related eating desires indicated satiation after 30 trials. These effects were comparatively large, in agreement with the findings by Larson et al. (2014), and hence expected. They also lend support for hypotheses H1.3 and H1.4.
Desires for something fatty or salty, i.e., stimuli-unrelated tastes, were expected to show either no change or possibly even an increase after many trials, when participants may have desired a flavour change. The opposite happened; namely, a decrease was particularly apparent after the many trials. Therefore, hypothesis H1.5 is rejected.
The free-text comments indicate that some participants seemed to intuitively grasp the nature of the repetitive stimuli, almost foreseeing the next study where differently coloured M&Ms are introduced.

Study 2
Study 1 validated our general experimental paradigm and fulfilled the first two research aims. Hence, study 2 targets the third research aim: investigating the role of food variety on the satiation response. Some participants had already commented on the visually repetitive stimuli used in the first study. The focus, then, is to investigate how visual food variety, independently of flavour variety, impacts imagined eating outcomes. Rolls et al. (1982) had already conducted a similar experiment using Smarties of various colours, and found a difference in the post-consumption decline of liking, but not food intake. However, this research was not based on mental imagery.
Assuming that satiation requires activation of the gustatory cortex (Larson et al., 2014;Simmons et al., 2005), the mere visual variety in the absence of differences in taste or flavour should not differentially impact imagined eating outcomes, compared to the visually monotonous stimulus of study 1. Thus, as in the previous study, participants should satiate to the stimulus after 30 trials.

H2.1. Visual variety has no impact on satiation.
In the mental imagery context, where no external information is sensed, perceptions must be inferred based on prior experience and beliefs. Hence, it seems reasonable to assume that individual participants' beliefs about a food's taste, rather than its actual taste profile, would determine imagined eating outcomes. If so, the effect of the experimental manipulation must then be moderated by whether participants mentally simulate varying or identical taste experiences. For instance, believing that differently coloured M&M's also taste differently should impair satiation, compared to believing they all taste the same.

H2.2. Stimulus taste belief significantly moderates eating behavioural outcomes.
METHODSAfter the application of study 1's inclusion and exclusion procedure, 398 valid responses remained (for details, see Table 1). Prolific participants from study 1 were prohibited from participating in this study.
The procedure mostly followed study 1. The stimuli were mixedcoloured M&M's, while the orange-coloured M&M remained for the post-assessment (see Fig. 2). Orange M&M's were excluded from the stimulus set to test for transferability across colours. While the different M&M's ought to be almost identical in taste and flavour, participants were asked if they thought this to be the case ("Do you believe that M&M's with different colours all taste the same?", VAS ranging from 0 to 100, anchored: 0 = "They all taste different", 50 = "I have no idea", 100 = "They all taste the same"). Data analysis for study 2 differed from study 1 only in terms of model selection, starting with models also including taste belief as an interacting factor. Otherwise, the procedure was as previous, i.e., a step-wise decrease in model complexity via likelihood-ratio tests. An additional analysis comparing studies 1 & 2 also followed this procedure.

Results
Participants in the 30-trials group spent less time on each trial than their counterparts in the 3-trials group (3 trials: median = 6.6 s, IQR = 5.9 s; 30 trials: median = 3.9 s, IQR = 3.6 s). The majority of participants considered M&M's to all taste the same (P 50 = 28; scale: − 50 = "They all taste different", 50 = "They all taste the same").
The desire for something fatty was significantly affected by the interaction of time and the number of trials (χ 2 (1) = 8.37, p = .004). The two experimental groups did not differ at the beginning of the experiment (p = 1), but the 30-trials group (vs. 3) showed significantly lower desire for Fig. 5. Study 2 results. Panels A, B and D show the participants' eating desires (general, sweet, fatty) before (pre) and after (post) the experimental manipulation (3 or 30 imagined eating trials). Panel C shows participants' desire for something salty before and after the manipulations, with the two experimental groups pooled due to no significant differences. Panels E and F show the post-experimental expected enjoyment and the desired number of M&M's between the two groups. The shaded area represents effect confidence intervals. Furthermore, panel F illustrates the dependency of the desired number of M&M's on the interaction of the number of trials and taste belief, with a low taste belief score indicating the expectation of differently coloured M&M's to all taste differently, and vice versa. something fatty at post (CE = − 6.22 ± 2.29, d = 0.53, t (526) = − 2.71, p = .028). There was no change after 3 trials (p = 1), but a significant decrease after 30 (CE = − 5.45 ± 1.18, d = 0.46, t (527) = − 4.63, p < .001).
The effect of number of trials on desired number of M&M's was significantly moderated by taste belief (Deviance (1) = 4.25, p = .04). At the taste belief extremes, participants in the 30-trials group that considered coloured M&M's to all taste the same desired significantly fewer M&M's than participants that thought they all taste different (CR = 0.81 ± 0.07, d = 0.18, z-ratio = − 2.27, p = .02). This was not the case for participants in the 3-trials group (p = .58). Conversely, for those participants that considered M&M's to all taste the same, 30 trials led to significantly fewer desired M&M's compared to 3 trials (CR = 0.82 ± 0.05, d = 0.16, z-ratio = − 3.45, p = .001). There was no group difference between 3 and 30 trials for participants that thought the M&M's have different tastes (p = .53). Fig. 5 visually summarises the results.

