You eye what you eat: BMI, consumption patterns, and dieting status predict temporal attentional bias to food-associated images

.


Introduction
The past few decades have seen a rise in the availability of high fat high sugar (HFHS) foods.Although convenient and often delicious, overconsumption of HFHS foods has been linked with obesity (Drewnowski & Popkin, 1997;Popkin et al., 2012), metabolic disorders (Eckel et al., 2005), cardiovascular disease (Haslam & James, 2005), cognitive deficits (Abbott et al., 2019), and certain types of cancers (Steele et al., 2017).The negative consequences of a HFHS diet on our health are widely known, but many people continue to consume foods high in refined sugar and saturated fat content, making lack of knowledge about these foods an unlikely mechanism to solely explain the overconsumption.Instead, understanding the cognitive biases individuals have toward HFHS foods may be key to uncovering the mechanisms underlying the persistence of HFHS consumption despite adverse consequences.
Modern theories about overconsumption of HFHS foods consider cognitive processes such as memory, attention, and inhibitory control to be key factors that drive consumption.Two such accounts explain why refraining from eating HFHS foods can be especially difficult.One account considers HFHS food intake to be a "vicious cycle" that impairs cognitive functioning, which in turn leads to further overconsumption (e.g., Kanoski & Davidson, 2011).According to this model, HFHS foods impair hippocampal dependent functions and cause dysregulations in feeding behaviours.Evidence from animal studies that experimentally manipulate dietary content support a key step in the vicious cycle model by demonstrating that HFHS foods can impair hippocampal dependent learning and memory (for a review, see Abbott et al., 2019).These diet-induced cognitive impairments can occur rapidly, over a period of days (Kanoski & Davidson, 2011).However, the deficits that emerge over the initial days and weeks are highly specific.For example, cognitive impairments from short-term HFHS diet affect some forms of spatial learning and memory but not others (e.g., Tran & Westbrook, 2015), depend on whether the memory was learned prior to the diet exposure (e.g., Tran & Westbrook, 2017), and leave intact object recognition memory (e.g., Tran & Westbrook, 2018).Although the vicious cycle model offers a promising framework for understanding overconsumption, further elaboration is required to explain the mechanisms through which such selective cognitive impairments (e.g., in spatial memory) during early exposure are able to sustain overconsumption and develop into widespread dysfunction of neural circuits that control feeding.
The second account proposed to explain overconsumption models food intake as a form of addiction (Berridge, 2009;Volkow, Wang, Fowler, Tomasi, & Baler, 2012).According to these models, HFHS foods act as reinforcers similar to drugs of abuse, whereby repeated exposure upregulates reward and motivation circuits while downregulating cognitive control circuits.The upregulation of motivational drive and downregulation of inhibitory control acts as a dual process that maintains consumption despite negative consequences (Volkow et al., 2012).Moreover, a consequence of having overactivated reward and motivational circuits is that individuals will develop a heightened sensitivity to food rewards, such that food stimuli will increasingly attract attention (Berridge, 2009); often termed "incentive salience" (Robinson & Berridge, 1993).A critical feature of these addiction models is that cues frequently associated with the reinforcer can themselves come to elicit psychological and physiological properties of the reinforcer, which drives craving even when the cues are encountered in the absence of the reinforcers themselves (Berridge, 2009;Volkow et al., 2012).
While the vicious cycle model can explain how overconsumption is maintained over extended periods, the addiction models offer a mechanism for overconsumption during early exposure periods through the development of incentive salience.Of course, it is possible that both processes play a role, with overconsumption driven in the short-and long-term by changes to motivation and attention, while overconsumption is compounded in the long-term by impairments to neural circuits that control feeding behaviours.Therefore, a critical step in understanding overconsumption is understanding the motivational and attentional priorities that HFHS foods elicit on individuals.Based on the ideas of incentive salience, individuals who overconsume HFHS foods should have greater attentional biases to images of food and foodassociated cues (i.e., environmental stimuli that commonly accompany foods when they are consumed such as brand logos and food packaging).However, previous research has produced equivocal findings on the link between BMI and attentional bias for food.For example, some studies have found that a higher BMI is associated with greater attentional biases to food-related cues (Castellanos et al., 2009;Nijs et al., 2010;Werthmann et al., 2011;Woltering et al., 2021); some have found that BMI is unrelated to attentional bias for food (Arumäe et al., 2019;Liu et al., 2022;Stamataki et al., 2019); some have even found that BMI predicts less attentional bias for food (Flack et al., 2022;Nummenmaa et al., 2011).Correspondingly, recent meta-analyses have failed to find an association between BMI and food-related attentional biases (Hagan et al., 2020;Hardman et al., 2021).
Mixed findings about the relationship between attentional biases towards food and overconsumption or obesity do not necessarily mean they are unrelated; instead, it may suggest that food-related attentional biases are moderated by factors that can obscure relationships with BMI.For instance, Castellanos et al. (2009) found that relative to healthy weight individuals, obese individuals had a larger gaze direction bias for food images only when sated; when fasted, these groups did not differ in attentional biases for food.More broadly, participants are more likely to exhibit attentional biases for food stimuli when hungry (Hardman et al., 2021), particularly if hunger is manipulated by having participants fast before the experiment (Mogg et al., 1998;Nijs et al., 2010;Piech et al., 2010;Stamataki et al., 2019).These findings have led theorists to propose that an attentional bias to food is not a stable characteristic of individuals, but a dynamic state that fluctuates over time in response to momentary evaluations of food-related cues (Field et al., 2016).Thus, in addition to being influenced by learning history (i.e., prior consumption patterns), attentional biases for food are also argued to be influenced by various state-related qualities such as hunger, craving, and motivation from competing goals such as dieting.
If attentional biases toward food are indicative of evaluative processes and/or incentive salience related to food, then we may expect to see attentional biases for food that an individual prefers or for foods high in fat/sugar since they are generally perceived as more appetising.Studies have found that people are quicker to respond to HFHS foods relative to foods low in fat and sugar (Harrar et al., 2011) and are more likely to initially orient to high-calorie foods over low calorie foods (Castellanos et al., 2009).Moreover, after eating a given food to satiety, the degree to which that food is perceived as less pleasant predicts the degree to which images of that food will no longer attract attention (di Pellegrino et al., 2011).However, several studies have also failed to find attentional preferences for high calorie/appetising foods relative to low calorie/appetising foods (Ballestero-Arnau et al., 2021;Nummenmaa et al., 2011), and food can attract attention independent of people's explicit liking for them (Arumäe et al., 2019;Liu et al., 2022).Thus, whether attentional biases for food vary as a function of how appetising they are remains unclear.
When considering discrepant findings in the literature, it is worth noting that the choice of experimental paradigm could be crucial.For example, Dondzilo et al. (2022) were unable to find attentional biases for high-calorie food over low-calorie food in a standard dot-probe task, but using a novel 'chase the food' task they demonstrated that restrained eaters exhibited faster attentional switching towards high-calorie foods an important finding given the lack of clarity on the link between eating restraint and food-related attentional bias (Watson & Le Pelley, 2021).More broadly, the emotional Stroop and dot-probe task are some of the most commonly used tasks for assessing attentional biases for food (Hagan et al., 2020;Hardman et al., 2021); yet these tasks have poor internal consistency and/or test-retest reliability (Ataya et al., 2011;Chapman et al., 2019;Rodebaugh et al., 2016;Schmukle, 2005).This highlights the need for robust attentional paradigms to reconcile conflicting findings.
