Sequential dependency for affective appraisal of food images

Our perception is shaped not only by current experiences but also by recent ones. This study explores how this sequential effect influences emotional responses to food images, a critical element in food marketing and online food services. Attractive food images can positively influence purchasing decisions, while aversive images have the opposite effect. The rise of online-to-offline (O2O) food delivery services and digital menu boards has made food imagery even more crucial. People also share food images extensively online, further highlighting the importance of understanding visual responses to food. While the emotional response to food images is well-established, the role of prior image exposure in shaping these responses is largely unknown. Emotional responses have a temporal component, not disappearing immediately after stimulus removal, implying that our emotional state to a given food image is likely a composite of recent experiences. Perception research indicates that our perception of a given stimulus depends on previous stimuli. This serial dependency can be positive (assimilation) or negative (repulsion). While such dependencies have been demonstrated for low-level visual features, their presence for amodal stimuli like emotions remains under-investigated. According to the circumplex model of affect, emotions have two primary dimensions: valence (pleasantness) and arousal (intensity). This exploratory study investigates whether sequential effects impact the valence and arousal ratings of food images, using a large, cross-cultural dataset of participants from 16 countries.

Previous research highlights the impact of food imagery on consumer behavior, with images influencing attitudes and expectations. The growing online food delivery market underscores the importance of visual appeal. While the emotional response to food images is acknowledged, the influence of preceding images is less understood. Studies on sequential dependencies in visual perception have focused on low-level features (color, orientation, motion), with some research showing effects for more complex stimuli like face attractiveness. These dependencies can be either positive (percept biased towards previous stimulus) or negative (biased away). The sign might reflect different processing levels, with negative dependencies reflecting low-level perception and positive dependencies reflecting higher-order processes. Studies on serial dependencies in emotion perception are scarce, with one study suggesting a positive dependence for facial expression judgments, possibly due to a visual or emotional aftereffect. This study aims to examine whether sequential effects affect the emotional appraisal of food images, considering the dimensions of valence and arousal.

This study re-analyzed data from 1278 participants (795 females; mean age 29.8 years; mean BMI 23.6) across 16 countries. Data from 44 participants were excluded due to age, height, or weight inconsistencies. Participants were recruited via social media, email, and online databases (Prolific, Crowdworks). The experiment used Gorilla software and a stimulus set of 60 food images (50 from CROCUFID, 10 from FRIDa). The images varied widely in valence, as determined in previous studies. Valence and arousal were measured using the EmojiGrid, a validated tool with emoji representing various emotional states. Participants viewed the 60 images in a randomized order, rating each using the EmojiGrid. The EmojiGrid is a 2D grid labeled with emojis expressing various facial expressions that reflect valence (horizontal axis, ranging from unpleasant to pleasant) and arousal (vertical axis, ranging from low to high). The EmojiGrid has been validated in previous studies and was proven effective in assessing food-evoked emotions. Each trial involved displaying a food image and recording the participant's emotional response using the EmojiGrid. The experiment lasted about 10 minutes on average. Instructions were given in English or their respective languages. For data analysis, an inter-trial analysis was conducted to investigate how valence and arousal ratings on a given trial were influenced by the ratings on preceding trials (up to four trials back). For each participant, a median valence and arousal rating was determined. Trials were categorized as having high or low valence/arousal ratings relative to their median. Then mean valence/arousal for each category was calculated, and differences indicated serial dependency. Repeated measures ANOVA examined the valence inter-trial effect across trial distances (1-4). One-sample t-tests assessed whether each inter-trial effect differed from zero. A similar analysis was performed for arousal. A further analysis was performed for each food image separately, again looking at the trial immediately preceding. Independent t-tests compared valence and arousal ratings between trials preceded by high vs. low ratings. Finally, ANCOVA examined the influence of gender, age, BMI, and nationality on the valence and arousal serial dependencies. BMI was calculated using standard methods (weight/height²).

The analysis revealed a significant positive serial dependence for both valence and arousal ratings of food images. For valence, the rating on trial *t* was significantly higher when the previous trial (*t*-1) had a high valence rating. A negative (repulsive) dependence was observed for the trial four steps back. For arousal, a positive dependence was observed up to three previous trials. The ANOVA results revealed a significant effect of trial distance on both valence and arousal inter-trial effects, confirming that the effect varied based on how many trials back we are looking at. The inter-trial effects were statistically significant for trial *t*-1 for both valence and arousal ratings. The results for arousal indicated a positive dependency for up to three trials back (*t*-1, *t*-2, *t*-3), while for valence, the dependency was positive only for the immediately preceding trial (*t*-1) and negative for the fourth trial back (*t*-4). Individual image analysis revealed that the majority of images exhibited a significant positive serial dependence for both valence and arousal, although not all showed significance, potentially because each image was shown only once per participant. For valence, the inter-trial effects were strongest for images rated as neutral and weaker at the extremes (high or low valence); for arousal, the effect was more consistent. ANCOVA analysis showed a significant effect of gender on arousal, with larger inter-trial effects for males. A trend towards significance was observed for gender's effect on valence. Age, BMI, and nationality did not significantly moderate the inter-trial effects. A subsequent analysis exploring gender's effect showed a significant interaction between previous rating and gender for both valence and arousal. Males showed larger inter-trial effects than females for both.

This study provides the first evidence of sequential effects on affective appraisal of food images, showing that prior images influence subsequent emotional responses. The positive serial dependence observed suggests a merging of emotional responses across trials, which might reflect residual emotional activity influencing the appraisal of subsequent stimuli. The longer-range effect for arousal compared to valence is intriguing and requires further investigation. The results suggest that sequential effects are not limited to low-level visual features but extend to amodal stimuli like emotions. The lack of nationality effect suggests a robust and generalized phenomenon. The significant gender effect is noteworthy. Potential explanations include differences in brain responses to food images (amygdala, insula activation) and memory consolidation processes for emotions. This may reflect response bias in males compared to females. While the magnitude of the sequential effect may seem small, it is potentially appreciable considering the response range. The study supports the hypothesis that recent emotional experiences shape current food image evaluations.

This study demonstrates that the affective appraisal of food images is influenced by previously viewed images, highlighting a significant sequential dependency. This finding has implications for the design of online food platforms and menus. Future research could explore different task demands to clarify whether an explicit emotional task is needed to observe the effect, investigate the neural mechanisms underlying the sequential dependency, and test the generalizability of this effect to other types of stimuli and contexts. The relatively small magnitude of the effect warrants further investigation to assess its practical implications.

The study relied on self-reported emotional ratings, which can be subjective. The use of a relatively small number of images per participant might have introduced noise. While the sample size was large and diverse, the use of convenience sampling might limit the generalizability of the findings to certain demographics. Further investigation is needed to determine whether this effect is limited to the emotional assessment of food stimuli.