Discussion
The design was successful in isolating the effect of visual variety, as the majority of participants considered coloured M&M's to be identical in taste.
The results are very similar to study 1, lending general support for H2.1. The desired number of M&M's was significantly increased in this study compared to study 1. However, the effect was very small, and the effect applied to both 3-and 30-trial groups. Hence, this result does not prove any priming effect of colour variety, as the pre-post difference between 30 trials was smaller than expected. There were no statistically significant between-study differences in any of the other outcome measures.
Taste belief significantly moderated the experimental effect only on the desired number of M&M's, but the effect size was small. The general lack of significant findings may in part be due to the very uneven distribution of taste belief. As already indicated, this was expected as a consequence of the design. The results, therefore, do not necessarily imply the irrelevancy of taste beliefs for mentally simulated eating. Rather, this exploratory investigation provides only very minor evidence towards H2.2.
On the whole, the results of study 2 suggest that colour variety in this experimental setup does not make a substantial difference for mental imagery-induced satiation.

Study 3
Study 3 tests the impact of flavour variety, in addition to visual variety on mental imagery-induced satiation. The flavour of food, as compared to food colour alone, is directly relevant to consumptive sensations. If imagined consumption of the same food satiates due to repetitive mental consumption simulations, then imagining various flavours should lead to more differential simulations and, thus, a primed state (cf. Groves & Thompson, 1970).

Methodology
Study 3 included 399 valid responses, following the same exclusion principles as previously stated (see Table 1 for demographic details). Participants from studies 1 and 2 were prohibited from taking part in the study.
The stimuli were mixed-coloured Skittles (Wrigley Co.) These are very similar in appearance to M&M's (see Fig. 2), yet each colour has a distinct flavour. Apart from the change of stimuli, study 3 was identical to study 2.
As previously, an additional analysis of the combined data of studies 2 and 3 tested for a between-study effect to isolate the effect of flavour variety.

Results
Participants in the 30-trials group spent less time on each trial than their counterparts in the 3-trials group (3 trials: median = 5.3 s, IQR = 4.4 s; 30 trials: median = 3.9 s, IQR = 4.0 s). On the question of whether differently coloured Skittles taste different or the same, the majority of participants considered Skittles to all taste different (P 50 = − 38; scale: − 50 = "They all taste different", 50 = "They all taste the same").
Desire for something sweet was significantly affected by the interaction of time and number of trials, as well as a taste belief main effect (χ 2 (1) = 4.58, p = .03). In post hoc testing, after Bonferroni-correcting for five comparisons, the taste belief main effect was no longer significant (p = .17). While there was no group difference in initial desire to eat something (p = 1), post-scores were significantly higher for the 3-compared to the 30-trials group (CE = 7.70 ± 2.57, d = 0.45, t (634) = − 3.00, p = .01). Participants showed an increase in scores after 3 trials (CE = 12.79 ± 1.76, d = 0.73, t (401) = 7.27, p < .001), but no change after 30 (p = .5).
There was no significant difference between studies 2 and 3 in any of the outcome measures (all p > .1).