In the current experiment, we tested the model that food can acquire incentive salience using a modified emotion-induced blindness task (e. g., Most et al., 2005).The basic premise of this design is that it can indicate an attentional priority in the way that individuals regard highly salient images by observing the degree to which these images involuntarily distract attention.Such effects have been demonstrated with similar tasks using perceptually salient (e.g., Asplund et al., 2010) and emotionally arousing images (e.g., Ciesielski et al., 2010) as distractors, whereby salient images disrupt processing of stimuli presented shortly after.For example, participants are less accurate at judging the orientation of a rotated target image in a stream of rapidly presented (e.g., 100ms) images if it is preceded by an emotionally arousing distractor, such as an image of a snake, compared with a neutral image (Most et al., 2005).Accuracy is most impaired when the gap between the distractor and the target is small, (e.g., 100-200ms; Kennedy & Most, 2015) and unimpaired at longer gaps (e.g., 800ms; Ciesielski et al., 2010), suggesting a fast-acting, transient, impact.Although distractors are task-irrelevant, participants display a limited ability to counteract their influence on target detection (Le Pelley et al., 2017;Most et al., 2007); as such, the paradigm permits the examination of stimulus-driven effects on attention.Importantly, the emotion-induced blindness paradigm has superior test-retest reliability to other common emotion-attention tasks, with Onie and Most (2017) reporting intraclass correlation coefficients (ICC) ranging from 0.33 to 0.91.For comparison, reported reliabilities for the dot-probe task, for example, included test-retest ICC values that range from -.48 to 0.38 (Rodebaugh et al., 2016) and -.12 to 0.44 (Jones et al., 2018), and split-half reliabilities ranging from -.29 to 0.59 (Chapman et al., 2019, Table 1); in either case, it is not uncommon for the assessed reliability index of the dot-probe task to be negative or approximate zero (Chapman et al., 2019;Rodebaugh et al., 2016).Modified versions of the emotion-induced blindness paradigm have been used previously to demonstrate the effect of hunger and sensory-specific satiety on food-related attentional biases (Davidson et al., 2018;Neimeijer et al., 2013;Piech et al., 2010).Thus, this paradigmwhich we label a "rotation detection task" for the purposes of our studyis a strong candidate for assessing the relationship between food-related attentional biases and diet-related variables.
To our knowledge, two previous studies have used a similar task to examine the relationship between BMI and food-related attentional bias (Arumäe et al., 2019;Kirsten et al., 2019).Arumäe et al. (2019) found no relationship between BMI and attentional biases to food-related stimuli, although a small sample size may have contributed to this null result (N = 39).Kirsten et al. (2019) did not explicitly analyse the relationship between performance and BMI although data they recently provided for a meta-analysis appears to indicate no relationship (Hardman et al., 2021).Nonetheless, the task intermixed various trial types, such as trials in which food stimuli was a target and not a distractor.This may have caused distraction from food stimuli that was not purely stimulus-driven if participants were expecting to identify food.The identification task for the target stimulus also differed: participants were required to report the category type of a blue-framed image, rather than the orientation of a rotated image.Moreover, Kirsten et al. (2019) did not replicate the common finding that attentional distraction is stronger at short lags (Ciesielski et al., 2010;Most et al., 2005).Thus, while previous research using a similar design indicates that the rotation detection task may reveal no relationship between BMI and food-related stimuli, low power or design demands in those previous studies may have obscured a possible relationship.
Food stimuli in the world come in many different forms.Although the sight and smell of food itself serves as cues for food availability, associations may also develop to the food brand's logo and packaging, which may themselves serve as conditioned cues (Spence & Velasco, 2018).For example, people associate different colours with certain flavours of potato chips due to colour-flavour packaging conventions used by companies selling the food products (Piqueras-Fiszman & Spence, 2011); these associations facilitate target detection in a visual search task (Velasco et al., 2015).Nevertheless, there is limited research examining whether weight, hunger, food preferences and other diet-related variables are predictive of attention given to food logos.Within the animal literature, organisms differ in whether they prefer interacting with cues associated with a reward (e.g., a lever predicting food delivery), or the reward itself; that is, whether they are sign-trackers or goal-trackers, respectively (Boakes, 1977;Colaizzi et al., 2020;Hearst & Jenkins, 1974).Sign trackers often exhibit maladaptive response patterns that resemble addictive behaviours in humans, such as persisting with a conditioned response after the reward is omitted (Ahrens et al., 2016;Colaizzi et al., 2020; although see Robinson et al., 2014, for an account of goal-trackers' susceptibility to addiction).There is evidence to suggest that similar learning phenotypes exist in humans, such that sign-trackers exhibit increased trait impulsivity (Garofalo & di Pellegrino, 2015), risky alcohol use (Albertella et al., 2019), and are less likely to be restrained eaters (Watson et al., 2021).Examining the attentional priority systems elicited by natural food versus food-associated cues for individuals of varying weight, hunger, and food preference statuses could further determine whether similar learning phenotypes exist in humans.Indeed, many theories of attentional biases in obesity and addiction propose that cues co-occurring with the reinforcer can elicit a response similar to the reinforcer itself via associative learning (Berridge, 2009;Volkow et al., 2012).
The aim of this study was to examine the factors that contribute to greater attention to HFHS foods.Specifically, if body mass index, dieting, hunger, consumption of HFHS foods and a preference for HFHS foods predict attention to food stimuli.We approached this aim by determining whether diet-related individual differences predicted attentional priority for food images and food logo images.This was done by including trials in which the distractor was an image of HFHS food itself, and then including trials in which the distractor was an image of logo branding related to HFHS food, such as the logo of a company that sells HFHS or the HFHS food itself wrapped in its branded packaging.
To determine how food and food logos are prioritised in attention, we had a large sample of healthy undergraduate students complete a rotation detection task that was modified from previous emotioninduced blindness studies (e.g., Most et al., 2005).Participants viewed a stream of rapidly presented (100ms) images containing landscape and architecture scenery and were required to report the orientation of a rotated target image.On some trials, task-irrelevant distractor images depicting food or food logos preceded a task-relevant target image by 200ms or 400ms.We anticipated that food and food logo distractors would impair target accuracy relative to trials with no distractors.We then determined whether the degree of food-related distraction was predicted by various diet-related variables.Indeed, previous research using similar designs indicate that participant-relevant stimuli can distract differently based on individual differences (e.g., individuals with eating disorder symptomatology are more distracted by certain body types, Berrisford- Thompson et al., 2021; individuals with post-traumatic stress disorder are more distracted by war-related imagery, Olatunji et al., 2013).
To ensure any associations with individual differences were specific to food stimuli, we carefully chose stimuli classes to compare performance of different food distractor types.For food stimuli, we compared performance with trials that included positively valenced distractors unrelated to food (e.g., a cute animal or a happy couple).Such stimuli should distract attention (e.g., Kennedy et al., 2020), but the degree of distraction should not be predicted by diet-related variables.For comparisons with food logo stimuli, we compared performance with trials that included distractor images depicting household product logos, since these would be perceptually similar to food logos but have no association with food.To further test the specificity of any associations between diet-related individual differences and attentional bias for food, we also included a self-report measure of attentional control (Derryberry & Reed, 2002).This was done to test if attentional biases to food were associated with a general deficit in attentional control rather than being related to individual differences specific to food.