Discussion
Participants expected Skittles to differ in flavour, again validating the choice of stimuli. The results from study 3 are very similar to study 2, i.e., combined colour and flavour variety did not affect any measures of satiation. There was no evidence to support hypothesis H3.1. This is in line with early studies on isolated sensory properties' effect on sensoryspecific satiety (B. Rolls et al., 1982;B. J. Rolls, Rolls, et al., 1981, 1984. Moreover, in contrast to the previous studies, the group difference (3 vs. 30) in the desired number of Skittles did not quite reach statistical significance. Nevertheless, the effect was similar in direction and size.
There are two competing overall explanations for the lack of an effect. The first explanation is that there simply was no difference in priming between the stimuli. The second explanation proposes that the task was sufficiently repetitious to eventually overcome initial priming differences. In other words, all participants satiated despite initial differences in priming.
Regarding the first overall explanation, it may be that the stimuli were inherently too similar to meaningfully impair response satiation. The response to similar stimuli would be expected to generalize, based on overlapping neurological activation (Groves & Thompson, 1970). However, it doesn't appear reasonable to argue that the stimuli per se Fig. 6. Study 3 results. Panels A-D show the participants' eating desires (general, sweet, salty, fatty) before (pre) and after (post) the experimental manipulation (3 or 30 imagined eating trials). Panels E and F show the post-experimental expected enjoyment and the desired number of M&M's between the two groups.
were too similar when both Redden (2008) and Galak et al. (2009) used jelly beansvery comparable to Skittlesto successfully introduce variety-effects.
The stimuli may just have been perceived as very similar. Redden (2008) had shown that satiation depends on stimulus categorizationor the granularity of perception. He had participants consume jelly beans, accompanied by either a general label ("jelly bean") or a flavour-specific label (e.g., "cherry"). While all participants consumed the same number and flavours of jelly beans, those that saw the flavour-specific labels experienced a lesser decrease in jelly bean enjoyment. The effect on jelly bean enjoyment was mediated by perceived repetition.
Relatedly, researchers demonstrated that attending to the various flavours contained in the very same food can lead to increasing, rather than decreasing, liking across consumption trials ("hedonic escalation"; Crolic & Janiszewski, 2016). For example, in one of the studies, the researchers guided participants unfamiliar with the experimental food (tortilla chips) to either discover new flavours with each successive piece eaten or repeatedly focus on the first identified flavour. Participants guided towards discovering new flavours experienced significantly more hedonic escalation.
Theoretically, stimulus perception should affect the degree of neurological activation overlap underlying stimulus generalization similar to actual differences in stimuli (Groves & Thompson, 1970). Hence, participants generically perceiving "Skittles" would have mentally simulated a generic flavour, rather than any specific flavour, e. g., "lemon Skittles". While Redden's (2008) participants tasted the food stimuli, this study's participants merely imagined them. They did also not see any explicit flavour cues. Therefore, this study's participants would be more likely to disregard actual flavour differences.
Mentally simulated tastes are not particularly intense stimuli. However, stimulus intensity is an important determinant of the priming response (cf. Groves & Thompson, 1970). There might also be an interaction between stimulus intensity and variety on response generalizationan organism may be more discriminating between stimuli at higher intensities (cf. generally Gibson, 1979). In other words, given the study's low-intensity stimuli, a more pronounced variety may have been required to cause spontaneous attention shifts.
It is also possible that participants simply did not or were not able to mentally simulate the varying flavours. This could have occurred despite conceptually perceiving the differences if relating more to their mental imagery ability. Basic tastes are possibly more readily imagined than flavours, which have an olfaction component. Research suggests that olfactory imagery is more difficult to invoke and generally less vivid, at least compared to visual imagery (Stevenson & Case, 2005). For instance, in a study by Krishna et al. (2014), physiological and behavioural responses to food-related olfactory imagery were moderated by mental imagery ability. Unfortunately, we didn't assess participants' mental imagery ability or vividness (c.f. Croijmans & Wang, 2022).
Regarding the latter overall explanation, participants may have perceived and simulated the varying Skittles flavours, yet 30 trials were sufficient for participants to satiate on the food despite the flavour variety. As reported by Larson et al. (2014), offering a much more varying stimulus set, satiation was not evident after 20 trials, but after 60 trials. Hence, there may have been a difference at earlier points in the experiment, e.g., after 10 trials.

General discussion
Overall, all three studies provided very similar results, showing that 3 imagined eating trials increase the desire for the stimulus, while 30 trials decrease it. Furthermore, stimuli-unrelated eating desires decrease after 30 imagined eating trials, and to a lesser extent also after 3 trials. The studies imply that mental imagery-induced sensory-specific satiety is very reproducible -at least with sweet food stimuli.
The colour variety introduced in study 2 was expected not to make a significant difference compared to study 1, on the basis that colour should be indifferent to mental food consumption simulations. Apart from a very small difference (d = 0.05) in the desired number of M&M's, there were indeed no differences between studies 1 and 2. In contrast, the introduction of flavour variety in study 3 was expected to lead to different results compared to study 2. Flavour differences should modify consumption-related mental simulations, leading to differential neurological activation and thereby priming (Groves & Thompson, 1970). The results do not reflect this expectation, as there were no significant outcome differences between the two studies.
Effects sizes are 35-79% smaller than those reported by Larson et al. (2014;d = 0.94). There are two possible reasons. First, Larson et al. (2014) used 60 trials on the high end, which may have driven satiation further, thereby exacerbating the group differences. Second, the participants did not consume any food. A recent in-house pilot study (unpublished) replicated the online study design but tested actual consumption enjoyment of M&M's before and after the imagined eating task. Here, some effects (d > 1) were even slightly larger than those in Larson et al. (2014). Fig. 7 visualizes the effect sizes and directions of the different studies. It is important to remember that the only statistically significant between-study difference in the figure, although visually minuscule, is in the desired number of food pieces between studies 2 and 3. The rank of the different studies is not consistent across the different outcome measures. Moreover, irrespectively of statistical significance, studies' effect sizes would have been expected to follow a rank pattern corresponding to increases in variety. This is also not the case.
In contrast, effect directions are very consistent, with no conflicts between the studies on any of the outcome measures. The priming effect of 3 trials on the desire to eat something and desire for something sweet is particularly uniform across all 3 studies. Similarly, the pre-post decrease of desire for something salty very consistently enlarges from 3 to 30 trials. The only deviation from this overall consistent pattern is the significant pre-post decrease of desire to eat something in study 1, compared to non-significant pre-post differences in studies 2 and 3.
As already discussed in study 1, and confirmed by the subsequent studies, it is striking and unexpected that the stimuli-unrelated eating desires (i.e., salty and fatty) are uniformly decreasing. There appears to be a dose-response relationship, with larger decreases after 30 (vs. 3) trials. This is unexpected, as some prior sensory-specific literature suggests that not-consumed food items or categories remain relatively unaffected (B. Rolls, Rolls, et al., 1981).
However, more recent research may help to make sense of this apparent discrepancy. Griffioen-Roose et al. found that carefully manipulating the taste of either a relatively large preload (10% of daily energy intake; 2010) or an entire day's diet (2012), sweet and savoury taste was not equipotent in their effects on sensory-specific eating desires. Both tastes equally decreased hunger and general desire to eat. However, after savoury food consumption, there was a marked relative preference for sweet food. In contrast, after sweet food consumption, sweet and savoury were equally preferred. In other words, savoury taste's satiation is more specific, while sweet taste's satiation is more general. This conjecture is reflected by the present result where eating desires uniformly after 30 sweet food imagined eating trials.

Ecological validity
Similar to prior studies, the lack of ecological validity is the primary limitation of the present work. It is possible to criticize the design based on the fact that stimuli were presented in a manner that does not at all resemble common online platforms, e.g., Instagram. However, the studies aimed at answering theoretical and fundamental questions, rather than an investigation in situ.
Another factor of ecological validity is limited stimulus variety. For instance, Instagram may be likened to a buffet, offering an abundant variety of food types, tastes, and flavours (Andersen et al., 2021). In the studies at hand, however, participants saw only a single food type, with some variation in colour and flavour. Prior mental imagery studies (Missbach et al., 2014;Morewedge et al., 2010) repeated the very same food (e.g., imagining eating an M&M's). Larson et al. (2014) showed food pictures of the same taste (e.g., only salty food), but various types and flavours.

Generalizability
The studies' findings are also limited by questions of generalizability. Despite our large sample sizes, it is conceivable that online recruitment off Prolific may not be representative of the general public. The studies employed also only sweet-tasting stimuli. Hence, it does not provide evidence of stimuli of other tastes' effects. However, other studies have found analogous results with savoury stimuli (Larson et al., 2014;Missbach et al., 2014).
Similar to procedures in prior literature, participants saw explicit cues to "vividly image" eating the depicted food. It is currently unknown how well the findings generalize to situations where people look at food pictures in the absence of explicit cuesas they would in realistic settings, e.g., when browsing through Instagram food images.
Generalizability may also have been limited by the imagined eating outcome measure. It can not a priori be taken for granted that changes in food liking or wanting in an imagined scenario are equivalent to sensation changes during actual consumption. Nevertheless, analogous research using actual eating outcomes saw similar results, suggesting that findings would generalize after all (Larson et al., 2014;Missbach et al., 2014;Morewedge et al., 2010).