Participants
322 undergraduate students from UNSW Sydney participated in the study as part of their Perception and Cognition (PSYC2071) course.Data from ten participants were not included as these participants did not complete the entire task (six participants did not complete the rotation detection task and an additional four participants did not complete the questionnaires).Data from another six participants were excluded from analyses due to their low performance on the rotation detection task (overall performance accuracy below 55%, using the same criterion as previous studies, Kennedy et al., 2020).The final sample comprised of 306 participants.Eighty-three participants identified as male, 219 identified as female, and 4 participants did not report their gender.Participants were aged between 18 and 40 years old, with a mean age of 20.54 years (SD = 2.60 years).All participants gave informed consent, and the experiment was approved by the UNSW Sydney Human Research Ethics Advisory Panel (approval number: 2518).

Apparatus
Participants completed the experiment in a computer lab classroom at UNSW Sydney.Participants completed the experiment during their scheduled class, which was comprised of approximately 20-30 students; each participant was allocated their own desktop computer in the B.L. Kennedy et al. computer lab to complete the task, and all computers were set up similarly with similar hardware, including 24-inch LED monitors set at 60Hz.Computers were arranged in a U-shape along the perimeter of a room with participants facing the walls; participants sat down at any computer available when they entered the room.Head position was not fixed, and participants sat at a comfortable distance from the computer screen.Participants wore headphones during the rotation detection task to hear auditory stimuli and to limit noise exposure from the room.

Rotation detection task
The rotation detection task was presented via the Psychtoolbox extension for Matlab (Brainard, 1997;Pelli, 1997).Every trial included a rapid serial visual presentation (RSVP) of 17 images presented at a rate of 100ms per image, such that every image immediately replaced the next (see Fig. 1).The RSVP streams were comprised of 320 × 240 pixel coloured images that appeared in the centre of the computer screen and against a black background.Each RSVP stream included one distractor image, one target image, and 15 "filler" images.
Distractor images appeared at serial position 3, 4, 5, 6, or 7 in the 17 item RSVP stream.Distractor images depicted content representative of one of five categories: food distractors, positive distractors, food logo distractors (i.e., food-associated cues), product logo distractors (e.g., household cleaning items), or baseline "no distractors".There were 32 distractor images per distractor category except for the baseline "no distractors", which came from the same bank of images as the "filler" images (see below).Natural food distractors were images of high fat high sugar foods that depicted one of eight categories: fried chicken, hamburgers, hot dogs/sausages, pizza, donuts, cake, chocolate, or ice cream.Positive distractors were images of babies, baby animals, and happy couples.Food logo distractors were images of high fat high sugar food-associated cuessuch that it depicted either the logo of a food brand or the food wrapped in its branded packaging.These included images of signs for KFC, Pizza Hut, and Krispy Kreme, as well as images of wrapped Snickers bars and unopened Doritos chip bags.Product logo distractors contained images with packaging and brand words, but did not depict food, and thus served as a comparison condition.These images depicted brands such as Colgate toothpaste, Finish detergent, and Windex cleaner.Distractor images were sourced from the internet using Google Images.
Previous emotion-induced blindness studies typically include a "neutral" distractor type as control comparison for emotional distractors.There was no obvious neutral condition in our design, so we chose distractor images that were similar in their content but not foodrelated properties as comparison control stimuli.The positive distractors served as a control comparison for the natural food distractors, since both image types depicted something positive and approachable, but differed in content otherwise.Similarly, product logo distractors served as a control comparison for the food logo distractors, since both depicted logo content related to well-known brands.We carefully chose foodrelated stimuli and their control comparisons to represent content that was generally well-known and easy to identify in Australian culture, were images that clearly depicted the subject of interest in every image (so that they could be seen at a fast rate), and that were similarly diverse in content within an image category. 2 252 images were used as the "filler" images and "no distractor" images.These images depicted landscape or architecture scenery and were the same as those used in previous emotion-induced blindness studies (e. g., Most et al., 2005).Filler images and no distractor images were essentially the samethe reason that they served as a "no distractor" condition was because they were the same as all other items in the stream, but were presented in lieu of another distractor type, as a way to establish performance on the task without a distractor image.
128 images served as target images.Target images depicted the same type of content as the "filler" and "no distractor" images, however this content was rotated 90 • to the left or 90 • to the right.The same 64 images were used for the left-rotated and right-rotated images, and these images were the same as those used in previous emotion-induced blindness studies (Most et al., 2005).On trials with distractors, the target always appeared either two (lag 2) or four (lag 4) images after the distractor image.Specifically, target images could appear at serial position 5, 6, 7, 8, 9, 10, or 11.On trials with baseline "no distractors", the target also appeared in serial positions possible for other targets throughout the experiment, but since there was no specific distractor on these trials, there was no lag variable on these trials.
The task consisted of eight blocks of 40 trials (320 trials total).In the first four blocks, each block included 16 trials with a natural food distractor, 16 trials with a positive distractor, and 8 baseline trials with no Fig. 1.Schematic of Rotation Detection Task Note.On every trial, participants viewed a rapid serial visual presentation of images and reported the direction of the one rotated target picture in a stream of otherwise upright images.This image depicts part of a lag 2 food distractor trial and includes images that are similar to those used in the experiment for illustration purposes.
2 Due to the nature of our research questions, we prioritised the content of the images when choosing stimuli.Nevertheless, for full representation of our image sets, we compared the image statistics of the distractor image types using the Natural Image Statistical Toolbox (Bainbridge & Oliva, 2015;Torralba & Oliva, 2003), and compared, spatial frequency energy in 10%, 30%, 50%, 70%, 90% frequency bands, color (RGB) and proportion of non-white image space.Natural food and positive distractors differed in spatial frequency energyespecially at higher spatial frequencies (ps ≥ .001),and in some color properties (i.e., in green, p = .014,but not red or blue, ps ≥ .062),but did not differ in their proportion of non-white image space (p = .537).Food logo and product logo distractors also differed in spatial frequency energyespecially at higher spatial frequencies (ps ≥ .037),and in some color properties (i.e., green and blue, ps < .001,but not red, p = .318),but they did not they differ in their proportion of non-white image space (p = .075).In both cases, food-related stimuli had more high spatial frequency energy and greenness compared to their image comparisons (positive or cleaning logo images).Therefore, if spatial frequency or color drove our effects, we would expect to see the direction of impairment to be the same in both image-type comparisons.However, like previous emotion-induced blindness findings (see the use of scrambled images in e.g., Most et al., 2005; see also Goodhew & Edwards, 2022 for a review), these low-level image differences could not explain our results: even though food-related images had more high spatial frequency energy and greenness in both conditions, natural food distractors were less distracting than positive images, whereas food logo distractors were more distracting than product logo distractors (see Results).
B.L. Kennedy et al. distractor.In the second four blocks, each block included 16 trials with a food logo distractor, 16 trials with a product logo distractor, and 8 baseline trials with no distractor.Natural food distractors (first four blocks) were always presented before food logo distractors (second four blocks, i.e., distractor order was not counterbalanced) because we predicted that natural foods would generate a stronger attentional bias compared to food logos and wanted to test this effect first.Further, counterbalancing distractor order was not essential as the analyses conducted did not directly compare across natural food versus food logo distractors.Trials were otherwise presented in a pseudorandom order.On any given trial, images were chosen randomly from their respective pool of images with no replacement, such that all images of an image category were presented once, then the whole pool was shuffled and presented again in the same manner if needed.
Before starting the RSVP task, participants completed six practice trials to get used to the task.Practice trials started at a rate of 200ms per image and slowly sped up to the experiment speed.No distractors were included in the practice trials.Participants were instructed to press the left arrow key if the image was rotated left, and the right arrow key if the image was rotated right.When participants made their response, they heard a "ding" noise if they answered correctly, and nothing if they answered incorrectly.A blank screen was presented for 500ms before the program advanced to the next trial, which started automatically.

Questionnaires
Questionnaires were programmed in Qualtrics.

Dietary fat and free SugarShort questionnaire.
The diet questionnaire (Francis & Stevenson, 2013) assessed the dietary habits of participants based on how frequently they consumed common foods.Instructions to participants were to "Think about the food you've eaten over the past year.Remember breakfast, lunch, dinner, and eating out.
Please select the option that best describes how often you have consumed each of the following food or drink items."Items included foods high in fat or sugar such as pizza, bacon, soft drink, and chocolate.
Participants indicated their consumption habits on a five-point scale: less than 1 per month, 2-3 per month, 1-2 per week, 3-4 per week, and 5+ per week.

Attentional control scale.
In the attentional control scale (Derryberry & Reed, 2002), participants respondedon a four-point scale of almost never, sometimes, often, or alwaysto 20 statements about the ability to control attention in different situations.In general, the questions probed how well they are able to concentrate and ignore distraction in everyday life, such as when switching from one task to another, or paying attention to a task when there are distractions going on in the same room.

Food liking questionnaire.
In the third questionnaire, participants rated their liking for the types of food that they saw in the experiment.Participants completed eight questions, where they were shown all of the images for each of the eight natural food types from the rotation detection task (fried chicken, hamburgers, hot dogs/sausages, pizza, donuts, cake, chocolate, and ice cream), and were asked, "On a scale from 1 (not at all) to 7 (very much), how much do you enjoy eating the foods shown above." 2.2.3.4.General information questionnaire.Finally, participants were asked to report their sex, age, height, weight, hunger status, and diet status.Height and weight were used to calculate approximate body mass index (BMI).For hunger, participants were asked "On a scale from 1 (not at all) to 7 (very much), how hungry are you right now?" and "How many hours ago did you eat your last meal?"For diet status, participants were asked "Are you currently dieting or restricting consumption of food and/or drink?" and were asked to indicate yes or no.