Potential confounders
The current studies investigated responses only after 3 and 30 trials. Assuming non-linearity (Groves & Thompson, 1970), they do not provide insight into the dose-response relationship beyond the specifically investigated number of trials. For example, 30 trials may or may not have maximized satiation. Estimating a non-linear dose-response curve would require at least a third measurement point. Moreover, the dose-response relationship may interact with stimulus variety. Increased stimulus variety may explain why Larson et al. (2014) saw a difference between 20 and 60 trials while Missbach et al. (2014) did not find any between 18 and 36 trials. In the current study design, the estimation of thispossibly more subtleinteraction was not possible.
The experimental groups were not equated for the cognitive workload. This design decision leaves open the possibility that the findings, e. g., decrease in eating desires, were due to generic boredom or cognitive load, rather than sensory-specific satietyespecially since the outcomes after 30 trials are relatively homogenous. However, the literature contains studies with (Missbach et al., 2014;Morewedge et al., 2010) and without workload-equated designs (Larson et al., 2014) that both yield similar results.

Practical significance
The reduction in desired number of food pieces across the studies is very small, e.g., with a reduction of less than two M&M's. This is less than the reduction reported by Morewedge et al. (2010), with an ad-libitum reduction after 30 repetitions, compared to 3 repetitions, of between 2 and 4 M&M's, depending on the specific experiment. Thus, it is possible that the employed explicit measure of desired number of food pieces underestimates actual consumption. The practical significance of a reduction even of 4 M&M's, equivalent of 15 kcal, is unclear.

Future research
Future research may want to more conclusively determine the cause for the lack of differentiating results between the three studies. As discussed, there are two main possible explanations. First, the stimuli may not have led to any priming-differences, possibly due to insufficiently divergent neurological activation (Groves & Thompson, 1970). As stimulus similarity per se is unlikely to be the cause, participants' perception and mental simulation ability remain possible factors to inspect. For example, manipulating the salience of stimuli's flavour or colour differences (e.g., with explicit labels, cf. Redden, 2008) should make it possible to determine whether the lack of an effect in studies 2 (vs. 1) and 3 (vs. 2) was due to either of these factors.
The second possible explanation is that, despite priming-differences, participants fully satiated after 30 trials. Put another way, any varietyrelated priming, particularly in study 3, may have been overcome by overwhelming stimulus-repetitiveness. However, in the ecological case the opposite may be possible, too; namely, the overwhelming stimulivariety overcoming any satiation from repeated exposure. It, therefore, appears relevant to systematically explore different combinations of stimulus variety and number of trials.
Future research may also seek to increase the ecological validity of the findings. There are at least two ways. The first is to increase food image variety, akin to how Missbach et al. (2014) extended Morewedge et al.'s (2010) findings to other food items. Similarly, the work at hand on sweet food stimuli complements Larson et al.'s (2014) on salty food. However, no study to date has conducted a complete taste cross-over study with both sweet and salty foods to test this conjecture in the context of mental simulations.
The second way to increase ecological validity is to investigate the extent to which satiation-relevant mental simulations occur spontaneously, i.e., in the absence of explicit cues. Neuroimaging evidence suggests eating-related mental simulations occur spontaneously, i.e., by the mere look at food pictures (Simmons et al., 2005). However, in one of Larson et al.'s (2014) studies, evaluating food image brightness (vs. taste) did not lead to satiation. This implies that goals and attention significantly affect cognitive processing (Custers & Aarts, 2005;Gibson, 1979). Social media platforms are, of course, not without implicit cues or incentive structures. Hence, it remains an open question whether users' appetites would spontaneously be primed or satiated.
In conclusion, the present work accomplished its three aims. First, it successfully replicated the prior finding that imagined eating can lead to satiationor at least explicit markers hereof. This finding was highly reproducible across three studies. Second, these studies revealed this effect on satiation to be sensory-specific and dose-dependent to the exposed stimuli. Excitingly, satiation responses can increase above (i.e., become primed) or decrease even below the baseline as a function of stimulus repetition. Third, and contrary to hypotheses, there was no evident effect of colour or flavour variety within the same food type. It would appear that more pronounced differences in food image stimuli are required to impede the imagery-induced satiation effect.

Funding
This research project (PN: 32764) was supported by the Graduate School of Technical Sciences, Aarhus University (GSTS), the Sino-Danish Center of Education and Research (SDC), and Aarhus University's Center for Innovative Food Research (CiFOOD). Funding sources were not involved in any part of the research or publication process.

Ethics statement
The research was conducted following the ethical principles stated in the Declaration of Helsinki. Participants gave informed consent before taking part and all data was collected anonymously. The non-invasive nature of the online studies did not require formal approval by Aarhus University's institutional review board.

Declaration of competing interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.