Procedure
Participants were given general verbal instructions at the start of their lab class about how to start and open the experiment in Matlab.Participants viewed the consent form and indicated consent via a keypress.They then completed the task independently, were told not to speak to one another during the experiment, and were proctored by a lab tutor who was instructed about how to run the experiment.
Participants first completed the rotation detection task, at their own pace, and were automatically directed to the questionnaires once they completed the task.Participants then completed the questionnaires and waited quietly until the rest of the class completed the experiment.Participants were debriefed afterward and had the goals of the study explained to them through a course lesson.

Analytic plan and hypotheses
We used percentage accuracy in target identification as the dependent variable for attentional distraction from task-irrelevant distractors.We first examined how participants were distracted from food distractors compared to positive distractors using a 2 (distractor type: food vs positive) × 2 (lag: lag 2 vs lag 4) repeated measures ANOVA.Note that we did not include baseline "no distractor" trials in this initial analysis, since they did not contain distractor images and therefore could not be classified according to lag.We conducted a similar 2 (distractor type) × 2 (lag) ANOVA using food logo and product logo trials.
In order to determine the extent of attentional distraction from all stimulus types, we conducted a series of paired samples t-tests comparing each distractor × lag condition with baseline "no distractor" performance.When comparing baseline "no distractor" performance with food and positive distractors, we used mean accuracy for baseline trials across the first four blocks, since food and positive distractor trials only occurred in these blocks.Likewise, since food logo and product logo distractor trials occurred in the last four blocks, we used mean accuracy for baseline trials across the last four blocks for comparisons with these distractor types.Bonferroni's procedure was used to correct for four comparisons (four comparing food/positive distractors with baseline accuracy across the first four blocks, and four statistically independent comparisons involving food logo/product logo distractors and baseline accuracy across the last four blocks), such that statistical significance was evaluated at p = .013.
To determine how individual differences predicted attentional distraction, we conducted a series of simultaneous regressions using the diet-related individual difference measures as predictors.Attentional bias scores were calculated by subtracting mean accuracy in each distractor condition from mean accuracy on the corresponding baseline "no distractor" trials.Since we expected attentional distraction to be more robust on lag 2 trials, we created attentional bias scores for each distractor condition at lag 2. Bias score served as the dependent variable in a simultaneous regression using BMI, hunger, time since last meal, dieting status, dietary fat and sugar score, attentional control, and food liking score as predictors.While we had predictions for how the individual difference variables might relate to attentional distraction, we had no a priori expectations of the relative contributions from each predictor or how they relate to each other.We therefore took an exploratory stepwise approach that involved testing a full model with food bias regressed on all the predictors; testing a reduced model with food bias regressed on any significant predictors from the full model; and then testing the full and the same reduced model on attentional bias scores calculated using the control stimuli (positive distractors, and product logos).Note that the reduced model predicting food bias and control bias scores include the same predictors, but these predictors are selected from full model predicting food bias score.Using this datadriven stepwise approached allowed us to explore the relationship between attentional bias and the individual difference predictors (using the full model) while checking the specificity of any significant B.L. Kennedy et al. relationships to the food stimuli compared with the control stimuli (using the reduced model).
The analytic plan and hypotheses were specified prior to data collection.We expected all distractor types to lower identification accuracy relative to baseline accuracy, but we anticipated that this effect would be stronger at lag 2 than lag 4. We also expected diet-related individual differences to predict attentional bias scores for food and food logo distractors, but not for positive and product logo distractors, reflecting an attentional bias specific to food-related stimuli.

Descriptive statistics and correlations
Descriptive statistics for the questionnaire variables are depicted in Table 1.Notably, most participants (77.4%; n = 226) had BMIs in the healthy weight range (18.5-24.9); in comparison, only 13.0% (n = 38) of participants had a BMI in the overweight or obese range, and 9.6% (n = 28) had BMIs in the underweight range.Since our stimuli deviated from previous emotion-induced blindness tasks, we examined the reliability of our task by calculating the intraclass correlation coefficients for each of the four distractor stimulus types presented in the rotation detection task (food, positive, food logo, product logo) across the four blocks of trials and two lag conditions.Intraclass correlation coefficients ranged from 0.66 to 0.71, based on a mean rating (k = 8), 2-way mixedeffects model.
Table 2 depicts the bivariate correlations between the independent variables.BMI was positively related to time since last meal and negatively related to dieting status, and hunger was positively associated with time since last meal.Food liking scores were negatively associated with attentional control, but were positively associated with hunger as well as dietary fat and sugar scores.For all regression models, collinearity was not an issue (tolerance >.7, variance inflation factor <1.5 for all variables across all analysed models; see Table 2 for correlations between predictors) and inspection of residual plots (actual vs predicted) suggested homoscedasticity and linearity were not violated as there was no funnelling or curve pattern in the residuals.

Individual differences associated with attentional bias
Regressing food lag 2 attentional bias scores (no distractor accuracy − distractor accuracy) onto BMI, hunger, time since last meal, dieting status, dietary fat and sugar scores, attentional control, and food liking score did not yield a significant model, F (7, 284) = 1.46, p = .181,R 2 = 0.035.Despite this, two of the predictors in the model were significant, BMI and dietary fat and sugar scores (see Table 3).Specifically, after controlling for other variables, a higher BMI significantly predicted decreased attentional bias on food lag 2 trials, and increased dietary fat and sugar scores were associated with increased attentional bias for food lag 2 trials.A follow-up simultaneous regression was analysed using only these two variables as predictors, and food lag 2 attentional bias scores as the dependent variable.The overall model was significant, F (2,289) = 3.56, p = .030,R 2 = 0.024.BMI was still a significant predictor in the reduced model, although dietary fat and sugar scores failed to reach statistical significance (see Table 3).Collectively, these results suggest that individuals with a lower BMI and higher consumption of dietary fat and sugar are more likely to experience attentional bias from food.
The results for the regression with positive lag 2 attentional bias scores as the dependent variable suggested that the association of BMI and dietary fat and sugar scores with attentional bias was specific to food-related stimuli.The overall model was not significant, F (7,284) = 1.11, p = .358,R 2 = 0.027, nor were any of the predictors in the model (ts ≤ 1.66, ps ≥ .098).Additionally, regressing positive lag 2 attentional bias scores onto BMI and dietary fat and sugar scores (reduced model) did not yield a significant model, F (2,289) = 1.25, p = .289,R 2 = 0.009.

Attentional priority for food logo and product logo distractors
Fig. 3 depicts mean accuracy for distractor trials at lag 2 and lag 4 for food logo and product logo stimuli and baseline "no distractor" trials.
A paired samples t-test comparing accuracy on baseline "no

Individual differences associated with attentional bias
The simultaneous regression using food logo lag 2 attentional bias scores (no distractor accuracy − distractor accuracy) as a dependent variable was not significant, F (7,284) = 1.47, p = .177,R 2 = 0.035.Nonetheless, there were significant predictors in the model (see Table 4): after controlling for other variables, a higher BMI was Note.Food Lag 2 attentional bias scores were calculated by subtracting mean accuracy in the food lag 2 distractor condition from mean accuracy on the corresponding baseline trials.Note.Food Logo Lag 2 attentional bias scores were calculated by subtracting mean accuracy in the food logo lag 2 distractor condition from mean accuracy on the corresponding baseline trials.
B.L. Kennedy et al. associated with a reduced food logo lag 2 attentional bias score, and being on a diet was significantly associated with a greater food logo lag 2 attentional bias score.A follow-up regression was conducted using BMI and dieting status as predictors, and food logo lag 2 attentional bias scores as the dependent variable.The overall model was significant, F (2,289) = 3.54, p = .030,R 2 = 0.024.A higher BMI was associated with lowered attentional bias from food logo stimuli at lag 2, although this effect failed to reach statistical significance when controlling for dieting status (see Table 4).Dieting status was still a significant predictor after controlling for BMI, such that being on a diet was associated with increased food logo attentional bias at lag 2 relative to non-dieters.
Collectively, these results indicate that those with a lower BMI and those who are on a diet are more likely to have their attention biased toward food logos.
The regression using product logo lag 2 attentional bias scores was not significant, F (7,284) = 0.27, p = .964,R 2 = 0.007, nor were any of its predictors (ts ≤ 1.26, ps ≥ .210);and regressing product logo lag 2 attentional bias scores onto BMI and dieting status (reduced model) did not yield a significant model, F (2,289) = 0.81, p = .446,R 2 = 0.006.Thus, the associations of BMI and dieting status with attentional bias are specific to food-related brand imagery presented 200ms before target presentation.

Exploratory post-hoc analysis
Dietary fat and sugar scores predicted increased attentional bias for food stimuli only, and dieting status predicted increased attentional bias for food logo stimuli only.These results suggest that these variables uniquely predict whether a person's attention is biased more by food stimuli or food logo stimuli.To test this, we created a new variable by subtracting food logo lag 2 attentional bias scores from food lag 2 attentional bias scores, such that higher scores indicate greater attentional bias for food stimuli relative to food logo stimuli.This variable was regressed onto dietary fat and sugar scores, dieting status, and attentional control (included as a control variable), yielding a significant model, F (3,300) = 3.28, p = .021,R 2 = 0.032.Dietary fat and sugar scores predicted increased attentional bias to food stimuli relative to food logo stimuli, b = 0.12, SE b = 0.06, t = 2.15, p = .032,and being on a diet predicted increased attentional bias for food logo stimuli over food stimuli, b = 4.42, SE b = 2.06, t = 2.15, p = .033.Attentional control score was not a significant predictor, b = 0.05, SE b = 0.10 t = 0.52, p = .602.This suggests that people can differ in whether their attention is biased more food or food logo stimuli, and that consumption of HFHS foods and diet may discriminate between these individual differences.

Discussion
The aim of this study was to examine the relationship between a range of diet variables and attentional bias for food and food-associated stimuli.We did so by having a large sample of healthy participants complete an attentional task that is robust for measuring attention directed toward different stimuli classes.In line with previous findings, distractors of all stimuli classes distracted participants more at lag 2 than lag 4 (Ciesielski et al., 2010;Most et al., 2005).A higher BMI was associated with weaker attentional biases for both food and food logo stimuli.Increased consumption of HFHS foods predicted stronger attentional bias for food stimuli, and being on a diet was associated with increased attentional bias for food logo stimuli.Notably, these associations were specific to food-related stimuli as none of the diet-related variables were associated with attentional biases for control distractors.Moreover, attentional control scores did not predict attentional biases for food-related stimuli, suggesting that biases for food are related to diet-specific individual differences rather than a general attentional control deficit.

The attentional power of food logo stimuli
Although it is well recognised that cues associated with a reinforcer can attract attention through associative processes (Berridge, 2009;Field et al., 2016;Hofmann & Van Dillen, 2012;Kavanagh et al., 2005), previous research has largely not considered the ability of food-associated cues to attract attention beyond the sight of food itself.A notable exception is Maloney et al. (2019): using a dot-probe task with eye-tracking, they found that frequent consumers of low-calorie sweetened beverages exhibited an attentional bias towards these stimuli.Since comparison stimuli included sweetened beverages, differences in attentional bias can be attributable to the branding differences associated with low-calorie sweetened beverages.We extend these results by demonstrating that food logo stimuli elicit attentional biases in the temporal domain, and that the extent of this attentional bias appears to be related to diet-related variables.
The need to consider the influence of food logo stimuli on attention is particularly important given the prevalence of HFHS food advertising.For example, 1 in 4 food advertisements on Australian television are for 'fast food' companies (Roberts et al., 2013).Research on the effect of HFHS advertising on adults is less common; however, in adolescents, striatal responses to HFHS food advertising predict later increases in BMI (Yokum, Gearhardt, Harris, Brownell, & Stice, 2014), and exposure to HFHS advertising increases acute food consumption in children (Boyland et al., 2016).Food logo, branding, and packaging stimuli feature prominently in advertising, and people associate the perceptual qualities of these stimuli with the taste and pleasantness of food (Lemos et al., 2020;Velasco et al., 2015).Consequently, understanding how these stimuli impact attention, and determining the implications of this relationship for HFHS food overconsumption, is pertinent.Our study represents an important first step, by demonstrating that food logo stimuli bias temporal attention in a manner similar to emotionally arousing imagery (Most et al., 2005).
Interestingly, an attentional bias for food logo stimuli was not associated with the same set of diet-related variables as an attentional bias for food stimuli, suggesting that individuals may differ in whether they attend primarily to food or food logo stimuli.This was supported by our exploratory post-hoc analysis, which found that dietary fat and sugar consumption and dieting status predicted participants' attentional bias for food versus food logo stimuli.Specifically, a person who does not diet and regularly consume HFHS food is more likely to exhibit attentional priority for HFHS food; whereas someone who is on a diet and does not regularly consume HFHS food is more likely to exhibit attentional priority for HFHS food logo imagery.This finding could mirror the distinction between goal-tracking and sign-tracking in the animal literature (Boakes, 1977;Hearst & Jenkins, 1974), although additional research should explore how food and food-associated cues are similar and diverge in the way they influence behaviour.

BMI and food-related attentional biases
The negative association between BMI and attentional bias for food and food logo stimuli is somewhat unexpected given that BMI is putatively linked to increased attentional bias for food (Nijs & Franken, 2012; although see Hardman et al., 2021), not less.Most measures of food-related attentional bias primarily measure spatial attention, such as the dot-probe task, whereas the design that we used relies on a temporal measure (Hardman et al., 2021;McHugo et al., 2013).There is evidence that food-related attentional biases may occur only in select components of attention, such as attentional switching rather than maintenance (Dondzilo et al., 2022), or early rather than late attention (e.g., Nijs & Franken, 2012; although see Hardman et al., 2021).Thus, our results may reflect a component of the attentional system that is not well measured by previous tasks and exhibits an attenuated food-related bias as BMI increases.Nonetheless, it is important to note that our sample had a narrow range of BMIs, with only a minority (13%) of individuals B.L. Kennedy et al. being overweight or obese.Studies which have found a negative association between BMI and attentional bias have also sampled narrow BMI ranges, using predominantly healthy weight (Nummenmaa et al., 2011) or overweight/obese participants (Flack et al., 2022;Gearhardt et al., 2012).Future research could use similar measures with a larger range of BMIs, to determine if our findings are in fact indicative of a unique attentional component.

High fat high sugar food intake and attentional biases
Those who reported eating more HFHS food in the past year had a greater attentional bias to food stimuli.Several theories suggest that food attracts attention in part because of the learning history an individual has with food (Berridge, 2009;Field et al., 2016;Hofmann & Van Dillen, 2012;Kavanagh et al., 2005).Although BMI is often used as a proxy measure for an individual's learning history with food, our results did not support this equivalence: BMI and past consumption of HFHS food both predicted attentional bias for food, but in opposite directions.However, having participants self-report their intake of HFHS food over the past year introduces the possibility of recall biasfor instance, those with greater food-related attentional bias may simply remember more events of HFHS consumption.In future studies, the recall period could be shortened, such as in 24-h dietary recalls (Thompson & Subar, 2017).Combining repeated 24-h dietary recalls with self-report questionnaires of consumption over longer timeframes has been shown to improve estimates of typical consumption patterns (Carroll et al., 2012).
Unlike for food stimuli, increased consumption of HFHS foods did not predict increased attentional bias for food logo stimuli.Food logo stimuli were largely associated with companies selling pre-packaged or "takeout" food; these stimuli may not have sufficiently represented the broader range of HFHS foods consumed by our participants.Future research may examine whether attentional bias occurs for food logo stimuli belonging to food companies that individuals buy from frequently (for an example with low-calorie sweetened beverages, see Maloney et al., 2019).

Dieting status and attentional biases for food logo stimuli
People on a diet exhibited greater attentional bias for food logo stimuli relative to non-dieters.Using a food-Stroop task, Tapper et al. (2008) found that among restrained eaters, those on a diet had less attentional bias for food than non-dieters.However, the food-Stroop task is more susceptible to strategy use by participants, whereas the attentional paradigm used in this study is largely impervious to participant attempts to ignore distractors (Kennedy et al., 2018;Le Pelley et al., 2017;McHugo et al., 2013).Thus, dieters may experience more stimulus-driven attentional bias towards food, whereas this relationship may differ in contexts where the individual is able to intervene and inhibit such biases.Interestingly, being on a diet only predicted enhanced attentional bias for food logo stimuli, and not food stimuli.This result may be because logos and packaging for HFHS food may be more easily recognisable as "problem stimuli" for dieters relative to the HFHS food itself, since the latter may be more visually similar to healthy foods.The fact that dieting status predicted attention only to food logo stimuli may be indicative of distinct learning phenotypes (Colaizzi et al., 2020).Alternatively, it could be that individuals with a greater bias and awareness of the packaging of unhealthy food are more likely to commit to a diet.In either case, overall, our results again demonstrate that food logo stimuli are not interchangeable with food stimuli, stressing the need to consider the attentional power of food-related cues beyond the food itself.

Hunger, time since last meal, and food preference
The current study did not find evidence that hunger, time since last meal, and food preferences were related to food-related attentional biases.Although meta-analytic evidence suggests that hunger predicts increased attentional bias for food (Hardman et al., 2021), whether hunger is manipulated appears to modulate this effect.Studies which manipulate hunger by comparing participants who have undergone fasting with sated participants usually find that hunger predicts increased food-related attentional bias (Castellanos et al., 2009;Mogg et al., 1998;Nijs et al., 2010;Piech et al., 2010;Stamataki et al., 2019; although see Arumäe et al., 2019).Studies which simply measure hunger at the time of testing like the current study have shown more mixed findings: some studies find hunger predicts an attentional bias for food (Davidson et al., 2018;Tapper et al., 2010), but many find null results (Ballestero-Arnau et al., 2021;Flack et al., 2022;Kirsten et al., 2019;Liu et al., 2022;Nummenmaa et al., 2011).Most participants in the current study had eaten within the last 2 h.Additionally, it is possible that those who had gone longer without food had simply fasted overnight, which may not have had an appreciable effect in our task given that hunger is naturally weaker in the morning despite overnight fasting (Scheer et al., 2013).It is possible that our task may reveal robust effects of hunger when extended, standardised fasting intervals are implemented.
The finding that food preferences did not predict food-related attentional bias is largely consistent with previous literature.Although there is some evidence that HFHS/appetising foods bias attention relative to healthier foods (Castellanos et al., 2009;Dondzilo et al., 2022;Harrar et al., 2011), there is little evidence that people's explicit preferences for food modulate attentional biases (Arumäe et al., 2019;Liu et al., 2022;although see di Pellegrino et al., 2011).Indeed, models of attentional bias to food often describe the incentive value/salience of stimuli in terms of implicit motivational value (its 'wanting') as separable from its explicit liking (Berridge, 2009;Davidson et al., 2018;Field et al., 2016;Hardman et al., 2021).Future research in this space could consider using established measures of both wanting and liking, such as the Leeds food preference questionnaire (Finlayson et al., 2006), to determine whether attentional bias is mediated by an individual's wanting for and/or liking of a given food.

Considerations
There are several aspects about our study that are important to consider in the interpretation of the results.Although not a limitation of the design itself, the findings revealed that only a small amount of variance in attentional bias was explained by our model (<3% for food and food logo stimuli).Effect sizes of this magnitude are relatively common in the literature; a recent meta-analysis on attentional bias for food found that hunger, craving and food intake each explained less than 2% of variation in attentional bias (Hardman et al., 2021).Notably, despite accounting for a small amount of variance, our model was specific to food-related stimuli and the individual difference predictors were not significantly related to attentional bias of control stimuli.Additionally, our results add to the existing literature by revealing similar attentional bias effects with food logo stimuli.Given that we found similar relationships with BMI across two different food-related stimulus types indicates that the relationship withstands across food stimuli types.Nevertheless, care should be taken not to overgeneralise the associations found in the current study due to their small effects.The important comparisons that we chose to make in this study were often experimental trials against baseline/"no distractor" trials.This comparison is common in similar studies to account for overall performance on the task, however it also common to use "neutral" distractors for such comparisons in similar attention designs.For example, in emotion-induced blindness studies, comparisons between emotional stimuli tend to be made against images that are otherwise emotionally neutral.Thus, one may argue that subtracting food-related distractor trial accuracy from baseline trial accuracy is not a pure measure of food-related attentional bias, since some of this variation could be due to novelty of seeing an image in a stream of otherwise landscape or architectural images.We made this decision because a natural "neutral" B.L. Kennedy et al. condition for food was not obvious, so chose instead to include other category images that were likely also distracting but different from food.These images were not the same as a "neutral" stimulus but still served as important controls, because although they similarly distracted attention, they did not predict the individual differences related to food in the way as the food stimuli.
The individual differences variables that we measured revealed several relationships.Future research could expand on these for further examination of how these variables related to attention bias.For example, we attempted to control for general deficits in self-control using the Attentional Control Scale.This scale is a fast and easy way to measure attentional control, however issues have been identified in this measurefor example, participants may not be able to accurately self-report their attentional control (Clarke & Todd, 2021).However, our findings cannot be reduced to a novelty effect or general deficits in attentional control, since regression models using positive and product logo stimuli did not yield significant results, even though these stimuli were also novel.Nevertheless, future research may want to include a different measure of attentional control.Other individual differences may also be of interest.For example, our study did not ask participants if they were vegetarian or vegan.We did ask for liking of different food stimuli, but future research may specifically investigate other aspects of related individual differences for more nuanced or targeted research questions.

Conclusion
The reasons for overconsuming HFHS foods are multifaceted, but understanding the way that individuals selectively attend to food and food-associated cues is key to understanding why overconsumption is maintained in the short-and long-term, and how "addiction"-like changes in attentional priority may flow into a vicious cycle of overconsumption.In this study, we found that BMI negatively predicted attentional biases to images of food and food-associated cues; HFHS food intake predicted greater attentional biases for food; and that being on a diet predicted attentional biases for food logos.These associations were specific to food-related stimuli, since there were no significant associations involving control stimuli or attentional control scores.Our study is one of the first to examine the influence of food logos on attentional bias; the findings suggest that food-associated cues are not interchangeable with natural foods in their distracting effects on attention, and that both classes of stimuli should be considered when examining attentional biases to food.Individuals also vary in the extent to which their attention is biased by HFHS food and branding associated with it, which may help explain why some people continue to overconsume HFHS foods despite knowing the negative consequences for their health.

Fig. 2 .
Fig. 2. Performance Accuracy for Food Distractor, Positive Distractor and Baseline "No Distractor" Trials at Lag 2 and Lag 4 Note.Since baseline trials had no distractor stimuli, we have plotted mean baseline trial accuracy at both lag 2 and lag 4 (the score has been duplicated) for visual comparison with the distractor stimuli.Errors bars represent withinsubject 95% confidence interval for the mean.

Fig. 3 .
Fig. 3. Mean Accuracy for Food Logo Distractor, Product Logo Distractor and Baseline "No Distractor" Trials at Lag 2 and Lag 4 Note.Since baseline trials had no distractor stimuli, we have plotted mean baseline trial accuracy at both lag 2 and lag 4 (the score has been duplicated) for visual comparison with the distractor stimuli.Error bars represent withinsubject 95% confidence interval for the mean.

Table 1
Descriptive statistics for questionnaire scores.

Table 2
Bivariate correlations between the diet-related variables.

Table 3
Regression coefficients of individual differences on food lag 2 attentional bias scores.

Table 4
Regression coefficients of individual differences on food logo lag 2 attentional bias scores.